Falls in care facilities and hospitals are common events that cause considerable morbidity and mortality for older people. This is an update of a review first published in 2010 and updated in 2012. To assess the effects of interventions designed to reduce the incidence of falls in older people in care facilities and hospitals. We searched the Cochrane Bone, Joint and Muscle Trauma Group Specialised Register (August 2017); Cochrane Central Register of Controlled Trials (2017, Issue 8); and MEDLINE, Embase, CINAHL and trial registers to August 2017. Randomised controlled trials of interventions for preventing falls in older people in residential or nursing care facilities, or hospitals. One review author screened abstracts; two review authors screened full‐text articles for inclusion. Two review authors independently performed study selection, 'Risk of bias' assessment and data extraction. We calculated rate ratios (RaR) with 95% confidence intervals (CIs) for rate of falls and risk ratios (RRs) and 95% CIs for outcomes such as risk of falling (number of people falling). We pooled results where appropriate. We used GRADE to assess the quality of evidence. Thirty‐five new trials (77,869 participants) were included in this update. Overall, we included 95 trials (138,164 participants), 71 (40,374 participants; mean age 84 years; 75% women) in care facilities and 24 (97,790 participants; mean age 78 years; 52% women) in hospitals. The majority of trials were at high risk of bias in one or more domains, mostly relating to lack of blinding. With few exceptions, the quality of evidence for individual interventions in either setting was generally rated as low or very low. Risk of fracture and adverse events were generally poorly reported and, where reported, the evidence was very low‐quality, which means that we are uncertain of the estimates. Only the falls outcomes for the main comparisons are reported here. Care facilities Seventeen trials compared exercise with control (typically usual care alone). We are uncertain of the effect of exercise on rate of falls (RaR 0.93, 95% CI 0.72 to 1.20; 2002 participants, 10 studies; I² = 76%; very low‐quality evidence). Exercise may make little or no difference to the risk of falling (RR 1.02, 95% CI 0.88 to 1.18; 2090 participants, 10 studies; I² = 23%; low‐quality evidence). There is low‐quality evidence that general medication review (tested in 12 trials) may make little or no difference to the rate of falls (RaR 0.93, 95% CI 0.64 to 1.35; 2409 participants, 6 studies; I² = 93%) or the risk of falling (RR 0.93, 95% CI 0.80 to 1.09; 5139 participants, 6 studies; I² = 48%). There is moderate‐quality evidence that vitamin D supplementation (4512 participants, 4 studies) probably reduces the rate of falls (RaR 0.72, 95% CI 0.55 to 0.95; I² = 62%), but probably makes little or no difference to the risk of falling (RR 0.92, 95% CI 0.76 to 1.12; I² = 42%). The population included in these studies had low vitamin D levels. Multifactorial interventions were tested in 13 trials. We are uncertain of the effect of multifactorial interventions on the rate of falls (RaR 0.88, 95% CI 0.66 to 1.18; 3439 participants, 10 studies; I² = 84%; very low‐quality evidence). They may make little or no difference to the risk of falling (RR 0.92, 95% CI 0.81 to 1.05; 3153 participants, 9 studies; I² = 42%; low‐quality evidence). Hospitals Three trials tested the effect of additional physiotherapy (supervised exercises) in rehabilitation wards (subacute setting). The very low‐quality evidence means we are uncertain of the effect of additional physiotherapy on the rate of falls (RaR 0.59, 95% CI 0.26 to 1.34; 215 participants, 2 studies; I² = 0%), or whether it reduces the risk of falling (RR 0.36, 95% CI 0.14 to 0.93; 83 participants, 2 studies; I² = 0%). We are uncertain of the effects of bed and chair sensor alarms in hospitals, tested in two trials (28,649 participants) on rate of falls (RaR 0.60, 95% CI 0.27 to 1.34; I² = 0%; very low‐quality evidence) or risk of falling (RR 0.93, 95% CI 0.38 to 2.24; I² = 0%; very low‐quality evidence). Multifactorial interventions in hospitals may reduce rate of falls in hospitals (RaR 0.80, 95% CI 0.64 to 1.01; 44,664 participants, 5 studies; I² = 52%). A subgroup analysis by setting suggests the reduction may be more likely in a subacute setting (RaR 0.67, 95% CI 0.54 to 0.83; 3747 participants, 2 studies; I² = 0%; low‐quality evidence). We are uncertain of the effect of multifactorial interventions on the risk of falling (RR 0.82, 95% CI 0.62 to 1.09; 39,889 participants; 3 studies; I² = 0%; very low‐quality evidence). In care facilities: we are uncertain of the effect of exercise on rate of falls and it may make little or no difference to the risk of falling. General medication review may make little or no difference to the rate of falls or risk of falling. Vitamin D supplementation probably reduces the rate of falls but not risk of falling. We are uncertain of the effect of multifactorial interventions on the rate of falls; they may make little or no difference to the risk of falling. In hospitals: we are uncertain of the effect of additional physiotherapy on the rate of falls or whether it reduces the risk of falling. We are uncertain of the effect of providing bed sensor alarms on the rate of falls or risk of falling. Multifactorial interventions may reduce rate of falls, although subgroup analysis suggests this may apply mostly to a subacute setting; we are uncertain of the effect of these interventions on risk of falling. Interventions for preventing falls in older people in care facilities and hospitals Review question Background Search date We searched the healthcare literature for reports of randomised controlled trials relevant to this review up to August 2017. Study characteristics Quality of the evidence Key results There was evidence, often from single studies, for a wide range of interventions used for preventing falls in both settings. However, in the following we summarise only the falls outcomes for four key interventions in care facilities and three key interventions in hospitals. Care facilities We are uncertain of the effect of exercise on the rate of falls (very low‐quality evidence) and it may make little or no difference to the risk of falling (low‐quality evidence). General medication review may make little or no difference to the rate of falls (low‐quality evidence) or the risk of falling (low‐quality evidence). Prescription of vitamin D probably reduces the rate of falls (moderate‐quality evidence) but probably makes little or no difference to the risk of falling (moderate‐quality evidence). The population included in these studies appeared to have low vitamin D levels. We are uncertain of the effect of multifactorial interventions on the rate of falls (very low‐quality evidence). They may make little or no difference to the risk of falling (low‐quality evidence). Hospitals We are uncertain whether physiotherapy aimed specifically at reducing falls in addition to usual rehabilitation in the ward has an effect on the rate of falls or reduces the risk of falling (very low‐quality evidence). We are uncertain of the effect of bed alarms on the rate of falls or risk of falling (very low‐quality evidence). Multifactorial interventions may reduce the rate of falls, although this is more likely in a rehabilitation or geriatric ward setting (low‐quality evidence). We are uncertain of the effect of these interventions on risk of falling. Studies of falls in nursing facilities show considerable variation in falls incidence rates but a “middle of the road” figure provided in a review of incidence rates is 1.7 falls per person‐year, compared with 0.65 falls per person‐year for older people living in the community (Rubenstein 2006). In a study conducted in 40 Canadian residential care facilities, 62% of participants fell over a one‐year period, with a falls rate of 2.51 falls per person per year (Kennedy 2015). It should be noted, however, that routine recording of falls incidents in standard reporting systems is likely to under‐estimate the incidence of falls (Hill 2010; Sutton 1994). In a prospective one‐year study in 528 nursing homes in Bavaria, Germany, about 75% of falls occurred in the residents' rooms or in bathrooms; 41% occurred during transfers and 36% when walking (Becker 2012). The fall rate was higher in men (2.8 falls per person year) than women (1.49 falls per person year), and falls were less common in people requiring the lowest and highest levels of care. Lord 2003 also found that fall rates were lower in frailer people who were unable to rise from a chair or stand unaided. In this group, increased age, male sex, higher care classifications, incontinence, psychoactive medication use, previous falls and slow reaction times were associated with increased falls. Systematic reviews have shown that in nursing homes, falls history, walking aid use, moderate disability, cognitive impairment, wandering, Parkinson's disease, dizziness, use of sedatives, antipsychotics, antidepressants and total number of medications used are associated with an increased risk of falling (Deandrea 2013; Muir 2012). In residents with dementia, age, use of psychotropic drugs, fair or poor general health, gait impairment and trunk restraint use are associated with an increased number of falls (Kropelin 2013). In hospital settings, a falls incidence of 5.71 falls per 1000 bed days has been found in 16 US general medical surgical and speciality units (Shorr 2012), 6.45 falls per 1000 bed days in 24 Australian medical and surgical wards (Barker 2016), 10.9 falls per 1000 bed days in eight Australian rehabilitation/geriatric units (Hill 2015) and 17.1 falls per 1000 bed days in psychogeriatric wards (Nyberg 1997). In elderly care wards in an UK district general hospital in 2004, the reported rate was as high as 18.0 falls per 1000 bed days (Healey 2004). A similar rate has been reported in some high‐risk wards in Australia (Barker 2016). Systematic reviews have shown that risk factors for falls in hospital inpatients are falls history, age, cognitive impairment, sedative and antidepressant use, gait instability, agitated confusion and urinary incontinence (Deandrea 2013; Oliver 2004). For older patients in rehabilitation hospital settings, risk factors include carpet flooring, vertigo, being an amputee, confusion, cognitive impairment, stroke, sleep disturbance, anticonvulsants, tranquillisers, antihypertensive medications, previous falls and need for transfer assistance (Vieira 2011). There is considerable mortality and morbidity associated with falls in care facilities and hospitals. A study in 24 Australian medical and surgical wards reported a fall injury rate of 2.36 per 1000 bed days (Barker 2016). A study in both these settings reported an incidence of 533 per 1000 person years for all injuries, 20 per 1000 person years for hip fracture, and 270 per 1000 person years for head injuries, for which 13% (14/107) required medical attention (Nurmi 2002). Overall, men were 1.5 times more likely to be injured than women. Older people who sustain a hip fracture while in hospital have been shown to have poor outcomes compared with people sustaining similar fractures in the community (Murray 2007). Falls have been reported to be the most common cause of death from an external cause in residents of care facilities (Ibrahim 2015). The majority of falls are caused by complex combinations of factors operating at the time of each fall event. Interventions may target risk factors in participants or target staff and clinicians with the aim of improving clinical practice or the organisation of care. In some studies, single interventions have been evaluated while in others, interventions with more than one component have been evaluated. Delivery of multiple‐component interventions may be based on individual assessment of risk (a multifactorial intervention) or the same components are provided to all participants (a multiple intervention). A taxonomy has been developed to describe and classify types of intervention (Lamb 2007; Lamb 2011). Key intervention categories include exercise, medication (drug target) interventions which include interventions targeting vitamin D and medication reviews, environment or assistive technologies including bed/chair alarms or the use of low/low beds, social environment interventions which target staff members and changes in the organisational system, knowledge interventions and multifactorial interventions. The majority of randomised controlled trials considered within this review provide a comparison with ‘usual care’ in the care facilities and hospitals involved. Typically, 'usual care' will include standard practices for managing commonly known, potentially modifiable, risk factors for falls and, moreover, the components of usual care will vary both over time and between settings. A systematic review is required to summarise evidence of the impact of purposeful interventions designed to prevent falls, in addition to the unknown impact of routine (and probably variable) care in care facilities and hospitals. Despite routine activities attempting to reduce falls, falls are common in these settings and they result in considerable mortality and morbidity. Results will inform healthcare professionals, researchers, policy makers, informal care givers and consumers. This review is an update of a Cochrane Review first published in 2010 (Cameron 2010), and previously updated in 2012 (Cameron 2012). To assess the effects of interventions designed to reduce the incidence of falls in older people in care facilities and hospitals. Types of studiesWe considered for inclusion all randomised trials, including quasi‐randomised trials (for example, alternation), cluster‐randomised trials and trials in which treatment allocation was inadequately concealed. Types of participantsWe included trials of interventions to prevent falls in older people, of either sex, in care facilities or hospitals. We considered trials for inclusion if the majority of participants were over 65 years or the mean age was over 65 years, and the majority were living in care facilities or were patients in hospital. We excluded trials conducted in places of residence that do not provide residential health‐related care or rehabilitative services, for example retirement villages or sheltered housing. Trials with participants resident in the community and in care facilities were included either in this review or in the Cochrane Review of interventions for preventing falls in older people living in the community (Gillespie 2012), depending on the proportion of participants in each setting. Inclusion in either review was determined by discussion between the authors of both reviews. Trials recording falls in both settings may be included in both reviews. We subdivided care facilities based on level of care provided. We defined high‐level care facilities as "establishments that are primarily engaged in providing inpatient nursing and rehabilitative services for long‐term care patients. The care is generally provided for an extended period of time to individuals requiring nursing care. These establishments have a permanent core staff of registered or licensed practical nurses that, along with other staff, provide nursing care in combination with personal care" (OECD 2011). We defined intermediate‐care facilities as "institutions which provide health‐related care and services to individuals who do not require the degree of care which hospitals or skilled nursing facilities provide, but because of their physical or mental condition require care and services above the level of room and board" (NLM 2012). Some facilities provided both these levels of care. For cluster‐randomised trials, the classification of the level of care was based on the description of the facility. For individually‐randomised trials where the level of care provided by the facility was clearly described, this description informed the classification. Where the inclusion/exclusion criteria of a trial selected patients who required high or intermediate level of care from a mixed‐care facility, the classification was based upon the care needs of the individual participants. For trials in hospitals, participants included staff or in‐patients. We excluded interventions that took place in emergency departments, outpatient departments or where hospital services were provided in community settings. We subdivided hospitals into those providing acute, and those providing subacute care. We defined subacute care as "medical and skilled nursing services provided to patients who are not in an acute phase of an illness but who require a level of care higher than that provided in a long‐term care setting" (NLM 2012). Studies recruiting participants post‐stroke were excluded as interventions to prevent falls in this population are reviewed in a separate Cochrane Review Interventions for preventing falls in people after stroke (Verheyden 2013). Types of interventionsAny intervention designed to reduce falls in older people compared with any other intervention, usual care or placebo. We grouped interventions using the fall‐prevention classification system (taxonomy) developed by the Prevention of Falls Network Europe (ProFaNE) (Lamb 2011). Interventions have been grouped by combination (single, multiple, or multifactorial), and then by the type of intervention (descriptors). Full details are available in the ProFaNE taxonomy manual (Lamb 2007). The possible intervention descriptors are: exercises, medication (drug target, i.e. withdrawal, dose reduction or increase, substitution, provision), surgery, management of urinary incontinence, fluid or nutrition therapy, psychological interventions, environment/assistive technology, social environment, interventions to increase knowledge, other interventions. Types of outcome measuresWe included only trials that reported raw data or statistics relating to rate or number of falls, or number of participants sustaining at least one fall during follow‐up (fallers). Trials that reported only those participants who had more than one fall were included. Trials that reported only specific types of fall (e.g. injurious falls) were not included. Trials that focused on intermediate outcomes such as improved balance or strength, and did not report falls or falling as an outcome, were excluded. Primary outcomes
Secondary outcomes
Electronic searchesWe searched the Cochrane Bone, Joint and Muscle Trauma Group Specialised Register (to 3 August 2017), the Cochrane Central Register of Controlled Trials (CENTRAL) (2017, Issue 8), MEDLINE (including Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations, Ovid MEDLINE Daily, Ovid MEDLINE and Versions) (1946 to 3 August 2017), Embase (1980 to 2017 Week 31), and CINAHL (1982 to 3 August 2017). We also searched ongoing trial registers via the World Health Organization's ICTRP Search Portal (3 August 2017) and ClinicalTrials.gov (3 August 2017). We did not apply any language restrictions. For this update, the search results were limited from 2012 onwards. The search update process was run in two stages: the first search was run in February 2016 and a second top‐up search was run in August 2017. Details of the search strategies used for previous versions of the review are given in Cameron 2012. In MEDLINE (OvidSP), subject‐specific search terms were combined with the sensitivity‐ and precision‐maximising version of the Cochrane Highly Sensitive Search Strategy for identifying randomised trials in MEDLINE (Lefebvre 2011). We modified this strategy for use in CENTRAL, Embase, and CINAHL (seeAppendix 1 for all strategies). Searching other resourcesWe also checked reference lists of articles and further trials were identified by contact with researchers in the field. For the first version of this review, we identified trials in care facilities and hospitals included in Gillespie 2003. Data collection and analysis were carried out according to methods stated in the published protocol (Cameron 2005), and subsequently amended to concur with updated methods in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a) as described in Differences between protocol and review. Data collection and analysis were carried out according to methods stated in the published protocol (Cameron 2005), which were based on the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a). Selection of studiesFrom the title, abstract, or descriptors, one review author screened all abstracts to identify potentially relevant trials for full review. Two review authors screened potentially relevant abstracts. From the full text, two review authors independently assessed potentially eligible trials for inclusion and resolved disagreement by discussion, or by adjudication with a third review author. Full‐text review was undertaken using Covidence. Disagreement was resolved by discussion and consensus or third party adjudication when necessary. We contacted trial authors for additional information if necessary to assess eligibility. Data extraction and managementPairs of review authors independently extracted data using a pre‐tested data extraction form for studies included to 2012. For this update,again pairs of review authors independently extracted data from the identified studies using Covidence. Multiple reports from the same study were linked as a single study in Covidence and evidence from all reports were reviewed in undertaking data extraction. Where data were unclear authors were contacted whenever possible for clarification. Disagreement was resolved by discussion and consensus or third party adjudication when necessary. Assessment of risk of bias in included studiesPairs of review authors independently assessed risk of bias for each included study based on recommendations in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b). Assessors were not blinded to author and source institution. Review authors did not assess their own trials. Disagreement was resolved by consensus, or by third party adjudication. We assessed risk of bias for the following domains: sequence generation (selection bias); allocation concealment (selection bias); blinding of participants and personnel (performance bias); blinding of outcome assessment (detection bias), incomplete outcome data (attrition bias), and selective reporting (reporting bias). Since all the outcomes collected in our review are susceptible to the same risk of bias, we have not assessed outcomes for risk of detection bias or completeness of outcome data separately. Additionally, we assessed bias in the recall of falls due to less reliable methods of ascertainment (Hannan 2010), and bias resulting from major imbalances in key baseline characteristics (e.g. age, gender, previous falls, medical status, dependency, cognitive function). Assessors rated the risk of bias as low, high or unclear for each domain. We established additional criteria within currently existing domains for assessing the additional risks of bias associated with cluster randomisation (Section 16.3.2; Higgins 2011b). Thus 'recruitment bias' was considered as a component of selection bias under allocation concealment; 'baseline imbalance' resulting from small numbers of clusters was considered in bias resulting from major imbalances in key characteristics; risk of bias resulting from 'loss of clusters' was considered under incomplete outcome data; and 'incorrect analysis' that failed to take into account the effect of clustering and that could not be satisfactorily remedied was considered under selective outcome reporting. We did not assess the risk of bias relating to the 'comparability with individually‐randomised trials' as a separate item as it is impossible to establish suitable criteria for an individual trial out of context. The potential for differences in effects between cluster‐ and individually‐randomised trials was considered in our assessment of the quality of the evidence and in our Discussion. Our criteria for 'Risk of bias' assessments are shown in Appendix 2. Measures of treatment effectWe have reported the treatment effect for rate of falls as a rate ratio (RaR) and 95% confidence interval (CI). For number of fallers and number of participants sustaining fall‐related fractures we have reported a risk ratio (RR) and 95% CI. We used results reported at discharge from hospital for trials that continued to monitor falls after discharge. Rate of fallsThe rate of falls is the total number of falls per unit of person time that falls were monitored (e.g. falls per person year). The rate ratio compares the rate of falls in any two groups during each trial. We used a rate ratio (for example, incidence rate ratio or hazard ratio for all falls) and 95% CI if these were reported in the paper. If both adjusted and unadjusted rate ratios were reported, we used the unadjusted estimate, unless the adjustment was for clustering. If a rate ratio was not reported but appropriate raw data were available, we used Excel to calculate a rate ratio and 95% CI. We used the reported rate of falls (falls per person year) in each group and the total number of falls for participants contributing data, or we calculated the rate of falls in each group from the total number of falls and the actual total length of time falls were monitored (person years) for participants contributing data. In cases where data were only available for people who had completed the study, or where the trial authors had stated there were no losses to follow‐up, we assumed that these participants had been followed up for the maximum possible period. Where there were no falls in one arm of a study, and a low total number of falls and/or participants (e.g. Beck 2016; Cadore 2014), the rate of falls cannot be determined. Such data were therefore not pooled, however the omission of these data from the pooled analysis is considered unlikely to change any estimate of effect. Risk of fallingFor number of fallers, a dichotomous outcome, we used a risk ratio as the treatment effect. The risk ratio compares the number of people who fell once or more (fallers) in the intervention and control arms of each trial. We used a reported estimate of risk (hazard ratio for first fall, risk ratio (relative risk), or odds ratio) and 95% CI if available. If both adjusted and unadjusted estimates were reported we used the unadjusted estimate, unless the adjustment was for clustering. If an odds ratio was reported, or there was no effect estimate and 95% CI, and appropriate data were available, we calculated a risk ratio and 95% CI using the csi command in Stata or in Review Manager. For the calculations, we used the number of participants contributing data in each group if this was known; if not reported, we used the number randomised to each group. Secondary outcomesFor the number of participants sustaining one or more fall‐related fractures, we used a risk ratio as described in 'Risk of falling' above. Unit of analysis issuesFor trials that were cluster randomised, for example by care facility or ward, we performed adjustments for clustering (Higgins 2011c), if this was not done in the published report. We used intra‐cluster correlation coefficients reported by Dyer 2004 (falls per person year 0.100, number of residents falling 0.071, and residents sustaining a fracture 0.026). For trials with multiple intervention groups, we either combined the groups or included only one pair‐wise comparison (intervention versus control) in any analysis in order to avoid the same group of participants being included twice. For trials that excluded the intervention period from the falls outcomes, we did not pool the outcomes data with other studies. Dealing with missing dataOnly the available data were used in the analyses; we did not impute missing data. Assessment of heterogeneityWe assessed heterogeneity within a pooled group of trials using a combination of visual inspection of the graph along with consideration of the Chi² test (with statistical significance set at P < 0.10), and the I² statistic (Higgins 2003). We based our interpretation of the I² results on that suggested by Higgins 2011a: 0% to 40% might not be important; 30% to 60% may represent moderate heterogeneity; 50% to 90% may represent substantial heterogeneity; and 75% to 100% may represent very substantial ('considerable') heterogeneity. Assessment of reporting biasesTo explore the possibility of publication and other reporting biases, we constructed funnel plots for analyses that contained more than 10 studies. Data synthesisWe classified interventions into those taking place in care facilities and those taking place in hospitals, and pooled these separately because participant characteristics and the environment warrants different types of interventions in the different settings, possibly implemented by people with different skill mixes. Within these categories, we grouped the results of trials with comparable interventions and participant characteristics, and compiled forest plots using the generic inverse variance method in Review Manager. This method enabled pooling of the adjusted and unadjusted treatment effect estimates (rate ratios or risk ratios) that were reported in the paper, or we calculated from data presented in the paper (seeMeasures of treatment effect). Where the total number of patients, rather than admissions, could not be determined, we did not pool these data with other studies. Where the reported trial outcomes did not include falls during the intervention period, we did not pool these data with those of other trials. Where appropriate, we pooled results of comparable studies using both fixed‐effect and random‐effects models. We chose the model to report by careful consideration of the extent of heterogeneity and whether it can be explained by factors such as the number and size of included studies, or the level of care provided. We used 95% CIs throughout. We considered, on a case by case basis, not pooling data where there was considerable heterogeneity (I² statistic value of greater than 75%) that could not be explained by the diversity of methodological or clinical features among trials. Where it was inappropriate to pool data, we still presented trial data in the analyses or tables for illustrative purposes and reported these in the text. Subgroup analysis and investigation of heterogeneityWe minimised heterogeneity as much as possible by grouping trials as described previously (using ProFaNE categories of interventions). We categorised broad interventions further by grouping subtypes of interventions according to ProFaNE (e.g. for exercise interventions). We explored heterogeneity by carrying out subgroup analyses based on level of care and level of cognition at enrolment in care facilities and hospitals where possible. We subdivided the care facilities into high, intermediate or mixed levels of care. The levels of care of the facilities reflect the levels of dependence of the participants. In hospitals, the level of care was subdivided by acute versus subacute or mixed levels of care. We also carried out subgroup analyses by stratification of intervention types according to ProFaNE (e.g. for exercise types, medication target interventions), and type of fracture. Subgroup analyses based upon the individual components of the multifactorial interventions was precluded by the study design and reporting. Data were inadequate for conducting a subgroup analysis by level of frailty of the participants in trials of exercise in care facilities. We grouped trials by level of cognition into those that included only participants with cognitive impairment versus those with no cognitive impairment, or a mixed sample at enrolment. We used the random‐effects model to pool data in all subgroup analyses testing for subgroup differences due to the high risk of false‐positive results when comparing subgroups in a fixed‐effect model (Higgins 2011d). We used the test for subgroup differences available in Review Manager to determine whether there was evidence for a difference in treatment effect between subgroups. Sensitivity analysisWhere there was substantial statistical heterogeneity we carried out a post‐hoc sensitivity analysis to explore the effect of removing trials from the analysis if visual inspection of the graph showed poorly overlapping confidence intervals. Where there was considered to be significant statistical heterogeneity for rate of falls but not risk of falling, sensitivity analyses were carried out to determine the likely effects of using random‐effects versus fixed‐effect meta‐analyses for the risk of falling (e.g. for exercise versus usual care in care facilities and multifactorial interventions in care facilities). We conducted post‐hoc sensitivity analyses for exercise in care facilities, excluding trials with 20 participants or less in each arm of the trial to explore the possibility of small‐trial effects, due to the observed asymmetry in the Funnel plots. We conducted a sensitivity analysis for exercise compared to usual care in care facilities including Cadore 2014, which had zero falls in the intervention arm, using one fall in the intervention arm to examine the likely effect of omitting this trial from the analysis. We also conducted a sensitivity analysis excluding one trial with a known non‐normal distribution of falls in the intervention arm from the analysis of general medication review in care facilities for the rate of falls outcomes. Sensitivity analyses according to study quality were not possible as most studies were at potential risk of bias. Economic issuesWe have noted the results from any economic evaluations (cost‐effectiveness analysis, cost‐utility analysis) incorporated in included studies. We also extracted from each trial reporting a cost analysis, cost description or analytic model, the type of resource use reported (e.g. delivering the intervention, hospital admissions, medication use) and the cost of the items for each group. Assessing the quality of the evidence and 'Summary of findings' tablesFor each comparison, we used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach to assess the quality of the body of evidence (Schünemann 2011) for each outcome listed in Types of outcome measures. For all comparisons where there were two or more trials, GRADE assessment was performed independently by two review authors and disagreement was resolved by discussion, or by adjudication with a third review author. We adopted a different approach for single trial comparisons, where we started with the assumption that the quality of evidence was likely to be very low. This reflected assumptions of downgrading at a minimum for serious risk of bias (typically performance and detection bias), for serious indirectness (trial being conducted was a single trial or setting), and for serious imprecision (failure to meet the 200 to 300 events optimal size criteria) (Guyatt 2011). Where these assumptions did not hold, we performed GRADE assessment as above. The quality rating 'high' is reserved for a body of evidence based on randomised controlled trials. We ‘downgraded’ the quality rating to 'moderate', 'low' or 'very low' depending on the presence and extent of five factors: study limitations, inconsistency of effect, imprecision, indirectness or publication bias. We used the GRADE approach to assess quality of evidence related to the primary and secondary outcomes listed in the Types of outcome measures. We prepared a 'Summary of findings' table for each of the main categories of interventions, for listed outcomes. We selected the following comparisons for presentation in 'Summary of findings' tables as these are the most common falls prevention activities considered and applied in clinical settings. In care facilities: exercise, vitamin D supplementation, medication review and multifactorial interventions; in hospitals: exercise, bed alarms and multifactorial interventions. Results of the searchFor this update we screened a total of 3989 records from the following databases: Cochrane Bone, Joint and Muscle Trauma Group Specialised Register (0 records); CENTRAL (127), MEDLINE (1104), Embase (1211), CINAHL (314) the WHO ICTRP (450) and Clinicaltrials.gov (783). We also found 29 potentially eligible studies from other sources. After removal of 503 duplicates, 3515 citations were screened for inclusion. Screening of the search update identified a total of 413 records for potential inclusion, for which full‐text reports were obtained. Thirty‐five new trials were included in this update, 27 new ongoing trials identified and seven new studies await classification. In addition, a new subgroup analysis (Stenvall 2012) from the Stenvall 2007 trial and a cost‐effectiveness analysis (Haines 2013) of Haines 2011 have been added. A flow diagram summarising the study selection process is shown in Figure 1. Overall, there are now 95 included trials, 105 excluded studies, eight studies awaiting classification and 31 ongoing trials. Due to the review size, not all links to references have been inserted in the text but can be viewed in Table 8. Included studiesThirty‐five additional trials have been included in this update, 28 trials in care facilities and seven in hospitals (seeTable 8). This review now contains 95 trials with 138,164 participants. Details of individual trials are provided in the Characteristics of included studies, and are briefly outlined below. DesignParticipants were individually randomised in 53 studies, whereas 42 studies used a cluster‐randomised design (seeTable 8). SettingsThe included trials were carried out in 23 countries (seeTable 8). Of the 71 studies (40,374 participants) in care facilities, 17 were in high‐level care facilities, 17 were in intermediate‐level care facilities and 37 were in facilities with mixed levels of care, or combinations of facilities that included both high and intermediate levels of care. Of the 24 studies (97,790 participants) in hospital settings, 10 were in an acute hospital setting, 12 were in subacute settings, and 2 were in both acute and subacute care settings (seeTable 8). Van Gaal 2011a and Van Gaal 2011b have been included as two separate trials although reported in the same paper as the participants were randomised separately in two settings (nursing homes and hospitals) and results are reported by setting. InterventionsUsing ProFaNE taxonomy, all studies were categorised by intervention and grouped by combination (single, multiple, or multifactorial) (seeAppendix 3). The first column of Appendix 3 shows the intervention classification (single, multiple, or multifactorial) and setting type (care facility or hospital). The components of included 'Exercises' interventions, 'Environmental/assistive technology' and 'Medication (drug target)' interventions are shown in Appendix 4, Appendix 5 and Appendix 6 respectively. In care facilities, 54 trials tested the effect of a single intervention only, three trials tested both single and multiple interventions (Huang 2016; Imaoka 2016; Sambrook 2012), one trial tested a multiple intervention only (Schnelle 2003), and 13 trials tested a multifactorial intervention. In hospitals, 18 trials tested the effect of a single intervention and six tested a multifactorial intervention. Seven studies tested the effect of two interventions (Faber 2006; Haines 2011; Huang 2016; Nowalk 2001; Sambrook 2012; Saravanakumar 2014; Tuunainen 2013), and one tested three interventions (Imaoka 2016) in comparison with usual care. Donald 2000 was a 2 x 2 factorial study of supervised exercises and flooring types that has been classified as two single interventions. In general, included studies compared an active falls prevention intervention with a control group comprising 'usual care', that typically would have included standard falls prevention activities. Often, however, standard practice in terms of falls prevention activities was not clearly described. Additional descriptions of the control groups provided for individual trials are provided in the Characteristics of included studies table, the 'Summary of findings' tables available for some comparisons, and the analyses headings and/or footnotes. A general description of the control arms for the main intervention categories is also given below. In care facilities, 17 trials of exercise provided a comparison with usual care, defined as no exercise, no change in previous lifestyle or exercise type or level unlikely to change physical performance and nine trials provided a comparison of two different exercise programmes (see Table 9). Trials of medication target interventions in care facilities more often provided a comparison with placebo (see Table 10). Trials of vitamin D supplementation in care facilities provided estimates of effect compared with usual care or placebo. In hospitals, multifactorial interventions were generally compared with a control group consisting of standard falls prevention activities. Whether or not the control arm included some of the multifactorial intervention components was not always clearly reported. Additional detail is provided in the description of individual studies in the results text and within the Characteristics of included studies table.
OutcomesThe source of data used for calculating outcomes for each trial for generic inverse variance analysis is shown in Appendix 7. Seventeen trials met our inclusion criteria but did not report data that could be included in pooled analyses. Reported results from these trials are presented in the text or additional tables. Raw data for rate of falls and number of fallers when reported or when they could be calculated are shown in Appendix 8. Twenty‐four trials reported data on fractures suitable for use in pooled analyses, other reported fractures data is presented in the text. Twenty‐nine trials clearly reported data on adverse events, but in many of these it was not clear if adverse‐event data were recorded systematically; for the majority of trials, this outcome was not reported. Excluded studiesOverall there were 105 excluded studies (seeCharacteristics of excluded studies for details). Of the 51 newly excluded studies (see Figure 1): five were excluded as they were not randomised; five were conducted in the wrong population (e.g. including participants post stroke); 10 were conducted in the wrong setting (in most of these, the majority of participants were living in the community); two studies of flooring interventions were excluded as the intent was to reduce fall injuries, rather than falls (Drahota 2013; {"type":"clinical-trial","attrs":{"text":"NCT01618786","term_id":"NCT01618786"}}NCT01618786); 22 studies were excluded as they measured falls as a potential adverse outcome of the intervention; two did not report falls outcomes; one study was excluded as it reported a specific type of falls only (Sahota 2014); three trials were discontinued and one had invalid falls data (DeSure 2013). Of the 54 studies excluded in the previous version of this review: 21 trials were excluded because the intervention they tested was not designed to reduce falls, rather falls were measured as a potential adverse outcome of an intervention with a different aim; in 11 trials the majority of participants were living in the community; eight excluded trials did not provide sufficient data on falls or fallers; seven included participants post stroke and seven were not randomised (Cameron 2012). Of note is that four trials that had been excluded in Cameron 2012 because they included participants with post‐stroke hemiplegia, have now either been retracted (Sato 2000; Sato 2005a; Sato 2005b; see Retraction Watch) or, for Sato 2011, likely to be retracted in future because of serious concerns about research misconduct as revealed in Bolland 2016. Ongoing studiesWe are aware of 31 ongoing studies, 14 set in care facilities and 17 in hospitals (seeCharacteristics of ongoing studies for details). The ongoing studies in care facilities include five exercise trials in care facilities (two of whole body vibration), one trial of a multiple intervention of exercise and nutrition, one of nutrition, three of medication review, one of vitamin D supplementation, three of service model changes, and one of a telesurveillance system; two trials are likely to have been completed, one of whole body vibration (JPRN‐UMIN000000555) and one of vitamin D supplementation (JPRN‐UMIN000008361). The ongoing studies in hospitals include three trials of medication review, four of exercise, one of an education intervention, five social environment interventions including one of student training, one psychological intervention, one of a sensor technology, one educational intervention, and one multifactorial intervention; five trials are likely to be completed, three of medication review (ISRCTN42003273; {"type":"clinical-trial","attrs":{"text":"NCT01876095","term_id":"NCT01876095"}}NCT01876095; {"type":"clinical-trial","attrs":{"text":"NCT02570945","term_id":"NCT02570945"}}NCT02570945), one of exercise (Hassett 2016), and one of telesurveillance ({"type":"clinical-trial","attrs":{"text":"NCT01561872","term_id":"NCT01561872"}}NCT01561872). Details of 'Risk of bias' assessment for nine items for each trial are shown in the Characteristics of included studies. Summary results for these items are shown in Figure 2, Figure 3 and Table 11.
The majority of included studies were considered at high risk of bias for at least one domain. In particular, there was a high risk of performance bias for the majority of studies due to lack of blinding. Only three trials were considered at low risk of bias for all or the majority of domains (Bischoff 2003; Broe 2007; Flicker 2005), these all examined vitamin D supplementation in comparison with placebo. However, for many other types of interventions, blinding was generally not feasible (e.g. exercise, bed alarms). The risk of bias was often unclear, in particular for risk of selection bias due to allocation concealment. Potential bias varied within comparison groups and it is difficult to judge whether any bias would result in an over‐ or under‐estimation of treatment effect. AllocationUnder half of included studies (39 in all) were considered at low risk of selection bias; this often reflected lack of clarity on the methods for allocation concealment. We assessed risk of bias in sequence generation as low in 66 trials, high in two trials that described inappropriate methods (Michalek 2014; Wald 2011), and unclear in the remaining 27 trials, usually because of a lack of reporting of methods. We judged methods for concealment of allocation prior to group assignment to carry low risk of bias in 43 trials, high in 14 trials and to be unclear in the remaining 38 trials, again typically due to lack of reporting. Barker 2016, a cluster‐randomised trial, is an example of a trial at high risk of selection bias due to lack of allocation concealment: although the initial cluster allocation was concealed, the subsequent recruitment of participants into the study (i.e. admission to the ward) was not. BlindingBlinding of participants and personnel was uncommon and indeed blinding of these was not feasible for many intervention types (e.g. exercise, multifactorial interventions). In all, 86 trials were at high risk of performance bias, with just seven trials being at low risk and the remaining two trials being judged at unclear risk of bias. The likelihood of detection bias in relation to the ascertainment of falls by outcome assessors was also high in 65 trials, generally as falls were ascertained by staff who were not blinded (e.g. Barker 2016). Risk of bias was low in 10 trials, most commonly in vitamin D trials where administration of a placebo was possible (e.g. Flicker 2005) and unclear in 20 trials. Incomplete outcome dataThe risk of attrition bias due to incomplete outcome data was assessed as high in 26 trials (the high risk of attrition in some trials is likely to be related to longer periods of follow‐up; e.g. 12 months for Juola 2015 and 16 months for Kennedy 2015). Risk of bias was low in 61 trials, where there was no loss to follow‐up (this occurred more frequently in a hospital setting: e.g. Barker 2016; Hill 2015) or losses were balanced between groups (e.g. Cadore 2014; Kerse 2008). Risk of bias was unclear in eight trials, which generally reflected unclear reporting (e.g. Van de Ven 2014). Selective reportingReporting bias was judged as unclear in 37 trials, generally as no protocol was identified (e.g. Healey 2004), and low risk in 50 trials where results were reported according to the protocol (e.g. Potter 2016), or all expected falls outcomes were reported (e.g. Law 2006). Eight trials were at high risk, usually where outcomes mentioned in the protocol or methods were not reported (e.g. Ang 2011). Other potential sources of biasThe method of ascertaining falls was judged to be at a low risk of bias for 45 trials, at high risk of bias for 27 trials, generally where falls were poorly defined (e.g. Healey 2004), and at unclear risk for 23 trials when methods were not reported (e.g. Sakamoto 2006). The risk of bias relating to imbalance in baseline characteristics was considered to be low in 51 trials, high in 26 trials, and unclear in 18 trials. Risk of baseline imbalance usually occurred in small trials (e.g. Buckinx 2014) or cluster‐randomised trials (e.g. Becker 2003; Choi 2005; Van Gaal 2011a; Van Gaal 2011b; Whitney 2017). Two trials were considered to be a high risk of other bias, this was due to the author being employed by the company producing the intervention (Clifton 2009), or the individual randomisation being to one of two clusters, hence the trial was not truly individually randomised (Michalek 2014). There was a low risk of other bias in 87 trials and unclear risk in six trials due to unusual study design (stepped‐wedge trial in Aizen 2015; Hill 2015; and including a non‐randomised patient preference arm in Streim 2012) or ongoing falls prevention activities (Aizen 2015; Ang 2011; Barker 2016; Cumming 2008). Cluster‐randomised trialsThere were a large number of included cluster‐randomised trials (44%, 42/95), many of which had a large number of participants (e.g. Barker 2016; Shorr 2012). Risk of bias particular to cluster‐randomised trials were considered within other domains (seeAssessment of risk of bias in included studies). However, it is worth noting that some of these trials contained a small number of clusters and hence were more prone to baseline imbalance (e.g. Choi 2005; Van Gaal 2011a; Van Gaal 2011b), and in some cases prediction of allocation concealment (e.g. Choi 2005; Koh 2009). Loss of whole clusters could also lead to a high risk of attrition bias (e.g. Cox 2008). See: Table 1; Table 2; Table 3; Table 4; Table 5; Table 6; Table 7
We present results by setting (care facilities or hospitals), combination (single, multiple, or multifactorial) and intervention type (categorised according to ProFaNE, Lamb 2011) in Appendix 3. Care facilities: single interventionsSingle interventions consist of one major category of intervention only and are delivered to all participants in the group. ExerciseTwenty‐five trials (2848 participants) investigated exercise as a single intervention (seeTable 9), four trials (986 participants) were cluster randomised (Choi 2005; Kerse 2008; Rosendahl 2008; Yokoi 2015), and the remaining 22 trials (1862 participants) were individually randomised. However, many of these trials were small (median 60 participants, range 16 to 682; seeTable 8). The types of exercise are shown in Table 9. The control arm of the different trials also varied. Four trials included three arms (Faber 2006; Nowalk 2001; Saravanakumar 2014; Tuunainen 2013). One was a cross‐over trial (Toulotte 2003). The trials are categorised below, both according to the ProFaNE exercise category (see Appendix 4) and the comparator arm of the trial. A summary of the evidence from exercise versus usual care for falls prevention in care facilities is provided in Table 1. Only two trials reported on the impact of exercise interventions on fractures (Rosendahl 2008, Sitja Rabert 2015). Nine trials reported on adverse events, while 16 trials did not report adverse‐event data. In seven trials, the reported data were incomplete and not suitable for pooling with other studies (Buettner 2002; Cadore 2014; da Silva Borges 2014; Imaoka 2016; Nowalk 2001; Serra‐Rexach 2011; Toulotte 2003); seeAnalysis 1.2 and Analysis 4.2). Falls data from Imaoka 2016 excluded the intervention period and thus are not presented in the forest plot. Comparison 1 Care facilities: Exercise vs usual care, Outcome 2 Rate of falls and number of fallers: trials with incomplete data.
Comparison 4 Care facilities: Comparisons of different exercise programs (see Appendix 4 for details), Outcome 2 Rate of falls and number of fallers: trials with incomplete data.
Exercise versus usual careSeventeen trials (2406 participants) compared an exercise intervention with usual care, defined as no exercise, no change in previous lifestyle or exercise type or level unlikely to change physical performance (e.g. seated flexibility exercise programme). Four trials (986 participants) of exercise in comparison with usual care were cluster randomised (Choi 2005; Kerse 2008; Rosendahl 2008; Yokoi 2015), the remaining 13 trials (1420 participants) were individually randomised. Faber 2006, included two exercise intervention arms, we combined the results from the two intervention groups in these analyses. As there is considerable clinical heterogeneity within these studies, we undertook analyses to explore heterogeneity, which are reported below. Rate of fallsTen trials (2002 participants) reporting on the impact of exercise in comparison with usual care in care facilities on the rate of falls had considerable statistical heterogeneity (I² = 76%, heterogeneity P < 0.0001). Nevertheless, as these trials were considered clinically similar in terms of the intervention, comparator, patient group and outcomes, these trials were pooled with a random effects meta‐analysis (Analysis 1.1: Rate ratio (RaR) = 0.93, 95% confidence interval (CI) 0.72 to 1.20). We are uncertain whether exercise reduces the rate of falls in care facilities as the quality of the evidence was assessed as very low (Table 1). Comparison 1 Care facilities: Exercise vs usual care, Outcome 1 Rate of falls. In a subgroup analysis by broad types of exercise, there was no evidence of a difference between subgroups (Analysis 2.1: test for subgroup differences P = 1.00). Comparison 2 Care facilities: Exercises vs usual care (grouped by type of exercise), Outcome 1 Rate of falls. To explore further the heterogeneity in these findings, we carried out a post‐hoc subgroup analysis by level of care (high or intermediate levels of care, or mixed levels). There was evidence of a difference between these subgroups that partially explained the heterogeneity (Analysis 3.1: test for subgroup differences Chi² = 6.39, I² = 69%, 2 df, P = 0.04). In studies of facilities providing mixed levels of care, the heterogeneity was no longer evident (I² = 0%, P = 0.41) and there was no evidence of an effect (Analysis 3.1.3 RaR: 1.08, 95% CI 0.92 to 1.28, 3 trials, 477 participants: I² = 0%). However, heterogeneity remained considerable for trials in a high or intermediate level of care (I² = 78%, P = 0.001). Comparison 3 Care facilities: Exercise vs usual care (grouped by level of care), Outcome 1 Rate of falls. Four additional trials (130 participants) reported outcomes on rate of falls with data not suitable for pooling (Analysis 1.2); all reported a reduction in falls. Risk of fallingPooled data from 10 trials (2090 participants) indicated exercise may make little or no difference to the risk of falling (risk ratio (RR) with random‐effects RR 1.02, 95% CI 0.88 to 1.18: I² = 23%; Analysis 1.3; low‐quality evidence, Table 1). Comparison 1 Care facilities: Exercise vs usual care, Outcome 3 Number of fallers. There were no subgroup differences in post‐hoc analyses for number of fallers between different levels of care (Analysis 3.2; test for subgroup differences P = 0.56) or types of exercise (Analysis 2.2; test for subgroup differences P = 0.71). Comparison 2 Care facilities: Exercises vs usual care (grouped by type of exercise), Outcome 2 Number of fallers. Comparison 3 Care facilities: Exercise vs usual care (grouped by level of care), Outcome 2 Number of fallers. Faber 2006 carried out a post‐hoc subgroup analysis and found that the intervention in frail participants may increase risk of falling (hazard ratio (HR) 2.95, 95% CI 1.64 to 5.32; 115 participants), while in the pre‐frail subgroup there was no strong evidence for a reduction in the risk of falling (HR 0.62, 95% CI 0.29 to 1.33; 105 participants) (test for subgroup difference P ≤ 0.10). Other trials did not provide data suitable for a post‐hoc subgroup analysis of the effectiveness of the intervention according to the frailty of the participants. Nowalk 2001 (N = 110) reported that there was no significant difference in the risk of falling between "Fit NB Free" individually‐tailored combination exercises, or the "Living and Learning/Tai Chi" in comparison with usual routine activities; data were not suitable for pooling (Analysis 1.2). Risk of fractureOne trial of functional exercises (Rosendahl 2008, 183 participants) found no strong evidence for a reduction in the risk of hip fracture (Analysis 1.4.1: RR 0.16, 95% CI 0.01 to 2.81; 3 fractures) or total fractures (Analysis 1.4.2: RR 0.88, 95% CI 0.25 to 3.14; 10 fractures). We are uncertain whether exercise reduces the risk of fracture as the quality of the evidence was assessed as very low (Table 1). Comparison 1 Care facilities: Exercise vs usual care, Outcome 4 Number of people sustaining a fracture. Adverse eventsTwo trials (833 participants) of exercise compared with usual care reported the rates of adverse event outcomes including aches, pains, fatigue, soreness and bruises. Kerse 2008 (639 participants) reported no differences in the level of adverse outcomes on negative binomial regression adjusted for clustering (aches and pains at six months exercise 46.7, 95% CI 39.3 to 54.9 versus usual care 51.1, 95% CI 43.8 to 58.4, P = 0.75). Mulrow 1994 (194 participants) found no difference in the proportion of participants reporting severe soreness (Analysis 1.7.1: RR 0.91, 95% CI 0.40 to 2.04), severe bruises (Analysis 1.7.2: RR 2.00, 95% CI 0.18 to 21.69) or severe fatigue (Analysis 1.7.3: RR 4.00, 95% CI 0.46 to 35.14); there were no injuries during the therapy sessions. One trial (16 participants) reported that there were no adverse events (Schoenfelder 2000). One trial (183 participants) reported a death due to a ruptured abdominal aortic aneurysm one week after the follow‐up tests of the exercise intervention for which association could not definitely be excluded by geriatric review (Rosendahl 2008). We are uncertain of the effects of exercise on adverse events as the quality of the evidence has been assessed as very low; Table 1). Comparison 1 Care facilities: Exercise vs usual care, Outcome 7 Adverse events: aches and pains. Sensitivity analysisAs a sensitivity analysis, the pooled analysis of rate of falls was conducted with a fixed‐effect model. This made little difference to the estimate of effect (RaR 1.01, 95% CI 0.91 to 1.13). The pooled analysis of the risk of falling with a fixed‐effect model also made little difference to the estimate of effect (RR 1.04, 95% CI 0.92 to 1.18). We also conducted a sensitivity analysis including Cadore 2014, which had zero falls in the intervention arm, calculated using one fall in lieu of zero in this arm. This had little impact on the effect estimate (RaR 0.85, 95% CI 0.63 to 1.13; I² = 81%). To further explore the heterogeneity in the results, outcomes for all trials excluding two trials (Schoenfelder 2000; Sihvonen 2004) with 20 participants or less in each arm of the trial were pooled (this chosen threshold was arbitrary but considered indicative of 'very small' trials). This did not reduce the heterogeneity for rate of falls (Analysis 1.5: I² = 70%), or change the overall pooled estimate of rate of falls (Analysis 1.5: RaR 0.91, 95% CI 0.72 to 1.15) or risk of falling (Analysis 1.6: RR 1.04, 95% CI 0.89 to 1.21; I² = 25%). Comparison 1 Care facilities: Exercise vs usual care, Outcome 5 Rate of falls, excluding studies with ≤20 participants in each arm. Comparison 1 Care facilities: Exercise vs usual care, Outcome 6 Number of fallers, excluding studies with ≤20 participants in each arm. Funnel plots testing for publication biasWe constructed funnel plots of trials of exercise versus usual care for both the rate of falls and risk of falling outcomes. The funnel plots appeared asymmetrical for both rate of falls and risk of falling (Figure 4 and Figure 5), which may indicate publication bias or lower methodological quality leading to spuriously inflated effects in the smaller trials. In addition to the trials included in the funnel plots, there were four other trials reporting a reduction in the rate of falls. Funnel plot of comparison: 1 Care facilities: Exercise vs usual care (grouped by level of care), outcome: 1.2 Number of fallers. NB One additional trial with data not suitable for pooling reported no significant reduction in the risk of falling. Comparisons of different exercise categoriesNine trials (584 participants) provided 12 comparisons of two different exercise programmes (Faber 2006; Fu 2015; Imaoka 2016; Kovacs 2012; Saravanakumar 2014; Shimada 2004; Serra‐Rexach 2011; Sitja Rabert 2015; Tuunainen 2013). All trials were individually randomised. Seven trials (nine comparisons; 505 participants) had data suitable for pooling (Faber 2006; Fu 2015; Kovacs 2012; Saravanakumar 2014; Shimada 2004; Sitja Rabert 2015; Tuunainen 2013). Two trials provided data on the effectiveness of additional balance exercises (Shimada 2004; Tuunainen 2013). All other comparisons included only single trials; the quality of evidence was considered very low for these comparisons. Rate of fallsFive trials (Faber 2006; Fu 2015; Saravanakumar 2014; Shimada 2004; Tuunainen 2013; 305 participants) with data suitable for analysis reported the effect of nine comparisons of different exercise programmes on the rate of falls (Analysis 4.1). For eight of these comparisons there was only a single trial with less than 200 participants; the quality of the evidence was considered very low so the relative effectiveness of these exercise programmes on reducing the rate of falls remains uncertain. Comparison 4 Care facilities: Comparisons of different exercise programs (see Appendix 4 for details), Outcome 1 Rate of falls. Pooled data from two trials (Shimada 2004; Tuunainen 2013) of additional balance exercises indicated a reduction in the rate of falls (Analysis 4.1.1: RaR 0.62, 95% CI 0.40 to 0.96; I² = 0%; 56 participants; 86 falls). We are uncertain of the effect of additional balance exercise on falls as the quality of the evidence has been assessed as very low (downgraded two levels due to serious risk of bias, and one level for imprecision). Serra‐Rexach 2011 (40 participants) compared training sessions of a combination of exercises in addition to usual physiotherapy and reported fewer falls in the intervention group (Analysis 4.2). Risk of fallingSix trials (Faber 2006; Imaoka 2016; Kovacs 2012; Shimada 2004; Sitja Rabert 2015; Tuunainen 2013; 327 participants) reported the effect of seven comparisons of different exercise categories on the risk of falling (Analysis 4.3). Six comparisons contained only a single trial and the quality of evidence for these comparisons was considered very low; the relative effectiveness of these exercise programmes on reducing the risk of falling remains uncertain. Comparison 4 Care facilities: Comparisons of different exercise programs (see Appendix 4 for details), Outcome 3 Number of fallers. Pooled data from two trials (Shimada 2004; Tuunainen 2013) of additional balance exercises did not show evidence of a strong effect on reducing the risk of falling Analysis 4.3.1 (RR 0.79, 95% CI 0.43 to 1.45; I² = 0%; 56 participants; 24 fallers). We are uncertain of the effect of additional balance exercise on falls as the quality of the evidence has been assessed as very low (downgraded two levels for risk of bias, and one level for imprecision). In Imaoka 2016, there was no strong evidence for a reduction in the risk of falling in the post‐intervention period with additional group exercise (RR 0.48, 95% CI 0.17 to 1.3). Risk of fractureSitja Rabert 2015 (159 participants) compared exercise performed on a whole body vibration platform to the same land based exercises and reported one fracture in the intervention group and none in the control group (Analysis 4.4: RR 2.89, 95% CI 0.12 to 69.07; 1 fracture). We are uncertain whether or not whole body vibration reduces the risk of fracture. Comparison 4 Care facilities: Comparisons of different exercise programs (see Appendix 4 for details), Outcome 4 Number of people sustaining a fracture. Adverse eventsFour trials (269 participants) comparing alternative exercise programmes reported on adverse events; no serious adverse events were reported. Saravanakumar 2014 (29 participants) reported an instance of a non‐injurious fall during a yoga intervention. Sitja Rabert 2015 (159 participants) comparing exercise on a whole body vibration platform with land‐based exercise reported that "statistical results showed no differences between groups (P=0.430)" and that "ten percent of participants in the exercise group and 16.3% in the whole body vibration plus exercise group presented a possible or probable relation of causality with the intervention, but this difference was not statistically significant (P =0.450)." The most commonly reported adverse events were pain (18%) and soreness (13%) but these data were not reported according to group allocation. Serra‐Rexach 2011 (40 participants), testing additional physiotherapy, reported a case of transient lumbalgia. Lastly, Kovacs 2012 (41 participants), which compared a multimodel exercise programme based on Otago plus osteoporosis exercises with osteoporosis exercises, reported that there were no adverse events. Medication (drug target) interventionsMedication reviewTwelve studies (7366 participants) examined the effect of medication review interventions in care facilities on falls (Crotty 2004a; Crotty 2004b; Frankenthal 2014; Garcia Gollarte 2014; Juola 2015; Frankenthal 2014; Houghton 2014; Lapane 2011; Patterson 2010; Potter 2016; Streim 2012; Zermansky 2006). Seven trials (4536 participants) were individually randomised (Crotty 2004a; Frankenthal 2014; Frankenthal 2014; Lapane 2011; Potter 2016; Streim 2012; Zermansky 2006), and five trials (2830 participants) were cluster randomised (Crotty 2004b; Garcia Gollarte 2014; Juola 2015; Houghton 2014; Patterson 2010). Two studies (1054 participants) did not report falls data suitable for pooling (Garcia Gollarte 2014; Streim 2012). The primary aim of all medication review is generally to reduce psychoactive medications. Therefore, all trials were considered clinically similar except for one study of medication review for hyponatraemia (Peyro Saint Paul 2013). Further details of the interventions and comparisons are provided in Table 10. A summary of the evidence for general medication review for falls prevention in care facilities is provided in Table 2. Rate of fallsSix trials (2409 participants) reporting data on the rate of falls in trials of general medication review were considered clinically appropriate to pool, despite considerable statistical heterogeneity. General medication review may make little or no difference to the rate of falls (Analysis 5.1.1: RaR 0.93, 95% CI 0.64 to 1.35, 6 trials, 2409 participants; I² = 93%; low‐quality evidence). Subgroup analyses by level of care were not conducted as all trials were conducted in mixed settings. Comparison 5 Care facilities: Medication review vs usual care, Outcome 1 Rate of falls. Garcia Gollarte 2014 (716 participants) conducted a cluster‐randomised trial of education of physicians on drug use in older people, plus medication review with feedback in 10% of patients. Data from this study were not pooled as falls during the six‐month intervention period were not reported. Over the three months following the intervention, after adjustment for clustering, the rate of falls (RaR 0.74, 95% CI 0.49 to 1.13) did not provide strong evidence for an effect. A post‐hoc sensitivity analyses was conducted excluding Potter 2016 (93 participants), in which 3 participants in the intervention group had more than 30 falls. The heterogeneity in this analysis remained high (Analysis 5.4: I² = 87%) and there was no strong evidence of a reduction in the rate of falls. Comparison 5 Care facilities: Medication review vs usual care, Outcome 4 Rate of falls post‐hoc sensitivity analysis (excluding Potter 2016). One additional small trial examined medication review to avoid hyponatraemia (Peyro Saint Paul 2013; Analysis 5.1.2: nine participants), we are uncertain whether medication review reduces falls in adults with chronic moderate hyponatraemia (serum sodium level 123 mEq/L to 134 mEq/L). Streim 2012 conducted a trial that included both randomised and a non‐randomised patient‐preference arm. The randomised arms of the trial (36 participants), examined deprescribing of antidepressants. The authors reported that "the discontinuation and continuation groups exhibited similar non‐significant increases in the odds of fall per week with an increase in odds of falls of 1.38 per week (95% CI 4.07 to 0.47); Z=0.59; p=0.55) in the discontinuation group and 1.50 per week (95% CI 0.55 to 4.07); Z=0.80; p=0.43) in the continuation group. The similarity in odds ratios corresponds to discontinuation only reducing the odds ratio of falls relative to the continuation ratio by approximately 10% (ratio of ORs=0.92 (95% CI=(0.21, 4.01); Z=0.11; p=0.91)." Risk of fallingPooled data from six clinically similar trials (5139 participants) reporting falls risk data indicated that general medication review may make little or no difference to the risk of falling (Analysis 5.2.1: RR 0.93, 95% CI 0.80 to 1.09; 5139 participants: I² = 48%). The quality of the evidence was considered low (downgraded one level for risk of bias and one level for inconsistency). Comparison 5 Care facilities: Medication review vs usual care, Outcome 2 Number of fallers. In Garcia Gollarte 2014 (716 participants), after adjustment for clustering, the risk of falling (RR 0.86, 95% CI 0.59 to 1.26) did not provide strong evidence for an effect over the three months following the intervention. We are uncertain of whether medication review reduces falls in adults with chronic moderate hyponatraemia (Analysis 5.2.2: RR 0.42, 95% CI 0.07 to 2.59: 1 trial; 9 participants). Risk of fracturePotter 2016 (93 participants) reported the effect of medication review on the risk of fracture (Analysis 5.3: RR 1.60, 95%CI 0.28 to 9.16; 5 fractures), we are uncertain of the effect of medication review on risk of fracture as the quality of the evidence has been assessed as very low. Comparison 5 Care facilities: Medication review vs usual care, Outcome 3 Number of people sustaining a fracture. Subgroup analysis by cognitive statusJuola 2015 provided data for subgroups according to cognitive status. After adjustment for clustering, the rate of falls was reduced for those with an Mini Mental State Examination (MMSE) greater than 15 (RaR 0.23, 95% CI 0.12 to 0.44; 49 participants) or an MMSE of 10‐15 (RaR 0.27, 95%CI 0.17 to 0.44; 45 participants) but not for those with an MMSE <10 (RaR 1.27, 95% CI 0.95 to 1.69; 95 participants). Adverse eventsTwo studies (102 participants) reported on adverse events; the remaining 10 studies did not clearly report on adverse events related to the intervention. In a study of deprescribing (Potter 2016; 93 participants), serious vascular events occurred in three control participants and one intervention participant, and two intervention participants experienced significant adverse medicine withdrawal reactions (symptomatic rapid atrial fibrillation and agitation) (Analysis 5.5.1: RR 1.07, 95%CI 0.23 to 5.01; 1 trial). Comparison 5 Care facilities: Medication review vs usual care, Outcome 5 Serious adverse events. Peyro Saint Paul 2013 (nine participants) reported one serious adverse event (a major gastrointestinal bleed) related to discontinuing a proton‐pump inhibitor in the intervention arm. We are uncertain of the effects of medication review on adverse events as the quality of the evidence has been assessed as very low (Table 2). Vitamin D supplementationEight studies (9278 participants) examined vitamin D supplementation administered in some form (Bischoff 2003; Broe 2007; Chapuy 2002; Flicker 2005; Grieger 2009; Imaoka 2016; Kennedy 2015; Law 2006). Six trials (5561 participants) were individually randomised (Bischoff 2003; Broe 2007; Chapuy 2002; Flicker 2005; Grieger 2009; Imaoka 2016) and two trials (3717 participants) were cluster randomised (Kennedy 2015; Law 2006). Four trials (4512 participants) tested the effect of vitamin D supplementation on falls (Bischoff 2003; Broe 2007; Flicker 2005; Law 2006), one trial (583 participants) tested the effect of vitamin D and calcium supplementation (Chapuy 2002), two trials (166 participants) tested multivitamin supplementation that included vitamin D plus calcium (Grieger 2009; Imaoka 2016), and one trial (4017 participants) tested an educational intervention aimed at increasing prescription of adequate levels of vitamin D, calcium and osteoporosis medications (Kennedy 2015). Seven of the eight studies reported serum vitamin D levels at baseline (Bischoff 2003; Broe 2007; Chapuy 2002; Flicker 2005; Grieger 2009; Imaoka 2016; Law 2006). Vitamin D levels were low or very low in these studies enrolling residents of care facilities. Baseline vitamin D levels for one trial (Kennedy 2015) were not reported. A summary of the evidence for vitamin D supplementation for falls prevention in care facilities is provided in Table 3. For the specific comparison of multivitamin supplementation including vitamin D and calcium versus placebo (Grieger 2009; Imaoka 2016), the quality of the evidence was considered very low. Rate of fallsPooled data from four trials (4512 participants) indicated that vitamin D supplementation probably reduces the rate of falls (Analysis 6.1.1: RaR 0.72, 95% CI 0.55 to 0.95; I² = 62%: moderate‐quality evidence). The type of vitamin D administered is indicated in the footnotes. Comparison 6 Care facilities: Vitamin D supplementation vs no vitamin D supplementation, Outcome 1 Rate of falls. We are uncertain whether multivitamin supplementation including vitamin D and calcium reduces the rate of falls as the quality of the evidence is very low (Analysis 6.1.2: RaR 0.38, 95% CI 0.20 to 0.71; 91 participants; 1 study). An education intervention aimed at increasing the prescription of vitamin D, calcium and osteoporosis medication (Kennedy 2015) may make little or no difference to the rate of falls (Analysis 6.1.3: RaR 1.03, 95% CI 0.85 to 1.25; 4017 participants; 1 study; low‐quality evidence, downgraded two levels due to risk of bias). Risk of fallingPooled data from four trials (4512 participants) indicated that vitamin D supplementation probably makes little or no difference to the risk of falling (Analysis 6.2.1: RR 0.92, 95% CI 0.76 to 1.12; I² = 42%; moderate‐quality evidence, downgraded one level for risk of bias). Comparison 6 Care facilities: Vitamin D supplementation vs no vitamin D supplementation, Outcome 2 Number of fallers. Vitamin D plus calcium supplementation (Chapuy 2002), probably makes little or no difference to the risk of falling (Analysis 6.2.2: RR 1.03, 95% CI 0.90 to 1.18; 583 participants; 1 study; moderate‐quality evidence downgraded one level for risk of bias). We are uncertain whether multivitamin supplementation including vitamin D and calcium reduces the risk of falling (Analysis 6.2.3: RR 0.82, 95% CI 0.40 to 1.66; 91 participants; 1 study). Imaoka 2016 (75 participants), conducted a four‐arm trial which found no strong evidence for an effect of daily nutritional supplementation including 900 IU vitamin D (including 400 IU vitamin D3 and 200mg calcium in a multivitamin supplement) in comparison with usual care over the six months following the three‐month intervention period (RR 0.58, 95%CI 0.20 to 1.68, N = 34). Outcomes data were not pooled with other studies as they excluded the intervention period; falls are for six months post‐intervention. An education intervention aimed at increasing the prescription of vitamin D, calcium and osteoporosis medication (Kennedy 2015) may make little difference or no difference to the risk of falling (Analysis 6.2.4: RR 1.05, 95% CI 0.90 to 1.23; 4017 participants; 1 study; low‐quality evidence, downgraded two levels for risk of bias). Risk of fracturePooled data from three trials of vitamin D supplementation showed little effect on fall related fractures (Analysis 6.3.1: RR 1.09, 95% CI 0.58 to 2.03; I² = 63%; 4464 participants; 178 fractures: very low‐quality evidence). Different trials reported different types of fractures; the type of fractures are shown in the footnotes to the analysis. We are uncertain whether vitamin D supplementation reduces the risk of fall related fractures as the evidence has been assessed as very low. Comparison 6 Care facilities: Vitamin D supplementation vs no vitamin D supplementation, Outcome 3 Number of people sustaining a fracture. We are uncertain whether vitamin D plus calcium supplementation reduces the risk of fall related fractures (Analysis 6.3.2: RR 0.62, 95% CI 0.36 to 1.07; 583 participants; 48 hip fractures; very low‐quality evidence, downgraded one level for risk of bias, one level for imprecision and one level as this review only includes a subset of the trials available reporting the effects of this intervention on fractures). An education intervention aimed at increasing the prescription of vitamin D, calcium and osteoporosis medication (Kennedy 2015; 4017 participants) reported that 1.5% of falls in control participants and 1.6% of falls in intervention participants resulted in a fracture, the study was not powered to detect a difference in fall‐related fractures, we are uncertain of the effects of this intervention on fractures (very low‐quality evidence, downgraded two levels for risk of bias and two levels for imprecision). Adverse eventsFour trials (1365 participants) reported adverse‐event data. Two of four trials (747 participants) of vitamin D supplementation reported on adverse events (Bischoff 2003, Flicker 2005); no serious adverse events were reported. Bischoff 2003 reported two cases of increased constipation in the intervention arm and no cases of hypercalcaemia (Analysis 6.4.1: constipation RR 4.84, 95%CI 0.24 to 98.80; 122 participants). Flicker 2005 reported that there were no adverse events. We are uncertain of the effects of Vitamin D supplementation (up to 1000 IU daily) on adverse events as the quality of the evidence has been assessed as very low (Table 3). Comparison 6 Care facilities: Vitamin D supplementation vs no vitamin D supplementation, Outcome 4 Adverse events. One trial of vitamin D and calcium supplementation (800 IU of vitamin D3 + 1200 mg calcium carbonate daily) reported a similar rate of gastrointestinal disorders in each arm of the study and three cases of hypercalcaemia in the intervention arm, we are uncertain of the effects on adverse events (Chapuy 2002; Analysis 6.4.2; gastrointestinal adverse events RR 0.82, 95% CI 0.45 to 1.48; 583 participants; very low‐quality, downgraded one level for risk of bias and two levels for imprecision). Grieger 2009, which tested multivitamin supplementation including vitamin D and calcium, reported there were no serious adverse events; the three adverse events reported were in the control arm of the trial (rash/vertigo, behavioural issues, indigestion), we are uncertain of the effects on adverse events (Analysis 6.4.2: RR 0.13, 95% CI 0.01 to 2.41; 91 participants, 40 events; very low‐quality evidence). Environment/assistive technologyIn a cross‐over trial, Clifton 2009 (43 participants) tested a wireless position‐monitoring device and found no strong evidence for a reduction in the rate of falls (Analysis 7.1: RaR 0.65, 95% CI 0.33 to 1.27; no adjustments for cross‐over design made in the analysis). There were no serious adverse events. We are uncertain whether or not wireless position monitoring has an effect on the rate of falls in care facilities (very low‐quality evidence). Comparison 7 Care facilities: Environmental interventions vs usual care, Outcome 1 Rate of falls. Social environmentSeven cluster‐randomised trials examined service change interventions in care facilities (13,127 participants in six trials Cox 2008; Chenoweth 2009; Meyer 2009; Van de Ven 2014; Van Gaal 2011a; Ward 2010, plus 982 facility beds in Colon‐Emeric 2013). These included three trials of staff training interventions (Colon‐Emeric 2013; and 7029 participants from Cox 2008 and Van Gaal 2011a) and four of a service model change (6098 participants; Chenoweth 2009; Meyer 2009; Van de Ven 2014; Ward 2010). These interventions target staff or caregivers and changes in the organisational system in which an intervention is delivered, rather than targeting patients directly. The rate of falls for these interventions were not pooled due to high clinical and statistical heterogeneity (test for subgroup differences: P = 0.0001, I² = 85.6%). Two studies (6516 participants) reported data on risk of fracture (Meyer 2009, Ward 2010). No studies reported on adverse events. Although there were only single trials for the comparisons within this category, the generally larger size of these trials meant that optimal information size criteria may be met and GRADE assessments were conducted by two review authors. Staff trainingCox 2008 (5637 participants) studied a half day education programme about fall and fracture prevention for managers, nurses and health care assistants, given by specialist osteoporosis nurses. There was no strong evidence for a reduction in the rate of falls, we are uncertain of the effects as the quality of the evidence was assessed as very low (Analysis 8.1.1: RaR 1.19, 95% CI 0.92 to 1.53; very low‐quality evidence, downgraded two levels for risk of bias and one level for imprecision). The intervention may make little or no difference to the rate of fracture (reported incidence rate ratio (IRR) for all fractures: IRR 0.94, 95% CI 0.71 to 1.26; for hip fractures: IRR 0.86, 95% CI 0.63 to 1.18; low‐quality evidence downgraded two levels for risk of bias). Comparison 8 Care facilities: Social environment vs usual care, Outcome 1 Rate of falls. The intervention in Van Gaal 2011a (392 participants) consisted of education to implement a patient‐safety programme directed at falls, urinary tract infection, and pressure ulcers based on available guidelines. There was no strong evidence for a reduction in rate of falls, we are uncertain of the effects on the rate of falls (Analysis 8.1.2: RaR 0.63, 95% CI 0.34 to 1.16; very low‐quality evidence, downgraded two levels for risk of bias, one level for indirectness and one level for imprecision). Colon‐Emeric 2013 (number of resident participants not reported, 497 staff participants, 982 facility beds) conducted a pilot cluster‐randomised trial testing a programme to improve staff connections, communication, and problem solving compared to usual care during implementation of a falls quality improvement programme. There was no strong evidence for an effect on the change in falls rate from baseline to post intervention periods between the two arms of the study, we are uncertain of the effects in reducing falls (RaR of change in falls rate 0.81, 95% CI 0.55 to 1.20; very low‐quality evidence, downgraded one level for each of risk of bias, indirectness and imprecision). Service model changeMeyer 2009 (1125 participants) found that use of a falls risk‐assessment tool in comparison with nurses' judgement alone probably makes little or no difference to the rate of falls or risk of falling (Analysis 8.1.3: RaR 0.96, 95% CI 0.84 to 1.10; Analysis 8.2: RR 0.99, 95% CI 0.85 to 1.16; both outcomes moderate‐quality evidence, downgraded one level for risk of bias). We are uncertain whether or not this intervention reduces the risk of fracture as the quality of the evidence was assessed as very low (Analysis 8.3.1: RR 0.96, 95% CI 0.57 to 1.63; 77 fractures in total; downgraded one level for risk of bias and two levels for imprecision). Comparison 8 Care facilities: Social environment vs usual care, Outcome 2 Number of fallers. Comparison 8 Care facilities: Social environment vs usual care, Outcome 3 Number of people sustaining a fracture. Two studies examined dementia care mapping, but data from Chenoweth 2009 were not suitable for pooling. Chenoweth 2009 (289 participants) reported that "... at follow‐up there were fewer falls with dementia‐care mapping than in usual care (p=0·02) and more falls in person‐centred care than in usual care (p=0·03)." Van de Ven 2014 (293 participants) delivered a four‐month dementia care mapping intervention twice during the 12‐month follow‐up period after baseline. The rate of falls at study endpoint was greater in the intervention arm of the study (Analysis 8.1.4: RaR 1.84, 95% CI 1.40 to 2.42). We are uncertain of the effects of dementia care mapping on the rate of falls as the quality of the evidence has been assessed as very low (downgraded two levels for risk of bias, one level for inconsistency and one level for imprecision). Ward 2010 (5391 participants) employed a practice nurse to encourage the adoption of best practice strategies and reported "0.13 fewer falls per 100 beds per month; 95% CI, −0.36 to 0.10; P = 0.259" for the intervention period. There was no difference in risk of hip fracture between intervention and control groups during the 17 months of intervention (Analysis 8.3.2; RR 0.95, 95% CI 0.63 to 1.44; 215 hip fractures). We are uncertain of the effects of this intervention on fractures as the quality of the evidence has been assessed as very low (downgraded two levels for risk of bias, and two levels for imprecision). Psychological interventionsTwo studies (163 participants) examined the impact of psychological interventions on falls (Huang 2016; Van het Reve 2014). Both trials were individually randomised, Huang 2016 is a three‐arm trial for which falls excluded the intervention period; findings are also discussed under "Care facilities: multiple interventions". Neither trial reported data on the risk of fracture or adverse events. In Van het Reve 2014 (114 participants) a computer‐based cognitive training programme focused on improving attention was combined with strength and balance training, and compared with strength and balance training alone. The intervention showed no strong evidence for an effect on falls rates (Analysis 9.1: RaR 1.22, 95% CI 0.78 to 1.92), risk of falling during the intervention period (Analysis 9.2.2; RR 1.35, 95% CI 0.23 to 7.88) or over 12 months post‐intervention (RR 1.38, 95% CI 0.76 to 2.51; data not shown). Comparison 9 Care facilities: Psychological interventions vs control, Outcome 1 Rate of falls. Comparison 9 Care facilities: Psychological interventions vs control, Outcome 2 Number of fallers. In a three‐arm study, Huang 2016 tested the effects of a cognitive‐behavioural intervention conducted by a trained facilitator in comparison with usual care in 49 participants. Over the three months following the intervention, there were 1.67 falls per person year in the usual care arm of the study (10 falls in seven fallers), but no falls in the cognitive‐behavioural intervention arm. Data were not pooled as falls excluded the intervention period. The quality of the evidence for both the rate and risk of falling was considered very low (downgraded one level for risk of bias, inconsistency and indirectness and two levels for imprecision), so we are uncertain of the effectiveness of psychological interventions in reducing falls. Other single interventionsThree trials (564 participants) examined other single interventions of lavender olfactory stimulation (Sakamoto 2012), sunlight exposure (Sambrook 2012), and multisensory stimulation in a Snoezelen room (Klages 2011); two trials (169 participants) were individually randomised (Sakamoto 2012; Klages 2011) and one (Sambrook 2012; 395 participants) was cluster randomised. The quality of the evidence was considered very low for all of these single‐trial comparisons. For one year, Sakamoto 2012 (145 participants) tested the effect of lavender olfactory stimulation by applying lavender patches or placebo patches to clothing near the neck daily. This intervention did not show strong evidence for a reduction in the rate of falls (Analysis 10.1: RaR 0.57, 95% CI 0.32 to 1.01) or risk of falling (Analysis 10.2: RR 0.67, 95% CI 0.40 to 1.12). The authors reported that there were no adverse events. We are uncertain of the effectiveness of lavender olfactory stimulation as the quality of the evidence is very low. Comparison 10 Care facilities: Other single interventions vs control, Outcome 1 Rate of falls. Comparison 10 Care facilities: Other single interventions vs control, Outcome 2 Number of fallers. In Sambrook 2012 (395 participants), a trial of increased sunlight exposure had low adherence to the sunlight intervention (Durvasula 2012). We are uncertain of the effects on falls as the quality of the evidence has been assessed as very low for all outcomes (downgraded one level for each of risk of bias, indirectness and imprecision; Analysis 10.1.2: RaR 1.05, 95% CI 0.71 to 1.56; Analysis 10.2.2: RR 1.09, 95% CI 0.88 to 1.36; Analysis 10.3: risk of fracture: RR 1.07, 95% CI 0.53 to 2.17, total 32 fractures). The authors reported no difference in the incidence rates of new skin cancers between arms of the trial and one fall on the way to a sunlight session. Adverse‐event data for this three‐arm trial are also reported below under Multiple interventions. Comparison 10 Care facilities: Other single interventions vs control, Outcome 3 Number of people sustaining a fracture. Klages 2011 (24 participants) compared the effect of multisensory stimulation in a Snoezelen room with control activities in people with dementia and reported, without providing data, that the "Group membership did not alter falls frequency". Adverse‐event data were not reported. We are uncertain of the effectiveness of multisensory stimulation as the quality of the evidence is very low. Care facilities: multiple interventionsIn multiple interventions, the same combination of single categories of intervention was delivered to all participants in the group. Three trials (652 participants) examined multiple interventions in care facilities (Sambrook 2012; Schnelle 2003; Huang 2016). One trial (412 participants) was cluster randomised (Sambrook 2012) and two trials (240 participants) were individually randomised. The quality of the evidence was considered very low for the single trial comparisons of exercise plus management of urinary incontinence and fluid therapy with usual care (Schnelle 2003), and cognitive‐behavioural therapy to address fear of falling with an exercise programme versus usual care (Huang 2016). In Schnelle 2003 (190 participants), participants engaged in supervised exercises and were offered fluids and regular toileting. There was no strong evidence for an effect in reducing the rate of falls (Analysis 11.1.1: RaR 0.62, 95% CI 0.38 to 1.01), risk of falling (Analysis 11.2.1: RR 0.62, 95% CI 0.36 to 1.05) or risk of fracture (Analysis 11.3.1: RR 4.26, 95% CI 0.48 to 37.55; total five fractures). Adverse events were not reported. We are uncertain of the effectiveness of this intervention as the quality of the evidence is very low. Comparison 11 Care facilities: Multiple interventions vs usual care, Outcome 1 Rate of falls. Comparison 11 Care facilities: Multiple interventions vs usual care, Outcome 2 Number of fallers. Comparison 11 Care facilities: Multiple interventions vs usual care, Outcome 3 Number of people sustaining a fracture. One intervention group in Sambrook 2012 (412 participants), which was based in Australia, tested the effect of increased sunlight exposure plus calcium supplementation, with low adherence to the sunlight intervention (Durvasula 2012). We are uncertain of the effects on falls as the quality of the evidence has been assessed as very low for all outcomes (downgraded one level for each of risk of bias, indirectness and imprecision; Analysis 11.1.2: RaR 1.03, 95% CI 0.85 to 1.25; Analysis 11.2.2: RR 0.96, 95% CI 0.77 to 1.19; Analysis 11.3.2: risk of fracture RR 0.78, 95% CI 0.36 to 1.67; total 31 fractures). The authors reported no significant difference in the incidence rates of new skin cancers between arms of the trial (18 new cancers total) and an increase in the adjusted all‐cause mortality in the calcium‐treated group compared with the UV alone group (HR 1.23 versus 0.76, P = 0.03; 40 deaths; adjusted for age, sex and season). There was a lack of evidence for a strong effect on increased death rates from myocardial infarction (age‐adjusted HR 3.83, 95% CI 0.97 to 15.27, P = 0.06; sex‐adjusted HR 4.17, 95% CI 0.69 to 25.16, P = 0.12; the authors reported that they did not record cardiovascular events prospectively). We are uncertain of the effects on adverse events as the quality of the evidence is very low (downgraded one level for each of risk of bias, indirectness and imprecision). In a three‐arm trial, Huang 2016 studied an intervention which combined cognitive‐behavioural therapy to address fear of falling with an exercise programme in comparison with usual care in 50 participants. In the three months following the eight‐week intervention the authors reported a reduction in falls in both the combined intervention and the cognitive‐behavioural intervention arm alone (reported Kruskal‐Wallis P < 0.001). There were 1.67 falls per person year in the usual care arm of the study (10 falls in seven fallers), and no falls in the cognitive behavioural plus exercise intervention arm; data were not pooled as falls excluded the intervention period. Adverse events were not reported. We are uncertain of cognitive‐behavioural therapy combined with an exercise programme as the quality of the evidence is very low. Care facilities: multifactorial interventionsIn multifactorial interventions, two or more categories of intervention are given, and these are linked to each individual's risk profile. An initial assessment is usually carried out by one or more health professionals and an intervention is then provided or recommendations given or referrals made for further action. A summary of the evidence for multifactorial interventions in comparison with usual care in care facilities is provided in Table 4. Thirteen trials (4226 participants) in care facilities studied multifactorial interventions (Beck 2016; Becker 2003; Dyer 2004; Jensen 2002; Kerse 2004; McMurdo 2000; Neyens 2009; Ray 1997; Rubenstein 1990; Salvà 2016; Shaw 2003; Walker 2015; Whitney 2017). Eleven trials were cluster‐randomised trials (Beck 2016; Becker 2003; Dyer 2004; Jensen 2002; Kerse 2004; McMurdo 2000; Neyens 2009; Ray 1997; Salvà 2016; Walker 2015; Whitney 2017; 3470 participants), and two were individually randomised (Rubenstein 1990; Shaw 2003; 756 participants). Whitney 2017 was also a cross‐over trial. None of these trials were sufficiently similar to allow analysis of subgroups of specific combinations of interventions. Two studies did not report data suitable for use in the quantitative analysis (Beck 2016; Ray 1997). Three studies (2160 participants) reported data on hip fractures (Becker 2003; Jensen 2002; Shaw 2003), and one reported total fractures (Salvà 2016). Three studies (312 participants) reported adverse‐event data (Beck 2016; McMurdo 2000; Whitney 2017). Rate of fallsDespite statistical heterogeneity between the trials for the rate of falls, trials were considered clinically similar enough for pooling to be meaningful. Pooled data from 10 trials (3439 participants) for rate of falls did not demonstrate strong evidence for a reduction in falls (Analysis 12.1: RaR random effects 0.88, 95% CI 0.66 to 1.18: I² = 84%). Beck 2016 (31 participants) reported falls outcomes in a cluster‐randomised trial of an exercise programme plus nutritional support. There were zero falls in the intervention arm and two in the control arm over an 11‐week period. Overall, we are uncertain of the effects of multifactorial interventions on the rate of falls in care facilities as the quality of evidence has been assessed as very low (Table 4). Comparison 12 Care facilities: Multifactorial interventions vs usual care, Outcome 1 Rate of falls. Risk of fallingPooled data from nine trials (3153 participants) for risk of falling (Analysis 12.2: RR random effects 0.92, 95% CI 0.81 to 1.05: I² = 42%) did not demonstrate strong evidence for a reduction in falls. Ray 1997 (482 participants) only recorded the number of people having two or more falls during follow‐up (recurrent fallers) and reported a reduction in the proportion of recurrent fallers (difference 19%, 95% CI 2% to 36%: P = 0.03). Overall, multifactorial interventions in care facilities may make little or no difference to the risk of falling (low‐quality evidence; Table 4). Comparison 12 Care facilities: Multifactorial interventions vs usual care, Outcome 2 Number of fallers. Risk of fracturePooled results for five studies (2160 participants) reporting risk of fracture did not show strong evidence for an effect (Analysis 12.3: RR 0.79, 95% CI 0.30 to 2.07: I² = 44%; 76 fractures). Data from three of the five trials (1695 participants) were for hip fracture (Becker 2003; Jensen 2002; Salvà 2016) and two trials (465 participants) reported total fractures (Shaw 2003; Whitney 2017). Two trials (1255 participants) included hip protectors as an intervention (Becker 2003; Shaw 2003). We are uncertain of the effects of multifactorial interventions on the risk of fracture as the quality of evidence has been assessed as very low (Table 4). Comparison 12 Care facilities: Multifactorial interventions vs usual care, Outcome 3 Number of people sustaining a fracture. Adverse eventsThree studies (312 participants) reported adverse‐event data. One trial reported an instance of a fall in the intervention arm (Whitney 2017), two studies reported that there were no adverse events (Beck 2016; McMurdo 2000). We are uncertain of the effects of multifactorial interventions on adverse events as the quality of evidence has been assessed as very low (Table 4). Subgroup analyses exploring heterogeneityTo explore the heterogeneity in these results, we carried out post‐hoc subgroup analysis by levels of care (high or intermediate or mixed levels of care). The test for subgroup differences showed a difference between subgroups for both the rate of falls (Analysis 13.1: P = 0.005, I² = 81%) and risk of falling (Analysis 13.2: P = 0.03, I² = 72%). Within care facilities providing either high or intermediate levels of care, statistical heterogeneity was not important and pooled data showed a reduction in both the rate of falls (Analysis 13.1.1: high‐level care: RaR 0.59, 95% CI 0.44 to 0.79; I² = 8%, P = 0.30; Analysis 13.1.2: intermediate‐level care: RaR 0.64, 95% CI 0.50 to 0.83; I² = 33%, P = 0.23), and the risk of falling (Analysis 13.2.1: high level care: RR 0.75, 95% CI 0.57 to 0.98; Analysis 13.2.2: intermediate level care: RR 0.75, 95% CI 0.60 to 0.94; I² = 0%, P = 0.44). However, heterogeneity remained high in studies of mixed levels of care (Analysis 13.1.3: RaR 1.23, 95% CI 0.85 to 1.77; I² = 77%, P = 0.001; Analysis 13.2.3: RR 1.01, 95% CI 0.88 to 1.15; I² = 24%, P = 0.26). Comparison 13 Care facilities: Multifactorial interventions vs usual care (grouped by level of care), Outcome 1 Rate of falls. Comparison 13 Care facilities: Multifactorial interventions vs usual care (grouped by level of care), Outcome 2 Number of fallers. We also carried out a subgroup analysis comparing trials recruiting people with cognitive impairment versus trials with participants with no cognitive impairment (based on inclusion/exclusion criteria) or a mixed sample. Two trials recruited residents with cognitive impairment only (Neyens 2009; Shaw 2003). In addition, two trials (Becker 2003; Jensen 2002) carried out pre‐planned subgroup analyses by levels of cognition, which are reported in Rapp 2008 and Jensen 2003, respectively. Cognitive impairment was defined differently in all four studies (see footnotes to Analysis 14.1 and Analysis 14.2). There was no evidence of subgroup differences between those with higher or mixed levels of cognition and those with lower cognition for both rate of falls (Analysis 14.1: test for subgroup differences P = 0.97, I² = 0%) and risk of falling (Analysis 14.2: test for subgroup differences P = 0.41, I² = 0%). Comparison 14 Care facilities: Multifactorial interventions vs usual care (grouped by level of cognition), Outcome 1 Rate of falls. Comparison 14 Care facilities: Multifactorial interventions vs usual care (grouped by level of cognition), Outcome 2 Number of fallers. Subgroup analysis based upon the individual components of the interventions was precluded by the study design. Sensitivity analysisConsidering statistical heterogeneity in the rate of falls, meta‐analyses with a random‐effects model was considered the most appropriate. However, there was only moderate heterogeneity in the risk of falling data, therefore trials were pooled using the fixed‐effect model as a sensitivity analysis. Pooled data from 10 trials (3439 participants) using a fixed‐effect model for rate of falls showed an RaR 0.87, 95% CI 0.79 to 0.97 (compare with Analysis 12.1: I² = 84%) and from nine trials (3153 participants) for risk of falling showed an RR 0.92, 95% CI 0.84 to 1.00 (compare with Analysis 12.2: I² = 42%). Funnel plots testing for publication biasA funnel plot of trials of multifactorial interventions in care facilities was conducted for the outcome of rate of falls (Figure 6). There was no obvious asymmetry on visual inspection. Hospitals: single interventionsExerciseThree individually‐randomised trials (244 participants) tested the effect of additional physiotherapy in rehabilitation wards (Donald 2000; Jarvis 2007; Treacy 2015). One study tested additional strengthening exercises (Donald 2000), one additional balance training (Treacy 2015), and one additional physiotherapy (Jarvis 2007). A summary of the evidence for exercise for falls prevention in hospitals is provided in Table 5. No data on the risk of fractures were reported. One trial (161 participants) reported that there were no adverse events (Treacy 2015), two studies did not report adverse‐event data. Pooled data did not provide evidence for a reduction in rate of falls (Analysis 15.1: RaR 0.59, 95% CI 0.26 to 1.34; 215 participants, 2 trials; I² = 0%; very low‐quality evidence). Pooled data from two trials (83 participants) showed a reduction in the risk of falling (Analysis 15.2: RR 0.36, 95% CI 0.14 to 0.93: I² = 0%: very low‐quality evidence). We are uncertain whether additional exercise reduces the rate or risk of falling or has adverse events as the evidence has been assessed as very low. Comparison 15 Hospitals: Additional exercises vs usual physiotherapy, Outcome 1 Rate of falls. Comparison 15 Hospitals: Additional exercises vs usual physiotherapy, Outcome 2 Number of fallers. Medication (drug target) interventionsTwo trials (319 participants) examined medication target interventions, one examined medication review (Michalek 2014), and the other vitamin D supplementation (Burleigh 2007). These comparisons were from single trials only and the quality of evidence was considered very low. Multiprofessional medication reviewMichalek 2014 (114 participants) conducted a quasi cluster‐randomised trial that examined the effect of review of suitability of medications for aged patients in comparison with usual care. After adjustment for clustering there was no strong evidence for an effect on the rate of falls (Analysis 16.1: RaR 0.14, 95% CI 0.00 to 6.63) or risk of falling (Analysis 16.2: RR 0.18, 95% CI 0.01 to 3.47). Adverse‐event data were not reported. We are uncertain of the effectiveness of medication review in hospitals as the quality of the evidence is very low. Comparison 16 Hospitals: Medication review vs usual care, Outcome 1 Rate of falls. Comparison 16 Hospitals: Medication review vs usual care, Outcome 2 Number of fallers. Vitamin D supplementationBurleigh 2007 (205 participants) conducted an individually‐randomised trial that investigated whether 800 IU of vitamin D plus 1200 mg of calcium supplements reduced falls compared with 1200 mg calcium supplements alone in participants with a median length of stay of 30 days. There was no strong evidence for an effect on risk of falling (Analysis 17.1: RR 0.82, 95% CI 0.59 to 1.14) or fractures (Analysis 17.2: RR 0.34, 95% CI 0.04 to 3.05; total four fractures). The rates of gastrointestinal complaints were similar between the arms of the trial (Analysis 17.3: RR 1.37, 95% CI 0.32 to 5.98). We are uncertain of the effectiveness of vitamin D supplementation in hospitals as the quality of the evidence is very low. Comparison 17 Hospitals: Vitamin D supplements vs no vitamin D supplements, Outcome 1 Number of fallers. Comparison 17 Hospitals: Vitamin D supplements vs no vitamin D supplements, Outcome 2 Number of people sustaining a fracture. Comparison 17 Hospitals: Vitamin D supplements vs no vitamin D supplements, Outcome 3 Adverse events. Environment/assistive technology interventionsSix trials (39,127 participants) examined environment or assistive technology interventions, two trials (11,153 participants) were of furnishing adaptations (Donald 2000; Haines 2010), and four (27,974 participants) were of communication aids (Mayo 1994; Shorr 2012; Tideiksaar 1993; Wolf 2013). Four trials (356 participants) were individually randomised (Donald 2000; Mayo 1994; Tideiksaar 1993; Wolf 2013), and two (38, 771 participants) were cluster randomised (Haines 2010; Shorr 2012). Donald 2000 was a 2 x 2 factorial design. The quality of the evidence was considered very low for the single trial comparisons of carpet in comparison with vinyl floors (Donald 2000) and identification bracelets for high‐risk fallers (Mayo 1994). Furnishing/adaptationsDonald 2000, in a factorial design with 54 participants, found that carpeted floors compared with existing vinyl floors in subacute hospital wards resulted in an increase in rate of falls (Analysis 18.1.1: RaR 14.73, 95% CI 1.88 to 115.35) and no strong evidence for an increase in the risk of falling (Analysis 18.2.1: RR 8.33, 95% CI 0.95 to 73.37). We are uncertain of the impact of carpeting on falls as the quality of the evidence is very low. Comparison 18 Hospitals: Environmental interventions vs usual care, Outcome 1 Rate of falls. Comparison 18 Hospitals: Environmental interventions vs usual care, Outcome 2 Number of fallers. In a cluster‐randomised trial, Haines 2010 (11,099 participants) examined an intervention which consisted of providing one low‐low bed per 12 existing beds in acute and subacute wards. There was no strong evidence of an effect on the rate of falls; we are uncertain of the effectiveness of low‐low beds as the quality of the evidence is considered very low (Analysis 18.1.2: RaR 1.39, 95% CI 0.22 to 8.78; very low‐quality evidence downgraded two levels for risk of bias, one level for indirectness and two levels for imprecision). Neither trial reported adverse event or fracture data. Communication aidsIdentification bracelet for high‐risk fallersMayo 1994 (134 participants) studied the effect of wearing a blue identification bracelet on falls in high‐risk patients in a subacute hospital setting. They found no reduction in rate of falls (Analysis 18.1.3: RaR 1.15, 95% CI 0.72 to 1.84) or risk of falling (Analysis 18.2.2: RR 1.34, 95% CI 0.76 to 2.36). In this study, there was no reduction in risk of falling in the subgroup with a Short Portable Mental Status Questionnaire (SPMSQ) score < 9 (low cognition) or the subgroup with SPMSQ score ≥ 9 (high cognition). Adverse events were not reported. We are uncertain of the effectiveness of identification bracelets for reducing falls in hospitals as the quality of the evidence is very low. Bed exit alarmsThree trials (28,717 participants) examined bed exit alarms in hospital (Shorr 2012; Tideiksaar 1993; Wolf 2013). One large trial (Shorr 2012) was cluster randomised. A summary of the evidence for bed exit alarms for falls prevention in hospitals is provided in Table 6. Shorr 2012 (27,672 participants) examined an educational intervention to support clinical judgement on the use of bed or chair exit alarms. Wolf 2013 (98 participants) enrolled patients with an increased risk of falling that required assistance with mobilisation during rest time. Pooled data from these two studies did not show a strong reduction in the rate of falls (Analysis 18.1.4: RaR 0.60, 95% CI 0.27 to 1.34: very low‐quality evidence) or risk of falling (Analysis 18.2.3: RR 0.93, 95% CI 0.38 to 2.24: very low‐quality evidence). We are uncertain whether bed exit alarms reduce the rate of falls or risk of falling as the quality of the evidence has been assessed as very low Tideiksaar 1993 (70 participants) studied bed exit alarms for preventing falls in hospital. During the nine‐month evaluation period, "There was no significant difference in the number of bed‐falls between the two groups (p = 1.00)." Two trials of bed alarms (27,742 participants) indicated that there were no adverse events (Shorr 2012; Tideiksaar 1993); we are uncertain of the effects of bed alarms on adverse events as the quality of the evidence has been assessed as very low (Table 6). Social environmentSocial environment interventions target staff members and changes in the organisational system, rather than targeting patients directly. Six trials (9074 participants) examined service model change interventions (Dykes 2010; Koh 2009; Mador 2004; Stenvall 2007; Van Gaal 2011b; Wald 2011). Three trials (8587 participants) were cluster randomised (Dykes 2010; Koh 2009; Van Gaal 2011b), and three (487 participants) were individually randomised (Mador 2004; Stenvall 2007; Wald 2011). Studies were not pooled as they were considered to examine clinically heterogenous interventions. One study reported data on risk of fracture (Stenvall 2007). None of the studies reported adverse‐event data. We are uncertain of the effects of all social environment interventions in hospitals as the quality of the evidence was assessed as very low. Service model changeTwo studies examined implementation of guidelines in acute care settings in hospitals. Koh 2009 (1122 participants) compared multifaceted fall‐prevention guideline implementation with routine dissemination. There was no strong evidence for an effect on the rate of falls (Analysis 19.1.1: RaR 1.82, 95% CI 0.23 to 14.55; very low‐quality evidence, downgraded two levels for risk of bias, one level for indirectness and two levels for imprecision). Van Gaal 2011b (2201 participants) studied the implementation of three guidelines (falls, urinary tract infection, pressure ulcers) targeting nursing staff in comparison with usual care. There was no strong evidence for an effect on the rate of falls (Analysis 19.1.2: RaR 0.67, 95% CI 0.17 to 2.59; very low‐quality evidence, downgraded two levels for risk of bias, and two levels for imprecision). We are uncertain of the effects of guideline implementation on falls as the quality of the evidence is considered very low. Comparison 19 Hospitals: Social environment vs control, Outcome 1 Rate of falls. Dykes 2010 (5264 participants) tested the effect of a computer‐based fall‐prevention tool kit in comparison with usual care. There was no strong evidence for an effect on the rate of falls (Analysis 19.1.3: RaR 0.55, 95% CI 0.02 to 16.29) or risk of falling (Analysis 19.2.1 RR 0.91, 95% CI 0.06 to 14.21). We are uncertain of the effectiveness of this intervention (very low‐quality evidence, downgraded two levels for risk of bias, and two levels for imprecision). Comparison 19 Hospitals: Social environment vs control, Outcome 2 Number of fallers. Wald 2011 (217 participants) compared providing care in an acute ward for the elderly with care in general medical wards to usual care. There was no strong evidence for an effect on the rate of falls (Analysis 19.1.4: RaR 0.72, 95% CI 0.10 to 5.10). Mador 2004 (71 participants) examined a new behavioural advisory service for people with confusion in comparison with usual care. There was no strong evidence for an effect on the risk of falling (Analysis 19.2.2: RR 2.44, 95% CI 0.85 to 7.02). Stenvall 2007 (199 participants) compared post‐operative care in a ward providing a comprehensive ortho‐geriatric service with usual care in an orthopaedic ward following surgery for hip fracture. This intervention achieved a reduction in the rate of falls (Analysis 19.1.5: RaR 0.38, 95% CI 0.19 to 0.74) and the risk of falling (Analysis 19.2.3: RR 0.41, 95% CI 0.20 to 0.83) at discharge. There were four new fractures in the control group but none in the intervention group (Analysis 19.3.1: RR 0.11, 95% CI 0.01 to 1.52). These findings also applied to the subgroup analysis of patients with dementia (64 participants), i.e. the rate of falls and risk of falling was reduced (RaR 0.07, 95% CI 0.01 to 0.57; RR 0.12, 95% CI 0.02 to 0.85). Comparison 19 Hospitals: Social environment vs control, Outcome 3 Number of people sustaining a fracture. Knowledge interventionsTwo trials (3028 participants) examined knowledge interventions in hospitals in individually‐randomised trials. Neither trial reported data on the risk of fracture. Haines 2011 reported that there were no adverse events from interaction with the education materials; Ang 2011 did not report on adverse events. Ang 2011 (1822 participants), testing an educational session by a trained research nurse targeting individual fall risk factors in patients at high risk of falling in an acute setting and achieved a reduction in risk of falling (Analysis 20.2: RR 0.29, 95% CI 0.11 to 0.74); however, we are uncertain of the effects of this intervention as the quality of the evidence has been assessed as very low (downgraded two levels for risk of bias, one level for indirectness and one level for imprecision). Comparison 20 Hospitals: Knowledge/education interventions vs usual care, Outcome 2 Number of fallers. Haines 2011 (1206 participants) evaluated two forms of multimedia patient education compared with usual care in a mixture of acute and subacute wards. One intervention consisted of written and video‐based materials plus one‐on‐one bedside follow‐up from a physiotherapist (complete programme) and the other intervention group received educational materials only. Neither intervention showed strong evidence of a reduction in the rate of falls (Analysis 20.1.1 complete programme RaR 0.83, 95%CI 0.54 to 1.27; very low‐quality evidence, downgraded one level for indirectness, one level for inconsistency and one level for imprecision; Analysis 20.1.2 educational materials only RaR 0.91, 95%CI 0.62 to 1.35; low‐quality evidence, downgraded one level for indirectness and one level for imprecision) or risk of falling (Analysis 20.2.2 complete programme RR 0.74, 95%CI 0.48 to 1.14; very low‐quality evidence, downgraded one level for indirectness, one level for inconsistency and one level for imprecision; Analysis 20.2.3 educational materials only RR 0.84, 95% CI 0.56 to 1.27; low‐quality evidence, downgraded one level for indirectness and one level for imprecision). In a post‐hoc subgroup analysis, in participants who were cognitively intact the authors reported that falls were less frequent in those receiving the complete programme, compared with those in the materials only group (adjusted hazard ratio (HR) for rate of falls 0.51, 95% CI 0.28 to 0.93; risk of falling 0.65, 95% CI 0.36 to 1.18; 626 participants) and the control group (adjusted HR for rate of falls 0.43, 95% CI 0.24 to 0.78; risk of falling 0.51, 95%CI 0.28 to 0.94; 590 participants) (test for subgroup differences P < 0.05). There was a higher risk of injurious falls in those with cognitive impairment with the complete programme (7.49 falls per 1000 patient days compared with 2.89 falls per 1000 patient days in the control group; 192 participants). We are uncertain of the effects of the complete educational programme with follow‐up on falls (very low‐quality evidence) but providing educational materials only may make little or no difference to the rate of falls or risk of falling (low‐quality evidence). Comparison 20 Hospitals: Knowledge/education interventions vs usual care, Outcome 1 Rate of falls. Other single interventionsNo included studies examined other single interventions in a hospital setting. Hospitals: multiple interventionsNo included studies examined multiple interventions in a hospital setting. Hospitals: multifactorial interventionsSix trials (45,416 participants) tested the effect of multifactorial interventions in comparison with usual care in a hospital setting (Aizen 2015; Barker 2016; Cumming 2008; Haines 2004; Healey 2004; Hill 2015). Five trials (44,790 participants) were cluster randomised (Aizen 2015; Barker 2016; Cumming 2008; Healey 2004; Hill 2015), and one (626 participants) was individually randomised (Haines 2004). Two trials used a stepped‐wedge design (Aizen 2015; Hill 2015). The categories of interventions for each trial are shown in Appendix 3 and further details are provided in the Characteristics of included studies. A summary of the evidence for multifactorial interventions for falls prevention in hospitals is provided in Table 7. Two studies (4625 participants) reported data on risk of fracture (Cumming 2008; Haines 2004). Four of six trials (39,763 participants) reported on adverse events (Aizen 2015; Barker 2016; Haines 2004; Hill 2015). We have shown whether the settings were acute or subacute in the footnotes of the analyses. Given most of these trials were large with important differences such as in the setting and in the format and delivery of their multifactorial intervention, we present some details of the individual trials first before reporting the pooled analyses. Aizen 2015 (752 participants) conducted a two‐stage (stepped‐wedge) cluster randomised trial in five geriatric rehabilitation wards. The multifactorial intervention included medical, behavioural, cognitive and environmental modifications with additional orientation guidance and mobility restriction for moderate‐risk patients and permanent personal supervision for high‐risk patients. The usual care arm included any activities undertaken by the participants recommended or administered by their treating team. The authors reported that "No significant difference was found in fall rates during follow‐up between intervention and control wards". The findings of this study were not pooled as some aspects of the study methodology and data collection could not be confirmed. Barker 2016 (35,264 participants, 46,245 admissions) investigated a “6‐PACK” intervention in comparison with usual care (which included standard falls prevention activities) with a cluster‐randomised trial in 24 acute medical or surgical wards and found no change in rate of falls or risk of falling. There was no evidence of effect on the rate of injurious falls (RaR 0.96, 95% CI 0.72 to 1.27). Data were determined based on admissions, some patients were admitted more than once. Cumming 2008 (3999 participants) examined an intervention in both acute and subacute wards in which a nurse and physiotherapist each worked for 25 hours per week for three months in all intervention wards. No trial interventions were delivered in the usual care arm. This trial also found no change in the rate of falls or risk of falling. The review authors consider both Barker 2016 and Cumming 2008 to be well‐conducted trials. The interventions they studied would be regarded as sound falls prevention practice including use of falls risk‐assessment tools and supervision for patients at risk but no effect on falls was observed. The multidisciplinary intervention in Haines 2004 (626 participants) took place in three subacute wards. The programme included a falls risk alert card with an information brochure, exercise, education programme, and hip protectors, in addition to usual care. In the control arm, patients received usual care but none of the interventions from the falls prevention programme; the study staff completed the risk assessment and generated recommendations but none of these recommendations were instituted. The authors reported that the difference in falls between the two groups was "most obvious after 45 days of observation", suggesting that this programme benefited people staying longer in hospital but it could also be explained by long staying frequent fallers in the control group. Healey 2004 (1654 participants) examined a risk‐factor reduction care plan for patients with a history of falls in a cluster‐randomised trial in eight acute and subacute wards. Interventions included assessment and interventions targeted at eyesight, medications, blood pressure management, mobility, urine testing, bed rail use, bed height, footwear, ward positioning, environmental causes and call bells. In the usual care arm, the care plan was not introduced and no changes to practice or environment relevant to falls prevention were made during the study. Hill 2015 conducted a stepped‐wedge cluster‐randomised controlled trial in eight hospital rehabilitation and geriatric wards (3121 participants, 3606 admissions), which tested the effect of an individualised multimedia education intervention (also tested in Haines 2011) provided to eligible patients with basic cognition, and staff, aiming to educate patients about falls prevention strategies and to motivate engagement in falls‐prevention strategies (ProFaNE categories of social environment and knowledge). Usual care included patient’s screening, assessment and implementation of individualised falls‐prevention strategies, ongoing staff training and environmental strategies. There was a reduction in the rate of falls (Analysis 21.1: RaR 0.60, 95% CI 0.42 to 0.94). There was also a reduction in the rate of injurious falls (adjusted RaR 0.65, 95% CI 0.42 to 0.88; data analysed by number of admissions rather than participants). Comparison 21 Hospitals: Multifactorial interventions vs usual care, Outcome 1 Rate of falls. In a pre‐specified subgroup analysis, Hill 2015 reported that the rate of falls was reduced in people without significant cognitive impairment who received the educational intervention (MMSE > 23/30; adjusted RaR 0.53, 95%CI 0.36 to 0.77, P < 0.001; 1930 participants), but there was no strong evidence for an effect in the subgroup of patients who were cognitively impaired (who did not receive the patient intervention, but may have benefited from the staff training intervention component; adjusted RaR 0.65, 95% CI 0.40 to 1.05; 1676 participants). Rate of fallsPooled results from five trials (44,664 participants) of multifactorial interventions showed a borderline reduction in the rate of falls, with a reduction overall of 20%; the 95% confidence intervals indicated this estimate of effect may range as high as a reduction of 36% or result in an increase in falls rates of 1%; (Analysis 21.1: RaR random‐effects 0.80, 95% CI 0.64 to 1.01; 5 trials: I² = 52%; low‐quality evidence, downgraded one level for risk of bias and one level for imprecision; Table 7). These findings were further explored in a subgroup analysis by setting (see below). Risk of fallingPooled data from three trials (39,889 participants) of the five trials pooled for the rate of falls outcome were generally consistent with the effect estimate for the rate of falls with a reduction in the risk of falling that did not reach statistical significance (Analysis 21.2: RR random‐effects 0.82, 95% CI 0.62 to 1.09; 3 trials: I² = 0%; very low‐quality evidence; Table 7). Notably Hill 2015 reported a reduction in the risk of falling (adjusted odds ratio (OR) 0.55, 95% CI 0.38 to 0.81) in a subacute setting; however, these data were analysed by number of admissions, rather than participants, so these data were not pooled. The choice of model for the pooled analysis did not affect the estimate of effect as the statistical heterogeneity was 0%. We are uncertain of the effects of multifactorial interventions on risk of falling in hospitals (very low‐quality evidence). Comparison 21 Hospitals: Multifactorial interventions vs usual care, Outcome 2 Number of fallers. Risk of fractureTwo trials (4625 participants; Cumming 2008; Haines 2004) reported fracture data suitable for pooling. There was no strong evidence for a reduction in the number of people sustaining a fracture (Analysis 21.3: RR 0.76, 95% CI 0.14 to 4.10: I² = 0%; nine fractures; very low‐quality evidence; Table 7). Comparison 21 Hospitals: Multifactorial interventions vs usual care, Outcome 3 Number of people sustaining a fracture. In Barker 2016, there were very few fractures in an acute setting, with 11 (0.06%) people experiencing a fall‐related fracture in the intervention arm and 13 (0.07%) in the control arm. In Hill 2015, there were six fractures in the control group (three hip fractures) and four in the intervention group (not hip) in a subacute setting; these data represent number of fractures and admissions rather than patients. The data from these two studies are not pooled; however, the results are consistent with the pooled estimate showing no strong effect on the risk of fracture. We are uncertain whether multifactorial interventions reduce the risk of fracture as the quality of the evidence has been assessed as very low. Adverse eventsNo adverse events were reported in the four trials (39,763 participants; Aizen 2015; Barker 2016; Haines 2004; Hill 2015) that reported this outcome. We are uncertain of the effects of multifactorial interventions on adverse events as the quality of the evidence has been assessed as very low (Table 7). Subgroup analysis by type of care (acute, subacute or mixed settings)A post‐hoc subgroup analysis was conducted for multifactorial interventions conducted in hospitals for acute care settings, subacute settings or mixed (both subacute and acute) settings. The test for subgroup differences indicated a possible difference between the settings (types of care) for rate of falls (Analysis 22.1, P = 0.04). Pooled data indicate a reduction in the falls rate in trials conducted in the subacute setting (Analysis 22.1.3: RaR 0.67, 95% CI 0.54 to 0.83), but not in the acute (Analysis 22.1.1: RaR 1.04, 95% CI 0.79 to 1.37) or mixed settings (Analysis 22.1.2: RaR 0.88, 95% CI 0.61 to 1.27). There were no differences between subgroups for pooled data by setting for risk of falling (Analysis 22.2, test for subgroup differences P = 0.75) or risk of fracture (Analysis 22.3, test for subgroup differences P = 0.56). One additional study reporting data for the risk of falling and fracture that were not pooled was conducted in a subacute setting (Hill 2015). Comparison 22 Hospitals: Multifactorial interventions vs usual care (grouped by type of care), Outcome 1 Rate of falls. Comparison 22 Hospitals: Multifactorial interventions vs usual care (grouped by type of care), Outcome 2 Number of fallers. Comparison 22 Hospitals: Multifactorial interventions vs usual care (grouped by type of care), Outcome 3 Number of people sustaining a fracture. Multifactorial interventions including targeted patient education may reduce the rate of falls in a subacute setting (low‐quality evidence, downgraded one level for risk of bias and one level for inconsistency due to some uncertainty in the subgroup analysis). Studies in participants with cognitive impairmentEleven trials reported findings specifically for patients with dementia or cognitive impairment. Care facilitiesIn care facilities, Juola 2015 (227 participants) included 93% of participants with a dementia diagnosis in a trial of nurse education on harmful medications. The intervention showed a reduction in the rate of falls in those with an MMSE score of 10 or greater, but no strong evidence of an effect in those with an MMSE of less than 10. In a trial of a multifactorial intervention (Whitney 2017; 191 participants), 97% of participants were cognitively impaired but the intervention did not show any strong evidence for an effect on the rate of falls or risk of falling.The effects of combination exercise, a multimodal exercise programme, a behaviour advisory service for people with confusion, dementia care mapping, and multisensory stimulation in a Snoezelen room have been examined in people with dementia in several studies (Chenoweth 2009; Klages 2011; Kovacs 2013; Mador 2004; Toulotte 2003; Van de Ven 2014). However, these interventions were tested in single small studies or the studies did not report data suitable for further analysis.Chenoweth 2009 and Buettner 2002 reported costs associated with interventions for participants with dementia in care facilities. HospitalsIn hospitals, a knowledge‐based intervention that did not show strong evidence for a reduction in the rate of falls overall showed a reduction in falls in those who were cognitively intact, but not in those with cognitive impairment in a post‐hoc analysis (Haines 2011). When the intervention was applied as a multifactorial intervention, only delivered to those with basic cognition, a reduction in both the rate of falls and risk of falling was observed (Hill 2015). In an acute hospital setting, Stenvall 2007 found that a multifactorial intervention including comprehensive geriatric assessment and rehabilitation for people with femoral neck fractures reduced falls in a subgroup with dementia, however the number of participants was low and the evidence assessed as very low quality, so we are uncertain of the effectiveness of this intervention. Economic evaluationsThe 11 studies reporting economic outcomes (nine in care facilities and two in a hospital setting) are summarised in Appendix 10. Only one study (Haines 2013), reported an economic evaluation in terms of the cost to prevent falls. In a subgroup of hospital inpatients who were cognitively intact, a falls patient education programme in a hospital setting had a cost of AUD 294 to prevent one fall and AUD 526 to prevent one faller (Haines 2013). This review now includes 95 trials (138,164 participants) of which 71 trials (40,374 participants; mean age 84 years; 75% women) were in care facilities and 24 trials (97,790 participants; mean age 78 years; 52% women) were in hospitals. Despite the addition of 35 trials (77,869 participants) to the previous review, many of the results from the pooled analyses remain inconsistent and inconclusive. Although 24 trials reported data on fractures suitable for use in the analyses, all fracture data were very low‐quality evidence and thus we are uncertain of the effects of any intervention on risk of fracture. Twenty‐nine trials clearly reported data on adverse events, although in several it was to report an absence of adverse events. There were very few serious adverse events and minor complications, where reported, were usually similar in the intervention and control groups. Overall, we are uncertain of the effects on adverse events as the quality of the evidence has been assessed as very low. Care facilitiesExerciseTwenty‐five trials in care facilities investigated exercise as a single intervention. Despite the large number of trials, many were small (< 100 participants). Only two trials reported the effects of exercise on risk of fracture and nine on adverse events. Seventeen trials compared an exercise intervention with usual care. A summary of the evidence for exercise in comparison with usual care in care facilities is provided in Table 1. Funnel plots of the pooled trials (10 trials each for rate of falls and risk of falling; plus positive findings in an additional four trials reporting rate of falls that could not be pooled) indicated potential publication bias for this comparison. In the 10 trials of exercise compared with usual care that were pooled reporting rate of falls, there was considerable heterogeneity in the results, which was only partially explained by a subgroup analysis grouping trials according to level of nursing care provided. We are uncertain whether exercise had an effect on the rate of falls in care facilities as the quality of the evidence has been assessed as very low. Subgroup analyses by type of exercise did not explain the heterogeneity. There was less statistical heterogeneity in the data on risk of falling for trials of exercise compared with usual care. Pooled data indicated exercise may make little or no difference to the risk of falling (low‐quality evidence). There was limited evidence for exercise types other than gait, balance and functional training or trials testing a combination of exercise categories in comparison with usual care. Whilst three trials tested Tai Chi programmes (which have been demonstrated to be effective at reducing the risk of falling in a community setting), data were not suitable for pooling. We are uncertain of the impact of exercise on the risk of fracture or adverse events (very low‐quality evidence). Nine trials provided 12 comparisons of two different exercise programmes. Comparisons of different types of exercise were all considered of very low quality so we are uncertain of the relative effectiveness of different types of exercise. While no clear effect on reduction in falls from exercise was identified within the current review, either overall or by subgroups according to level of care or type of exercise, there was a high degree of heterogeneity between the studies. The range of different types of exercise, populations and settings investigated plus the small size of many trials has resulted in only limited evidence being available for any particular combination of these factors. Importantly, the limited evidence does not represent convincing evidence of a lack of effect and the possibility of some types, intensity or duration of exercise being effective for specific populations remains. Medication (drug target)Medication reviewTwelve studies examined medication review in care facilities. One study reported on the risk of fracture. Two studies reported instances of adverse events. A summary of the evidence for general medication review in care facilities is provided in Table 2. Pooled results from five trials of general medication review indicated that this intervention may make little or no difference to the rate of falls or risk of falling (low‐quality evidence). We are uncertain of the effect of general medication review on risk of fracture or adverse events as the quality of the evidence has been assessed as very low. Vitamin D supplementationEight trials examined vitamin D interventions in care facilities. Five trials examined the effect of vitamin D supplementation, two trials investigated the effect of daily multivitamin supplementation which included vitamin D and calcium and one tested an education intervention aimed at increasing prescription of adequate levels of vitamin D, calcium and osteoporosis medications. Only three trials reported data on the risk of fracture and five on adverse events. A summary of the evidence for vitamin D supplementation in care facilities is provided in Table 3. Vitamin D supplementation probably reduces the rate of falls (moderate‐quality evidence) but vitamin D supplementation (with or without calcium) probably makes little or no difference to the risk of falling (moderate‐quality evidence). The 28% reduction in falls rate observed (RaR 0.72, 95% CI 0.55 to 0.95) is substantial. Average serum vitamin D levels at baseline were reported to be low or very low in seven of eight studies (including the five studies of vitamin D with or without calcium supplementation), indicating that these results are applicable to residents of care facilities with low vitamin D levels. Based on other studies, the reduction in the rate of falls may be related to improvement in muscle function (De Spiegeleer 2018). We are uncertain of the effect of vitamin D supplementation (up to 1000 IU daily) on the risk of fall‐related fractures or adverse events as the quality of the evidence has been assessed as very low. These studies represent only a subset of the studies evaluating the effect of vitamin D on fractures. We are uncertain whether multivitamin supplementation including vitamin D and calcium reduces the rate or risk of falling based on two studies as the quality of the evidence is very low. One study of an education intervention aimed at increasing the prescription of vitamin D, calcium and osteoporosis medication may make little or no difference to the rate of falls or risk of falling (low‐quality evidence). Environment/assistive technologyThere were no large trials of this type in care facilities. We are uncertain of the effect on rate of falls of wireless position monitoring in care facilities (very low‐quality evidence). Social environmentSeven trials in care facilities targeted staff training or implemented service model changes. Two studies reported data on the risk of fracture and no studies reported adverse‐event data. None of the interventions showed strong evidence for a reduction in falls. These interventions included staff education on fall and fracture prevention, a project nurse facilitating best‐practice falls injury prevention strategies, guideline implementation (falls, urinary tract infection, and pressure ulcers), dementia care mapping, a risk‐assessment tool versus nurses' judgement and a programme to improve staff connections, communication, and problem‐solving. Results were inconsistent in two trials of dementia care mapping. Use of a falls risk‐assessment tool in comparison with nurses' judgement alone probably makes little or no difference to the rate of falls or risk of falling (moderate‐quality evidence). We are uncertain of the effect on falls of a half‐day education programme about fall and fracture prevention for staff (very low‐quality evidence). We are uncertain of the impact of the other social environment interventions on falls. Knowledge/educationThere were no trials of knowledge interventions in care facilities. Psychological interventionsTwo studies in care facilities evaluated the effect of psychological interventions on falls. Neither trial reported data on the risk of fracture and adverse‐event data were not reported. One trial examined a cognitive‐behavioural intervention with a focus on falls‐risk reduction, the other examined a computer‐based cognitive training programme focused on improving attention combined with strength and balance training, compared with strength and balance training alone. We are uncertain of the effects of psychological interventions on rate of falls or risk of falling as the quality of the evidence is very low. Other single interventionsThree trials (564 participants) examined other single interventions. We are uncertain whether lavender olfactory stimulation, multisensory stimulation in a Snoezelen room or sunlight exposure reduces falls as the quality of the evidence has been assessed as very low. Multiple interventionsAn intervention for incontinent residents in high‐level nursing care facilities that included exercise, offering regular fluids and toileting, showed no strong evidence for an effect and we are uncertain of the effectiveness as the quality of the evidence is very low (Schnelle 2003). Increased sunlight exposure plus calcium supplementation had low adherence to sunlight exposure; we are uncertain of the effects on falls or adverse events as the quality of the evidence is very low (Sambrook 2012). There was no difference in the incidence rates of new skin cancers, but an increase in the adjusted all‐cause mortality in the calcium‐treated group compared with the UV alone group (hazard ratio (HR) 1.23 versus 0.76, P = 0.03). Despite documented concerns about increased risk of cardiovascular events, in particular myocardial infarction, with calcium supplementation (Bolland 2010), there was a lack of evidence for a strong effect on increased death rates from myocardial infarction, so the biological reason for the observed increase in all‐cause mortality is uncertain. We are uncertain of the effects on adverse events as the quality of evidence is very low. Multifactorial interventionsIn multifactorial interventions, two or more categories of intervention are given, and these are linked to each individual's risk profile. An initial assessment is usually carried out by one or more health professionals and an intervention is then provided or recommendations given or referrals made for further action. All trials compared a multifactorial intervention with 'usual care', which in many cases included some falls‐prevention activities. These standard care practices may have changed over time; however, the degree to which the comparator arm does or does not include components of the intervention activities is not clear enough to base any additional analysis on. A summary of the evidence for multifactorial interventions in comparison with usual care in care facilities is provided in Table 4. This review included 13 multifactorial trials in care facilities. Five studies reported data on risk of fractures. One study reported an instance of a fall as an adverse event, two studies reported that there were no adverse events, and the remaining studies did not report on adverse events. The interpretation of pooled data from multifactorial interventions is problematic because of variation in components between trials, and variation of combinations of components delivered to individuals in the trials. Pooled results did not show strong evidence for a reduction in the risk of falling or risk of fracture; however, there was considerable statistical heterogeneity. Multifactorial interventions may make little or no difference to the risk of falling in care facilities (low‐quality evidence). We are uncertain of the effects of multifactorial interventions in care facilities on the rate of falls or risk of fractures as the quality of evidence has been assessed as very low. A post‐hoc subgroup analysis based on high, intermediate or mixed levels of nursing care showed a statistical difference between subgroups, with a reduction in falls in high‐ and intermediate‐level care facilities, but not in studies or facilities with a mixed level of care. As there is no clear external evidence that could explain these subgroup results, and the finding is not completely consistent across studies, the finding is not considered credible (Guyatt 2011a), and no conclusion based on these subgroups is made. Subgroup analysis by level of cognition did not explain the heterogeneity. HospitalsExerciseThree trials in hospitals (244 participants) investigated exercise as a single intervention. Two of these were small, including less than 60 participants. Only one trial reported on adverse events. The three trials tested the effect of additional physiotherapy in rehabilitation wards (Table 5); however, we are uncertain of the effect of this intervention on rate of falls or whether it reduces risk of falling as the quality of the evidence has been assessed as very low. There were no data available on fractures and the one study reporting on adverse events found none. Medication (drug target)Medication reviewIn hospitals, we are uncertain of the effects of medication review on either rate of falls or risk of falling; this was tested in only one trial (very low‐quality evidence). Vitamin D supplementationOne trial in an acute geriatric unit found no strong evidence of an effect of vitamin D supplementation on risk of falling, despite the low levels of vitamin D at baseline. The median length of stay was only 30 days. We are uncertain of the effects of vitamin D in hospitals on rate of falls or risk of falling, rate of fracture or adverse events as the quality of the evidence has been assessed as very low. Environment/assistive technologySix trials in hospitals investigated environment/assistive technology interventions. Pooled data from two trials (28,649 participants) were available on the use of bed alarms in hospitals (Table 6). The larger trial, which was a cluster‐randomised trial with 28,551 participants, of bed/chair alarms was an education intervention to support judgement on their use. We are uncertain of the effects of bed alarms on the rate of falls, risk of falling or adverse events as the quality of the evidence has been assessed as very low. We are uncertain whether carpet flooring, tested in one small trial, increases the rate of falls and risk of falling compared with vinyl flooring (very low‐quality evidence). We are uncertain of the effects on rate of falls or risk of falling of using identification bracelets for patients at high risk. A large trial of the use of one low‐low bed per 12 existing beds in hospitals had no effect on rate of falls. However, large confidence intervals indicate a lack of precision in the estimate and we are uncertain of the effect of providing low‐low beds on the rate of falls (very low‐quality evidence). Social environmentSix trials in hospitals targeted staff training or implemented service model changes. One trial in a hospital setting reported data on the risk of fracture. No studies reported adverse‐event data. Trials tested a comprehensive post‐operative ortho‐geriatric service in a geriatric ward for patients with proximal femoral fracture surgery compared with usual care in an orthopaedic ward, guideline implementation, fall‐prevention toolkit software, a new acute care service for elderly patients, and a new behavioural advisory service for people with confusion. We are uncertain of the effects of these interventions on falls as the quality of the evidence has been assessed as very low. Knowledge/educationTwo trials examined knowledge interventions in hospitals. Neither trial reported data on the risk of fracture and one study reported that there were no adverse events. We are uncertain of the effects of an educational session based on identified risk factors and usual fall‐prevention care in acute medical wards as the quality of the evidence was assessed as very low. In a mixture of acute and subacute wards, a trial providing patients with educational materials alone and educational materials with professional follow‐up did not show strong evidence for a reduction in the rate of falls (Haines 2011). Providing patients with educational materials alone may make little or no difference to the rate of falls or risk of falling (low‐quality evidence). In a post‐hoc subgroup analysis, educational materials with professional follow‐up showed a reduction in falls in participants with no cognitive impairment in comparison with usual care. There is moderate credibility for this post‐hoc subgroup analysis (Guyatt 2011a); however, we are uncertain of the effectiveness of this intervention in reducing the rate of falls as the quality of the evidence has been assessed as very low. Due to the contrast between the effectiveness of providing this intervention as a single intervention and its effectiveness when provided as a multifactorial intervention targeted at cognitively intact participants (Hill 2015; which further supports the credibility of the result found in the subgroup analysis within Haines 2011), no conclusion on the effectiveness of this intervention when delivered as a single intervention is made as this is likely to result in difficulty in interpretation. Psychological interventionsThere were no trials of psychological interventions in hospitals. Other single interventionsThere were no trials of other single interventions in hospitals. Multiple interventionsThere were no trials of multiple interventions in hospitals. Multifactorial interventionsIn multifactorial interventions, two or more categories of intervention are given, and these are linked to each individual's risk profile. An initial assessment is usually carried out by one or more health professionals and an intervention is then provided or recommendations given or referrals made for further action. All trials included a comparison with 'usual care' that in many cases included some falls prevention activities. These standard care practices may have changed over time; however, the degree to which the comparator arm does or does not include components of the intervention activities was not clear enough to explore this. This review included six multifactorial trials in hospitals. Five trials provided data suitable for pooling for the rate of falls, three for the risk of falling. Two studies reported data on risk of fractures. Four studies reported adverse‐event data, there were no adverse events. The evidence for multifactorial interventions in hospitals is summarised in Table 7. Pooled results showed a borderline reduction in the rate of falls with a point estimate of a reduction of 20%; the 95% confidence intervals indicated this estimate of effect may range as high as a reduction of 36% or result in an increase in falls rates of 1% (Analysis 21.1: RaR random effects 0.80, 95% CI 0.64 to 1.01; 5 trials: I² = 52%); however, there was moderate heterogeneity. The interpretation of pooled data from multifactorial interventions is problematic because of variation in components between trials, and variation of combinations of components delivered to individuals in the trials. A subgroup analysis based on the setting demonstrated a likely significant difference between subgroups. Pooled data from two trials in a subacute setting showed that multifactorial interventions, both included targeted patient education, may reduce the rate of falls (RaR 0.67, 95%CI 0.54 to 0.83; low‐quality evidence). Pooled results on the risk of falling included only three of the five trials that were pooled for the rates of falls, but the overall effect estimate was generally consistent with the rate of falls, giving a point estimate of a 18% reduction in the risk of falling, with wider 95% confidence intervals indicating this may range between a 38% reduction and a 9% increase (Analysis 21.2: RR random‐effects 0.82, 95% CI 0.62 to 1.09; 3 trials: I² = 0%). This did not achieve statistical significance, but one of the additional trials that was not pooled also reported a reduction in the risk of falling based on admissions in a subacute setting (Hill 2015; 3121 participants). No difference between subgroups by setting was observed. We are uncertain of the effects on risk of falling as the quality of the evidence was assessed as very low. We are uncertain of the effect of multifactorial interventions on the risk of fracture or adverse events as the quality of the evidence has been assessed as very low. Subgroup analyses by level of care partly explained the heterogeneity, but due to variations in study design there is some uncertainty if findings are due to the setting or other factors, including the specific combination of interventions provided. Multifactorial interventions that include targeted patient education may reduce the rate of falls in a subacute setting (low‐quality evidence). A cost‐effectiveness analysis from one trial of multifactorial interventions is to be published (Hill 2014 protocol for Hill 2015). Studies in participants with cognitive impairmentThere is limited evidence for interventions to reduce falls in people with cognitive impairment where these people are a clearly defined group. Although only 11 trials reported findings specifically for patients with dementia or cognitive impairment, many participants in care facilities trials, including those testing interventions that probably or may reduce falls (e.g. vitamin D supplementation), had cognitive impairment. Economic evaluationsA cost‐effectiveness analysis of a patient education programme reduced falls in a subgroup of hospital patients who were cognitively intact (Haines 2011). In this subgroup the intervention, which consisted of written and video‐based materials plus one‐on‐one bedside follow‐up from a trained health professional, cost AUD 294 to prevent one fall and AUD 526 to prevent one person falling (2008 dollars; reported in Haines 2013). No conclusions can be drawn from the other 10 trials reporting economic outcomes. Although we have included 95 trials in this review, these have tested a very wide variety of interventions, sometimes with different comparators rather than control or usual care, in various types of facility. Approximately three quarters of included trials were conducted in care facilities, however many of these were small. In this review, we have reported results from care facilities and hospitals separately to improve applicability of the interventions to each setting. Careful consideration of the context of effective interventions is required. As Becker 2010 points out, the type of care provided in care facilities differs between countries and healthcare systems. Also, consideration needs to be taken of cultural and organisational contexts when generalising the results from this review. Unfortunately, the level of care and case mix in each facility in this review was often not clearly defined. In addition there is striking variability in type, targeting, intensity and duration of the falls prevention programmes that were studied. Reports of trials in hospitals are also unlikely to adequately describe the complex interaction that is likely to occur between the intervention and the usual falls‐prevention practices occurring within hospitals. Twenty‐five trials of exercise in care facilities were included, 17 of which tested exercise with usual care. However, many of these were small and whilst there were a number of trials examining balance, gait or functional training exercise programmes, there were few trials on flexibility, strength/resistance training and 3D exercise (including Tai Chi). There were several comparisons of different exercise programs; however, there was generally only one small trial for each comparison so the data were too few to be informative. The quality of available evidence for vitamin D supplementation was reasonable (moderate‐quality evidence). However, there were few studies of vitamin D supplementation taken in the form of a multivitamin. Trials of environmental/assistive technologies and social environment (e.g. staff training, service model changes) generally studied clinically different interventions, precluding pooling of trial results. Whilst there was a very large trial of bed alarms conducted in hospitals, this trial was of education, training and support for their use and there were no trials of bed alarms in care facilities. Medication review is generally aimed at reducing psychoactive medications. There were a number of trials of medication review in care facilities considered clinically similar enough to justify pooling. However, there was a large degree of inconsistency in the trial findings. The interpretation of the multifactorial interventions is complex because of the variation in components, duration and intensity of the intervention, and how the interventions were implemented. The study design does not allow evaluation of individual components of the interventions in either care facilities or hospitals. Only one trial specifically assessed the benefit of using a validated falls risk‐assessment tool in comparison with clinical judgement in a care facility (Meyer 2009) and none did in hospital, although this approach is widely used in both settings. Some multifactorial trials (e.g. Barker 2016) used validated falls risk‐assessment tools to determine the application of appropriate interventions, but the effects of the falls risk‐assessment tool cannot be separated from that of the interventions. This lack of evidence calls into question the wide use of these tools internationally and further trials examining the effectiveness of the tools are warranted. Few trials have incorporated interventions relating to the circumstances of falls, e.g. assistance with toileting, rather than targeting individual risk factors, as in the continuous quality improvement model used to develop a falls‐prevention programme in Lohse 2012. The comparator in many trials is 'usual care'. Frequently, what falls‐prevention activities are included as a component of usual care is not clearly reported. This hinders interpretation of how 'usual' care may change over time and any potentially useful subgroup analyses based on this. In terms of outcomes, 30 of the included trials did not report usable data for calculating rate of falls and 36 trials for risk of falling (seeAppendix 7). Many studies reporting data suitable for pooling reported data for one but not both of these outcomes. This may explain some of the inconsistency between the findings. Even fewer studies reported the impact of the interventions on fractures or adverse events. Within those studies that did report on adverse events, it was often unclear if these data were recorded systematically. Studies that reported data on fractures reported outcomes for different types of fractures (e.g. hip fractures only versus total fractures). Other studies not eligible for inclusion in this review may provide additional evidence for the impact of the interventions on fractures. In particular, whilst a larger proportion of included studies reported data on the risk of fracture following vitamin D supplementation, it is important to consider that these trials represent only a subset of the studies evaluating the effect of vitamin D on fractures available. In addition, some trials of interventions that may increase falls during the intervention period (exercise, medication review) only reported falls during the post‐intervention period. Other studies report only a subset of falls (e.g. bedside falls in Sahota 2014), and therefore do not meet the inclusion criteria for this review. Many cluster‐randomised trials did not adjust for clustering, therefore this was performed post‐hoc by the review authors (as indicated by a "c" in Appendix 7, for details see Unit of analysis issues). Vitamin D supplementation in care facilities reduced the rate of falls but not the risk of falling. This discrepancy might be explained by differential effects on multiple fallers (i.e. those falling more than once over the study period). However, too few of these trials reported data on multiple fallers to enable meaningful analysis of this outcome. Only Haines 2011 included a cost‐effectiveness evaluation of their hospital patient education programme in terms of falls prevented to inform the value for money for the intervention tested. An economic evaluation of the intervention tested in Hill 2015 is still to be published. Many of the interventions studied would be difficult to sustain in usual clinical practice due to competing factors in the clinical environment. In aged‐care settings, vitamin D supplementation is relatively cheap, and once it commences as part of a person’s regular medication regimen it can be continued indefinitely. In hospital settings, educating staff and patients regarding falls prevention would be regarded as good clinical practice and is sustainable in the long term provided the necessary resources are available. There is scope for realigning clinical practice with less emphasis on use of scales to assess falls risk (because there is no convincing research evidence of their effectiveness) and encouraging clinical staff to focus on factors that may be more effective, for example educating patients and families about falls and how to avoid them. This review containing 95 trials (138,164 participants) does not provide robust evidence regarding effective interventions for reducing falls in the settings considered. We assessed the quality of the evidence using the GRADE approach which considers the risk of bias, inconsistency, indirectness, imprecision and other biases (including publication bias) for the evidence for each outcome of the main comparisons. The GRADE assessments are reported in Table 1 to Table 7 and the findings are cross‐referenced in the relevant results sections. The GRADE quality of evidence for many outcomes was low or very low. This largely reflects the risk of bias in the individual studies and also the significant heterogeneity and imprecision in many of the pooled study estimates. Despite the addition of 35 trials in this update, this has generally not improved the robustness of the results compared with the previous version of this review (Cameron 2012). Although there are now a number of trials conducted for some interventions types (e.g. exercise, medication review and vitamin D supplementation in care facilities and multifactorial interventions in hospitals), the overall quality of the evidence was low to very low for all outcomes and comparisons except for rate and risk of falling for vitamin D supplementation, and use of a falls risk‐assessment tool, all in care facilities. There was also evidence indicating potential publication bias in trials of exercise conducted in care facilities. Studies in this review varied widely in their risk of bias (seeTable 11). The majority of included studies all contained some risk of bias. The included studies illustrated the wider problems of variation in the methods of ascertaining, recording, analysing, and reporting falls described in Hauer 2006. Many trials have used a single approach for ascertaining the number of falls, the limitations of this have been demonstrated in a study of falls data derived from a large hospital based randomised controlled trial (Hill 2010). For some aspects of study design, minimisation of bias is difficult. For example, it is not possible to blind participants and treatment providers for exercise, bed alarms and other types of interventions. Falls were generally recorded by nursing or care home staff who were frequently not blinded to the intervention. In addition, not all studies met the contemporary standards of the extended CONSORT statement (Schulz 2010), including the extensions for cluster‐randomised trials (Campbell 2004), non‐pharmacological trials (Boutron 2008), and pragmatic randomised trials (Zwarenstein 2008), so reporting was unclear in many instances, particularly for allocation concealment or selective outcome reporting when no protocol could be identified. There is a potential for differences between individually‐ and cluster‐randomised trials. This review included a large proportion of cluster‐randomised trials (44%). Within this review, in general trials were more likely to be cluster randomised or not depending on the intervention being investigated and the setting. Thus, whilst five of six trials of multifactorial interventions in hospitals (enrolling 99% of participants), and 85% of those conducted in care facilities (82% of participants) were cluster randomised, in contrast for trials of exercise in care facilities, 88% of trials with 65% of participants were individually randomised. Similarly, for trials of vitamin D supplementation in care facilities, 75% of trials (with 60% of participants) were individually randomised. Although it has been reported that contamination, or 'herd effects' in individually‐randomised trials conducted in facilities may result in decreasing the estimate of effect (Hahn 2005), this is considered unlikely to have had a major impact on the estimates of effect or conclusions for this review. The reasons for this according to the major categories of intervention are described below. For trials of exercise in care facilities, the estimates of effect of the three cluster‐randomised trials that contributed to pooling (Kerse 2008; Rosendahl 2008; Yokoi 2015), did not appear to differ to the range of estimates for the individually‐randomised trials. For vitamin D in care facilities, as the single cluster‐randomised trial contributing to the pooled result (Law 2006) had a smaller estimate of effect compared to the individually‐randomised trials, this indicates that contamination of the control group was unlikely to have played a role in the estimate of effect, which increases the confidence in the effect estimate. For medication review in care facilities, there was a more even balance of individually‐ and cluster‐randomised trials; 58% of trials (62% of participants) were individually randomised. The estimates of effect from the trials were inconsistent within both the cluster‐ and individually‐randomised trials, thus the high inconsistency of findings between trials for this intervention cannot be explained by the type of randomisation used. Two cluster‐randomised trials contributed only 18% of the participants for the evidence for multifactorial interventions in care facilities, the estimates of effect in these two trials were similar to that for the pooled overall effect estimates. All trials of additional exercise in care facilities were individually randomised. In trials of bed exit alarms in hospitals, only two trials contributed to pooled data; 96% of participants were enrolled in one trial that was cluster randomised, thus consideration of the findings of trials that were individually in comparison with cluster randomised is uninformative. Similarly, comparisons of individually‐ and cluster‐randomised trials within multifactorial interventions in hospitals are not feasible given 99% of participants were enrolled in cluster‐randomised trials. There was significant unexplained heterogeneity in the findings for the rate of falls for several comparisons (exercise, medication review and multifactorial in care facilities), which limited the confidence in the results (seeTable 1, Table 2 and Table 4), and was reflected in the generally low quality of evidence. The heterogeneity may be due to variations in intervention components, duration, intensity and settings as well as variations in the populations. The evidence for some ProFaNE categories of interventions contained a degree of indirectness, where the intervention was a recommendation for, or education on, use of the intervention, rather than implementing the intervention for all participants (e.g. Kennedy 2015 for vitamin D, Shorr 2012 for bed alarms). In addition, where evidence was from a single trial or setting, it was likely to be considered to have a degree of limited applicability, or indirectness to other settings, (e.g. Sambrook 2012 which examined sunlight exposure in Australia). There was also imprecision in some estimates, where the number and size of trials was small (seeTable 5) or in particular for the risk of fracture where few trials reported this outcome and events were infrequent (e.g. vitamin D Table 3). There was some evidence for likely publication bias for trials in exercise, where the included studies appeared to include a disproportionate number of small studies with positive findings (seeFigure 4, Figure 5) . We attempted to minimise publication bias in the review by searching multiple databases, and drew on the handsearch results published in the Cochrane Library in the Cochrane Central Register of Controlled Trials (CENTRAL). We also contacted authors of studies identified in trials registers that were completed, but for which full reports had not been identified, studies where only conference abstracts were identified, and many studies where it was unclear whether or not they met the inclusion criteria. We placed no foreign language restrictions in our search strategy; two studies were published in languages other than English (Peyro Saint Paul 2013; Salvà 2016), correspondence with authors provided information on study methods and results. However, despite these efforts, evidence of likely publication bias in trials of exercise conducted in care facilities remained. Although the majority of screening of search citations for potentially eligible studies in this update was performed by only one author, we suggest this was not a source of bias given that the screening was over‐inclusive with the onus being given to obtaining full‐text reports for all potentially eligible studies. We observe also that where screening was undertaken by two review authors, the progression to full‐text review was reduced. Five newly published studies that were identified in the top‐up search in August 2017 await classification (Dever 2016; Hewitt 2014; Raymond 2017; Van der Linden 2017; Wylie 2017). This was a pragmatic decision taken in view of the delay that would have resulted from their likely inclusion and after consideration of the potential impact of these trials on review findings. We concluded that our decision to postpone the inclusion of these five trials was not an important source of bias. Whilst we strictly applied a priori inclusion and exclusion criteria to the selection of studies for this review, which should minimise bias, this does result in the inclusion of a subset of the available evidence and this applies in particular to risk of fracture outcome. All included studies were required to present data on the overall rate of falls or risk of falling, those reporting only a subset of falls (e.g. injurious falls, bedside falls) were excluded. We also excluded 22 trials reporting falls as adverse effects, although in some instances the intervention might plausibly have reduced falls. For a more comprehensive systematic review of the effect of vitamin D supplementation on fractures, see Avenell 2014. For single‐trial comparisons, we took a different approach to GRADE assessment where a single rater checked whether the trial findings for each outcome met pre‐specified criteria for downgrading the evidence. The criteria were established before this alternative assessment took place. For 26 single‐trial comparisons these criteria were met. For 18 comparisons in 16 trials these criteria did not apply, generally because of a large trial size, and GRADE assessment was conducted in duplicate. For these assessments, in two trials (three outcomes), the quality of the evidence was considered moderate (Chapuy 2002; Meyer 2009), and in three trials (five outcomes) the quality of the evidence was considered low (Cox 2008; Haines 2011; Kennedy 2015); for all other comparisons and outcomes the quality of the evidence was considered very low. There are potential biases within the data included in the review in terms of non‐normal distribution of falls rates in the included studies (as seen in Potter 2016), missing data including the loss of clusters within some trials, selective outcome reporting (seeTable 11), decisions regarding pooling of studies where there is high heterogeneity and selection of models used for meta‐analyses where there is heterogeneity for one falls outcome, but not another (e.g. high heterogeneity for rate of falls but not risk of falling). The potential biases due to these factors are captured by the GRADE assessments of the overall quality of evidence (Table 1 to Table 7). There are also potential biases in decisions to conduct post‐hoc subgroup and sensitivity analyses (e.g. Analysis 5.4; seeSubgroup analysis and investigation of heterogeneity and Sensitivity analysis). This has been taken into account in conducting GRADE assessments (e.g. confidence in the credibility of subgroup analysis is considered in the inconsistency rating for the subgroup analysis by setting for multifactorial interventions in hospitals), making cautious interpretations of the findings (e.g. considering findings based on subgroup analysis by setting for multifactorial interventions in care facilities of low credibility) and transparently reporting these analyses under Differences between protocol and review. We explored the possibility of publication bias by constructing funnel plots of trials of exercise in care facilities and multifactorial interventions in care facilities (Figure 4, Figure 5, Figure 6). There was some asymmetry in the falls outcomes for trials of exercise in care facilities indicating potential publication bias. Using the generic inverse variance method in this review enabled us to pool results as reported by trial authors with our own calculated from raw data, and results adjusted for clustering. The ProFaNE falls prevention taxonomy enabled us to pool similar interventions in the analyses using a systematic approach. However, classification of some interventions according to this taxonomy was unclear and required judgement in some cases. We consulted with the ProFaNE authors when necessary. We searched for other systematic reviews of falls prevention initiatives in care facilities and hospitals published since 2012 within our search described in Appendix 1. We compared our review results with the Cochrane Review 'Interventions for preventing falls in older people living in the community' (Gillespie 2012), and identified six other systematic reviews incorporating meta‐analyses (Chan 2015; Le Blanc 2015; Sherrington 2017; Silva 2013; Stubbs 2015; Vlaeyen 2015). Comparison with trials in community‐living older peopleIn contrast to the findings in this review for residents of care facilities and hospital inpatients, the evidence is clear that falls can be prevented using exercise in older people living in the community (Gillespie 2012). The effectiveness of group‐based and home‐based exercise programmes and Tai Chi in particular is well established in the community setting. There is the potential for falls to be reduced in care facilities using the same multiple‐component exercise programmes, but despite 25 trials in this review testing exercise programmes in care facilities, the results were inconsistent. Only three trials examined exercises in hospitals; the quality of the evidence was considered very low. Vitamin D supplementation may reduce falls in community‐living people with lower vitamin D levels (Gillespie 2012). This is consistent with the finding in this review that vitamin D is effective in reducing falls in care facilities as most residents have low vitamin D levels (Pilz 2012). The effects of multifactorial approaches are inconsistent between trials and settings. In the community setting, multifactorial interventions, including falls‐risk assessment, reduced the rate of falls but not the risk of falling (Gillespie 2012). Similarly, multifactorial interventions overall may make little or no difference to the risk of falling in care facilities. However, findings on the rate of falls were inconsistent. In hospitals, multifactorial interventions (that include targeted patient education) may reduce the rate of falls in a subacute hospital setting. There is some evidence that falls prevention strategies in the community can be cost saving (Gillespie 2012), but there were no economic evaluations conducted within the care facilities and only one in hospital trials (Haines 2011) to provide information on value for money for effective interventions. Supplementary reviewNyman 2011 conducted a supplementary review of the 41 trials included in Cameron 2010 with specific reference to people’s recruitment, retention in the trial, and adherence to intervention components. Adherence was high for individually‐targeted and group‐based exercise (72% to 89%) and for medication interventions (68% to 88%). The authors reported that adherence was related to treatment effectiveness in three studies testing medication and multifactorial interventions in care facilities. They estimated that by 12 months, on average, only a third of care‐facility residents are likely to be adhering to falls prevention interventions. The current review was not able to comment on adherence or retention. Nyman 2011 provides an important perspective giving context to interpretation of the research. ExerciseChan 2015 conducted a systematic review of exercise interventions for older adults with cognitive impairment, only three of seven trials in a pooled analysis enrolled participants living in a care setting. Two of these studies were included in this review (Toulotte 2003 and Rosendahl 2008), but Chan 2015 included unpublished subgroup data for Rosendahl 2008, and Rolland 2007 and was excluded from this review as falls were monitored as adverse events. Sherrington 2017 conducted a systematic review and meta‐analysis of exercise interventions to prevent falls in older adults. This review included 14 RCTs (15 comparisons) of exercise interventions in care settings and found no significant effect on the rate of falls. These authors observed possible asymmetry in the funnel plot, which was not statistically significant on Egger's test. Three of the trials included in Sherrington 2017 were excluded from this review (DeSure 2013; Resnick 2002; Rolland 2007; seeCharacteristics of excluded studies). Two of the trials included in the pooled estimate in Sherrington 2017 were considered as multiple interventions under the ProFaNE classification system in this review (Huang 2016, ; seeAppendix 3). Data reported for one study were considered not suitable for pooling in this review (Toulotte 2003). All other trials were included. Silva 2013 included 12 studies of exercise in care facilities. This review pooled studies of exercise as a single intervention with studies of exercise as a component of a multifactorial intervention. The authors found a significant reduction in the risk of falling (RR 0.71, 95% CI 0.64 to 0.92, I² = 72%). There was no significant effect on the risk of fracture (RR 0.57, 95% CI 0.21 to 1.57). All of the included trials were included in our review. Lee 2017 included 21 studies of exercise in care facilities, 15 with exercise as a single intervention, six with exercise combined with one or more interventions. Data were pooled from studies comparing exercise with other interventions, usual care or placebo. In the current review, comparisons of alternate exercise programs were not pooled with trials of exercise in comparison with usual care (for details see Table 9). Three of the trials included in Lee 2017 were excluded from this review (DeSure 2013; Lord 2003b; Wolf 2003); two of these were considered to be conducted in a community setting. Data from one trial were not pooled in our review as there were zero falls in the intervention arm (Cadore 2014); this study has a weighting of 0.4% in the meta‐analysis in Lee 2017. Pooled data of trials of exercise as a single intervention in Lee 2017 found no difference in the rate of falls or risk of falling, consistent with the findings of our review. The current review found inconsistent effects for exercise in care facilities and is broadly consistent with Silva 2013 and Sherrington 2017 although pooling combinations differed. Our review contrasts with Chan 2015 as Chan 2015 pooled trials across both community and care facility settings and much of the impact observed in their meta‐analysis may have been from trials conducted in the community. Vitamin D supplementationA systematic review conducted for the US Preventative Services Task Force (Le Blanc 2015), examining trials conducted in both institutionalised or community settings, found that vitamin D significantly reduced the number of falls per person but did not significantly reduce the risk of falling, consistent with the findings in care facilities in this review. The authors reported that sensitivity analysis based on institutionalised status "resulted in similar estimates". The two included studies conducted in institutionalised settings are included in this Cochrane Review. The authors concluded that "Treatment of vitamin D deficiency in asymptomatic persons might reduce mortality risk in institutionalised elderly persons and risk for falls but not fractures." Bolland 2014 pooled outcomes from six randomised trials conducted in care facilities or hospitals and found no significant reduction in falls with vitamin D supplementation with or without calcium supplementation (RR 0.96, 95% CI 0.88 to 1.05). The authors concluded that supplementation with vitamin D does not reduce risk of falling by a 'clinically relevant' threshold of 15% or more and that future trials are unlikely to alter this conclusion. One study included as institutional in the Bolland 2014 review was excluded from this review as 51% of participants were residing in the community (Graafmans 1996); all other studies were included in this review. This Cochrane Review has analysed studies conducted in care facilities or hospitals separately and found that whilst vitamin D supplementation did not reduce the risk of falling, it did reduce the rate of falls in care facilities. Our analysis included data on the rate of falls in care facilities from the same four studies pooled for the risk of falling and whilst there was heterogeneity for the pooled rate of falls outcome (I² = 62%), it was lower than observed in Bolland 2014 when pooling studies in either setting (I² = 92%). Other recent systematic reviewsVlaeyen 2015 included 13 randomised controlled trials of fall‐prevention programmes conducted in nursing homes. The authors found no significant effect of the interventions overall on the number of falls (10 studies) or risk of falling (six studies). They reported that multifactorial interventions significantly reduced the number of falls (four studies) and the number of recurrent fallers (four studies), but not the risk of falling (four studies). They reported that staff training and education had a significant harmful effect on the number of falls (two studies). All trials were included in our review. Stubbs 2015 conducted an umbrella review of meta‐analyses in care facilities and hospitals and concluded that there was consistent evidence that multifactorial interventions reduce falls in care facilities and hospitals and reported that there was consistent evidence that exercise and vitamin D reduces falls in care facilities, based on the inclusion of nine individual meta‐analyses including Cameron 2012, Bolland 2014 and Sherrington 2011 (Sherrington 2017 is discussed above). Other meta‐analyses included in Stubbs 2015 and published since 2012 were Choi 2012, Guo 2014 and Santesso 2014. Choi 2012 pooled three studies conducted in care settings, all of which were included in this review: a vitamin D trial (Broe 2007), a multifactorial trial (Neyens 2009), and Rapp 2008, which is included as a subgroup analysis of Becker 2003 in our review. Guo 2014 conducted an 'exploratory meta‐analysis' examining fall‐prevention interventions for those with or without cognitive impairment in institutionalised and non‐institutionalised settings. Eight trials included in Guo 2014 were not considered for our review as they had been assessed as being conducted in the community setting: all eight trials were considered in Gillespie 2012, seven of which were included (Conroy 2010, Davison 2005, Haines 2009, Hendriks 2008, Latham 2003, Lightbody 2002, Lord 2005) and one of which was excluded because falls were reported as adverse events (Vogler 2009). Santesso 2014 conducted a meta‐analysis of hip protectors; as we consider hip protectors are intended to reduce fractures rather than falls, this intervention is not included in our review. Implications for practiceWe found evidence of effectiveness for some fall‐prevention interventions in care facilities and hospitals, although for many the quality of the evidence was considered low or very low. For all interventions, we are uncertain of their effects on fractures and on adverse events as the quality of the evidence for both outcomes was assessed as very low. For each setting, the summary is structured by the main categories of interventions evaluated in at least one setting in the review: exercise, medication (medication review; vitamin D supplementation); psychological interventions, environment/assistive technology, social environment, interventions to increase knowledge, other interventions, multiple interventions and multifactorial interventions. There was a lack of evidence on surgery, management of urinary incontinence, or fluid or nutrition therapy in both settings.
Implications for researchFurther research, primarily randomised controlled trials, is warranted to help inform decisions in this key area. We suggest the following guide to help discussions on future priorities.
Other aspects, including research methods, that need to be adopted in all future studies are as follows.
SummaryAccording to the Cochrane Handbook for Systematic Reviews of Interventions (1), systematic reviews summarize the results of controlled healthcare trials to provide a high level of evidence on the effectiveness of healthcare interventions. This evidence is then used to enlighten judgments about the evidence and to inform practice recommendations. While fall prevention in hospitals has been studied for several decades, much of the evidence has been inconclusive. Our team read this review with great interest hoping that it would inform practice recommendations (2). Unfortunately, this most recent fall prevention systematic review is again inconclusive and furthermore adds to the confusion related to the benefits of fall prevention interventions in hospital settings. However, the reported limitations relate to inaccurate assumptions and errors in calculations made by the systematic review authors, rather than the quality of the studies and associated evidence. For example, in relation to our team’s study, Cameron et al reported the following: "Dykes 2010 (5264 participants) tested the effect of a computer‐based fall prevention tool kit in comparison with usual care. There was no strong evidence for an effect on the rate of falls (Analysis 19.1.3: RaR 0.55, 95%CI 0.02 to 16.29) or risk of falling (Analysis 19.2.1 RR 0.91, 95% CI 0.06 to 14.21)." To calculate the confidence interval for rate of falls, Cameron et al used a formula that did not consider the effect of matching in their analyses. As noted by Imai (2009), matching greatly reduces the standard error in cluster‐randomised experiments (3). Based on our data, the RaR 0.55 has 95% CI 0.36 to 0.83 and the p‐value for that test that pRaR=1 was p =.005 (this is the same p‐value for the test that the rate difference = 0 as published in our results paper (4)). In addition, Cameron et al incorrectly reported the relative risk of falling for patients in our study as .91, rather than .61 (see Table 3 of our results paper [(34/2755)/(51/2509) = .61]). Perhaps this was a typographical error or a miscalculation but given the difference in results found by the systematic review authors compared to those reported by our team in Journal of the American Medical Association (4), Cameron and colleagues should have contacted us to clarify their results, rather than risk including inaccurate data in their systematic review and meta‐analysis. The inaccurate assumptions and miscalculations associated with our clinical trial call into question the rigor and accuracy of Cameron et al’s systematic review since the authors calculated rate ratios, risk ratios, and 95% confidence intervals for many of the included studies and therefore they may have made similar errors in other calculations that they made for other studies they evaluated. The inaccurate assumptions and calculation errors in this systematic review further perpetuate the myth that patient falls in hospitals are not preventable. It is our hope that Cochrane will refine its systematic review methodology to include a process for systematic review authors to check their assumptions with study authors to ensure that accurate results are reported that can be used to enlighten judgments and to inform practice recommendations. Cochrane reviews carry substantial weight; they should have high methodological standards. 1. Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA (editors). Cochrane Handbook for Systematic Reviews of Interventions version 6.0 (updated July 2019). Cochrane, 2019. Available from www.training.cochrane.org/handbook. 2. Cameron ID, Dyer SM, Panagoda CE, Murray GR, Hill KD, Cumming RG, Kerse N. Interventions for preventing falls in older people in care facilities and hospitals. Cochrane Database of Systematic Reviews 2018, Issue 9. Art. No.: {"type":"entrez-nucleotide","attrs":{"text":"CD005465","term_id":"30322203","term_text":"CD005465"}}CD005465. DOI: 10.1002/14651858.CD005465.pub4. 3. Imai K, King G, and Nall, C. The essential role of pair matching in cluster‐randomized experiments, with application to the Mexican universal health insurance evaluation. Statistical Science 2009;24(1):29‐53. 4. Dykes PC, Carroll DL, Hurley A, Lipsitz S, Benoit A, Chang F, et al. Fall prevention in acute care hospitals: A randomized trial. JAMA 2010;304(17):1912‐8 ReplyThank you for your feedback. We agree that Cochrane reviews hold to a high standard for evidence quality to inform formal recommendations. The authors of this review share your frustration that the evidence is, on the whole, insufficient to be certain of the effects of the various and often complex interventions used for falls prevention in the settings covered in our review. Cochrane reviews are undertaken using standard methods that need to be applied to all studies. For cluster randomised trials, these methods also involve adjustment for clustering. The approach taken in our review is detailed in the methods section under Unit of analysis issues. Of note is that these methods do not include consideration of matching of clusters in the results. Your trial, Dykes 2010, was added to the review in the previous 2012 update. We can confirm that the analysis of the trial was done in an appropriate manner that was consistent with all other cluster randomised trials within the review. As the results for number of fallers was not reported as an adjusted risk ratio for falling within the trial report, adjustment was necessary and performed. The adjustment for clustering requires rounding to whole numbers of participants for determination of the risk ratio, and thus the reported 0.91 rather than 0.61 is not an error but a consequence of rounding following adjustment for a relatively small number of clusters (8 clusters, adjusted values intervention 1/58 versus control 1/53; RR 0.91). For all versions of this review, the authors have approached trial investigators for missing data and clarification where necessary; your trial report was considered sufficiently reported not to need this action. We hope this explanation will restore your faith in our review. ContributorsFeedback from: Patricia C Dykes1,2, Jason S Adelman3, Michael Bogaisky4, Ann C Hurley1, David W Bates1,2 and Stuart R Lipsitz1,2. (1Center for Patient Safety, Research and Practice, Brigham and Women’s Hospital, Boston, MA; 2Harvard Medical School, Boston, MA; 3Columbia University Irving Medical Center/New York‐Presbyterian, New York, NY; 4Albert Einstein College of Medicine/Montefiore Medical Center, Bronx, NY)
Protocol first published: Issue 3, 2005
The authors would like to acknowledge the considerable contributions of Leslie Gillespie and Clare Robertson to earlier versions of the review, and Clare Robertson for support on data management and statistical calculations. The authors would like to thank Lindsey Elstub and Joanne Elliott for their support at the editorial base. We thank the following for their useful and constructive comments on this version of the review: Dr Joanne Elliott, Dr Helen Handoll, Prof Finbarr Martin, and Prof Cameron Swift. We thank the following for their useful and constructive comments on earlier versions of the protocol and/or review: Assoc Prof Jacqueline Close, Dr Simon Gates, Dr Helen Handoll, Prof Peter Herbison, Prof Finbarr Martin, Prof Cathie Sherrington, and Dr Janet Wale. We are grateful to Prof Sarah Lamb, Prof Clemens Becker and Dr Klaus Pfeiffer for their assistance with use of the ProFaNE taxonomy, and to Prof Peter Herbison for his advice on statistical issues in previous versions of the review. We are also grateful to Prof William Gillespie and Dr Mohit Arora for assistance in assessing the risk of bias for some previously included studies and Dr Mohit Arora for screening trial registry records in the top‐up search. We thank Geraldine Wallbank of the George Institute for Global Health, Sydney for her assistance in completing Appendix 7 for previously included studies. This project was supported by the National Institute for Health Research (NIHR) via Cochrane Infrastructure funding to the Cochrane Bone, Joint and Muscle Trauma Group. The views and opinions expressed therein are those of the review authors and do not necessarily reflect those of the Systematic Reviews Programme, the NIHR, the National Health Service (NHS) or the Department of Health. Funding for immediate open access was received from the NIHR Cochrane review gold open access scheme. Edited (no change to conclusions)
In this version of the review, we now exclusively assess risk of bias of each included study based on the recommended tool described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b). We also assessed bias in the recall of falls due to less reliable methods of ascertainment (Hannan 2010). We now use the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach to assess the quality of the body of evidence. We prepared 'Summary of findings' tables for each of the main categories of interventions, for the listed outcomes. The risk of bias has been assessed according to the Cochrane tool for assessing risk of bias, plus two items relating to method of ascertaining falls and baseline imbalance. Where the reported trial outcomes did not include falls during the intervention period, we did not pool these data with those of other trials. In addition to subgroup analyses by intervention types according to the Prevention of Falls Network Europe (ProFaNE) fall‐prevention taxonomy (Lamb 2007; Lamb 2011), we conducted sensitivity analyses of exercise trials excluding those with 20 participants or less in each arm of the trial. We also conducted a sensitivity analysis of medication review excluding one trial with three participants with more than 30 falls in the intervention arm of the trial. In the previous version of this review, subgroup analyses were conducted according to level of cognition and level of care in care facilities. In this update, we have added subgroup analysis by level of care (setting) in hospitals. We have conducted a sensitivity analysis for the rate of falls analysis for exercise versus usual care in care facilities to test the exclusion of one trial with zero falls recorded in the intervention arm of the trial. Upon further consideration, we have re‐categorised some interventions across different ProFaNE categories that fall within the social environment classification. Stenvall 2007 has been reclassified as a social environment intervention (previously multifactorial). Koh 2009 and Van Gaal 2011b remain classified within the social environment ProFaNE category but are considered as organisational service model change rather than staff training as these interventions are primarily to introduce new guidelines and staff training was secondary. Trials including only participants after stroke were excluded as a protocol for a Cochrane Review on interventions for preventing falls in people after stroke has been published (Verheyden 2010). We reported the results for care facilities and hospitals separately as the primary analyses because this is likely to be more useful to the users of this review. Interventions will be organised differently in these two types of settings and there may be different effectiveness of similar interventions between the two settings. The protocol was completed and submitted for publication prior to the general release of RevMan 5 and the supporting version of the Cochrane Handbook for Systematic Reviews of Interventions (version 5.0) in February 2008. In the protocol, we stated that we would assess methodological quality using the 11‐item tool used in Gillespie 2003. For this version of the review, we used three criteria from the Cochrane tool for assessing risk of bias: 'Random sequence generation', 'Allocation concealment', and 'Blinding of outcome assessment', and eight items from the 11‐item tool (seeAppendix 2). The items relating to allocation concealment and blinding of outcome assessors have not been used (now redundant). Also, the item relating to appropriateness of duration of clinical surveillance was not used due to very poor agreement between assessors during preparation of the first version of this review. Interventions were classified using the Prevention of Falls Network Europe (ProFaNE) fall‐prevention taxonomy (Lamb 2007; Lamb 2011). Subgroup analyses were conducted to explore heterogeneity where appropriate. ID Cameron, the guarantor for this review, conceived and designed the review and for this update contributed to assessment of retrieved studies against inclusion criteria, carried out 'Risk of bias' assessment, data extraction and assessment of GRADE quality of the evidence, assisted with categorisation of trial interventions using the ProFaNE taxonomy, and commented on drafts of the review. SM Dyer for this update co‐ordinated the review, carried out trial registry searches, screened search results and obtained papers, screened retrieved papers against inclusion criteria, carried out 'Risk of bias' assessment, data extraction and assessment of GRADE quality of the evidence, managed data and carried out statistical calculations, entered data into Review Manager, and drafted the review. CE Panagoda screened search results and obtained papers, screened retrieved papers against inclusion criteria, carried out 'Risk of bias' assessment and data extraction, and commented on drafts of the review. GR Murray carried out 'Risk of bias' assessment, data extraction and assessment of GRADE quality of the evidence, assisted with categorisation of trial interventions using the ProFaNE taxonomy, and commented on drafts of the review. KD Hill carried out 'Risk of bias' assessment and data extraction, and commented on drafts of the review. RG Cumming carried out 'Risk of bias' assessment and data extraction, and commented on drafts of the review. N Kerse carried out 'Risk of bias' assessment and data extraction, and commented on drafts of the review. See Appendix 12 for 'Contribution of authors' for the previous version of this review.
|