Population‐based interventions for preventing falls and fall‐related injuries in older people
Abstract
Objectives
This is a protocol for a Cochrane Review (intervention). The objectives are as follows:
To assess the effects of population‐based interventions for preventing falls and fall‐related injuries in older people. Population‐based interventions are defined as community‐wide initiatives to change the underlying societal, cultural, or environmental conditions of risk of falling.
Background
Description of the condition
After road traffic injuries, falls are the second leading cause of accidental death globally, with those aged 65 and over experiencing the majority of fatal falls (World Health Organization 2018). A fall can be defined as “an unexpected event in which the participant comes to rest on the ground, floor or lower level” (Lamb 2005). Falls and fall‐related injuries in older people can be attributed to multiple risk factors. Physiological risk factors include reduced muscle mass and strength, limited mobility, balance deficit and bone fragility, resulting from metabolic changes and low levels of physical activity (Kinney 2004; Uusi‐Rasi 2017). Cognitive risk factors, such as deficits in cognitive domains related to executive function (e.g. slower processing speed and reaction times), are also predictive of injurious falls (Welmer 2017). In addition, Clemson and colleagues have identified environmental factors, including poor footwear, inappropriate lighting and uneven surfaces, as contributing to falls risk (Clemson 2019).
A World Health Organization report on the magnitude of falls in older age groups estimates that approximately one third of community‐dwelling people aged 65 years and over fall annually (World Health Organization 2007). Falls are recorded more frequently in older people living in long‐term care institutions, with an estimated 30% to 50% falling each year. An UK‐based study estimated that falls in older people resulted in 20% to 30% of mild to severe injuries requiring medical attention (Scuffham 2003). Furthermore, falls account for two‐thirds of deaths due to unintentional injuries in older people (Rubenstein 2006). Injury resulting from falls can range in severity, from relatively minor bruising and abrasions to debilitating fractures and head injuries that can have long‐term consequences for health and quality of life, and can cause death (Gillespie 2012). Falls are associated with reduced physical functioning, loss of independence and fear of future falls, which can lead to reductions in physical activity and social engagement (Frieson 2018; Gillespie 2012). Reduced physical activity can in turn impact an individual’s strength and balance, increasing falls risk (Jørstad 2005), while reduced social engagement may result in depression and a reduction in quality of life (Delbaere 2010). There are also financial consequences associated with falls, as serious injury in older people is a major risk factor for subsequent hospitalisation and long‐term care (World Health Organization 2018). In the UK, for instance, falls cost the National Health Service approximately GBP 2.3 billion per year (NICE 2013). As such, falls represent a serious public health problem, particularly in the context of population ageing.
Description of the intervention
Interventions to prevent falls and fall‐related injuries usually target individuals, such as those with known modifiable risk factors or a history of falls (Gillespie 2012). Most fall prevention interventions targeting at‐risk individuals include one or more of the following components: exercise (e.g. strength, balance, general physical activity); medication (e.g. vitamin D supplementation, medication review); medical intervention (e.g. cataract surgery, management of urinary incontinence, fluid or nutrition therapy); environmental intervention (e.g. home adaptation, mobility aids); psycho‐social interventions (e.g. cognitive behavioural therapy, home care services); and educational intervention (e.g. written material, videos, lectures) (Hopewell 2018). Selective approaches to falls prevention can be more readily evaluated in the form of randomised controlled trials, and there is a large body of evidence supporting the effectiveness of these kinds of interventions in at‐risk groups (Gillespie 2012; Hopewell 2018; Sherrington 2019).
Population‐based interventions differ from selective approaches to falls prevention. The description of population‐based interventions advanced here is based on both the Prevention of Falls Network Europe (ProFaNE) taxonomy, and the definition of population‐based approaches put forward by Geoffrey Rose (Rose 1985). ProFaNE defines population‐based interventions as "approaches in which the entire population of older people are targeted” (Lamb 2007). For the purposes of this review, we expand on this definition by drawing on Rose, who developed the idea of population‐based approaches. Rose 1985 defined population‐based approaches in public health as those that prioritise changing the conditions that lead to the distribution of risk in specific populations. A key premise on which the population‐based approach is based, is that the distribution of risk in a particular population is influenced by contextual conditions. As such, interventions should focus on changing these conditions instead of targeting the risk profiles of individuals. A common misconception regarding population‐based interventions is that they are simply programmes or policies designed to impact a large number of people (Frohlich 2014). This view may not always be accurate, as the emphasis is on interventions that focus on individual factors (e.g. health knowledge), and involve changing the individual attributes of many people. Rather, population‐based interventions should attempt, through programmes and policies, to change the underlying societal, cultural or environmental conditions of risk for the whole population (e.g. smoking bans in public places or promoting exercise). Population‐based interventions can be viewed as ecological interventions, rather than interventions delivered at the level of the individual. The size or scope of the population (or community – these terms are used interchangeably for the purposes of this review) depends on the type of intervention and could involve a large catchment population within a geographic location, or entire villages, towns, cities or regions.
Population‐based intervention strategies in falls prevention might include, for example:
-
government policies surrounding vitamin D supplementation that might apply to entire states, regions or municipalities;
-
local council or local government providing general recommendations or maintenance programmes (at the population level) for hazard reduction in homes (e.g. good lighting; non‐slip surfaces) or in public places (e.g. care and maintenance of public walkways; railings on steps) for villages, towns and cities;
-
public health initiatives in which communities are offered information or access to interventions, such as strength and balance exercise, regardless of risk status and without assessment of individual risk (e.g. leaflet campaign targeting an entire city that provides general information on the importance of strength and balance training and details of accredited local training programmes); or
-
implementation of public health programmes to enable fall prevention behaviours, such as engagement in physical activity at the UK Chief Medical Officers’ recommended levels (Dept Health and Social Care 2019; McClure 2010; Skelton 2005) (e.g. all gyms in a town providing free membership for people over 60).
A central requirement for population‐based interventions is that the focus of the intervention is on the community rather than the individual. As such, the use of randomised controlled trial (RCT) designs in programme evaluation is often precluded due to difficulties with blinding and ensuring members of a control group are not exposed to intervention material (Kempton 2000). Instead, separate communities, towns, cities or regions with comparable demographic attributes, can be used as intervention and control areas in assessing programme effectiveness. Study designs using cluster randomisation ‐ such as stepped wedge, where clusters receiving no intervention (control condition) transition to the intervention condition at different time points, or multi‐community cluster randomised controlled trials ‐ can be used to evaluate the effectiveness of population‐based interventions (Hussey 2007).
How the intervention might work
Population‐based interventions can be viewed as attempts, through programmes and policies, to change the social and environmental contexts that influence health (Fuller 2012). Where practical, for an intervention to be considered for inclusion in a population‐based programme, it should have demonstrated effectiveness in a randomised controlled trial as a single measure and addressed a key risk factor for falls (Campbell 2010). For example, there is evidence to suggest that vitamin D supplementation is effective in reducing the rate of falls in people with insufficient vitamin D levels (Gillespie 2012). Similarly, there is evidence that vitamin D plus calcium reduces risk of fracture in older people (Avenell 2014). As such, government policies relating to provision of vitamin D and calcium supplements for whole regions or communities may impact the rate of falls and fracture risk. Population‐based interventions use an 'upstream' approach to reduce intrinsic risk factors for falls across the whole population before they manifest as proximal risk factors requiring clinical intervention (McClure 2010). Population‐based interventions work by reducing risk exposure in the cohorts of people within the setting under investigation. Thus, it is an approach that differs notably from interventions that target individuals experiencing a particular problem and attempt to assist them one at a time (Hawe 2012).
Selective approaches (targeting specific individuals, usually considered to be at high risk) and non‐selective, population‐based approaches should be considered complementary to each other rather than in competition. Non‐selective approaches have no means of ensuring tailored and appropriate implementation of a recommended intervention (Skelton 2005). Likewise, targeting only high risk groups may not be sufficiently effective, as risk factors for falls are widespread in the population and falls can often impact otherwise healthy older people (McClure 2010). Given the relative prevalence of interventions targeting those at high risk, there is a need for complementary population‐based programmes that can reach whole communities.
Why it is important to do this review
High‐quality evidence on the effectiveness of population‐based approaches for preventing falls and fall‐related injuries is limited. To our knowledge, no one has conducted a systematic review to determine whether population‐based approaches are effective in preventing falls. A Cochrane Review of population‐based interventions for the prevention of fall‐related injuries was published in 2005 (McClure 2005). The authors concluded there were no relevant randomised controlled trials available, but reported consistency in the reduction of fall‐related injuries across five prospective controlled community studies included in the review. An update on the effectiveness of population‐based approaches in the prevention of fall‐related injury and an expansion to include fall incidence is warranted. Furthermore, since the 2005 review, study designs, such as stepped wedge designs (a form of cluster‐crossover randomised trial where clusters transition between control and intervention conditions at different time points, the order of which is determined using a random process), have increased in popularity and can be considered to provide evidence comparable with other randomised designs (Haines 2018). We propose to conduct a review of high‐quality, population‐based controlled studies as well as other high‐quality studies using other designs as appropriate, in order to further evaluate the effectiveness of population‐based strategies in falls prevention.
Objectives
To assess the effects of population‐based interventions for preventing falls and fall‐related injuries in older people. Population‐based interventions are defined as community‐wide initiatives to change the underlying societal, cultural, or environmental conditions of risk of falling.
Methods
Criteria for considering studies for this review
Types of studies
Although we will aim to include RCTs, as observed above, RCT designs are often precluded when evaluating population‐based interventions due to difficulties with blinding and ensuring control groups are not exposed to the intervention (Kempton 2000). Given this, we will also include non‐randomised trials as well as controlled before‐after studies that evaluate the effects of population‐based interventions on incidence of falls, fall‐related injuries, or both, in entire communities or large parts of communities. Controlled before‐after studies are defined as studies “in which observations are made before and after the implementation of an intervention, both in a group that receives the intervention and in a control group that does not” (EPOC 2017). The term 'community' is defined as any arbitrary geographic location (e.g. villages, towns, cities, regions) or large catchment population within a geographic location. Where available, we will include cluster randomised controlled trials, and trials using a stepped wedge design, that similarly evaluate the effects of population‐based interventions in entire communities or large parts of communities. We will only include studies with at least two intervention sites and two control sites. We will exclude before‐after studies without a control group, interrupted time series studies, and any study using historic controls, or where the intervention selectively targets individuals deemed to be at‐risk of falling due to the presence of intrinsic risk factors other than their age.
Types of participants
We will include participants aged 60 years and over. As participants will be selected based on their living in a particular geographic location, they may comprise both those living independently in the community or those residing in institutions (e.g. residential care homes, assisted care facilities, sheltered housing, elder or retirement communities) which are considered to be part of the broader community. Participants should not be selected on the basis of specific disease, condition or risk status (e.g. intervention studies solely targeting care home residents will not be included).
Types of interventions
The review will include population‐based interventions aiming to reduce the incidence of falls, fall‐related injuries, or both, in older people.
Our primary umbrella comparison is any population‐based intervention targeting entire communities compared with no intervention or usual care (control). Control groups may include communities, towns, cities, or regions with comparable demographic attributes that received no intervention (or usual care), or a delayed intervention, thus providing a comparison group for a fixed period of time.
Our planned secondary comparisons are based on categorising the specific fall or fall‐related injury prevention strategy employed in the population‐based intervention tested in individual studies. The six broad groupings, targeting entire communities, are as follows:
-
exercise and physical activity (these types of interventions may promote evidence‐based falls prevention exercises, such as strength and balance or tai chi (Sherrington 2019), or physical activity generally (e.g. promotion of community‐based falls prevention exercise classes or free gym membership for older people);
-
medication or nutrition – these types of interventions provide a medical or nutritional intervention to the entire community (e.g. vitamin D and calcium supplementation);
-
environmental – these types of interventions involve local council or governments providing general recommendations or maintenance programmes to entire communities for falls hazard reduction in homes or public places (e.g. care and maintenance of public walkways);
-
educational – these types of interventions may inform the community of risk factors and consequences of falls, or ways to prevent falls (e.g. television, radio, social media, poster or leaflet campaigns with general information on falls prevention);
-
other initiatives targeting whole communities; and
-
multi‐component population‐based interventions that include more than one of the previous intervention types.
We will also consider further sub‐categorisation based on interventions within these categories, where the overarching population‐based intervention related to the specific intervention is likely to differ. For example in the environmental category, home hazard reduction as distinct from public health hazard reduction. We will also consider whether the specific interventions for reducing falls or fall‐related injury in individuals are evidence‐based by checking relevant Cochrane Reviews.
Types of outcome measures
Primary outcomes
Similar to Sherrington 2019, our primary outcomes will be:
1. Rate of falls (number of falls; falls per person‐year);
2. Number of fallers (number of people experiencing one or more falls);
3. Number of people who experienced one or more fall‐related injuries.
Secondary outcomes
Our secondary outcomes will be:
1. Number of people who experienced one or more fall‐related fractures;
2. Number of people who experienced one or more falls that resulted in hospital admission;
3. Number of people who experienced one or more falls that required medical attention;
4. Health related quality of life (HRQoL), measured using a validated scale. We will report on this where information is available;
5. Fall‐related mortality;
6. Fear of falling, measured using a validated scale. Fear of falling is predominantly measured by the Falls Efficacy Scale International (FES‐I) (Yardley 2005). We will report on this where information is available;
7. Number of people who experienced one or more adverse events (e.g. increased falls or fall‐related injuries, heart attack, or death). We expect that this will vary according to the type of intervention.
Other outcomes
If an economic analysis has been conducted, we will extract health economic data on cost utilisation and cost effectiveness, where available.
Search methods for identification of studies
Electronic searches
We will search the following electronic databases to the present date: the Cochrane Bone, Joint and Muscle Trauma Group Specialised Register, the Cochrane Public Health Group Specialised Register, the Cochrane Central Register of Controlled Trials (CENTRAL), the Cochrane Register of Studies Online (CRSO), MEDLINE Ovid (from 1946 onwards), EMBASE Ovid (from 1980 onwards), CINAHL (Cumulative Index to Nursing and Allied Health Literature) (from 1982 onwards) and PsycINFO (from 1967 onwards). There will be no limitations based on language or publication status.
We will search the following trials registries for ongoing and recently completed studies: the World Health Organization International Clinical Trials Registry Platform and ClinicalTrials.gov.
We have devised a draft search strategy for MEDLINE which is displayed in Appendix 1. This will be used as the basis for search strategies for the other databases listed.
Searching other resources
We will investigate the reference lists of other systematic reviews, including McClure 2005, and will also aim to identify ongoing and unpublished studies by contacting researchers in the field.
Data collection and analysis
Selection of studies
Two independent reviewers will screen titles and abstracts for relevance. Two independent reviewers will assess for eligibility the full texts of studies considered potentially relevant. Any disagreements will be resolved through discussion with a third independent reviewer. We will record reasons for excluding studies, and we will illustrate the selection process using a PRISMA flowchart.
Data extraction and management
Two independent reviewers will extract data from the included studies using a pre‐defined data extraction form. Data to be extracted are based on previous reviews and protocols of falls interventions (Clemson 2019; Sherrington 2019), adjusted for population‐based interventions, and will include the following items.
1. General information: authors; publication year; date of data extraction; study objectives.
2. Study details: design; location; setting; population size; inclusion and exclusion criteria; comparability of control and intervention groups or sites; length of follow‐up; funding source.
3. Characteristics of population: population composition by age, sex, ethnicity, residential status (e.g. living independently in the community, residential care homes, assisted care facilities, sheltered housing, elder or retirement communities), and socio‐economic status.
4. Interventions: experimental and control interventions; timing of intervention; mode of delivery; and information on uptake and adherence, if available.
5. Outcomes measured: rate of falls; number of fallers; number of people who experienced one or more fall‐related injuries; number of people who experienced one or more fall‐related fractures; number of people who experienced one or more falls that resulted in hospital admission; number of people who experienced one or more falls that required medical attention; HRQoL measured using a validated scale; fall‐related mortality; fear of falling measured using a validated scale; and number of participants experiencing one or more adverse events.
6. Other details: data on cost utilisation and cost‐effectiveness will be retrieved from studies, where available.
Any disagreements will be resolved through discussion with all reviewers. If data are insufficient for any given study, we will contact the authors for additional information.
Assessment of risk of bias in included studies
Two independent reviewers will conduct risk of bias assessments and disagreements will be resolved through discussion between all review authors. In the case of non‐randomised controlled studies, we will use the Effective Public Health and Practice Project (EPHPP) (EPHPP 2010) tool recommended by Cochrane Public Health (CPH), and adapt it for specific confounders as appropriate. EPHPP assesses the following domains: selection bias; study design; presence of confounders (group differences in the following prior to intervention: age; sex; ethnicity; residential status; socioeconomic status; and health status); blinding; validity and reliability of data collection methods; withdrawals and dropouts; intervention integrity; and data analysis. We will create the overall risk of bias assessment following EPHPP guidance and we will rate studies as strong, moderate or weak. We will conduct the assessment at the study level, with rate of falls as the outcome to be assessed.
In the case of randomised controlled trials, we will use the Cochrane’s ‘Risk of Bias’ tool (version 1.0) (Higgins 2011). Using this, we will assess the following domains: random sequence generation (selection bias); allocation concealment (selection bias); blinding of participants and personnel (performance bias); blinding of outcome assessment (detection bias: rated separately for falls, injuries and medical attendance outcomes, and self‐reported outcomes such as HRQoL); incomplete outcome data (attrition bias: rated separately for falls, injuries and medical attendance outcomes, and self‐reported outcomes such as HRQoL); and selective outcome reporting (reporting bias). Specifically, for trials using a cluster randomised design, we will consider the following as described in chapter 16 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011): risk of additional bias relating to recruitment; baseline imbalance; loss of clusters; incorrect analysis; and comparability with individually randomised trials. In the case of stepped wedge designs, we will adapt the tool to assess the following (Eldridge 2016): bias arising from the randomisation process; bias arising from identification or recruitment of individual participants within clusters; bias due to deviations from intended interventions; analytical biases; and chance imbalance. We will consider extending this tool to assess the additional risk of contamination across treatment conditions for stepped wedge designs (e.g. the intervention condition may take longer to embed in practice than planned; or there may be a delayed assessment of outcome in a sample with long exposure to the intervention condition).
Measures of treatment effect
Similarly to Clemson 2019, we will report treatment effects for rate of falls, fractures, falls requiring medical attention, and fall‐related mortality as a rate ratio (RaR) with 95% confidence intervals (CIs). For number of fallers, number of people who experienced one or more fall‐related injuries, number of people who experienced fall‐related fractures, and number of peole who experienced falls that required medical attention, we will report risk ratios (RR) and 95% CIs.
The rate of falls is measured as falls per person‐year (the total number of falls per unit of person‐time that falls were monitored). The RaR compares the rate of falls in any two sites during each study. We will use an RaR (e.g. incidence RaR or hazard ratio for all falls) with a 95% CI if these were reported. If both adjusted and unadjusted RaRs were reported, we will use the unadjusted estimate unless the adjustment was for clustering. If an RaR was not reported but appropriate raw data are available, we will calculate an RaR and 95% CI. We will use the reported rate of falls (falls per person‐year) in each site and the total number of falls for participants contributing data, or we will calculate the rate of falls in each site from the total number of falls and the actual total length of time falls were monitored (person‐years) for participants contributing data.
For number of fallers, a dichotomous outcome, we will use the RR as the treatment effect. The RR compares the number of people who fell once or more (fallers) between sites. We will use 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 are reported, we will use the unadjusted estimate, unless the adjustment is for clustering. If an odds ratio is reported, or an effect estimate and 95% CI is not reported, and appropriate data are available, we will calculate an RR and 95% CI using the number of participants contributing data at each site if this is known. We will use the same approach for the number of people sustaining fractures, the number of people experiencing falls requiring medical attention, and the number of people experiencing adverse events.
For continuous outcomes (health‐related quality of life and fear of falling), where the same measure is used, we will report the mean difference with 95% CIs. Where different measures are used, we will report the standardised mean difference with 95% CIs.
Unit of analysis issues
For trials that are cluster randomised (e.g. by community), we will perform adjustments for clustering as described in Higgins 2011 if this was not done in the published study. We will use an intraclass correlation coefficient (ICC) of 0.01 (Smeeth 2002). For studies with multiple arms, we will include multiple pair‐wise comparisons (intervention versus control) in analyses, but to avoid the same group of participants being included twice, we will 'split' the control group by distributing the number of control group participants to each analysis in proportion to the number of participants in each intervention group. We will be alert to the unit of analysis issues relating to outcome reporting at different follow‐up times and the presentation of outcomes, such as adverse events, by the number of outcomes rather than participants with these outcomes.
Dealing with missing data
Where we encounter missing data, we will attempt to contact the study authors for clarification. We will also conduct sensitivity analyses to explore the impact of missing data on the treatment effect. If a study does not report standard deviations (SDs) for continuous outcomes, we will calculate these from standard errors, CIs, or exact probability (P) values where possible. We will not impute missing SDs.
Assessment of heterogeneity
To determine whether or not to combine study results, we will assess clinical and methodological heterogeneity. If study designs are considered sufficiently homogenous, we will conduct meta‐analyses and assess statistical heterogeneity using the Chi2 test and the I2 statistic. We will base our interpretation of the I2 results on that suggested by Higgins 2011: 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 biases
If possible, we will use funnel plots to investigate the possibility of reporting biases. Where there is sufficient data (10 or more studies), we will produce funnel plots.
Data synthesis
Once we have completed a systematic literature search and obtained all studies found to comply with the inclusion criteria, we will tabulate a detailed description of each study and their important characteristics (design, population size, risk of bias, relevant outcome data). We will base decisions for pooling interventions ‐ both for the primary umbrella comparison and within the same specific intervention category (see Types of interventions) ‐ on an examination of the interventions and settings to see if these can be considered sufficiently comparable to justify pooling. We will analyse separately those studies where the intervention does not adequately fit the predefined categories. We will include studies not eligible for meta‐analysis in a narrative analysis dependent on the available information and outcome of interest. We will compare the narrative analysis results with the meta‐analysis to see if they support the latter’s conclusions. RCTs are considered to be gold standard study design and are more likely to report key information. However, they may be rare and include a smaller, very specific, study population, thus reducing their wider generalisability. We will take this, and the risk of bias scores, into account in any final decisions relating to data synthesis. As part of the descriptive statistics, we will produce a flow diagram outlining the numbers of studies from screening to analysis, to reflect the number of studies present at each stage and in each analysis performed.
For each study, we will make every effort to derive a suitable comparison effect and estimate of variance for the primary and secondary outcomes. When considered appropriate, we will pool the results of comparable studies using both fixed‐effect and random‐effects models. The choice of the model to report will be guided by careful consideration of the extent of heterogeneity and whether it can be explained, in addition to other factors such as the number and size of included studies. We will use 95% CIs throughout. We will consider not pooling data where there is considerable heterogeneity (I² ≥ 75%) that cannot be explained by the diversity of methodological features among studies. A minimum of two (ideally four or more) suitable studies will be required to pool within each intervention category. We will consider pooling of data to be inappropriate in situations where there is limited evidence; incompletely reported outcome/effect estimates, or different effect measures used across studies; and bias in the evidence (McKenzie 2020).
For studies not eligible for meta‐analysis, we will attempt to synthesise data for our outcomes as per the synthesis without meta‐analysis (SWiM) reporting guidelines (Campbell 2019). SWiM analysis will follow the same structure as the meta‐analysis. The SWiM reporting guidelines provide a nine‐item checklist that outlines the standardised metrics used for synthesis, the synthesis method, presentation of data, a summary of findings, and limitations of the synthesis. The synthesis method will involve a summary of effect estimates where variance parameters are missing, combining P values, or vote counting based on direction of effect (McKenzie 2020). For each narrative method, we will use an appropriate reporting technique, such as descriptive statistics (e.g. median), a box‐whisker plot, combined P value, albatross plot, or a harvest or effect direction plots. We will aim to present data from all substantial single trials in addition to the narrative summary. Where possible, we will stratify the results by the risk of bias in order to highlight any change in conclusions due to inclusion of studies considered to be high risk. We will describe all studies in a structured table of results that will include study design, population size, risk of bias, and reported or derived results.
Subgroup analysis and investigation of heterogeneity
Where sufficient data are available, within all outcomes as outlined, we will explore heterogeneity by carrying out subgroup analyses on participant‐related characteristics (age, gender, ethnicity, residential status, and socioeconomic status). We will use the test for subgroup differences available in Review Manager to determine whether there is evidence for a difference in treatment effect between subgroups.
Sensitivity analysis
Where possible, we will perform sensitivity analyses to explore the impact of study design, missing data, the inclusion of unpublished data, the inclusion of studies at high risk of bias, the choice of statistical model for pooling (fixed‐effect versus random‐effects), and the effect of different intraclass correlations (ICCs) for cluster RCTs. Where possible, we will perform sensitivity analysis to explore the impact of fear of falling on falls outcomes. If quantitative sensitivity analysis is deemed inappropriate, we will narratively report observed patterns in the data.
Assessing the quality of the evidence and 'Summary of findings' tables
We will use the GRADE approach to assess the quality of evidence as it relates to the primary and secondary outcomes listed in the 'Types of outcome measures' section (Schünemann 2020). The quality rating ‘high’ is reserved for a body of evidence based on randomised controlled trials. We may downgrade 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 and publication bias. All non‐randomised studies start at low certainty evidence. There are rare circumstances in which authors can upgrade evidence from non‐randomised studies to moderate, or even high certainty. These include: a large estimated effect (e.g. RR > 2 or RR < 0.5) in the absence of plausible confounders, or a very large effect (e.g. RR > 5 or RR < 0.2) in studies with no major threats to validity; the presence of a dose‐response gradient; or the presence of plausible biases that may lead to an underestimation of an apparent effect. Where evidence is sufficient, we will prepare summary of findings tables for the following outcomes: rate of falls, number of fallers, number of people who experienced one or more fall‐related injuries, number of people who experienced one or more fall‐related fractures, number of people who experienced one or more falls that resulted in hospital admission, number of people who experienced one or more falls that required medical attention, and number of people who experienced one or more adverse events.
