Search strategy and study selection

We searched the Cochrane Library, PubMed, Embase, and CINAHL to identify relevant studies published in peer reviewed journals from inception to July 2013. In the Cochrane Library the search terms were: (fall) and (exercise or tai chi or training or physical activity) excluding (Alzheimer or Parkinson’s or dementia or nursing home or protocol or athletes) in the title or abstract of trials, with the word “prevention” in the text and the word variations option enabled. We also selected references in relevant reviews for screening.^{7 8 10} Two investigators (FEK, PDM) independently screened the titles, abstracts, and full text of identified papers to determine their eligibility for inclusion (see supplementary file). Discrepancies were resolved by discussion.

Inclusion criteria were randomised controlled trials of fall prevention exercise interventions, published in English or French, targeting community dwelling adults aged over 60 years, and providing quantitative data on injurious falls, serious falls, fall related injuries, or fall induced fractures. We included studies where exercise was compared with no intervention (usual activity or usual care) or a placebo control intervention (for example, general health education classes, social visits, or a low intensity exercise programme not designed to modify the risk of falling). We excluded studies in which exercise was part of a multifactorial programme such that participants received other interventions (for example, home management, visual treatment) in addition to exercise, and when participants were selected for a specific neurodegenerative disease or any other characteristic that greatly affected the risk of falling but was not correctable by exercise (such as severe visual impairment).

Data extraction and quality assessment

We used a form designed for this review to extract data on study and intervention characteristics, quality assessment, and outcomes (see supplementary file). The taxonomy for fall prevention interventions developed by the Prevention of Falls Network Europe (ProFaNE)¹² was used to describe the characteristics of the interventions provided (for example, participants’ selection criteria, type of exercises, and intervention procedures). This tool uses internationally agreed criteria to evaluate systematically the content and format of fall prevention interventions. We also extracted the definitions and methods used to collect falls and to classify fall related injuries, as well as the number, rate, or risk ratio of injurious falls and any available data on the nature of the injuries. We contacted authors of included articles to obtain more detailed data on the outcomes of injurious falls (for example, if authors reported the number of participants with an injurious fall but not the total number of injurious falls, or data on falls resulting in fractures but not data on other injurious falls).

After reviewing the case definitions used in the selected studies, we sought to group definitions of injurious falls into more homogeneous categories to allow results to be compared across studies and the data to be pooled. The ProFaNE group recently proposed a standardised classification of injurious falls to be used in future randomised controlled trials.¹⁰ As a foundation for developing a retrospective classification of the definitions of injurious falls found in the studies selected for this review, we used the ProFaNE classification along with the standardised classification of Campbell and Robertson,¹³ which is the classification most often used in published trials of these interventions.¹⁴ We also recorded any reports of adverse effects associated with interventions.

We followed the recommendations of the Cochrane Collaboration¹⁵ to assess risk of bias in the following domains: random sequence generation (selection bias), allocation concealment (selection bias), blinding of the assessment of falls and injurious falls (detection bias), and incomplete outcome data (attrition bias). We also assessed bias in the recall of falls owing to unreliable methods of ascertainment,¹⁶ using the criteria developed for the Cochrane review of fall prevention trials.⁷ The methods used to confirm serious injurious falls were also examined: we judged self reports from participants to be at a high risk of bias, whereas we considered the use of medical records or radiography reports (for fractures) to be at low risk. Two authors (FEK, PDM) independently assessed the risks of bias and extracted data. Disagreement was resolved by consensus or adjudication by a third party.

Characteristics of included studies and interventions

The 17 studies in the meta-analysis included a total of 2195 participants randomised in the exercise groups and 2110 randomised in the control groups, with sample sizes in individual studies ranging from 53 to 486 participants. The mean age of the overall population was 76.7 years, and around 77% were women. Seven studies selected participants based on a higher risk of falls—that is, history of falling, age over 80 years, or physical limitations (as measured by simple functional tests).^{13 19 20 21 22 23 24} Fourteen trials delivered the exercise intervention in a group setting, and six of those supplemented the group sessions with home exercise. In the other three trials, the intervention consisted of individualised exercises delivered at home.^{13 22 25} Tai Chi was the exclusive exercise intervention in two of the studies^{26 27}; the rest of the interventions included a gait, balance, and functional training component. Most also included strength/resistance training exercises. One study compared two similar multiple component exercise interventions, one with added endurance training, with the control group.²⁰ In the analysis, the two exercise groups were combined to create a single pairwise comparison (intervention versus control). Table 1​1 summarises the characteristics of the included studies.

Classification of injurious falls

The definition and classification of injurious falls varied substantially and most trials did not provide a reference for their definition. Injurious falls usually included diverse consequences, ranging from relatively minor injuries such as bruises or abrasions to fractures or other serious injuries requiring hospital admission.^{19 25 26 28} Most often the definition referred to either the presence of symptoms or the use of medical care.^{13 19 25 26 28} In other cases, injurious falls meant simply any self reported physical consequence of a fall, without any details.^{20 21 24 29 30} Some definitions specifically required the use of medical care,^{21 23 25 28 31 32} by using non-specific terms such as “fall for which medical care was sought,” “falls requiring medical care/medical attention,” or “medical consultations/visits.” When serious injuries were distinguished or specifically reported, their definition was more homogeneous across studies. Such injuries usually included fractures, severe soft tissue injuries requiring suturing, or other injuries leading to hospital admission.^{22 24 25 27 31} Some studies reported only fractures.^{33 34}

Based on our review of case definitions used in the 17 studies, we distinguished four categories of injurious falls: those resulting in any reported consequences, including specific symptoms (ranging from bruises and cuts to more serious injuries such as fractures) or medical care; those resulting in medical care; those resulting in serious injuries such as fractures, head trauma, soft tissue injury requiring suturing, or any other injury requiring admission to hospital; and those resulting in fractures.

These categories represent increasingly specific subgroups of all injurious falls, which can also be considered to correspond to increasing levels of severity (except for those resulting in fracture, which is simply a specific type of serious injury).

Depending on the available data, a given study could contribute data relevant to one or more categories of injurious falls. Table 2​2 gives the definitions of injurious falls reported in each selected article (as a direct quotation), the category or categories of injurious falls in which it was classified for this review, and the rate ratio used in the corresponding analysis. For two studies, the rate ratio could not be calculated because the authors provided only the number of participants who had an injurious fall (rather than the number of such falls). In these cases, we used the ratio of the risk of at least one injurious fall in both groups instead. Of note, the outcomes of injurious fall in those studies were severe injuries²⁶ and fractures,³⁴ two outcomes that are relatively rare, so that the risk ratio was likely to be close to the rate ratio.

Methodological quality

Table 3​3 shows the results of the analysis of risk of bias. Although few studies were judged at high risk in any one domain, the quality of some studies could not be judged with any certainty in several domains. Prospective daily calendars returned monthly are the preferred method for recording falls,¹⁶ and most of the trials used this method. However, only six of the 11 trials that reported data on serious injuries used medical records to confirm the injury.

Effect of exercise on injurious falls

Figure 2​2 shows the forest plots of the exercise effect estimate by category of injurious fall. Most of the exercise interventions tended to reduce injurious falls in all categories. Ten trials provided figures for all injurious falls; the pooled estimated rate ratio was 0.63 (0.51 to 0.77, I²=50%, P=0.04). The pooled estimate for falls resulting in medical care was 0.70 (0.54 to 0.92, I²=20%, P=0.27, eight trials) and for falls resulting in serious injuries was 0.57 (0.36 to 0.90, I²=46%, P=0.09, seven trials). Exercise seemed to significantly decrease the rate of falls resulting in fractures as well, with a pooled effect of 0.39 (0.23 to 0.66, I²=0%, P=0.96, six trials).

Open in a separate window

Fig 2 Forest plots of studies for four injurious fall categories: A, all injurious falls; B, falls resulting in medical care; C, falls resulting in serious injuries; and D, falls resulting in fractures

Sensitivity and subgroup analyses

To explore the possible impact of risk of bias on the exercise effect, we removed from the analyses studies for which the risk of bias was unclear for at least three of the quality components considered,^{19 27 31 34} or for which the risk was at least unclear for one component and high for another.^{22 24 30 32}. This barely changed the pooled estimate of the effect of exercise in any of the four injurious fall categories, but it greatly reduced the heterogeneity between studies included in the analysis of all injurious falls (rate ratio 0.69, 95% confidence interval 0.58 to 0.82, I²=26%, P=0.22, eight trials).

In the subgroup analysis of the effect of exercise in reducing injurious falls based on risk of falls at enrolment, no significant difference was found in pooled estimates between trials with participants selected for their higher risk of falling^{13 19 20 21 24} and trials with unselected participants,^{25 26 28 29 30} in any of the four categories.

Publication bias

The funnel plot constructed from the 10 trials included in the analysis of all injurious falls shows a barely asymmetrical scatter (fig 3​3).).

Open in a separate window

Fig 3 Funnel plot of studies included in analysis of all injurious falls

Comparisons with other reviews

One group of researchers conducted an individual level data meta-analysis of the results of four trials conducted by their fall prevention research group to estimate the overall effect of the Otago individualised strength and balance exercise programme on falls and injuries.¹⁴ The study demonstrated a 35% reduction in the risk of an injurious fall (moderate or serious), a result similar to that presented here, but it failed to show a significant reduction in serious fall related injuries. A more recent review including three additional trials of the Otago exercise programme,⁵² failed to support the previous finding that the programme significantly reduced the risk of injurious falls, possibly because of a lack of access to individual level data or because of differences in the implementation and supervision of the programme (as suggested by the significantly lower compliance rates reported in the three additional trials than in the first four).

The recent Cochrane review of fall prevention interventions conducted a specific analysis to examine the effect of exercise interventions on the risk of fall related fractures and obtained similar results (pooled relative risk 0.34, 95% confidence interval 0.18 to 0.63).⁷ Of the six trials included in the Cochrane analysis, five are also included in our analysis. We excluded the one study because participants were recruited in the hospital while in the acute phase of an injury (hip fracture) so that the results of this trial may not be comparable to those not recruiting on the basis of an injury.⁵³ Inversely, we included one study²⁹ that compared an exercise intervention including balance, gait, and strengthening and flexibility exercises to a “wellness” exercise programme designed not to affect the risk of falling, whereas this study was not considered in the Cochrane analysis of fracture risk.

Strengths and limitations of this review

Heterogeneity was notable among studies included in the analysis of all injurious falls (I²=50%, P=0.04). Although no significant heterogeneity was detected in the analysis of severe injurious falls, the inconsistency was also moderately large (I²=46%, P=0.09). Pooled data from studies with significant heterogeneity require cautious interpretation because this indicates the possibility of meaningful differences between these studies. Although this review focused on interventions based solely on exercise and targeting specifically community dwelling elderly people, the selected interventions are still quite diverse, in particular in terms of intervention components (type of exercises, intensity, frequency, mode of delivery, and total duration) and inclusion criteria for participants. Similar or even larger measures of heterogeneity have also been reported in other meta-analyses of fall prevention exercise interventions.^{7 54}

To explore possible reasons for heterogeneity, we carried out pre-planned subgroup analyses based on fall risk at enrolment and found no difference in pooled estimates between trials with participants selected for their higher risk of falling versus lower risk (unselected), for any of the four categories of injurious falls. Among intervention components, the type of exercise seems to be a key factor that influences effectiveness against falls: interventions including a balance training component seem to be more effective.^{7 54 55} One study found that the intensity of the balance training component was also important: interventions that include moderate to high challenging exercises (that is, standing exercises in which people sought to stand with their feet closer together or on one leg, to minimise use of their hands to assist, and to practice controlled movements of the centre of mass) are more effective in reducing falls than interventions that include less challenging balance exercises.⁸ All interventions included in this review include a balance training component, hence we performed an additional subgroup analysis by comparing trials where the exercise programme provided a moderate or high challenge to balance (based on Sherrington’s definition),^{13 19 20 21 23 24 25 26 27 31 33} versus a low challenge.^{22 28 29 30 32 34} No significant difference in exercise effect was found between the two subgroups with regard to the reduction in all injurious falls or falls resulting in medical care. For severe injurious falls and falls resulting in fractures, there were not enough studies to perform subgroup analyses by intensity of balance training. The relatively small number of studies included in this review did not allow us to perform additional subgroup analyses exploring the effect of other components within interventions or other factors related to the way interventions were implemented that might have affected results.

Although we tried to reduce heterogeneity in the definitions of injurious falls as much as possible by grouping them in more homogeneous categories, remaining inconsistency in definitions may also help explain the heterogeneity observed between studies. In the category of all injurious falls, in particular, several studies used imprecise definitions of injurious falls^{20 21 24 29 30} but were nevertheless included in the analysis since this category is the largest and refers to any types of physical consequences of a fall. We performed a sensitivity analysis excluding from the analysis of this category studies that used imprecise definitions: the resulting pooled rate ratio was similar to that obtained in the main overall analysis (0.69, 0.53 to 0.90, I²=49%, P=0.10) and indicated that our results are robust to differences in outcome definitions between studies. Results of the sensitivity analysis excluding trials judged to be at higher risk of bias in all four categories of injurious falls barely changed the pooled effect estimates and indicates that our results are also robust to key risks of bias.

The funnel plot of the 10 trials contributing to the analysis of all injurious falls showed a barely asymmetrical scatter. Asymmetrical funnel plots may indicate publication bias or may be due to exaggeration of treatment effects in small studies of low quality.¹⁵ However, the funnel plot in this analysis showed no clear evidence of “small study effects.” Note that almost all the trials included in this review were designed to prove that exercise has an effect on the fall rate and not on the rate of injurious falls. Among the 17 studies considered in this review, eight did not demonstrate that exercise had a significant effect on falls. Moreover, the pooled effect of exercise on the rate ratio of falls was 0.68 (95% confidence interval 0.61 to 0.77), which is similar to the effect of exercise reported in the Cochrane review.⁷ These points suggest that the trials included in this review, because they presented data on fall related injuries, do not represent a special selection of fall prevention exercise trials biased towards “positive” trials for all falls. None the less, we cannot exclude the possibility that trials are more likely to report injurious falls when there tend to be a positive effect on injurious falls. It is also possible that some trials did not report data on injurious falls simply because these data were not collected.

Only minor adverse reactions were reported in two of the 17 included studies. However, close to half of the selected studies did not even mention intervention adverse effects. It has also been suggested that exercise fall- prevention programmes may have adverse psychological effects that may affect the quality of life (for example, through self imposed activity restriction).^{56 57} Some studies have reported the effect of the intervention on fear of falling, physical activity levels, or other dimensions of quality of life.^{19 21 24 26 27} Their results show either no effect or a tendency towards a beneficial effect of the intervention on these outcomes, in particular a reduction in fear of falling. More complete data on adverse physical reactions as well as on psychological and quality of life outcomes would improve our ability to judge the overall benefit of exercise fall prevention interventions. Another limitation of the included studies is that they often lack information on “intermediate” outcomes such as gait, balance, and other physical or cognitive functions, which would help us to understand how exercise “works” and design optimum programmes. Future trials should also provide more detailed descriptions of implementation procedures, whether planned or unplanned, so that readers can judge the applicability of the programme in different settings or on a larger scale.