Association of exercise with risk of mortality (A) and hospitalization (B). Weights are from random effects analysis.
Association of exercise with risk of becoming a faller (A) and becoming a multiple faller (B). Weights are from random effects analysis.
Association of exercise with risk of becoming an injurious faller (A) and sustaining a fracture (B). Weights are from random effects analysis.
eAppendix 1. Study Protocol
eAppendix 2. Electronic Searches
eAppendix 3. Flow Chart of Study Selection
eAppendix 4. Risk of Bias
eAppendix 5. Description of Exercise Interventions
eAppendix 6. Meta-analyses and Meta-regressions for the Outcome “Mortality”
eAppendix 7. Meta-analyses and meta-regressions for the outcome “Number of people hospitalized
eAppendix 8. Meta-analyses and meta-regressions for the outcome “Fallers”
eAppendix 9. Meta-analyses and meta-regressions for the outcome “Fallers with multiple falls”
eAppendix 10. Meta-analyses for the outcome “Injurious fallers”
eAppendix 11. Meta-analyses and meta-regressions for the outcome “People with fractures”
eAppendix 12. Number of participants and events included in the main analysis per study and per outcome
Customize your JAMA Network experience by selecting one or more topics from the list below.
de Souto Barreto P, Rolland Y, Vellas B, Maltais M. Association of Long-term Exercise Training With Risk of Falls, Fractures, Hospitalizations, and Mortality in Older Adults: A Systematic Review and Meta-analysis. JAMA Intern Med. 2019;179(3):394–405. doi:10.1001/jamainternmed.2018.5406
What is the association of long-term (≥1 year) exercise with the risk of falls, fractures, hospitalizations and death in older adults?
In this meta-analysis of 40 long-term randomized clinical trials (RCTs) of 21 868 participants, exercise significantly decreased the risk of being a faller and injurious faller but did not significantly reduce the risk of fractures. Exercise did not diminish the risk of multiple falls, hospitalization, and mortality.
Long-term exercise, particularly moderate intensity, multicomponent training with balance exercises, performed 2 to 3 times per week, appears to be a safe and effective intervention for reducing the risk of being a faller/injurious faller in older populations.
Long-term exercise benefits on prevalent adverse events in older populations, such as falls, fractures, or hospitalizations, are not yet established or known.
To systematically review and investigate the association of long-term exercise interventions (≥1 year) with the risk of falls, injurious falls, multiple falls, fractures, hospitalization, and mortality in older adults.
PubMed, Cochrane Central Register of Controlled Trials, SportDiscus, PsychInfo, and Ageline were searched through March 2018.
Exercise randomized clinical trials (RCTs) with intervention length of 1 year or longer, performed among participants 60 years or older.
Data Extraction and Synthesis
Two raters independently screened articles, abstracted the data, and assessed the risk of bias. Data were combined with risk ratios (RRs) using DerSimonian and Laird’s random-effects model (Mantel-Haenszel method).
Main Outcomes and Measures
Six binary outcomes for the risk of falls, injurious falls, multiple falls (≥2 falls), fractures, hospitalization, and mortality.
Forty-six studies (22 709 participants) were included in the review and 40 (21 868 participants) in the meta-analyses (mean [SD] age, 73.1 [7.1] years; 15 054 [66.3%] of participants were women). The most used exercise was a multicomponent training (eg, aerobic plus strength plus balance); mean frequency was 3 times per week, about 50 minutes per session, at a moderate intensity. Comparator groups were often active controls. Exercise significantly decreased the risk of falls (n = 20 RCTs; 4420 participants; RR, 0.88; 95% CI, 0.79-0.98) and injurious falls (9 RTCs; 4481 participants; RR, 0.74; 95% CI, 0.62-0.88), and tended to reduce the risk of fractures (19 RTCs; 8410 participants; RR, 0.84; 95% CI, 0.71-1.00; P = .05). Exercise did not significantly diminish the risk of multiple falls (13 RTCs; 3060 participants), hospitalization (12 RTCs; 5639 participants), and mortality (29 RTCs; 11 441 participants). Sensitivity analyses provided similar findings, except the fixed-effect meta-analysis for the risk of fracture, which showed a significant effect favoring exercisers (RR, 0.84; 95% CI, 0.70-1.00; P = .047). Meta-regressions on mortality and falls suggest that 2 to 3 times per week would be the optimal exercise frequency.
Conclusions and Relevance
Long-term exercise is associated with a reduction in falls, injurious falls, and probably fractures in older adults, including people with cardiometabolic and neurological diseases.
Exercise training is an intervention of utmost importance for older adults’ health leading to benefits on multiple systems and functions, including muscle and bone health, the cardiometabolic system,1,2 as well as physical1,3,4 and potentially cognitive (results still mixed5,6) functions. Recent meta-analyses of randomized clinical trials (RCT) have shown that exercise reduces the number of incident falls in older adults,7-9 a major adverse event for this population.
Nevertheless, important gaps about the association of exercise with decreased risk of developing serious adverse outcomes still remain unclear. Most studies included in meta-analyses7-9 were short- to medium-term exercise interventions, evidencing a paucity of long-term (≥1 year) RCTs. Other important gaps are: a lack of evidence on exercise effects on death and hospitalization in diverse older adult populations, and the best exercise prescription (ie, type, intensity, frequency, session duration) for decreasing the risk of serious adverse events. The recent findings of the LIFE study,10 the largest and longest exercise trial performed to date among older people, showed, unexpectedly, increases in both hospitalization and mortality among exercisers compared with controls (differences statistically nonsignificant), raising doubts about safety issues of exercise for older individuals. Furthermore, meta-analyses have obtained mixed results for the effects of exercise in preventing fractures.2,11,12 Regarding falls, to the best of our knowledge, no meta-analyses have investigated the association of long-term exercise with falls-related outcomes, particularly multiple falls, which are common in older adults.13
The objectives of this systematic review of RCTs with preplanned meta-analysis were to investigate the association of long-term exercise interventions with the risk of mortality, hospitalization, becoming a faller, a faller with multiple falls, a faller with injurious falls, and sustaining a fracture.
This systematic review and meta-analysis was registered in PROSPERO (CRD42018090757) and follows the PRISMA guidelines.14 The protocol is available in the Supplement.
One author (M.M.) performed the electronic searches between February 20 and March 5, 2018, using a search strategy approved by all authors, from inception until the date of search in the following databases: PubMed, Cochrane Central Register of Controlled Trials, SportDiscus, PsychInfo, and Ageline. Full search strategies are available in the Supplement. Language restrictions were not applied. Two authors performed title/abstract screening independently. After that, the full-text of potentially eligible studies was accessed by 2 authors (P.S.B. and M.M.) for finally determining eligibility and, then, proceeding to data extraction. The reference list of previous systematic reviews7-9,12,15,16 were scrutinized. Divergences between authors on articles’ eligibility were resolved in an in-person meeting (100% consensus on articles’ eligibility was reached).
To be included in this review, studies had to meet the following criteria: (1) RCT design with exercise length of 1 year or longer (or ≥12 months or ≥48 weeks); (2) the study compared the effects of at least 1 exercise intervention against a comparator group (ie, no intervention, attention or active controls). Studies operationalizing cointerventions were eligible if the sole difference between intervention and comparator was the exercise training. All kinds of intervention structure (eg, home-based or group-based) were eligible, with unsupervised exercises being included only when a personalized exercise plan had been used; (3) participants had to be 60 years or older at baseline or the mean population age should be 60 years or older.
Six binary outcomes including mortality; hospitalization: number of individuals admitted to the hospital (eg, inpatient hospitalization, ≥24-hour hospitalization); fallers: people who fell at least once; fallers with multiple falls: people who fell at least twice; injurious fallers: people who suffered an injurious fall (eg, fall with wound, head trauma, medical care, fracture, or hospitalization) according to original investigators; and fractures: number of people who sustained a fracture.
Two raters (P.S.B. and M.M.) made the data abstraction independently using a standard data collection form specifically designed for this review. Divergences were solved in an in-person meeting (100% consensus reached). In case of doubts or insufficient data/information reported, original investigators were contacted by email.
When extracting data for the meta-analysis, we prioritized comparisons in which the sole difference between groups was the exercise intervention. In studies with multiple exercise groups vs a control group, we selected for the meta-analysis the group with higher amount of exercise sessions performed.
Two authors (P.S.B. and M.M.) independently coded the risk of bias in the 7 domains of the Cochrane Collaboration’s tool.17
Data on death, people hospitalized, fallers, and people with fractures were obtained from baseline until the end of the intervention period; data from observational follow-up were not used. Estimates of the outcomes were combined using the risk ratio (RR). Regarding fractures, for 5 studies without data on the number of people with fractures, we assumed the number of people with fractures was the same as the number of fractures (multiple fractures representing a relatively uncommon outcome, which would not severely bias our estimations); a sensitivity analysis removing these studies was undertaken. As prespecified in the review protocol (Supplement), we employed DerSimonian and Laird’s random-effects model18 (with Mantel-Haenszel method). Heterogeneity was evaluated using the I2 statistics, with an I2 greater than 50% representing substantial heterogeneity.19 Potential bias was evaluated using the Egger’s test, with P < .10 indicating substantial asymmetry, and funnel plots. Randomized clinical trials with attrition rates of more than 40% and those with low compliance (<30%) to the exercise intervention did not enter into the primary analyses, but were added in sensitivity analyses. Trials with no data on a given outcome were removed from meta-analysis of the specific outcome. For cluster RCTs, we used appropriate intracluster correlation (ICC) values (from the study, from another similar study included in the review, or from external databases20-22) to estimate the effective sample size using the design effect. If no appropriate estimate was available, we presented unadjusted estimates and ran a sensitivity analysis by removing cluster RCTs.
Other sensitivity/subgroup analyses were undertaken as prespecified in the protocol: by using a fixed-effect model when I2 was less than 50%, by restricting the analysis to RCTs with a low risk of attrition bias, and by stratifying analysis according to study population (clinically specific or disease specific vs nonclinically specific). We further performed analysis restricted to trials that have randomized more than 203 participants (median of study population across included RCTs), and by removing the 2 trials23,24 in which the average baseline age of participants was around 59 to properly address this deviation to the protocol.
When the number of studies was 10 or more, exploratory metaregressions were undertaken in an attempt to find which aspects of the exercise regimen would be associated to the effect size (log-transformed) of the outcomes. The following variables were tested: exercise frequency (3 times per week or more than 3 times per week compared with twice per week or less) as well as effective exercise frequency (weekly frequency multiplied by exercise compliance: between 2-3 times per week or more than 3 times per week compared with less than twice per week), volume (product of intensity and session duration: between 120 minutes per week and 180 minutes per week or 180 or more minutes per week compared with less than 120 minutes per week), intensity (vigorous compared with moderate), and type (aerobic, strength, or other exercise type compared with multicomponent training); given the importance of balance for all outcomes of this review, we also compared multicomponent training comprising a balance component vs all other exercise types combined.
All analyses were performed using STATA statistical software (version 14, StataCorp).
Forty-eight articles,10,13,23-68 representing 46 studies, met eligibility criteria and were included in the review. The flowchart for study selection is provided in the Supplement. The Table provides the characteristics of the included RCTs. The 46 studies have randomized 22 709 (median of 203 participants; range from 20 to 6420) participants and have been conducted mainly in Europe (n = 15), North America (n = 13, with 11 studies in the United States), and Oceania (n = 10). Participants had a mean age of 73.1 years, 66.3% were women (2 RCTs included only men45,46 and 11 solely included women13,27,34,37,43,48,49,53,55,56,64); mean intervention length was 17 months (median, 12 months). Sixteen trials23,24,29,33,40,41,45,46,50-52,57,58,61,62,64 were conducted in clinically specific or disease specific populations (eg, mild cognitive impairment or dementia [n = 7] and cardiac diseases [n = 4]). Thirty-five studies were parallel-group RCTs, whereas 11 were cluster RCTs30-32,35,36,47,51,52,65,67,68; most of the trials involved community dwellers (n = 35). Two studies in which participants had a baseline age of 59 years23 and 59.324 were included because their follow-up length was among the largest (12023 and 30.124 months, respectively), meaning that the average age increased to above 60 still at the beginning of the study. Two cluster RCTs were removed from primary quantitative analysis because very few participants exercised (26%30) or a very small fraction of exercise sessions (24%31) were attended.
As for any behavioral intervention, most trials had a high risk of bias related to blinding participants. The risk of concealment allocation was mostly unclear (n = 24), whereas incomplete data (n = 15) and blinding of outcome assessors (n = 11) may have been an issue for several studies (Supplement).
The most used exercise was a multicomponent training (multiple exercises; eg, aerobic plus strength plus balance training; 29 RCTs), followed by aerobic (8 RTCs)23,24,28,41,46,49,61,63 and strength (5 RTCs)27,31,38,48,64 training (Supplement). Exercise regimen was: mean frequency of 3 times per/week, about 50 minutes per session, at a moderate intensity (n = 35 RCTs). Average compliance was about 65%. Group-based supervised (n = 29) and a mix of group-based supervised and home-based unsupervised exercises (n = 12) were the most common format of exercise delivery. Comparator groups were often active controls, ranging from attention controls to more intensive interventions (eg, stent angioplasty). Thirteen trials had more than two study arms,26,27,31,37,38,40,48,52,56,57,62,66,67 all of them with two exercise groups except two studies,27,48 which had three exercise groups. All exercise arms in these studies have been scrutinized qualitatively. ICC values32,60 for mortality (ICC = 0.001) and fractures (ICC = 0.03) applied for cluster-RCTs performed in long-term care facilities (LTCF). No other suitable ICC values were found.
Original investigators of 28 studies were contacted; 18 responded, and 9 provided new data and/or information.
Thirty-nine studies10,13,23,24,27-52,54,56-58,61-65 provided available information on death for 19 670 participants. Among them, 8 studies27-29,40,41,46,62,64 had no deaths during the trial. Twenty-nine RCTs10,13,23,24,32-39,42-45,47-52,54,56-58,61,63,65 were included in the primary analysis (2 excluded owing to low compliance),30,31 totaling 11 441 participants in the model; 406 of 5677 (7.1%) and 453 of 5764 (7.9%) people died in the exercise and control groups, respectively. Figure 1A10,13,23,24,32-35,37-39,42-45,47-52,54,56-58,61,63,65 displays the forest plot of the effects of exercise on mortality. Exercise had no effects on mortality (P = .51). Heterogeneity was low (I2 = 0%) and Egger’s test suggested no important asymmetry (P = .76). The funnel plot (Supplement) suggested the presence of some bias, with small studies presenting inflated RRs (exercisers at increased risk of death). All sensitivity analyses provided similar results (Supplement), except analysis restricted to clinically specific or disease-specific populations, which found that exercise tended to reduce the risk of mortality (10 RCTs; RR, 0.70; 95% CI, 0.49-1.00; P = .05; I2 = 15%). Meta-regressions found associations for exercise frequency of 3 times per week compared with 2 or fewer times per week (exp[b], 0.42; standard error [SE], 0.14; P = .01) and effective frequency of between 2 and 3 times per week compared with fewer than 2 times per week (exp[b], 0.35; SE, 0.15; P = .03) with a reduced mortality risk. All the other meta-regressions found nonsignificant associations.
Fourteen studies10,23,24,30,32,33,38,40,41,46,47,50,54,57 had investigated hospitalization, but 1 RCT32 had no usable data on the number of people hospitalized; in this trial, exercisers and controls did not differ in terms of mean hospitalizations (0.87 vs 0.74, respectively). From the remaining thirteen studies (12 059 participants), 12 (5639 participants) entered into the primary analysis (1 exclusion owing to low compliance30): 1242 of 2822 (44%) and 1257 of 2817 (44.6%) people have been hospitalized in the exercise and control groups, respectively. Figure 1B23,24,33,38,40,41,46,47,50,54,57,59 shows that exercise has not reduced the risk of being hospitalized (P = .51), but heterogeneity was substantial (I2 = 59.2%). Even though the Egger’s test (P = .39) did not have evidence of small study effects, the funnel plot (Supplement) showed some asymmetry, with small studies having exaggerated large RR (increased risk of hospitalization among exercisers). Sensitivity and subgroup analyses (Supplement) provided unchanged nonsignificant findings. All meta-regressions provided nonsignificant associations of exercise variables with the risk of being hospitalized (Supplement).
Twenty-eight studies13,25,26,31,33-44,47,48,50,53,55-57,63,65-68 investigated falls; 7 had no usable information on the number of fallers,25,26,33,34,36-38 with 5 studies finding nonsignificant differences in number of falls across study groups, 134 showing significant effects favoring the exercise group, and 1 having no event across groups.38 Twenty-one studies contributed information on the number of fallers for 5220 participants. Of the 4420 people included in the primary analysis (20 RCTs13,35,39-44,47,48,50,53,55-57,63,65-68; 1 excluded31 owing to low compliance), 951 of 2207 (43.1%) and 1066 of 2213 (48.2%) became a faller in the exercise and control groups, respectively. As shown in Figure 2A,13,35,39-44,47,48,50,53,55-57,63,65-68 exercisers had a reduced risk of 12% to become a faller compared with controls (P = .02); heterogeneity was moderate (I2 = 50.7%). Although the Egger’s test did not evidence small study effects (P = .92), the funnel plot (Supplement) showed some asymmetry, with small studies having exaggeratedly large RRs (increased risk to be a faller in the exercise group). Sensitivity and subgroup analyses (Supplement) provided similar findings, with RRs for exercisers varying from 0.81 to 0.91, even if most of them did not reach statistical significance. All meta-regressions provided nonsignificant associations, except that exercise frequency (exp[b], 1.35; SE, 0.19; P = .05) as well as effective frequency (exp[b], 1.60; SE, 0.25; P = .01) more than 3 times per week were associated with increased risk of becoming a faller.
For fallers of multiple falls, 13 studies (3060 participants)13,35,39,42,44,47,48,50,53,56,63,65,67 composed the primary analysis: 329 of 1526 (21.5%) and 374 of 1534 (24.4%) individuals have fallen at least twice in the exercise and control groups, respectively. Exercise had no significant effect (Figure 2B)13,35,39,42,44,47,48,50,53,56,63,65,67 in decreasing the risk of being a faller with multiple falls (P = .20); heterogeneity was moderate (I2 = 60.2%). Egger’s test did not find small study effects (P = .96), but the funnel plot (Supplement) showed that small studies had inflated RRs (increased risk to be a multiple faller in the exercise group). Sensitivity analysis provided similar results. Meta-regressions found no significant associations.
Fourteen RCTs13,37-43,47,56,57,59,67,68 provided information on injurious falls, with 2 having no usable data for meta-analysis. These 2 studies found mixed results, with 1 showing more falls39 and the other37 reporting an average number of falls lower in exercisers compared with controls. Twelve RCTs gave data for 4972 participants, with 3 trials38,40,41 having no injurious fallers. Nine studies (n = 4481)13,42,43,47,56,57,59,67,68 composed the primary analysis, with 370 of 2192 (16.9%) and 471 of 2289 (20.6%) injurious fallers in the exercise and control groups, respectively. As displayed in Figure 3A,13,42,43,47,56,57,59,67,68 exercisers had a reduced risk of 26% to becoming injurious fallers compared with controls (P = .001); heterogeneity was moderate, but not substantial (I2 = 40%). Egger’s test (P = .22) and the funnel plot (Supplement) did not evidence any substantial asymmetry, even if there was a lack of small-to-medium scale studies collecting information on injurious falls (suggesting potential publication bias). All sensitivity and subgroup analyses (Supplement) provided unchanged results (except analysis restricted to disease-specific population; 1 RCT), with significant RRs favoring exercisers (range, 0.74-0.79). All 9 studies included in the primary analysis have operationalized a moderate-intensity, multicomponent training comprising balance exercises.
Twenty-three RCTs10,23,24,31-34,37-41,43,44,47,48,50,51,56,57,62,65,67 had information on fractures for 9701 individuals. Nineteen trials10,23,24,32-34,37,39,43,44,47,48,50,51,56,57,62,65,67 (no events occurred in 3 trials38,40,41 and 1 study was excluded owing to low compliance31) were entered into the primary analysis (8410 participants): 221 of 4138 (5.3%) and 270 of 4272 (6.3%) people in the exercise and control groups, respectively, have sustained a fracture. Figure 3B10,23,24,32-34,37,39,43,44,47,48,50,51,56,57,62,65,67 shows that exercise was not effective in reducing the number of fractures (P = .054). Heterogeneity was low (I2 = 0%), and the Egger’s test (P = .34) and funnel plot (Supplement) did not evidence any substantial asymmetry. Sensitivity analysis using a fixed-effect model found a significant effect of exercise for reducing the number of people with fractures by 16% (RR, 0.84; 95% CI, 0.70-1.00; P = .047). Excluding studies for which we assumed the number of fractures was equivalent to the number of people sustaining a fracture provided similar results (RR, 0.85; 95% CI, 0.71-1.03; P = .10). All the other analysis (Supplement) provided similar findings (favoring exercisers without reaching statistical significance). Meta-regressions found no significant association of exercise-related variables with this outcome.
This systematic review and meta-analysis showed that long-term exercise had modest but significant association with reduced risk of becoming a faller and an injurious faller, but not a faller with multiple falls, in older adults. Moreover, exercise was associated with a nonsignificant reduction in the risk of sustaining a fracture. Exercise benefits occurred without increasing the risk of mortality and hospitalization.
This is the first meta-analysis focusing on the benefits of long-term exercise (≥1 year), which may potentially lead to longer-term positive effects, against major adverse events in older adult populations. Our findings corroborate the results of recent meta-analyses focusing on falls that showed exercise (any intervention length) benefited several fall-related outcomes.7-9,16,69 The magnitude of associations for the risk of becoming a faller in our study (RR, 0.88; 95% CI, 0.80-0.98) was small but consistent with those (RR, 0.89; 95% CI, 0.81-0.97 and RR, 0.83; 95% CI, 0.70-0.99) recently reported by Guirguis-Blake et al69 and Tricco et al,16 respectively; the similarities in the findings across these meta-analyses represents compelling evidence of the positive effects of exercise against fall-related outcomes because in the present work we have used different eligibility criteria (particularly regarding intervention length ≥1 year), leading, then, to the inclusion of different studies. Furthermore, our study further extends current knowledge by examining for the first time the association of exercise with the risk of being a faller with multiple falls: we did not evidence a positive association of exercise with this outcome. However, multiple falls were not reported in some of the largest, well-conducted original studies (because it was not an endpoint of those RCTs) that provided data on fallers and injurious fallers,43,59 resulting in a small number of participants in the analysis; this issue, alongside the asymmetry found in the funnel plot, suggest publication bias could have affected estimations. For injurious fallers, we provide herein the most comprehensive evidence on the topic, gathering information from more than 4000 people in 9 RCTs, going beyond previous meta-analyses16,43,69; we consistently (across sensitivity and subgroup analysis) found that exercise decreased the risk of injurious falls by about 26%. Regarding fractures, our study contributes to this still not well-established field by showing that exercise seems to protect against fractures; although the primary finding was not statistically significant, the absence of both heterogeneity across studies and asymmetry (small study effects and publication bias), alongside the positive results from the fixed effect, suggest that long-term exercise might lead to a reduction in the risk of fractures. Even though we overcame a major issue of previous meta-analyses,2,11,12,16,69 which have included only a small number of studies and participants, analysis incorporating data from future well-conducted long-term RCTs is still needed before solid conclusions can be drawn on the protective effects of exercise against fractures in older adults. To our knowledge, our study is the largest meta-analysis ever done on the association of long-term exercise with mortality and the first one reporting about the association of exercise with hospitalization in older populations. Exercise had no statistically significant association with mortality or hospitalizations, which corroborates the mortality findings of Guirguis-Blake et al69; this may mean that mortality and hospitalizations are difficult-to-change outcomes, probably because they are determined by multidimensional parameters that may be beyond the scope of exercise-induced benefits. Lack of power may also have affected our results for mortality. Importantly, exercise tended to decrease mortality risk in clinical populations (mostly people with cognitive decline or cardiac disease), which reinforces the role of exercise as a core therapeutic element for treating prevalent diseases in older people.
Meta-regressions found that vigorous-intensity is as safe as moderate-intensity exercise. Exercise frequency of between twice and thrice a week was associated with decreased mortality, whereas more than 3 times per week was associated with increased risk of being a faller; therefore, the best exercise frequency seems to be 2 to 3 times per week, lower frequencies probably resulting in less effective outcomes whereas higher frequencies would augment the risk of adverse events. The association between exercise frequency and risk of becoming faller might be dependent on the fall-related vulnerability of the population, with higher risks in more vulnerable participants70; indeed, among studies with exercise frequency of 4 or more times per week, whereas Sherrington et al39 showed a higher risk of being a faller in the exercise group in people at increased risk of falling (mean age about 81 years, about 70% had fallen in the past 12 months), Kemmler et al34 found that exercise reduced fall rates in a population at lower risk for falls (young women aged on average 69 years) and Von Stengel et al37 found a trend in fall rates favoring exercisers in a similar low-risk population. It is possible that the dose-response idea implying that “more exercise is always better” might not fully apply for the most vulnerable older adults. The potential mechanisms involved require further investigation, but it could be related to overtraining: excessive exercise leads to diminished immunity and energy metabolism according with animal models71 and is associated with reduced calorie intake, worse sleep, and negative psychological patterns in young and middle-aged adults.72 Our findings on the best exercise frequency, alongside the observation that all RCTs included in the injurious falls analysis had a similar exercise structure, suggests that the best exercise regimen for protecting older people against diverse adverse events would be moderate-intensity, multicomponent training comprising balance exercises, performed 2 to 3 times per week; a session duration of 30 to 60 minutes (average of 50 minutes, according to studies on injurious falls analysis) should be safe and effective.
This review has limitations. First, several studies have not clearly reported exercise adherence, which impeded us to calculate the exact exercise volume performed by participants. Second, for examining the effects of long-term exercise, we arbitrarily established the 1-year length as the minimum intervention follow-up, which could lead to losing important studies that employed shorter follow-ups. A longer follow-up length would probably result in more precise data, particularly on less frequent events such as fractures, but it would have undermined the feasibility of the meta-analysis by reducing the number of eligible studies to 13. Third, owing to the several analyses performed, multiplicity may have elevated the chances of type I error. Finally, a high heterogeneity in terms of study population was found (from people with Alzheimer disease to overall healthy participants), but the small number of studies for most outcomes prevented us from performing stringent subgroup analysis. In an attempt to reduce such a bias, we performed subgroup analyses separately for populations with and without clinically and disease-specific profiles.
Exercise is associated with a modest decrease in the risk of becoming a faller, an injurious faller, and potentially sustaining a fracture in older adults. Exercise should be performed 2 to 3 times per week. Studies showing positive effects of exercise for reducing the risk of becoming an injurious faller operationalized moderate intensity, multicomponent training with balance exercises (eg, balance, strength training for the lower limbs, and aerobic exercise [eg, walking]), for about 50 minutes per session.
Corresponding Author: Philipe de Souto Barreto, PhD, Gérontopôle de Toulouse, Institut du Vieillissement, 37 Allées Jules Guesde, F-31000 Toulouse, France (firstname.lastname@example.org).
Accepted for Publication: August 16, 2018.
Published Online: December 28, 2018. doi:10.1001/jamainternmed.2018.5406
Author Contributions: Drs Philipe de Souto Barreto and Mathieu Maltais had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Barreto, Rolland, Maltais.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Barreto, Rolland, Vellas.
Critical revision of the manuscript for important intellectual content: Rolland, Maltais.
Statistical analysis: Barreto.
Study supervision: Rolland, Vellas, Maltais.
Conflict of Interest Disclosures. Dr Rolland reports grants from Novartis and Lactalis and fees from Nestlé and Daxter. Dr Vellas reports Scientific Board Memberships for Acadia, Accera, Alzhéon, Biogen, Lilly, Lundbeck, MSD, Nestlé, Roche, Sanofi, Transition Therapeutics, Takeda, Otsuka, and Schwabe; and also Gerontopôle Research Grants from Abbvie, Lilly, Lundbeck, MSD, Otsuka, Pierre-Fabre, Régénéron, Roche, Sanofi, Alzheon, Astra-Zénéca, LPG Systems, and Nestlé. No other conflicts are reported.
Additional Contributions: We thank the following investigators who contributed additional data or information to this study: Kaisu Pitkala (MD, PhD, University of Helsinki), Jennifer Hewitt (BAppSc, MHealthSc, University of Sydney), Jack Rejeski (PhD, Wake Forest University), Stefan Mustata (MD, University of Calgary), Stephen R. Lord (PhD, Neuroscience Research Australia), Stephen Messier (PhD, Wake Forest University), Kirsti Uusi-Rasi (PhD, UKK Institute for Health Promotion Research), Saija Karinkanta (PhD, UKK Institute for Health Promotion Research), Ari Heinonen (PhD, UKK Institute for Health Promotion Research). They were not compensated.