Physical Activity and Falls among a National Cohort of Older Veterans

Abstract

The more than 20 million U.S. veterans have a history of physical activity engagement but face increasing disability as they age. Falls are common among older adults, but there is little evidence on veterans’ fall risk. We conducted a retrospective cohort study using 48,643 observations from 14,831 older (≥ 65 years) Americans from the 2006–2014 waves of the Health and Retirement Study. Veterans reported more non-injurious falls (26.6% vs. 24.0%, p<0.002), but fewer fall-related injuries (8.9% vs 12.3%, p<0.001) than non-veterans. In adjusted analyses, for each 5-year increase in age, the odds of a non-injurious fall were greater for veterans (OR: 1.05, 95% CI: 1.01, 1.10) and, among those with regular physical activity, the odds were lower for veterans compared to non-veterans (OR: 0.89; 95% CI: 0.81, 0.99). For veterans, physical activity engagement may prove a particularly effective mechanism for reducing the aging-related risks associated with falls and fall injuries.

INTRODUCTION

Numbering over 20 million, U.S. veterans of the armed services are older and in poorer health but more physically active than other Americans (Hoerster et al., 2012; Littman, Forsberg, & Koepsell, 2009; Wu & Lewis, 2015). Nearly half of the veteran population are adults ages 65 and older compared to 8 percent of the civilian population (Hoerster et al., 2012). Given strong links between aging-related disability and physical engagement with falls (Deandrea et al., 2010; Florence et al., 2018; Sherrington et al., 2017), veterans may have a unique profile for exposure to falls, a growing threat to older adults well-being and independence (Hartholt, Lee, Burns, & van Beeck, 2019).

While risk factors for falls are generally well-understood, the interactions of such risks for the veterans’ population are poorly understood. On the one hand, because of a high prevalence of disability, veterans may be more greatly exposed to falls than other individuals. Research on disablement suggests that weakened functional status is the strongest predictor of falls (Verbrugge & Jette, 1994) and veterans may have therefore have greater function-related fall risk. On the other hand, due to their history of service that includes demanding physical activity and higher lifetime rates of activity (Littman et al., 2009; Sparling, Howard, Dunstan, & Owen, 2015), veterans may be at lower risk for falls, after accounting for differences in functional status. The literature is generally indicative of a protective (Finnegan, Seers, & Bruce, 2019; Sherrington et al., 2017) effect of physical activity on falls (Orwoll et al., 2018), and veterans may be more active when they do engage in physical activity. Given that disability may increase while the safety of physical activity may decrease over time (Orwoll et al., 2018; Tinetti & Kumar, 2010), veterans’ fall risk profiles may also change as veterans age.

There is an extremely limited literature on veterans and falls (Luther, French, Powell-Cope, Rubenstein, & Campbell, 2005; C. Soncrant, Neily, Bulat, & Mills, 2019; C. M. Soncrant et al., 2018; Zubkoff et al., 2018; Zubkoff et al., 2016) that has primarily focused on service-connected veterans, an older group with more disabilities than the general veteran population (Quigley, Palacios, & Spehar, 2006). These studies have relied on small samples of service-connected, hospitalized patients and have not provided comparisons to the non-veteran population. We build on this limited work by assessing, for veterans and non-veterans, the overall prevalence of falls as well as whether fall risk is influenced by respondents’ physical activity levels. Specifically, the aim of this study was to compare risks of non-injurious falls (NIFs) and fall-related injuries (FRIs) for the broader (service-connected and non-connected) veteran and non-veteran populations, including whether risks differed according by physical activity status and across age categories. With this approach, our findings will provide clinicians with novel insights into falls and fall risk factors for a large but previously unexplored older adult population. Given the disabilities that are strongly associated with veteran status, we hypothesized a priori that veterans would have higher overall risk for NIFs and FRIs than non-veterans, especially at older ages when relative disparities in functional status might expand between these groups. We also hypothesized a priori that, after adjusting for health and functional status, veterans would be more protected from falls due to physical activity given their service-related histories of physical fitness that may result in greater intensity and efficacy of physical activities (Littman et al., 2009).

RESARCH DESIGN

Data Sources and Study Population

This was a retrospective secondary data analysis using national survey data. Prior work examining veterans has typically used claims data for veterans treated by the Veterans Healthcare Administration (VHA), a population (~6 million) representing just 30% of all veterans who use VHA care annually (National Center for Veterans Analysis and Statistics, 2017). These veterans, often eligible for the health benefit from the U.S. Department of Veterans Affairs (VA) through a service-connected disability, are a unique subset of the overall veterans’ population (approximately 40% of all veterans are VA-enrolled) (CRS, 2014). Reliance on VA claims means that studies on broader veterans’ population (for whom claims in the VA system or Medicare data are not typically available) are limited. To address this gap, we assessed patient-level information from the 2006–2014 waves of the biennial Health and Retirement Study (HRS). The HRS is a nationally representative survey with sociodemographic, economic, health, functional, and social status, insurance, and veteran status information for older adults. Eligible HRS respondents are interviewed approximately each two years by telephone.

Our initial sample included 54,845 observations for HRS respondents. We excluded observations without complete information and those that were not connected to an adjacent survey wave (in order to include lagged variables, which are described below). For instance, we excluded a 2010 observation if the same respondent did not also have survey information from either of 2008 or 2012. This left us with a final analytic dataset of 48,643 observations including both veteran and non-veteran respondents ages 65 years and older, representing 14,831 unique respondents. This resulted in a pooled dataset in which respondents could contribute more than one observation if they were involved in multiple HRS survey waves – for instance, for an individual aged 74 and 76 in two adjacent surveys. This approach makes maximal use of the available data in the surveys, increasing sample size by employing these person-waves (Margolis & Verdery, 2017).

We constructed two study cohorts using HRS data—one for veterans one for non-veterans. To identify individuals with veteran status, we used responses to the question, “Have you ever served in the active military of the United States?” as has been done previously (Gould, Rideaux, Spira, & Beaudreau, 2015; Taylor, Ureña, & Kail, 2015). We also identified physical activity levels using responses to two HRS questions about how often respondents take part in each of moderate and vigorous physical activities. Using these responses, we constructed a dichotomous variable indicating whether respondents reported having regularly engaged (once per week or more) in either of moderate or vigorous physical activity. Moderate physical activities included gardening, walking at a moderate pace, stretching exercises, and home repairs; vigorous physical activity included jogging, swimming, tennis, and heavy housework.

Primary Outcomes

Self-reported non-injurious falls (NIFs) and FRIs were the primary outcomes examined. Specifically, respondents were asked whether or not, in the time since their previous interview (or in the prior two years, for those respondents interviewing for the first time with the HRS), they had fallen. Those responding “Yes” were considered to have had a fall. Respondents with FRIs were identified by identifying those individuals who answered “Yes” when asked whether, in that fall, they had injured themselves seriously enough to need medical attention. Respondents answering “No” to having required medical attention were coded as having had a NIF.

Other Variables

The Disablement Process model suggests that falls can result from a pathway reflecting chronic conditions, functional limitations, and other impairments (Verbrugge & Jette, 1994). Additionally, access to care (e.g., co-payments for medical treatment for FRIs) and respondents’ propensity to self-report a NIF or FRI might influence observed fall patterns. Because of potential differences in these disablement process, access to care, and propensity for self-report factors among veterans and non-veterans, we included representative characteristics that could proxy for these factors in risk-adjusted models. Demographic measures (age measured in 5-year bands, sex, race/ethnicity, educational level, and marital status that are known to predict FRI reporting accuracy) (Hoffman et al., 2018) were obtained from the HRS, as were economic indicators (household income and wealth, Medicaid status that could reflect access to care, such as co-payments for medical treatment), and health and functional status indicators that could confound observed relationships between veteran status, physical activity, and fall risk (Deandrea et al., 2010; Schwartz et al., 2008; Sennerby et al., 2009; Tinetti, Doucette, Claus, & Marottoli, 1995). Health and functional status indicators included activities of daily living (ADL, a count of 0–5 limitations), instrumental ADLs (IADL, a count of 0–3 limitations), counts of chronic conditions and depressive symptoms, and self-reported health (with dummy variables for very good or excellent health and for good health, with a reference category of fair or poor health). All time-varying variables (e.g., age, income, ADLs) were lagged by one survey wave (~two years) to account for potential reverse causality. To create lagged variables, we measured predictor variables in the wave prior to (wave t-1) the one in which the fall outcome was measured (wave t), before pooling all waves of data. We did not use lagged data from earlier survey waves (e.g., wave t-2) if information from the prior wave was missing.

Statistical Analysis

We first compared respondent characteristics, including demographic, health and functional status, physical activity status, and fall outcomes by veteran status, using t-tests and chi-square tests for continuous and categorical data and 2-tailed significance cut-offs at p<0.05. Next, we examined each of our study questions using multivariable logistic regression models with separate sets of models for each of the two outcome variables, NIFs and FRIs. To account for potential loss of precision in model estimates due to repeated observations of individuals in multiple surveys, we used cluster-robust standard errors in all models. In sensitivity checks, we re-estimated models using generalized estimating equations (GEEs) with independent working correlation structures (Heagerty & Comstock, 2013). Model standard errors and p-values were nearly identical when using GEEs; therefore, we report results from logistic regression models.

First, we assessed whether the odds of falling (NIFs and FRIs) varied according to veteran status. Next, in the second set of models, we explored the role of age in the veterans-falls relationship, by introducing an interaction term, Age x Veteran, into the original model. In a third set of models we explored the role of physical activity in the veterans-falls relationship, by introducing a separate interaction term, Activity Level x Veteran, into the original model. Finally, we introduced a triple interaction, Age x Activity Level x Veteran, to the original model to assess whether the veterans-falls association varied simultaneously by age and physical activity levels.

To facilitate interpretation of model results, given the difficulty in interpreting interaction terms in logistic regression models (Norton, Dowd, & Maciejewski, 2018), we used model estimates to compute predicted probabilities (or, “predicted risks”) using Stata’s post-estimation margins command. These are probabilities of the outcome for each category of the exposure (veteran status), while averaging over the remaining covariates. For instance, for the second set of models examining fall risk by veteran status and age, we allowed the predictions to vary across ages 65–90, in 5-year increments (e.g., 30% vs. 31% risk of a fall for a veteran compared to a non-veteran at age 65). Finally, given that 97% of veterans were male, and men were less likely than women to have an FRI (8.4 vs. 13.5%), we ran sensitivity analyses where we included interaction terms between the predictors of interest and a male dummy variable as well ran stratified models in which we only examined male respondents. In an additional sensitivity analysis, we used an alternative categorization of regular physical activity, indicated by more than once per week (as opposed to at least weekly) of moderate or vigorous physical activity.

This study was determined to be not regulated by the University of Michigan Institutional Review Board.

RESULTS

Unadjusted—Sociodemographics

We identified 11,841 (24.3%) veteran and 36,710 (75.7%) non-veteran observations. Substantial sociodemographic and health differences were observed (Table 1), including sex (97.4% of veterans were men, compared to 22.6% of non-veterans, p<0.001). Veterans had higher education levels (28% vs. 18% with ≥college degrees, p<0.001) and were more often married (78.2% vs. 56.3%, p<0.001) and of non-Hispanic White race/ethnicity (85.9% vs. 74.0%, p<0.001). Veterans reported better health (42.4% vs. 38.2% in very good/excellent health, p<0.001), but more mean (SD) chronic conditions (1.7 (1.2) vs. 1.5 (1.1), p<0.001).

Table 1.

Descriptive Statistics of Older Adults (≥ 65 Years), by Veteran Status, 2006–2014 (n = 48,643)

	Non-Veteran (n = 36,776)	Veteran (n = 11,867)
Mean Age (SD)*	73.10 (7.36)	74.36 (7.07)
Male Sex (%) *	22.6	97.45
Education (%) *
Less than High School	24.84	13.41
GED or High School Graduate	37.80	35.28
Some College	19.87	23.48
College and Above	17.50	27.83
Relationship Status (%) *
Married	56.25	78.17
Separated/Divorced	40.90	19.87
Never Married	2.85	1.96
Race/Ethnicity (%) *
Non-Hispanic White	74.02	85.87
African American	14.57	8.66
Hispanic	9.41	4.00
Other	1.46	2.00
Mean Household Income ($) (SD) *	52.20 (327.69)	64.31 (108.04)
Mean Household Wealth ($) (SD) *	461.06 (1,181.25)	591.73 (1,219.25)
Self-Reported Health-Status (%) *
Very Good/Excellent	38.22	42.42
Good	33.63	33.65
Fair/Poor	28.15	23.93
Mean No. Chronic Conditions (SD)*	1.50 (1.12)	1.70 (1.15)
Mean No. ADL Difficulties (SD)*	0.31 (0.83)	0.23 (0.69)
Mean No. IADL Difficulties (SD)*	0.11 (0.41)	0.09 (0.36)
Mean No. Depressive Symptoms (SD)*	1.46 (1.93)	1.00 (1.59)
Medicaid Status (%) *	9.61	3.45
Regular Physical Activity (%) *	48.6	55.7
Any Fall (%)	36.2	35.5
Non-Injurious Fall (%) *	24.0	26.6
Fall-Related Injury (%) *	12.2	8.9

Note:

^*p <0.001;

ADL = activities of daily living; IADL = instrumental ADL

Unadjusted—Physical Activity and Falls

As shown in Table 1, compared to non-veterans, veterans were more regularly physically active overall (55.7% vs. 48.6%, p<0.001) as well as across each of three age categories: 64.3 vs. 56.0% in the 65–74 (p<0.001), 59.3 vs. 50.6% in the 75–84 (p<0.001), and 50.3 vs. 38.5% in the ≥85 (p<0.001) age group (Figure 1).

Figure 1. Unadjusted Percentages of Older Adults (≥65) Engaging in Regular Physical Activity and Reporting Non-Injurious Falls and Fall-Related Injury, by Age, 2006–2014.

Notes: * indicates p<0.05. Using chi-square tests with α=0.05, differences between veterans and non-veterans for each of physical activity and fall-related injuries were statistically significantly different across age categories (p<0.001 for all age categories); differences between veterans and non-veterans for non-injurious falls were significantly different for the 65–74 and 75–84 age categories (p<0.001) but not for the 85+ age category.

In unadjusted comparisons, no overall differences in falls were observed (36.2 non-veterans vs 35.5% veterans, p=0.13) (Table 1). However, as shown in Figure 1, there were differences according to type of fall. Veterans reported more NIFS (26.6% vs. 24.0%, p<0.002), but fewer FRIs (8.9% vs 12.3%, p<0.001) overall (Figure 1). These patterns were consistent across age groups. Veterans had more NIFs in the 75–84 (29.3 vs. 26.0%, p<0.001) and ≥85 (35.7 vs. 29.0%, p<0.001) age groups. They had fewer FRIs in each of the 65–74 (6.7 vs. 10.1%, p<0.001), 75–84 (10.1 vs. 14.4%, p<0.001) and ≥ 85 (17.2 vs. 22.4%, p<0.001) age groups.

Adjusted Results: Veteran Status and Falls

As shown in Table 2, in multivariable analyses, after adjustment for all model covariates, no differences between veterans and non-veterans were observed overall in the odds of a NIF (OR: 1.06; 95% CI: 0.98, 1.16; Model 1a) or an FRI (OR: 0.96; 95% CI: 0.86, 1.08; Model 1b).

Table 2.

Odds of Self-Reported Falls, by Veteran Status, Age, and Physical Activity Level for U.S. Older Adults (≥ 65 Years), 2006–2014 (n =48,643)

Non-Injurious Fall
	OR	95% CI	p	OR	95% CI	p	OR	95% CI	p	OR	95% CI	p
	Model 1a			Model 2a			Model 3a			Model 4a
Characteristic
Vet	1.06	0.98, 1.16	0.16
Vet × Age				1.07	1.03, 1.12	0.001
Vet × Phys Act							0.90	0.81, 1.00	0.05
Vet × Age × Phys Act										1.00	0.93, 1.07	0.97
Fall-Related Injury
	Model 1b			Model 2b			Model 3b			Model 4b
Vet	0.96	0.86, 1.08	0.52
Vet × Age				1.03	0.97, 1.09	0.31
Vet × Phys Act							0.90	0.77, 1.05	0.18
Vet × Age × Phys Act										1.05	0.94, 1.16	0.41

Note: Vet = veteran; Phys Act = regular physical activity. Four separate multivariable regression models, for each of the any fall and fall-related injury outcomes respectively, were estimated—that controlled for respondent-level sociodemographic, functional status and health characteristics and included cluster-robust standard errors. Model 1 included veteran status as the primary predictor of interest, along with all other model covariates. Models 2 and 3 replicated the first model, but separately added an interaction term for veteran × age and an interaction term for veteran × physical activity, respectively. Model 4 replicated the first model but also included an interaction term for veteran × age × physical activity. The letter a represents models with any fall as the outcome while the letter b represents models with a fall-related injury as the outcome.

Adjusted Results: Age and Fall Outcomes, by Veteran Status

In multivariable analyses, for each 5-year increase in age, the odds increased more for more veterans compared to non-veterans for NIFs (OR: 1.07, 95% CI: 1.03, 1.12) but not for FRIs (OR: 1.03, 95% CI: 0.97, 1.09) (Table 2, Models 2a, 2b).

To facilitate interpretation of these model interaction terms, we computed predicted probabilities to illustrate differences in fall risks across age groups. As shown in Figure 2, while similar at ages 65–69 (21.5 vs. 22.8%), the risks of a NIF were greater for veterans compared to non-veterans at older ages (28.2 vs. 25.5% at ages 80–84; 30.7 vs. 26.4% at ages 85–89), even after controlling for a number of potential fall risk factors. This translated to [(28.2−25.5)/25.5=] 11% and [(30.7−26.4)/26.4=] 16% risk differences for NIFs at these oldest age groups.

Figure 2. Probability of a Non-Injurious Fall and Fall-Related Injury by V eteran Status and Age among Older Americans, 2006–2014 Health and Retirement Study.

Notes: Predicted probabilities for each of non-injurious falls and fall-related injuries according to age, physical activity level, and veterans status were computed from four separate multivariable regression models that each controlled for respondent-level sociodemographic, functional status and health characteristics and that included cluster-robust standard errors.

Adjusted Results: Physical Activity and Fall Outcomes, by Veteran Status

In multivariable analyses, when comparing by physical activity status, both veterans and non-veterans benefited from physical activity. However, there was evidence that veterans benefited more in terms of avoiding NIFs (OR: 0.90, 95% CI: 0.81, 1.00), but not FRIs (OR: 0.90, 95% CI: 0.77, 1.05) (Table 2, Models 3a, 3b). Predicted probabilities are plotted in Figure 3, which illustrates that, among inactive respondents, the risk of a NIF was higher (28.8 vs. 26.4%) for veterans compared to non-veterans; however, among active respondents, the risk of a NIF was similar for veterans compared to non-veterans (22.9 vs. 22.6%). This translated to a marginal risk reduction in NIFs of 2.5 percentage points (p=0.005) for veterans, or approximately 10% decrease in NIF risk, but no change in risk (0.5 percentage points, p=0.36) for non-veterans, when engaged versus not engaged in regular physical activity.

Figure 3. Probability of a Non-Injurious Fall and Fall-Related Injury by Veteran Status and Regular Physical Activity Status among Older Americans, 2006–2014 Health and Retirement Study.

Notes: * indicates p<0.05. Predicted probabilities for each of non-injurious falls and fall-related injuries according to age, physical activity level, and veterans status were computed from four separate multivariable regression models that each controlled for respondent-level sociodemographic, functional status and health characteristics and that included cluster-robust standard errors.

Adjusted Results: Physical Activity, Age, and Fall Outcomes, by Veteran Status

There were no significant differences in the fall-veteran relationship by age according to physical activity status (Table 2, Models 4a, b), after controlling for respondent characteristics.

Sensitivity Analyses

When we included interaction terms in each of the models testing for whether results varied according to male sex, the coefficients were in the direction of reduced risk of the outcomes for males versus females, but none of the terms were statistically significant (e.g., the interaction of male x physical activity x veteran’s status for the FRI outcome had an OR of 0.58, p=0.18; results available upon request). When we ran stratified analyses only including male respondents, increasing age was no longer statistically significantly associated with increased NIFs for veterans, although the magnitude of the coefficient was similar (OR: 1.04, 95% CI: 0.98, 1.10) (Supplement, Table S1). The marginal effect of the veteran status x physical activity term was also slightly smaller and non-significant. All other findings remained unchanged. We also used an alternative definition of regular physical activity (more than once per week) and found that veterans were more highly active than non-veterans (11% vs. 10%, p<0.001); also, veterans no longer had greater benefits from physical activity compared to non-veterans in an adjusted model.

DISCUSSION

In this study of veterans and fall risk, we had three novel findings. First, veterans had 11% more non-injurious but 28% fewer injurious falls than non-veterans. Put into context, were rates lowered to those of the group with lowest risk, we estimate that there would be 750,000 fewer veterans with non-injurious falls (NIFs) and 1 million fewer non-veterans with fall-related injuries (FRIs) each two years. Second, the risk of a NIF increased more with age for veterans than non-veterans, with a 10–15% greater relative risk for the oldest old veterans. Third, physical activity was protective against a NIF for veterans but not non-veterans, with a ~10% lower risk for veterans engaging in at least one day per week of moderate or vigorous physical activity. Collectively, these results are indicative of a U.S. older veterans’ population of more active older individuals who are more exposed to less serious fall risks than other older Americans.

In one of the few prior studies on veterans and falls, Quigley et al. (2006) (Quigley et al., 2006) observed a positive correlation between fall risk scores and age for 1,810 older veterans treated in VA nursing home. The current study extends existing knowledge of veterans and falls, showing that veterans have more NIFs (especially at older ages) but fewer serious injuries from falls that are associated with morbidity, institutionalization, and even death (Burns & Kakara, 2018; Hartholt et al., 2019; Kuehn, 2018). Because falls without injuries do not present these same threats to health and independence as FRIs and given the myriad health benefits to older adults of more active lifestyles (even if slightly increasing fall risk), these veterans’ fall patterns may illustrate a worthwhile trade-off between safety and independence.

Previous researchers have shown that regular physical activity is generally effective in preventing falls (Cauley et al., 2013; Finnegan et al., 2019; Gregg, Pereira, & Caspersen, 2000; Quach & Burr, 2016; Sherrington et al., 2017), but have also cautioned that the risks of falling can increase for highly active individuals (Orwoll et al., 2018). A recent Cochrane systematic review (Sherrington et al., 2019) also suggests that decreases in risk of up to 25% can be obtained through physical exercise programs. This study’s finding of a 10% overall reduced risk of NIFs is consistent with this prior work, in that veterans may have more active lifestyles that incur moderate risks for minor falls while avoiding more serious injury; at the same time, as veterans age, the relative benefits for veterans compared to non-veterans of such regular activity no longer appear to overcome certain intrinsic (health and functioning-related) risks for falling.

Military service itself may provide an explanation for this unique fall profile. Physical fitness and regular fitness tests are prerequisites for active-duty members (Littman et al., 2009), and veterans may retain these health habits into later life. For this reason, the intensity and frequency of regular physical activity may be different for veterans and non-veterans, even among those who report regular engagement in such activity; this was supported in our data, as more veterans than non-veterans reported being in the most active category. Further, when we compared the benefits of regular activity when only assessing highly active respondents (more than once per week of activity), we no longer observed relative fall risk benefits for veterans.

Our study has several limitations. First, self-reported data on falls can be unreliable or incomplete (Hoffman et al., 2018), resulting in conservative estimates of fall prevalence. However, unless reporting bias varies by veteran status over time, this is unlikely to substantially alter our findings. In prior work using the HRS and linked Medicare data, we observed that accuracy of self-reports of an FRI were generally higher for women compared to men and for those with more compared to fewer functional limitations or chronic conditions. However, by controlling for these factors in our models, these factors should not have influenced our risk-adjusted fall estimates. To the extent that residual risk remained after risk-adjustment, it is possible that falls may have been underreported by veterans to the extent they were healthier than non-veterans (perhaps at younger ages), but relatively overreported as their health declined. However, given model risk-adjustment and the opposite directions of these gender and health-related reporting biases for the two cohorts, these reporting issues should have had a limited effect on our reported outcomes.

Second, we examined adults ages 65 and older, so were unable to assess whether there were additional protective effects from physical activity for younger veterans. The HRS does not ask about fall history for respondents younger than age 65, however, and the majority of falls among Americans occur after age 65. Third, because men are less likely to fall (Deandrea et al., 2010) and also report higher levels of physical activity compared to women (Azevedo et al., 2007; Koeneman, Verheijden, Chinapaw, & Hopman-Rock, 2011), our results could reflect the predominantly male makeup of the veterans’ cohort (Gould et al., 2015; Taylor et al., 2015). Our data were somewhat consistent with this explanation, as 56% of men compared to 47% of women in the study were regularly active, and men also reported better overall health and fewer functional limitations compared to women. Moreover, in stratified analyses only including male respondents, we no longer observed absolute risk increases for a fall with increasing age or regular physical activity, for veterans compared to non-veterans. However, absolute risk differences in stratified models are dependent on baseline fall risks (which varied for male veterans and non-veterans). To ascertain relative differences in model relationships between men and women, it is more appropriate to use interaction terms (Sun, Briel, Walter, & Guyatt, 2010), as significant findings in sub-group analyses can be due to differences in sub-group sample compositions or due to chance. When we examined relative effects across sub-groups by sex, by including interaction terms in each model, we did not observe significant interactions. Therefore, we did not observe strong evidence that fall risk-related relative differences between veterans and non-veterans were due to sex compositional differences in the two cohorts. Future work should more fully examine this question.

Finally, our physical activity measure was less granular that the U.S. Department of Health and Human Services’ recommended activity levels for older adults (U.S. Department of Health and Human Services, 2008). However, a sensitivity analysis we included was able to demonstrate differences between respondents at the highest activity levels versus those with more modest regular activity—suggesting that the relative benefits of regular physical activity for veterans compared to non-veterans were most likely due to veterans being more active exercisers.

These limitations notwithstanding, our observation that veterans’ fall profile represent a classic trade-off between activity and safety can inform clinical practice for veterans the broader U.S. older adult population. First, for the oldest old veterans who are on a trajectory of increased falls, many (perhaps 6 in 10) are unlikely to have access to care through the VA; clinicians outside the VA may be unfamiliar with the unique risks of this cohort, however, and therefore may not optimally understand how to help veterans address changing, complex trade-offs between mobility and safety as veterans age.

More broadly, the findings support the need for patient-centered support to better understand and address trade-offs associated with falls, so that older adults do not simply avoid activity due to a fear of falling (Delbaere, Crombez, Vanderstraeten, Willems, & Cambier, 2004; Painter et al., 2012). A patient-centered approach will recognize the heterogeneity among older individuals in terms of fall risk and physical activity levels, and provide individual-specific plans accordingly (Tuvemo Johnson, Martin, Anens, Johansson, & Hellström, 2016). On the one hand, greater activity levels for older adults can improve functioning and the likelihood of longer-term independence, even as it presents more opportunities for trips, stumbles, and minor falls (Growdon, Shorr, & Inouye, 2017; Tinetti & Kumar, 2010); conversely, sedentary behavior can counterintuitively prevent falls (by restricting opportunities for falling), even as it harms longer-term functioning and autonomy. Therefore, aiming for no falls is not necessarily an optimal strategy, if the goal is older adult autonomy. To this end, it is concerning that half of respondents did not regularly engage in physical activity given the protective benefits for falls and for overall older adult health (Sparling et al., 2015).

On the other hand, the nearly 10% reduced risk of falls for veterans who regularly exercised has clinical meaningfulness. According to recent work, home-based exercise programs largely tested in clinical trials have obtained ~25% fall risk reductions among older adults (Sherrington et al., 2019); that more than one-third of that risk reduction was observed in the general veteran population without clinical intervention is impressive. Put another way, given the costs of widespread, community-based interventions and challenges of getting individuals to take-up exercise, it might cost billions of dollars to obtain similar results in the non-veteran population. Therefore, better understanding why veterans are able to engage in and benefit from exercise could help bolster broader prevention efforts.

Future work should further explore whether physical activity in this broader veterans’ population confers lasting health benefits in terms of functional independence beyond falls, such as differences in nursing home entry, support from informal caregivers, and acute and long-term care health care expenditures. It may be that early accumulation of health capital can explain later life habits that influence fall risk; measures of health capital and identification strategies to separate selection effects of veterans’ cohorts will be needed to pursue such analyses.

CONCLUSION

Collectively, veterans have a unique risk profile that appears to reflect an active population. Compared to other older adults, veterans have more minor falls but fewer injurious ones, which is consistent with a largely male, highly active, and relatively healthy population. However, veterans’ fall risk increases with age, a change in risk that is not offset by increased physical activity, suggesting that clinical efforts should help older veterans manage the complex trade-offs between maintaining independence through active lifestyles while ensuring their safety in older ages.