Postponed Development of Disability in Elderly Runners:  A 13-Year Longitudinal Study FREE

DISABILITY AMONG individuals in aging populations has become an extremely important public health concern and is arguably the largest health problem in developed countries. Because modifiable risk factors have been identified for cardiovascular disease and certain cancers, some fear that increased survival as a result of preventive measures will lead to an increasing number of frail and disabled elderly persons in the population. A longitudinal study¹ of a cohort of runners and community control subjects initially aged 50 to 72 years demonstrated that participation in regular aerobic physical activity is associated with significantly less disability. In that 8-year study, differences in disability levels between runners' club members and control subjects were sustained and continued to grow apart over time. A definite survival advantage was also conferred to those in the runners' club group, confirming the results of previous longitudinal studies.^2- 3

The benefits of running on disability and survival can be strongly inferred from these studies through an average age of 67 years, but it is not yet clear that these benefits can be sustained into older ages. According to the compression of morbidity hypothesis,^4- 6 we would expect running groups to accrue disability at an increasing rate and the difference in disability between runners and control subjects to narrow as the cohorts continue to age. Similarly, we would expect to see convergence of mortality curves as members begin to achieve their biological limit of life expectancy, regardless of the influence of exogenous factors.⁷ Essentially, compression of morbidity requires that cumulative lifetime disability, ascertained over all lifetimes of a cohort, is reduced in runners compared with comparison cohorts. Compression becomes more predictable as cohorts age, but is directly ascertainable only after long-term study.⁸

To our knowledge, the effects of exercise on cumulative disability and on the duration of disability postponement have not been studied. We present data from a 13-year longitudinal analysis of the runners' club and control cohorts, conducted to examine the extent to which these benefits are maintained, to determine the magnitude of the postponement of disability, and to test the compression of morbidity hypothesis. Because of the increased number of deaths since the last analysis, we also are able to compare the causes of death in the 2 groups. We hypothesized that those engaging in regular aerobic physical activity would have lower disability, increased survival, and fewer deaths due to cardiovascular disease and other diseases associated with lifestyle factors.

Subjects in this study were recruited in January 1984 to participate in a longitudinal study of the effect of long-distance running on various outcomes, including disability, mortality, and the development of osteoarthritis. Long-distance runners 50 years and older were enrolled from the 50+ Runners Association, a running club with members across the United States. Control subjects who had been enrolled in the 1972 Lipid Research Clinics study were recruited from the Stanford University community.⁹ Study descriptions were initially sent to all 1311 runners' club and 2181 Stanford University community members 50 years and older. Of these subjects, 654 runners' club and 568 control members expressed interest in the study and satisfied its age (≥50 years), education (high school graduate or above), and language (English as their primary language) requirements. The initial questionnaire and consent forms were sent to this group; the 538 runners and 423 controls returning initial questionnaires were enrolled in the study. In addition to obtaining information on medical history, exercise habits (running and other forms of aerobic exercise), history of musculoskeletal injuries, and demographic variables, all subjects were administered the Health Assessment Questionnaire (HAQ) yearly to assess levels of disability. The typical runners' club member had run an average of 25 851 km over 10.8 years before study enrollment. In comparison, the typical subject in the community control group had run 1374 km over 2.2 years before study enrollment. After 13 years, 370 runners' club members (72% of living subjects) and 249 community controls (70% of living subjects) continued to participate in the study. Dropouts and continuing participants closely resembled each other in many characteristics, such as educational level, body mass index (BMI), and baseline levels of exercise, including running. However, those who continued in the study tended to be younger and had lower disability levels than those who dropped out.

The HAQ assesses functional ability in 8 areas: rising, dressing and grooming, hygiene, eating, walking, reach, grip, and activities (such as running errands). Each area is scored from 0 to 3 (0 indicates no difficulty; 1, some difficulty; 2, much difficulty; and 3, unable to perform), and accounts for the need of special aids or devices or the assistance of another person. The score in each of the 8 areas is averaged to obtain the HAQ disability index (score, 0-3). This instrument has been widely used, is sensitive to change, and has been extensively validated.^10- 13

The principal longitudinal analysis focused on the 370 runners' club members and 249 community controls who continued to participate in the study in 1997. In addition to the runners' club and community control groups, we formed groups of ever runners and never runners based on the following question, "Have you ever run for exercise for a period greater than 1 month?" The ever-runner group, thus, includes subjects who may have run in the past but discontinued before study onset, and was created to reduce self-selection bias and the dilution of exercise effects due to many exercisers in the community control group. This procedure shifted 94 participants from the community control group to the ever-runner group. Moreover, we further explored the characteristics and potential effect of those who withdrew from the study by analyzing disability levels for all initial participants and for those who continued through the end of the present study in 1997.

The progression of disability over time for the runners' club and control groups was expressed as a rate, or slope, under the assumption that the rate of progression of disability is linear, remaining constant over time. To account for the correlation among the repeated disability index measurements made for each patient, a series of general linear mixed models was fitted to the data.¹⁴ The degree of correlation between any 2 disability scores from the same patient was assumed to be constant (a compound symmetric correlation structure).

The postponement of disability was estimated as the difference between groups in the average time before a specified level of disability was attained. By using the estimates of a linear mixed model, the average time until a member of the runners' club reaches a given level of disability, d₀, is estimated as follows: t_r(d₀) = (d₀ − a_r)/b_r, where a_r is the average baseline disability and b_r is the average annual change in disability. Similarly, t_c(d₀), the average time until a member of the community control group reaches a given disability level, can be estimated. The postponement of disability between the runners' club and the control groups is, thus, estimated as the difference: postponement of disability(d₀) = t_r(d₀) − t_c(d₀). A 95% bias-corrected and accelerated confidence interval (CI) was formed around this estimate using 1000 bootstrap replications.

The postponement of disability was estimated with and without adjustment for the potential confounders of age, sex, BMI, smoking exposure, and initial disability level. Interactions among these covariates were then selected using a manual backward selection procedure. A mixed model was specified using the selected covariates, time, and group status. From these equations, t(d₀) was predicted for each subject. The postponement of disability was estimated as the difference in the average t(d₀) between the runners' club and control groups. Again, a 95% CI was derived using 1000 bootstrap replications.

Differences between groups during follow-up were compared using χ² and 2-tailed t tests. Comparisons were considered statistically significant at P≤.05.

Deaths were confirmed for all original subjects through the National Death Index. At data analysis, death certificates were obtained for 21 of the 26 deceased persons in the runners' club group and for 58 of the 67 deceased persons in the community control group. Next of kin and friends were contacted by telephone for the 14 subjects for whom death certificates were not available. Information regarding the cause of death was obtained for 10 of these subjects. Causes of death were assessed blinded to group status.

Survival was analyzed for runners' club members vs community controls using all subjects enrolled at study inception. The Kaplan-Meier method was used to obtain crude survival estimates. Adjustment for the covariates of age, sex, BMI, weekly aerobic exercise other than running (measured in minutes), smoking exposure, and alcohol consumption was made using the Cox proportional hazards regression method. The BMI and aerobic exercise time (in minutes) were treated as time-dependent covariates in the model.

The characteristics of subjects at study onset in 1984 are shown in Table 1. Important differences between the groups were observed. The runners' club group contained substantially more male subjects than the community control group, and was on average 2.7 years younger than the control group. These differences were also reflected in the ever-runner/never-runner breakdown of groups. Runners smoked less, consumed less alcohol, and, as expected, continued to run more frequently and for a longer distance. They also had lower BMIs and lower initial disability scores. No difference between the groups was found in educational level.

Participation in other vigorous forms of exercise, including bicycling, swimming, brisk walking, aerobic dance, and racquet sports, was not significantly different between the 2 groups at baseline (Table 1). After 13 years, 55% of runners' club members and 57% of ever runners continued running. However, the amount of time spent in other forms of exercise substantially increased in both groups (Table 2), with runners now spending on average 96.6 minutes more on other vigorous exercise and 144.8 minutes less on running. The control group also had increases in other forms of vigorous exercise and decreases in time spent running. Table 2 also shows greater increases in BMI in the runners' club group than in the control group. Declines in smoking and alcohol consumption were observed in both groups. Disability scores increased in both groups, but increases were lower in magnitude among the runners' club members.

The progression of mean disability levels over 13 years of observation is shown in Figure 1. Figure 1 A shows the progression using available data from all initial participants (538 runners' club members and 423 community controls) and from the 370 runners' club members and 249 community controls who continued participation through 1997. A significant (P<.01 at all data points) difference in the progression of disability was seen between the study groups. A significant difference also existed between dropouts (excluding deaths) and study completers in the community control group at all data collection points (P<.01). In addition, dropout rates in the community control group were somewhat higher than in the runners' club group. During the first 8 years of this study, from January 1, 1984, through December 31, 1992, 93 subjects (22%) dropped out of the control group; a further 81 control subjects (25%) dropped out in the ensuing 5 years. Among runners' club members, 87 (16%) withdrew during the first 8 years of the study, with a further 81 (18%) withdrawing in the next 5 years. At study inception, initial disability was significantly (P = .01) lower in control participants who eventually completed the study. This may be in part due to the significantly (P = .02) lower baseline age among the completers. If this trend is considered as a selection bias, it would have the effect of minimizing the differences between the runners' club and community control groups, leading to more conservative estimates.

The progression of disability among the 464 study completers classified as ever runners and the 155 classified as never runners is compared in Figure 1 B. Significant differences in disability were present at study onset and were maintained at each data collection point (P<.001). The difference in mean disability score increased over time. This display is intended to demonstrate the effects of any bias whereby less healthy individuals might have stopped running before study onset. No such effects were seen, with the results in Figure 1 A and B being closely comparable.

Mean disability levels among male and female participants over time are shown in Figure 2. Disability is maintained at low levels among runners' club members. No differences in disability are seen between male and female runners' club members over time. Male controls have a significantly higher level of disability than male runners' club members over time (P<.05 at each data collection point except 1986). Greater still is the difference in disability levels between female control subjects and female runners' club members. Except for the first 2 years of the study, each of these differences is statistically significant (P≤.02 at each data point), and increases with time.

The effect of running and exercise intensity on the development of disability over time is shown in Figure 3. Running intensity was classified based on minutes run per week, with none being 0 min/wk; low to moderate, less than 200 min/wk; and high, 200 min/wk or more. Figure 3 A and B portrays mean disability over time as a function of running intensity for women and men, respectively. No statistically significant (P≥.24) difference is seen between those engaging in low to moderate running intensity and those engaging in high-intensity running. Nonrunners, however, accrue disability at a greater rate. In a similar fashion, the effect of aerobic exercise other than running was examined (Figure 3 C and D). Exercise intensity was classified based on the number of minutes per week spent in nonrunning aerobic exercise, with none being 0 min/wk; low, less than 160 min/wk; moderate, 160 to 299 min/wk; and high, 300 min/wk or more. Although a low level of aerobic exercise is associated with a lower level of disability compared with no exercise at some points during the study period, the difference is not consistently maintained (Figure 3 C and D). Disability levels are lower and tend to be maintained at lower levels only in the moderate- and high-intensity exercise groups.

Figure 4 is an examination of disability levels according to participant age, counting each subject once in each applicable age group. Although disability levels are different between runners' club members and community controls at all ages, the difference seems to widen after the age of 79 years, and suggests that the community control group may thereafter have a more accelerated increase in disability.

The rate of progression of disability during the 13-year observation period using linear mixed regression models is shown in Figure 5. The rate of progression of disability was lower in the runners' group (0.004 disability units per year) than in the community control group (0.017 disability units per year) (P<.001). After 13 years of follow-up, the average disability for the runners was 0.097, while the level for the community controls was 0.30, and the time required to reach a particular level of disability was significantly longer for the runners' club members. To avoid extrapolation beyond the known limits of the data, we set a hypothetical threshold at 0.075 disability units. Adjusting for potential confounders, the time required to reach a disability level of 0.075 was 1.6 years from study onset for community control subjects. On the other hand, those in the runners' club reached this disability level after 10.3 years, a difference of 8.7 (95% CI, 5.5-13.7) years. Thus, there was a significant postponement in the development of disability among runners' club members. Based on this model, the postponement increases as the chosen level of disability increases. For example, at a disability level of 0.050, the projected postponement is 4.6 (95% CI, 2.5-7.3) years, while at a level of 0.100, it becomes 12.8 (95% CI, 8.3-20.6) years. At each selected level of disability, the effect of postponing disability is more pronounced among women than men (Table 3).

The Kaplan-Meier plot of unadjusted survival estimates for the runners' club and community control groups is shown in Figure 6. The significant difference in mortality between runners' club members and controls is maintained (P<.001) over 13 years and increases over time. The 13-year survival of runners since study inception is 96%, vs 83% for community controls. The average age of the runners and controls in 1997 was 70.9 and 73.6 years, respectively. Adjustment for age, sex, BMI, aerobic exercise (other than running) time (measured in minutes per week), amount of smoking, and alcohol consumption was made using the Cox proportional hazards regression model. The BMI and alcohol consumption did not meet the statistical criteria to enter the model; the effect of age was of borderline statistical significance (P = .06), but was included as a clinically important covariate. Table 4 shows the adjusted variable estimates and death rate ratios in the multivariate model. The death rate ratio was higher for men vs women, and the rate of death increased by a factor of more than 2 for every pack per day smoked. On the other hand, being a runners' club member was strongly protective. Engaging in aerobic exercise was also a protective factor, an effect seen over and above running club status.

The causes of death in study participants are enumerated in Table 5. In 1997, there were 93 deaths identified in the cohort, which had accrued 12 493 person-years of observation (6994 person-years for the runners and 5499 person-years for the controls). Twenty-six deaths occurred in the runners' club members and 67 in the community controls. Among ever runners, there were 39 deaths; and among never runners, 54. Overall, the rate of death was approximately 3.3 times higher among community controls vs runners' club members. The death rate for controls was higher in each major disease category listed, although differences were not statistically significant, given the relatively few deceased persons. Striking increases in death rates among control subjects were seen for myocardial infarction, stroke, colon cancer, and esophageal cancer. There were 7 cases of lung cancer among control subjects, and no cases in the runners' group. The category of other cancers included bladder, renal cell, ovarian, and brain cancers and cancers of unknown primary origin. Neurological causes included Parkinson disease, Alzheimer disease, and amyotrophic lateral sclerosis. Infectious causes included pneumonia, bacteremia, and infective endocarditis; other causes of death included chronic obstructive lung disease, pulmonary fibrosis, cirrhosis, gastrointestinal bleeding, unintentional injuries, homicide, suicide, and acquired immune deficiency. Rates among control subjects were higher in each of these categories. Only for prostate cancer were there more deaths among runners. Considering deaths most strongly related to lifestyle factors, such as coronary artery disease, stroke, and lung cancer, there were 114.40 deaths per 100 000 person-years in runners' club members and 454.64 deaths per 100 000 person-years among controls, for a rate ratio of 3.97.

The average age of death was 72.6 years among controls and 74.9 years among runners' club members (P = .23). Among controls, there were 67 deaths, 44 (66%) of which were men. Among runners' club members, there were 26 deaths; all were men, illustrating the survival advantage for the women, particularly in the runners' group.

Studies^15- 17 in diverse populations of elderly persons have demonstrated the benefit of various forms of aerobic exercise on disability. Possible explanations for this effect include a combination of increases in skeletal muscle mass,¹⁸ metabolic adaptations of muscle,¹⁹ and improved aerobic capacity.¹⁷ Recent observations^20- 21 have confirmed that such improvements positively impact dependence and mortality. This study provides further insight into the influence of running and other aerobic exercise on disability and mortality during a prolonged period, during which the average age of study subjects increased from 59 to 72 years. We show that the lower disability levels observed in those engaging in aerobic exercise, including running, are maintained through 13 years of observation. Even minimal disability is significantly postponed in runners' club members, and this postponement increases as higher levels of disability are considered. Although postponement of disability through exercise in elderly persons has been examined,²¹ to our knowledge, this study is the first to attempt to quantify its degree longitudinally. These data also show that the lower mortality rates among runners' club members included not only decreases in diseases strongly related to lifestyle factors but were observed across all disease categories.

Because self-selection bias may be inherent in observational studies that lack an external intervention, care has been taken throughout this study to account for potential differences between groups that may influence outcomes, as described in a previous publication.¹ Briefly, validation studies of self-report have been conducted in this population in consideration of possible reporting bias, in which runners might report their function as being better than it actually was, and have shown that no reporting difference exists between the groups.²² We also considered the possibility of differences in familial predilection to disability. Thus, all screened candidates who reported a family history of prematurely disabling medical conditions were excluded. Radiographs were obtained during the study to examine for differences in the development of arthritis, a leading cause of disability; again, no differences between the groups were found.²³ Statistical adjustment was made for the covariates that were different between the groups at study outset, including age, sex, and baseline disability level. The runners' club group was on average 2.7 years younger than the control group, and contained more women. These differences would tend to operate in opposite directions in the 2 groups for cardiovascular-related mortality, but both of these factors would operate against the control group for disability. Because some subjects in the control group may have initiated running at an intermediate point in their life, and subsequently stopped running because of disability or discomfort, we formed and analyzed ever-runner and never-runner groups by including anyone who had run for 1 month or longer in the runners' group. This approximates an intent-to-treat analysis, and tends to yield more conservative estimates. We also noted that differences in study dropouts may act over time to bias the results, so analyses were conducted in consideration of this (Figure 1). The most severely disabled among the controls had the highest rates of discontinuation. Because this would leave those who were relatively less disabled to continue in the study, the effect is again to push the results toward the more conservative side. In view of these multiple considerations and adjustments, differences between the groups still persisted over the entire period of observation.

Also tending to make these results conservative is the fact that both groups had a relatively high level of education and a higher socioeconomic status; both groups were also predominantly white. Because it is well established that higher levels of education,^24- 27 a higher socioeconomic status,^28- 30 and white race^30- 31 all contribute to increased life expectancy, favorable outcomes in runner and control groups would be expected. Despite this, a significant benefit for disability and mortality associated with running and other aerobic exercise was demonstrated. Thus, given the low baseline risks in the study population, these may be considered conservative estimates of the effect of regular physical activity on disability, and although not derived from a strict cross section of the general population, may imply that those at higher risk for disability may derive an even greater benefit from exercise regimens. This hypothesis can only be tested through direct study of such populations.

In considering the features of the disability curve over time, the overall scores in the runners' club group remain relatively low. Even in the most elderly segment of the group, aged 85 to 90 years, disability scores only approximate 0.30 (Figure 4). This is equivalent to having some functional difficulty in 2 of the 8 areas of daily function assessed on the HAQ or much difficulty in 1 area. In contrast, the most elderly control subjects have levels reaching 0.80 and above. Although disability herein is lower than that observed in patients with a severe chronic disabling disease, such as rheumatoid arthritis (which has average HAQ disability scores of ≥1.2), it is substantial and reflects significant impairment in several daily activities comparable to disability in patients with osteoarthritis.¹³

The progression of disability among the study cohorts was assumed to be linear to apply the simplest analytical model that would simultaneously account for repeated measurements and the intrinsic correlations between observations over time (Figure 5). As suggested in Figure 4, however, disability trends may assume a nonlinear form, especially as a person approaches biological limits to life expectancy. Data in Figure 4 consist of relatively fewer observations toward the more advanced ages; disability is also plotted against patient age, rather than year of study. Both of these may explain the differences in the shape of the curves. Continued longitudinal observation will be required to refine the model of the shape of the disability progression curve.

An accelerated increase in disability levels in the runners' club group has not been observed. The compression of morbidity hypothesis^4- 6 suggests that in individuals with healthy lifestyles, disability can be postponed until a few years before death, when it develops at an increasing rate. In this cohort, the control subjects seem to be entering into a phase of accelerated disability, as has been observed in other elderly populations with low health risks. It is expected that a similar occurrence will be seen among the runners' club group. To be directly observed, however, this phenomenon will require continued longitudinal study. Nevertheless, the postponement of disability seen herein, even to minimum levels of impairment, is significant and compels us to consider the role of exercise in the elderly population as an important element of a disability-free life.

Significant differences in mortality also persist during the 13-year period. As with disability, we do not yet see acceleration of mortality rates in the runners' club group with convergence of the 2 curves, but would expect such as subjects approach their biological "limit" to life, independent of environmental factors such as diet and exercise.⁷ Runners' club members die less frequently of conditions strongly related to lifestyle, such as coronary artery disease, stroke, and lung cancer. Runners' club membership and engaging in aerobic exercise other than running were identified as protective factors and male sex and smoking as detrimental factors to survival in a multivariate analysis. The BMI, in contrast to other studies,^2,23- 34 did not enter the model on statistical grounds, nor did alcohol intake. It is possible that the distribution of these variables within this study population was too narrow for an effect to be detected. The attenuating effect of a J- or U-shaped relationship of these variables with either disability or mortality was excluded; an analysis showed no such relationship to be present (data not shown).

Because aerobic exercise itself may increase longevity by increasing muscle strength,²² cardiovascular reserve,^35- 36 bone mineral density,^37- 38 and glucose tolerance,³⁹ running may also be a surrogate for other positive lifestyle factors, ranging from healthy eating habits and abstinence from cigarette smoking to the increased use of automobile safety belts.^31,40 While it is clear that mortality due to diseases strongly associated with lifestyle factors is decreased among members of the runners' club, mortality is also decreased in every other disease group. It is probable that still-undescribed factors are operating to confer advantages to the runners' club group. Whether these are related to other lifestyle and socioeconomic factors, awareness of health issues, association with healthy peers, and access to health care or whether there are advantageous biological (such as genetic) factors present is not directly addressable by these data but warrants further study.

These data also suggest that women have a greater potential benefit from exercise. Even though women entered the study with higher initial disability levels than men, significant differences in disability between female runners' club members and controls became apparent within 3 years of study inception. These differences were greater in magnitude than those seen among male subjects. Furthermore, postponement of disability among female runners is more pronounced at each disability level compared with male runners.

Not only are deaths prevented, disability levels are decreased and the development of disability is postponed in association with running and other aerobic exercise. These, in turn, may be associated with other healthy lifestyle practices; the strong contribution of nonsmoking is suggested by these data. These data support the recommendation that individuals even in midlife participate in regular aerobic exercise of at least moderate intensity, and suggest that quantity and quality of life can be augmented through primary prevention measures. Improved risk profiles and overall health may also lead to reduced costs to the health care system.^41- 44 As modifiable health risks are identified, modifications advocated, and policies implemented, the goal of healthier aging and less prevalent illness for the increasing proportion of elderly persons in our population may be approached.

Accepted for publication February 13, 2002.

This study was supported by grant AR43584 from the National Institute of Arthritis and Musculoskeletal and Skin Diseases and the National Institute on Aging, National Institutes of Health, Bethesda, Md; and by the Canadian Institutes for Health Research, Ottawa, Ontario (Dr Wang).

This study was presented in part at the Annual Scientific Meeting of the American College of Rheumatology, Boston, Mass, November 14, 1999.

Corresponding author and reprints: Benjamin W. E. Wang, MD, FRCPC, Department of Medicine, The University of Tennessee Health Science Center, 956 Court Ave, Room G326, Memphis, TN 38163 (e-mail: bwewang@utmem.edu).