Physical activity, particularly aerobic physical activity, decreases mortality rates [1-3]. Its effects on morbidity and disability, however, have not been shown as clearly. On the one hand, vigorous physical activity (such as aerobic running) could accelerate the development of osteoarthritis and disability, and repetitive injuries to soft tissues might increase disability as a result of cumulative trauma [4-6]. On the other hand, increased fitness and training, leading to increased cardiovascular reserve, increased bone density, and increased strength from regular exercise could delay or prevent disability [7-13].

The effect of exercise on the development of disability is thus an important public health issue, but clinical studies in this area are difficult to design and perform. Experimental trials comparing exercise with no exercise over prolonged periods are not practical. Cross-sectional studies may be inconclusive because increased physical activity could be the self-selected result of good health, rather than good health the result of the physical activity. In longitudinal studies, on the other hand, we expect that if a self-selection bias is present, a baseline difference in disability between exercising and nonexercising groups would narrow with time, and, if physical activity leads to development of disability, that the trend lines eventually would converge.

For the past 9 years, the Stanford Arthritis Center has conducted prospective longitudinal studies of the effects of long-distance running, a popular form of vigorous physical activity, on patient outcomes in 537 members of a runners' club with participants at least 50 years old and 423 community controls. In addition to longitudinal study, our controls for self-selection bias included analysis of multiple variables at baseline and during the study, including family history of arthritis, body mass index, weight, history of arthritis, cigarette smoking, history of fractures, and history of lower-extremity abnormality. We have also studied presence of comorbid disease, progression as shown on x-ray films, comparisons of progression between hand and knee on x-ray films, bone density, and frequency of fractures. We performed longitudinal analyses with "intent to treat" analyses in which initial group membership is retained although individual members may have decreased the level of or stopped running.

Preliminary results have indicated, at baseline and at 2 years, that running was associated with substantially reduced disability, a marked increase in lumbar bone density, and no differences in the development of radiologic osteoarthritis [14-16]. At 5 years, radiologic osteoarthritis was developing with similar frequencies in running and nonrunning groups. Effects on associated lumbar bone density persisted for 5 years but were sensitive to changes in running, increasing in those who increased activity and decreasing in those who decreased activity [17-20]. Fracture rates were more common, however, in the running groups, with the increase seen in sports-related and fall-related fractures; fractures of wrist, hip, and spine were more common in nonrunners [21]. Presence of osteoarthritis was nearly identical in dominant and nondominant hands, supporting the argument that greater use is not necessarily associated with increased radiologic change [22]. Extensive validation studies showed absence of differential self-report bias in runners and nonrunners [14].

We report the results of our 8-year longitudinal study of progression of disability scores between runners and nonrunners, with additional analyses by age and sex.

Methods

Top Methods Results Discussion Author & Article Info References

Sample Selection

The 50+ Runners Association, with members across the United States, provided access to many long-distance runners who were 50 years or older. Runners' club members voluntarily joined an organization that has, as one of its purposes, participation in studies of the health effects of long-distance running. The typical runners' club member had run 16 869 miles during an average of 12.4 years before entering the study. The Lipid Research Clinic study, begun in 1972, provided access to a sample, similar in age, from the Stanford University community [23]. In January 1984, study descriptions were sent to and received by 1345 runners and control participants. Those persons who met age (50 years or older), education (high school graduate or additional education), and language requirements (English as their primary language) were sent questionnaires and consent forms. The 537 runners and 423 controls returning initial questionnaires were enrolled in the study. All participants provided information concerning exercise history, musculoskeletal injuries, medical history, dietary history (especially calcium and fat), and other variables. They also completed the Stanford Health Assessment Questionnaire. Since then, similar questionnaires have been sent to participants each year. At 8 years, 84% of the runners' club members and 78% of the community controls continue to participate in the study. Drop-outs closely resembled those in their groups who continued in the study for nearly all variables. Users of alcohol in all groups tended to drop out at a higher, but nonsignficant, rate compared with those who did not use alcohol. Initial disability in runners' club completers was 0.026 (0.004) and 0.029 (0.01) in drop-outs (P = 0.77) and 0.079 (0.01) in control completers compared with 0.152 (0.02) in drop-outs (P = 0.008). Exercise levels were similar in completers and in drop-outs, as were age, sex, body mass index, and education level. Eight runners' club members (1.5%) and 30 community controls (7%) have died. Deaths in the relatively few participants lost to follow-up were estimated using the National Death Index.

The Health Assessment Questionnaire has been the subject of multiple validation studies and is used widely [24-29]. The self-administered questionnaire assesses functions in eight areas: dressing and grooming, arising, eating, walking, hygiene, reach, grip, and activities over four categories scored from 0 to 3 [no difficulty, some difficulty, much difficulty, and unable to perform, respectively]. Three components are evaluated for each area: 1) ratings of the degree to which daily functions are difficult to perform; 2) reported use of special aids or devices; and 3) activities for which assistance of another person is required. The scores on each of the 8 functional areas are averaged to obtain the disability index (0 to 3 units). For reference, a typical patient with symptomatic osteoarthritis has a score of 0.6 to 0.8.

Statistical Analysis

We compared differences in participants at onset and during follow-up using chi-square and t-tests. We analyzed survival using the Kaplan-Meier method. We adjusted for covariates using analysis of covariance and Cox proportional-hazards models. Principal longitudinal analyses focused on the 330 community controls and 451 runners' club members who remained in the study through 1992. We maintained participants in their initial groups (formed in 1984), regardless of subsequent change in running status. We developed groups of ever-runners and never-runners from the entire group because many participants in the control group also exercised vigorously. These groups were formed on the basis of responses to the question, Have you ever run for exercise for a period greater than 1 month? Because the ever-runner group thus includes runners who discontinued this activity before study onset, we hypothesized that if self-selection bias were operating, differences between these groups would be less striking than with the parent groups. This reclassification shifted 83 persons to the ever-runner group from the community controls group used in the intent-to-treat analysis (see Table 1 for group sizes).

Results

Top Methods Results Discussion Author & Article Info References

Table 1 lists characteristics of study participants at study onset in 1984; these participants were classified as runners' club members and community controls, and also as ever-runners and never-runners.

We found substantial differences between the participant cohorts. The running groups are younger by approximately 3 years, had substantially lower disability scores at study onset, have lower body mass indices, are less frequently smokers, and, of course, exercise much more vigorously and run many more miles per year. Runners have lower systolic and diastolic blood pressures, visit their physicians less frequently, and use fewer medications. The runners have a less frequent history of arthritis and fewer other major medical problems. Runners report less frequent chronic joint swelling or chronic joint pain. On the other hand, runners report more fractures. We found no differences between groups in education level, alcohol intake, or history of other major musculoskeletal injury. These differences, at the onset of longitudinal study, may be alternatively interpreted as representing self-selection into the runner group or as representing beneficial effects from the average 12 years of running before the study onset, or both. These alternatives cannot be separated based on cross-sectional study, as represented in Table 1.

Table 2 shows characteristics of the same participants in 1992. The differences between the groups noted in Table 1 (data from 1984) persist 8 years later. Only 56% of the ever-runners and 63% of the runners' club members are still running after 8 years. The frequency of other vigorous exercise, however, although not significantly different between groups at baseline, has increased substantially in all groups but particularly in the running groups. Those runners who stopped running usually switched to other forms of vigorous exercise and maintained their exercise practice. Runners' club members who stopped running reported 262 minutes a week of vigorous exercise including swimming, bicycling, brisk walking, aerobic dance, and racquet sports, whereas community controls reported only 118 minutes per week of such exercise. Physician visits per year, similar across groups in 1984, had changed by 1992 so that runners had about 20% fewer physician visits. To assess congenital abnormalities, reported more frequently by runners, we included a query about whether one leg was shorter than the other, an observation perhaps better known to runners.

Figure 1 shows progression of disability over time in the runners club and community control groups for the 8 years from 1984 to 1992. In the top panel, data include all available data points on the initial 537 runners' club members and 423 community controls and also on those 451 runners' club members and 330 community controls who were continuing participants during the entire 8-year study period. We made minor modifications to the questionnaire between 1989 and 1990, resulting in increased sensitivity. Essentially no differences exist between the curves regardless of whether study drop-outs are included or excluded, but substantial and enlarging differences exist between the runners' club members and the community controls. Subsequent figures refer only to the 1992 continuing participants, so exactly the same persons are included for 1984 and 1992.

View larger version (21K): [in this window] [in a new window]   Figure 1. Progression of disability over time. Top. Runners' club members compared with community controls. Data from all patients and all data points were compared with data from continuing 1992 participants (completers). Bars represent 95% confidence limits. No significant differences existed between those completing 8 years of the study (1992 participants) and data including all data points, but significant differences between runners' club participants and community controls were observed at all time points (P < 0.01). Bottom. Ever-runners compared with never-runners. Similar differences, significant at all time points, are seen when ever-runners, those who have ever run for 1 month or more, were compared with never-runners. Bars represent 95% confidence limits (1992 participants).

Figure 2 presents a separate display of men and women in the runners' club and the community controls. Both male and female runners' club participants maintain low disability levels throughout the study. In contrast, female controls have far greater degrees of disability compared with any other group, and male controls also have significantly greater disability than do the men in the runners' club (P < 0.05). To control for baseline differences in disability, we repeated these analyses for those runners (325 men, 64 women) and community controls (130 men, 84 women) who did not report any disability in 1984. Again, runners maintained low disability levels for 8 years of follow-up but the controls showed increasing disability, so the differences became and remained significant after 5 years (P < 0.05).

View larger version (21K): [in this window] [in a new window]   Figure 2. Progression of disability over time by sex. Runners' club compared with community controls. Disability progression is less in the runners' club members. Significant differences were noted for men and women, with the effects more pronounced in women (P < 0.05). Bars represent 95% confidence limits (1992 participants).

Figure 3 shows cross-sectional data for disability levels by age at the final data point in 1992. Disability increased progressively with age in all groups, with the slope of the disability curve increasing in the oldest participants. Because only 13 runners and 30 community controls were older than 80 years, the groups cannot be statistically separated at that point. Overall, however, differences exist between community controls and runners at all ages, and the differences are maximal at ages 75 to 79 years.

View larger version (18K): [in this window] [in a new window]   Figure 3. Disability levels by age. The figure presents a cross-sectional analysis of disability in runner's club, community control, and all participants in 1992. Differences between runner's club members and community controls are significant (P < 0.05) at all ages except for the youngest and the oldest. Bars represent 95% confidence limits.

View larger version (19K): [in this window] [in a new window]   Figure 4. Survival analysis. Kaplan-Meier survival curves are presented for runners' club members and for community controls. Bars represent 95% confidence limits. These curves present the crude mortality data; Table 4 shows data adjusted for relevant covariates.

Discussion

Top Methods Results Discussion Author & Article Info References

The principal analytic problem of observational studies of this type is the ability to remove self-selection bias from the data. In addition to using longitudinal data, we considered and tried to control for five specific types of self-selection bias. First, we tried to identify covariates that were different at study outset between groups and found many. Generally, the differences were not unexpected, with fewer reports of arthritis, chronic disease, or smoking and with lower body mass in the runners. We adjusted statistically for all identified variables that were different between the groups, and the overall results persisted after this adjustment. Second, a bias could be present in which runners reported their functional abilities as better than they were, whereas the less enthusiastic and less self-selected controls reported more accurately. We ruled out this possibility of bias by doing validation studies comparing the two groups [14]. Third, we considered that there might have been differences in genetic characteristics, such as family history of arthritis or congenital abnormality. Here, however, there were few differences between the groups. Fourth, we considered that those persons who had begun to run at some intermediate point in their lives but who had pain or excessive difficulty during the process might have then stopped running, leaving the runners' club to inherit a self-selected group who were comfortable with continued running activity. To eliminate this effect, we created the "ever-runner" and "never-runner" groups, so that those who had ever run for more than 1 month would be included as "runners," although they may have stopped running many years before. This approximates an "intention to treat" analysis and might generally be considered conservative. However, differences between the ever-runners and never-runners were even larger than those between the runners' club and community control groups. Fifth, we considered that with time, selective dropouts might bias the results. But, as shown in Figure 1, the effect of including or excluding data from drop-outs is small. Further, drop-outs were similar in initial characteristics to completers, except for controls, for which the most disabled initially tended to drop out. Had these persons not dropped out, presumably the results reported here would have been strengthened.

Despite these efforts to investigate possible artifacts of bias between runners and controls, persons who choose to run may indeed be different in their makeup, in characteristics that we do not capture in our analysis but are linked to both running and disability rates of progression. However, we believe our results contribute to the evidence that the link between exercise and positive health outcomes is causal. A recent study of changes in quality-of-life and disability measures in 194 previously sedentary persons 50 to 65 years old participating in endurance exercise showed substantial improvement at 12 months, consistent with our results, and documented a dose-response relation with the amount of exercise [30].

In a subset of these participants, we reported changes shown on x-ray films after 5 years. All groups showed an increase in osteoarthritic changes in the knees, hands, and lumbar spine with time, but there were no differences between runners and nonrunners [17]. Therefore, we believe that the reduced levels of disability and disability rates we observed in runners are not related to a positive effect of running on development of osteoarthritis, but rather are related to other factors, such as improved physical abilities with improved conditioning, and lower rates of comorbidity from other conditions, such as cardiovascular disease, for which lack of exercise is an established risk factor.

Differences in mortality rates associated with running and other aerobic exercise have been reported by other investigators and are confirmed here. The differences are substantial, although only a few deaths have been recorded in these groups. Again statistical adjustment for differences in covariates, including age, failed to account for these differences.

Lifestyles incorporating long-distance running, and presumably other regular aerobic exercise activity, is associated with preservation of good physical function in the later years of life compared with more sedentary lifestyles. Important differences exist in mortality and morbidity rates. These findings have substantial implications for strategies, directed to increase regular lifetime physical exercise, to improve the quality of life of the growing older population.