Physical Activity and Life Expectancy With and Without Diabetes: Life table analysis of the Framingham Heart Study

The Framingham Heart Study cohort consisted of 5,209 respondents (46% male) aged 28–62 years residing in Framingham, Massachusetts, between 1948 and 1951. The Framingham cohort is primarily white and has been followed extensively for 46 years for the occurrence of cardiovascular disease and death through surveillance of hospital admissions, death registries, and other available medical sources. Examination of participants, including an interview, a physical examination, and laboratory tests, has taken place biennially. Further description of the Framingham Heart Study can be found elsewhere (24).

To calculate transition rates by level of physical activity, we pooled three nonoverlapping follow-up periods of 12 years. Each follow-up period started with a measurement of physical activity. In the present investigation, the follow-up periods started at the original exams 4 (1956–1958), 11/12 (1969–1973), and 19/20 (1985–1989). Using the pooling of repeated observations method (25), follow-up information over three follow-up periods was pooled, yielding a total of 9,773 observation intervals. The same participant may thus be followed during three periods until the event (first onset of diabetes or death) occurs or the subject is censored. Follow-up time and physical activity status may differ in each interval. After exclusion of subjects with diabetes at exam 1 to exclude juvenile diabetes and missing information on physical activity, we studied 4,176 subjects at exam 4, 3,286 subjects at exam 11/12, and 1,660 subjects at exam 19/20, yielding a total of 9,122 follow-up observation intervals.

Diabetes was considered present after a random blood glucose level ≥200 mg/dl or when the subject was treated with a hypoglycemic agent (insulin and/or oral hypoglycemic agent).

Participants were asked about their time spent resting or engaged in light, moderate, or heavy physical activity on an average day. Time spent at each activity in hours per week was multiplied by its metabolic cost (based on the oxygen consumption required for that activity) as described before by Kannel et al. (26). A weight of 1.0 was used for an activity with oxygen consumption of 0.25 l/min, for example sleep. Other weights were 1.1 for being sedentary, 1.5 for light activity, 2.4 for moderate activity, and 5 for heavy activity. The weight factor corresponds to a metabolic equivalent task. These weighted hours were added up to get a total daily physical activity score. The minimum physical activity score is 24, which is equivalent to 24 h of rest/sleep. Based on tertiles of the daily physical activity scores, we grouped the participants in three levels: low (<30), moderate (30–33), and high (>33) physical activity level.

Potential confounders were measured at the start of each follow-up period, except for education. All analyses were adjusted for age and sex. Confounders considered were education (eighth grade or less/higher than eighth grade), smoking (never, ever, or current smoking), marital status (single, married, widowed, or separated/divorced), diseases present at baseline (any of the following: cardiovascular disease, cancer, left ventricular hypertrophy, arthritis, ankle edema, or any pulmonary disease), total cholesterol, family history of diabetes (diabetes in parents and/or siblings), and the exam of start follow-up (exam 4, 11/12, or 19/20). The exam of start follow-up was included to correct for a potential cohort and period effect. Hypertension and BMI were not considered as confounders but as intermediate factors, as physical activity may cause individuals to have lower BMI and blood pressure, which in turn reduces the risk of diabetes and mortality. Adjustment for BMI and hypertension may result in an underestimation of the true beneficial effect of physical activity. Hypertension was defined as systolic blood pressure ≥140 mmHg or diastolic blood pressure ≥90 mmHg (27). For BMI, four categories were defined: BMI <18.5 kg/m², 18.5 ≤ BMI < 25 kg/m², 25 ≤ BMI < 30 kg/m², and BMI ≥30 kg/m². Because the information on alcohol consumption was not available for all follow-up periods, alcohol was excluded from the analyses. For the final analysis, only participants who had information on all confounders, BMI, and hypertension were included (3,978, 3,067, and 1,561 subjects at exam rounds 4, 11/12, and 19/20, respectively).

To calculate the life expectancy with and without diabetes, we created a multistate life table (MSLT) (28). This MSLT, which combines information of people at different ages and from different birth cohorts, included three states: “free of diagnosed diabetes,” “diabetes,” and “death.” The possible transitions were from free of diagnosed diabetes to diabetes or to death and from diabetes to death. No back flows were allowed, and only the first entry into a state was considered. Age 50 years was chosen as the starting age to include the effect of physical activity at adult ages, while avoiding unstable rates due to limited number of events below age 50 years. Another advantage is that at this age, where the prevalence of type 2 diabetes is still low, the outcomes of the MSLT for the total and diabetes-free population are virtually the same.

To obtain transition rates by levels of physical activity, we first calculated the overall sex- and age-specific transition rates for each transition (29). Next, we calculated hazard ratios of physical activity using Poisson regression. We analyzed the effect of potential confounders on the hazard ratios by first adjusting for age and sex only and next by including the following in addition: education, marital status, smoking, baseline diseases, and exam of start follow-up. Subsequently, adjustments for BMI and hypertension were added to the previous model to give information about the effect of these possible intermediates on the hazard ratios. Additionally, adjusting for BMI did not substantially change the hazard ratios. The small effect of adjusting for BMI may reflect a small risk of self-selection of heavy subjects for lower physical activity levels because physical activity includes physical activity at work. We did not adjust for total cholesterol (as continuous variable) and family history of diabetes in the final model, as these variables did not change the hazard ratios of physical activity and had a large number of missing values.

Finally, to calculate transition rates by level of physical activity, we used the overall transition rates, the prevalence of the different levels of physical activity in the population (stratified by 10-year age-group, sex, and diabetes), and the hazard ratios of physical activity adjusted for potential confounding but not for intermediates (BMI and hypertension).

Separate MSLTs were created for each level of physical activity for men and women, incorporating each of the three transitions. The MSLT started at age 50 years and was closed at age 100 years. The measures available from the MSLT include life expectancy without diagnosed diabetes (i.e., the average number of years lived free of diabetes before the onset of diabetes or death) and life expectancy with diabetes (i.e., the average number of years lived with diabetes) for populations with either low, moderate, or high physical activity and free of diabetes at age 50 years.

All statistical analyses were done using STATA version 8.2 for Windows (Stata, College Station, TX). We calculated CIs for all life expectancies and differences in life expectancies using Monte Carlo simulation (parametric bootstrapping) (30). To calculate the CIs, we used @RISK (Anonymous 2000; MathSoft) 10,000 runs.

In the categories low and moderate physical activity, the proportion of women was higher than in the category with high physical activity (63 and 62%, respectively, compared with 46%) (Table 1). Participants in the low physical activity group tended to be older (mean age 62 years) compared with the participants of the moderate and high activity groups (mean age 58 and 59 years, respectively); also, the level of baseline diseases was higher among participants with low physical activity. Blood pressure and total cholesterol showed a tendency toward an inverse relationship with physical activity (Table 1).

Higher levels of physical activity were associated with lower rates of incident diabetes (corrected for age and sex) (online appendix [available at http://care.diabetesjournals.org]). After additional correction for selected confounders, the effect slightly increased and the hazard ratio remained significant (Table 2).

The risks of mortality in subjects without diagnosed diabetes (corrected for age and sex) were inversely related to the level of physical activity (Table 2). Correcting for confounders slightly attenuated the effect, but all hazard ratios remained significant. The mortality rates in subjects with diabetes were also lower in active subjects than in sedentary subjects. After adjusting for confounders, this association was only significant for high physical activity.

Figure 1 shows survival curves, total and free of diabetes, by level of physical activity. The area under these survival curves is total life expectancy and life expectancy without diabetes, respectively. The difference between the total and diabetes-free life expectancy is life expectancy with diabetes. At age 50 years, total life expectancy was 27.2 (men) and 33.8 (women) years, of which 25.5 and 32.1 years, respectively, were spent without diabetes and 1.7 years (both sexes) with diabetes. In contrast, life expectancy of 50-year-old individuals with diabetes is 23.8 and 24.3 years for men and women, respectively.

Life expectancy of 50-year-old men in the moderate and high activity groups was 1.8 (95% CI 0.8–2.9) and 4.1 (2.8–5.4) years longer, respectively, than of men in the lowest activity group. For women, these differences were similar: 1.7 (0.8–2.8) and 3.7 (2.6–4.9) years (Table 3). This larger total life expectancy was composed of more years lived without diabetes and fewer years lived with diabetes. Both moderately active men and women lived 2.3 (1.2–3.4) more years without diabetes than those in the lowest activity group (Table 3). In the highest activity group, men and women lived 4.2 (2.9–5.5) and 4.0 (2.8–5.1) years longer, respectively, without diabetes than their sedentary counterparts. Moderately active men and women lived ∼0.5 (men: 95% CI −1.0 to 0.0, women: −1.1 to −0.1) years less with diabetes, compared with those with low physical activity. For high activity groups these differences for men and women were 0.1 and 0.2 years, respectively, but were no longer significant (Table 3).

Moderately and highly active people live longer and spend more years without diabetes than subjects with low physical activity levels. At age 50 years, life expectancy free of diabetes is 2.3 years longer for moderately active men and women and at least 4 years longer for highly active men and women. The effect of physical activity on life expectancy without diabetes reflects both the lower incidence of diabetes and the lower mortality of nondiabetic individuals associated with increasing physical activity levels. Life expectancy with diabetes is at least 0.5 and 0.1 years less for moderate and highly active people, respectively, compared with those with low physical activity. This reflects two opposing effects: 1) lower incidence of diabetes in the active group reducing the time spent with diabetes and 2) lower mortality in diabetic subjects, increasing the time spent with diabetes. The net result is that while moderate and highly active people live longer, they do not spend more years with diabetes.

The reported hazard ratios found in our study fall well within the range of the published measures of the effect of physical activity on incident diabetes (4–6,8,9,31–33) and mortality of diabetic subjects (15,16,19,20). However, comparison with prior studies is difficult because the measurement scales and definitions of physical activity used differ. Most studies published on the subject (4,6,8,9,31,32) have found dose-response relations between physical activity and the incidence of diabetes. We similarly found a dose-response relation between physical activity and the mortality rates among nondiabetic and diabetic subjects. However, similar to a few other studies (5,33,34), we found that the degree of protection against diabetes was virtually the same in those with either vigorous or moderate physical activity levels. Additional analyses (data not shown) suggest that, in particular, the oldest subjects are responsible for this lack of a clear dose-response relation. Our data suggested that the effect of physical activity could be different in those aged >80 years. A possible explanation is that at this age, the lower physical activity group still at risk for diabetes is more selected due to higher risks of diabetes and mortality earlier in their life than those with higher levels of physical activity. However, the study was underpowered to detect any true difference in effect, and it is unlikely that such a difference would have affected our conclusions. Similar to Gregg et al. (20), the effect of moderate physical activity on the transition of diabetes to death did not reach statistical significance after full adjustment. Additional analyses showed that using a hazard ratio of 1.00 for this transition in the moderate active group would only strengthen our results.

A strength of this study is the use of data from a prospective, well-organized study, with long-term follow-up. Another advantage is that the glucose levels as well as other risk factors are measured at regular, biannual intervals. In our study, diagnosis of diabetes was based on glucose tests or the use of hypoglycaemic agents instead of self-report. In studies based on self-reported diabetes, many subjects with diabetes remain undiagnosed. In this study, there could be underdiagnosis only if subjects were not present at one or more exams (or had a false-negative test). As most subjects only missed one or a few subsequent exams, it becomes more a matter of delayed diagnosis than underdiagnosis.

Some limitations should be mentioned. The present study is an observational study and not a randomized trial. Consequently, bias may occur if diseases at baseline are responsible for inactivity (reverse causation) and if other factors confound the association between physical activity and the transition rates. There are two approaches to avoid reverse causation: exclusion of subjects with known diseases at baseline or adjustment for baseline diseases in the analysis. We used the second option, since we considered that by excluding subjects with diseases at baseline, there would be a selection of healthy people, and therefore the results would not be applicable to the whole population.

Residual confounding cannot entirely be ruled out, but as we examined the potential effect of a large set of confounders (age, sex, education, presence of diseases, marital status, smoking, exam of start follow-up, cholesterol, and family history of diabetes) and included those that affected the association between physical activity and the transitions, we do not expect that this would have biased our results.

Another limitation of our study is that in the Framingham Heart Study, physical activity levels were evaluated by self-report, which may introduce misclassification of exposure. However, this misclassification is likely to be nondifferential, which can only attenuate our results and fade a stronger association. We maximized the power of our study by using 12 years of follow-up. As a long period of follow-up reduces the effect of selection, but increases the risk of misclassification of exposure, the optimal follow-up time is unknown. Since it has been reported that levels and effects of physical activity change with time (35), we evaluated the effect of length of follow-up on the relation between physical activity and the transitions. These sensitivity analyses showed that with a follow-up period of 8 or 10 years instead of 12 years, our main conclusions did not change (data not shown).

The added value of this study is the combination of the observed effects of physical activity on incidence of diabetes and mortality in a large prospective study and the translation into the population health measures (life expectancy with and without diabetes). This study shows that physically inactive people have shorter lives, and, moreover, they live fewer years without diabetes and more or an equivalent number of years with diabetes compared with people with higher levels of physical activity. These results underline the public health importance of increasing physical activity levels in the population. Moreover, as Reunanen et al. (2) found that total costs of medications for people with diabetes were 3.5 times greater than those for nondiabetic control subjects, our findings are also important for the health care sector. When people live longer, but do not spend more years with diabetes, they do not put an extra demand on diabetes-related health care.

As far as the associations reflect causal relationships, our study suggests that if sedentary people could be stimulated to be at least moderately active, they could extend their lives and increase their lifetime spent without diabetes without spending more years with diabetes.

This study was supported by grants from the Netherlands Organization for Scientific Research (ZON-MW grants 014-91-054 and 904-68-493) and part of the Master of Science Program in Public Health at the Netherlands Institute for Health Sciences.

The authors acknowledge the Framingham Heart Study coordinators for access to the original dataset. The Framingham Study is conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with the Framingham Heart Study investigators. This manuscript has been reviewed by NHLBI for scientific content and consistency of data interpretation with previous Framingham Heart Study publications and significant comments have been incorporated prior submission for publication. We also thank Jan Barendregt for his assistance in making confidence intervals with @risk.