Physical fitness and activity as separate heart disease risk factors: a meta-analysis

Recognition of physical inactivity as a major public health problem was an important recent milestone in exercise science. The Centers for Disease Control and Prevention/American College of Sports Medicine (7), National Institutes of Health (4), U.S. Surgeon General (8), and American Heart Association (3) are among leading organizations that have published reports on physical activity and public health. A consensus recommendation derived from these reports is that all adults should accumulate at least 30 minutes of moderate intensity physical activity each day. An additional feature of these reports is that the recommendations are based on a presumed curvilinear, inverse dose-response relation between physical activity and health outcomes. The shape of this dose-response curve indicates that a given dose of physical activity will have relatively greater health benefits at the lower end than at the higher end of the physical activity spectrum. Dr. Williams disagrees with this conclusion and states that “This presumption is based largely on studies of cardiorespiratory fitness.” We have been involved in the development of the recent reports on physical activity and health cited above, and it is not true that the recommendations are “based largely on studies of cardiorespiratory fitness.” The various committees and groups making the recommendations considered evidence from pathophysiological investigations, randomized clinical trials, and prospective epidemiological studies of both physical activity and cardiorespiratory fitness in their reviews and deliberations. However, we certainly agree that studies on cardiorespiratory fitness and health outcomes were considered in making the recommendations. Dr. Williams’ fundamental approach to evaluate the dose-response relation between activity and health is to compare data from separate meta analyses of recent epidemiological studies on physical activity or cardiorespiratory fitness and the outcomes of coronary heart disease (CHD) or cardiovascular disease (CVD). We agree that it is important to investigate more thoroughly the shape of the dose-response association between physical activity and health outcomes, and compliment Dr. Williams for addressing this important and difficult issue. There are a few additional issues that should be considered when interpreting the results of the meta-analyses. One of the main points emphasized by Dr. Williams is that his Figures 2–4 illustrate that there is a precipitous drop in risk beyond the 20th or 25th percentile for the fitness studies and that there is a linear decline in risk for physical activity. It is important to remember that the data points plotted in these figures result from simulations and estimates from data published in the respective reports. There were widely differing methods used in the various studies to assess either fitness or activity; however, fitness tests yield ratio data and activity assessments generally provide categorical data. Dr. Williams refers to this issue in his discussion on how the samples “were partitioned into intervals.” He discounts this problem by assuming that the physical activity referent categories were defined broadly because of no trend in the interval. We question this assumption because for the studies that used categorical data it is impossible to evaluate any trend within an interval. In fact, even studies with presumed ratio scale data have limitations. For example, suppose that the physical activity measure is miles walked per day. The referent category might be <1 mile per day; however, this does not mean that all individuals in that category did no physical activity. It is unlikely that all of these individuals were confined to bed; therefore, there could be considerable variation in actual daily energy expenditure that would be totally missed because of limitations in the physical activity questionnaire. Depending on the activity questionnaire, it is possible to have a sizable proportion of the study population classified as sedentary yet still have much misclassification because of limitations in the questionnaire used. We are not aware of any research that defines a questionnaire’s sensitivity and specificity to classify participants correctly as active or sedentary. These limitations generally do not hold for cardiorespiratory fitness assessments. Cardiorespiratory fitness test data are typically expressed as V̇O2max or maximal METs attained, which are ratio scales. The reference category for these measures is not interpreted as an absence of fitness, and there is a large volume of data from cardiovascular fitness tests that is available to interpret reference categories. Few investigators present detailed analyses on dose-response issues. This may be because the principal focus of earlier studies has been on determining whether there was any significant association of activity or fitness to health rather than attempting to characterize the shape of a dose-response relation. However, a few investigators have presented data that explore the shape of the association between activity or fitness and health. Paffenbarger (6) and colleagues were the first to present detailed analyses on this issue. Their classic report in 1978 clearly shows an inverse, curvilinear association between total weekly kcal of physical activity and CHD, and subsequent reports from the same study confirm this early observation. Our own initial report on cardiorespiratory fitness and mortality, which was included in Williams’ meta-analysis, presented the primary analyses by categories of fitness (2). In these analyses, the least fit quintiles of women and men were used as reference categories, and the results for CVD mortality are as reported in his Figure 1. We also reported on the association of fitness to all-cause mortality, and the results were similar to those for CVD mortality. The upper limits of the referent category for these analyses, expressed as maximal METs attained on the exercise test, are for women—8.1 for those 20–39 yr, 7.5 for 40–49 yr, 6.5 for 50–59 yr, and 5.7 for 60+ yr; and for men—10.5, 9.9, 8.8, and 7.5 across the same age groups. In the 1989 report, we also included age-adjusted death rates by maximal METs (in single units) attained on the exercise test. The data for women are sparse, with only 43 deaths, and the pattern of death rates is unstable when spread across MET levels from 6 to 12+. However for men, where there were 240 deaths, the rates are more stable and show a curvilinear gradient across MET levels. This is evident at the low end of the spectrum. For example, the death rate for men with a maximal capacity of 6 METs is approximately 150 per 10,000 man-years of follow-up. The death rates for 7, 8, and 9 METs are approximately 70, 50, and 40 per 10,000 man-years, respectively. These death rates show a definite inverse association with cardiorespiratory fitness within the low fitness group. As noted above, 6 and 7 METs are within the low fitness category shown in Williams’ Figure 1 for the ACLS, even for the oldest men. Therefore, we do not believe that Dr. Williams’ assertion of no gradient within the low fit group and a precipitous drop after the 20–25th percentile is accurate. Dr. Williams takes issue with the notion that cardiorespiratory fitness tests may provide a better indication than physical activity questionnaires of a person’s habitual physical activity. He bases his assumption on test-retest correlations for physical activity and for correlations between activity and fitness measures (his Table 3). Dr. Williams interprets the correlations in Table 3 as “. . .cardiorespiratory fitness seems unlikely to be a more accurate, less biased estimate of physical activity than the activity measures themselves, for two reasons.” The two reasons cited are 1) the validity correlations (fitness vs questionnaire) should be higher than the test-retest correlations (stability reliability estimate); and 2) the validity correlations should be near 1.0 after the effects of measurement error have been removed (corrected for attenuation). These views are not congruent with classical measurement theory (1). A test-retest reliability estimate is the ratio of true score and total variance where true score variance is total variance minus error variance. The limitation of self-reported physical activity on questionnaires is not just measurement error, but also other sources of systematic variance. True score variance does not just measure individual differences in a person’s activity habits but also other unique systematic variance unrelated to physical activity behavior. A major source of bias is the scaling method used by these questionnaires. It is a rare instance when a questionnaire meets the minimal criterion of interval scaling. The equation for correcting for attenuation used and provided in Table 3 in Dr. Williams’ paper demonstrates a well-known principle that validity coefficients (fitness vs questionnaire) are a function of the reliability of both measures. The highest validity coefficient is always lower than the highest reliability coefficient. Correcting for attenuation eliminates the influence of measurement error providing a corrected correlation that is the maximum correlation possible between two variables if we assume perfect reliability. The true score variance of both measures would need to be identical to have a correlation near 1.0. Dr. Williams’ assertion that fitness correlations should be higher than the test-retest correlation is not consistent with test validation theory and common practice. Those who have used self-report physical activity questionnaires are well aware of the measurement limitations of these instruments. They simply are not perfect measures of the respondent’s true level of physical activity. In practice, validity coefficients do not exceed reliability coefficients. We think that the true correlation between fitness and habitual activity is considerably higher than is suggested by the data in Williams’ Table 3. The low to moderate correlations shown in the table are due in large part to the crude and imprecise estimates of activity provided by questionnaires, which lead to substantial misclassification into activity categories. We agree that many activity questionnaires have acceptable reliability (as shown in Table 3), but that their validity is less impressive. All of the activity questionnaires used for the data in Table 3 require recall of physical activity participation over prior days, weeks, or months. We believe that an ongoing recording of activity over time would provide a more accurate and valid measure of activity. We have published such data (5). Members of the Cooper Fitness Center enter into computer files specific information about the details of their exercise session that day. This ongoing electronic diary should provide a more accurate and valid record of physical activity than any recalled self-report of physical activity. Participants in our study are all professionals or executives who are largely sedentary in their daily lives apart from their planned exercise sessions. We compared exercise records of 107 women and 475 men over the 6 months before their completing a maximal exercise test on a treadmill. The correlations between their average weekly exercise records and exercise test tolerance ranged from 0.66 to 0.83 across groups of younger and older women and men. Thus, the association between exercise and fitness is strong, if comprehensive exercise data are available for the analysis. We continue to argue that cardiorespiratory fitness provides a more objective measure and a better indicator than self-report questionnaires of habitual physical activity. We agree with Dr. Williams that low levels of cardiorespiratory fitness suggest high risk for CHD and CVD. However, we do not agree with his implication in the first paragraph of his discussion that the association between low fitness and increased risk is more likely to be due to the presence of subclinical disease or genetic factors than to inactivity. In observational studies, it is never possible to completely rule out these factors as potential confounders of observed associations. This caveat also applies to activity studies as well as to fitness studies. In fact, subclinical disease is less likely to be a confounder in fitness studies than it is in activity studies. All fitness studies require direct observation and examination of study participants, whereas many of the activity studies are conducted entirely by questionnaire methods. For example, in our studies, all participants undergo a 3- to 5-h medical evaluation that includes a physical examination by a physician and multiple clinical and laboratory tests. Although even this detailed and thorough evaluation almost certainly misses some cases of subclinical disease, the likelihood of such an occurrence is surely much less than in activity studies that rely exclusively on self-reported data. Furthermore, we observe virtually identical patterns of association between fitness and mortality in our participants who are healthy at baseline and those who already have chronic disease. Dr. Williams’ final question “when is it appropriate to screen for and intervene upon low levels of physical fitness, irrespective of physical activity status?” is relevant and important. We do not advocate widespread assessment of cardiorespiratory fitness, although we believe that is reasonable to consider such testing as part of a periodic health examination. We think, and suspect that Dr. Williams agrees, that the public health emphasis should be on getting sedentary individuals to become more physically active. It is clear that some activity is better than none at all, and that within limitations, more is better than less. More precise physical activity recommendations must await additional observational and experimental studies designed to characterize more precisely the dose-response relation of physical activity to health. Additional important issues include a more complete evaluation of the role of exercise intensity and the extent to which the daily exercise dose can be fractionated, if at all. We applaud Dr. Williams’ important contributions to the study of exercise and health and look forward to his future work and to continuing our discussions.