Physical activity types, variety, and mortality: results from two prospective cohort studies

Study population

The Nurses' Health Study started in 1976 when 121 700 female registered nurses aged 30-55 years were enrolled. The Health Professionals Follow-Up Study began in 1986 when 51 529 male health professionals aged 40-75 years were recruited. In both cohorts, participants reported their personal characteristics, medical history, and lifestyle information on enrolment, and provided biennial updates by completing follow-up questionnaires. The cumulative response rate in both cohorts was >90%.

Our study included participants with complete physical activity information in 1986. To reduce the potential reverse causation influence, we applied a four year lag time between assessing physical activity and the time at risk of death. We also excluded participants with a diagnosis of diabetes, cardiovascular disease, cancer, respiratory disease, or neurological disease at baseline to alleviate potentially strong confounding by existing diseases. Hence we included 111 467 participants in the analysis of physical activity levels; 70 725 from the Nurses' Health Study and 40 742 from the Health Professionals Follow-Up Study. Analysis of physical activity variety was limited to 111 373 participants; 70 725 women and 40 648 men who reported engaging in any physical activity.

Assessment of physical activity

Detailed information on leisure time physical activity was first collected in 1986 and updated biennially. Up to 15 repeated assessments were available for each participant (median 13, interquartile range 9-14), based on the same question: “During the past year, what was your average time per week spent at each activity?”. In both cohorts, data on walking, jogging (<10 min/mile (1.6 km)), running (≥10 min/mile), bicycling (including stationary machine), lap swimming, rowing or callisthenics, and tennis, squash, or racquetball were recorded from 1986. Information about weight training or resistance exercises was added to the Nurses' Health Study in 2000 and to the Health Professionals Follow-Up Study in 1990. Lower intensity exercises (eg, yoga, stretching, and toning) were added to the Nurses' Health Study in 1992 and to the Health Professionals Follow-Up Study in 2010. Also, other vigorous activities (eg, lawn mowing) were included in the Nurses' Health Study from 1992 onwards. In the Health Professionals Follow-Up Study, moderate outdoor work (eg, yardwork and gardening) was asked from the beginning of 2004, and heavy outdoor work (eg, digging and chopping) from 1988. Participants were also asked: “How many flights of stairs (not individual steps) do you climb daily?”. The average weekly time spent on climbing flights of stairs was estimated based on their responses to this question, assuming each flight takes eight seconds to ascend.

Based on a compendium of physical activities,16 each activity was assigned a metabolic equivalent task (MET) score, which represents the metabolic rate for that specific activity divided by the resting metabolic rate. The MET hours/week for each physical activity was derived by multiplying the average time spent on the activity (in hours/week) by its MET score. Total physical activity was defined as the sum of specific MET hours/week across all physical activities in each questionnaire cycle. The reproducibility and validity of the self-administered physical activity questionnaires used in both cohorts have been reported previously.17–20 For example, the estimated correlations between the physical activity questionnaire and true level of total activity were 0.54 in women and 0.60 in men; for moderate-to-vigorous activity, the correlations were 0.60 and 0.69, respectively.

For individual physical activities, our pooled analyses included nine physical activities recorded in both cohorts: walking, jogging, running, bicycling, swimming, rowing or callisthenics, tennis, squash, or racquetball, climbing flights of stairs, and weight training or resistance exercises. Given the limited follow-up time for the lower intensity exercises in the Health Professionals Follow-Up Study, we examined its association with mortality only in the Nurses' Health Study. Other cohort specific activities were analysed in the Nurses' Health Study (other vigorous activities) or Health Professionals Follow-Up Study (moderate outdoor work and heavy outdoor work), respectively.

To measure the variety of physical activities, we constructed a score by summing the number of individual physical activities in which participants consistently engaged, counting each physical activity as one if it met the predefined threshold and zero otherwise. The thresholds were set at five flights/day for stair climbing and 20 min/week for other activities, and were converted into MET hours/week by multiplying the reported duration by the specific MET value assigned to each activity (online supplemental table 1). The maximum number of individual physical activities was 11 in the Nurses' Health Study and 13 in the Health Professionals Follow-Up Study. This score was derived from the updated measure of physical activity levels for each follow-up cycle and then cumulatively averaged to represent the long term physical activity variety.

Covariates

Information on age, ethnic group, weight, family history of diseases, medical history, smoking status, menopausal status, and postmenopausal hormone use was assessed and updated biennially. Body mass index was calculated, and diet and alcohol intake were assessed with validated semi-quantitative food frequency questionnaires every two or four years. The modified alternate healthy eating index, excluding the alcohol component, was calculated to reflect overall quality of the diet.21 To reduce potential reverse causality, we stopped updating dietary information when participants developed chronic diseases (including diabetes, cardiovascular disease, cancer, respiratory disease, and neurological disease) during follow-up. Cumulative averages of dietary intake were used to represent habitual intake. Social integration was measured with a seven item index, covering marital status, social network size, contact frequency, religious participation, and involvement in other social groups.22 23

Ascertainment of death

Deaths were identified by searches of the National Death Index or reports from participants' next-of-kin or postal service, and >97% of deaths were determined for both cohorts.24 25 The cause of death was determined by review of medical records and death certificates by the study physicians. We used the ICD-8 (international classification of diseases, eighth revision) codes in the Nurses' Health Study and ICD-9 (international classification of diseases, ninth revision) codes in the Health Professionals Follow-Up Study, which were widely used at the start of the cohorts to distinguish deaths from cardiovascular disease (ICD-8: 390-458, 795; ICD-9: 390-459, 798), cancer (ICD-8: 140-207; ICD-9: 140-208), respiratory disease (ICD-8 and ICD-9: 460-519), and other causes.

Statistical analysis

Person years was calculated from baseline to the date of death or end of follow-up (June 2018 for the Nurses' Health Study and January 2020 for the Health Professionals Follow-Up Study), whichever occurred first. During follow-up, missing data for individual physical activities and covariates were replaced with valid values from the preceding questionnaire for one cycle. Person years were not accrued during cycles with missing physical activity data; participants contributed person years only in cycles where physical activity data were available.

For covariates, the missing indicator method was used to deal with any remaining missing data. The proportions of missing observations were 1.6% for alcohol consumption, 1.6% for total energy intake, 1.7% for the modified alternative healthy eating index, 2.0% for baseline body mass index, 3.0% for smoking status, 5.1% for postmenopausal hormone use (among women), and 12.0% for the social integration index. In previous analyses conducted within our cohorts, the missing indicator method gave results that were largely consistent with those obtained from multiple imputation.26 To enhance the statistical power of the study, we pooled individual level data from both cohorts in the primary analyses, harmonising covariates to ensure consistent definitions and measurements across the two studies. Cohort specific associations were also examined as secondary analyses.

To minimise reverse causation bias, where individuals might reduce physical activity because of illness, we stopped updating physical activity if participants reported a diagnosis of diabetes, cardiovascular disease, cancer, respiratory disease, or neurological disease, and carried forward the last physical activity value before diagnosis for subsequent follow-up cycles. Repeated measures of physical activity levels were cumulatively averaged to represent long term engagement and to reduce within person random measurement error. The relation between the amount and variety of physical activity was examined with Spearman's correlation coefficients, adjusting for age. We categorised total and individual physical activity levels and physical activity variety scores according to their distributions within each follow-up cycle and cohort: total physical activity levels and physical activity variety scores were divided into five equal groups; levels of walking and climbing flights of stairs were divided into four equal groups; for other individual physical activities, because of the skewed distribution towards zero, those who did not participate in the specific activity were categorised as the reference group and the remaining participants were categorised into three equal groups.

Cox proportional hazards models with age as the time scale and grouped by calendar time and cohort were used to estimate hazard ratios and 95% confidence intervals (CIs). Potential confounders were selected a priori based on the literature13 27 28 and adjusted for in the models: ethnic group (white or non-white participants); family history of myocardial infarction or cancer (yes or no); baseline body mass index categories (<23.0, 23.0-24.9, 25.0-29.9, 30.0-34.9, or ≥35.0); postmenopausal hormone use in women (premenopausal, never, former, or current use); smoking status (never smoked, past smoker, currently smoke 1-14, 15-24, or ≥25 cigarettes/day); alcohol intake (0, 0.1-4.9, 5.0-9.9, 10.0-14.9, 15.0-29.9, or ≥30.0 g/day); total energy intake and modified alternate healthy eating index (in five equal groups); social integration index (in four equal groups); and baseline hypertension or hypercholesterolaemia (yes or no). Values for dietary intake, postmenopausal hormone use, smoking status, and social integration index were updated during follow-up. For the analysis of individual physical activities, we also adjusted for the total levels of all other physical activities except the type being studied (continuous). For physical activity variety, we further adjusted for total physical activity levels (continuous). Tests for trend were conducted by modelling the physical activity categories as ordinal variables. The proportional hazards assumption was examined with a likelihood ratio test comparing models with and without multiplicative interaction terms between age and categorical physical activity variables, and no violation was found for the main exposure. Also, we fitted two multivariable adjusted models: one with total physical activity levels (continuous) and the other with total physical activity levels and physical activity variety (categorical) and then used the likelihood ratio test to examine whether the model including physical activity variety had a better fit than that including total physical activity levels only. We also examined the joint associations of total physical activity levels and physical activity variety with mortality: participants were grouped into nine subgroups based on combinations of total physical activity levels (in three equal groups) and physical activity variety score (in three equal groups). Participants in the lowest group for both categories were considered the reference group. Interactions were tested by the likelihood ratio test comparing models with and without product terms between total physical activity level and physical activity variety categories.

To explore the dose-response relation between total and individual physical activity levels and mortality, we fitted restricted cubic spline regressions with four knots,29 adjusting for the same covariates as in the main analysis and by using the SAS macro % LGTPHCURV9. Data were truncated at the 99th centile to limit the influence of extreme values. For total physical activity, walking, and climbing flights of stairs, we set the fifth, 35th, 65th, and 95th centiles as the four knots. For other individual physical activities, because of the skewed distribution towards zero, we set the fifth, 35th, 65th, and 95th centiles among the non-zero values. The minimum values of physical activity levels were set as the reference for each spline curve. We used the likelihood ratio test for tests for non-linearity, comparing the model with only the linear term with the model with the linear and cubic spline terms.

To test whether our findings were influenced by stopping updating physical activity after an intermediate outcome, we reanalysed the data with continuously updated physical activity data. Because recent physical activity practices might affect mortality more than past practices,30 we examined the associations based on the simple updated physical activity levels. All statistical analyses were performed with SAS software, version 9.4 (SAS Institute). A two sided P value <0.05 was considered significant. We applied the Benjamini-Hochberg false discovery rate adjustment to account for multiple testing in the association analyses between total physical activity, individual physical activities, and physical activity variety, and mortality.

Patient and public involvement

We did not have the infrastructure, resources, funding, or time to involve the public in study design, interpretation of results, or publication.

Descriptive characteristics

In the pooled dataset, participants with higher total physical activity levels had a lower baseline prevalence of hypertension, hypercholesterolaemia, and body mass index, higher alcohol and energy intake, better diet quality, higher social integration and physical activity variety, and were less likely to smoke (table 1). Online supplemental table 2 shows the cohort specific characteristics of participants in the two studies. The Spearman coefficient between total physical activity levels and variety was 0.75 in the Nurses' Health Study and 0.70 in the Health Professionals Follow-Up Study. For individual physical activities, the Spearman coefficients were small to modest, except for the correlation between jogging and running which was relatively higher (0.42 in Nurses' Health Study and 0.47 in Health Professionals Follow-Up Study; online supplemental table 3). Walking was the most frequently engaged activity in both cohorts, and participants in the Health Professionals Follow-Up Study were more likely to jog and run than those in the Nurses' Health Study (online supplemental figure 1). For most individual physical activities except for swimming, participants in the highest group (group 4) had a lower body mass index and baseline prevalence of hypertension than those in the lowest group (group 1; online supplemental table 4).

Physical activity amount and mortality

During 2 431 318 person years of follow-up, 38 847 deaths were recorded, with 9901 from cardiovascular disease, 10 719 from cancer, and 3159 from respiratory disease. Total physical activity was associated with lower mortality (online supplemental table 5). The lower risk of death from cardiovascular disease, cancer, and respiratory disease largely plateaued after reaching about 20 MET hours/week of total physical activity, whereas the lower risk of death from other causes was weaker (all P<0.001 for non-linearity, figure 1).

Compared with those in the lowest group (group 1) for the specific physical activity, the multivariable adjusted hazard ratios for all cause mortality for participants in the highest group (group 4) were 0.83 (95% CI 0.80 to 0.85) for walking, 0.89 (0.85 to 0.94) for jogging, 0.87 (0.80 to 0.93) for running, 0.96 (0.93, 0.99) for bicycling, 1.01 (0.97, 1.05) for swimming, 0.85 (0.80 to 0.89) for tennis, squash, or racquetball, 0.90 (0.87 to 0.93) for climbing flights of stairs, 0.86 (0.84 to 0.89) for rowing or callisthenics, and 0.87 (0.82 to 0.91) for weight training or resistance exercises (P_trend<0.001 for all except swimming (P_trend=0.18); table 2).

Figure 2 shows that the attainable levels of energy expenditure varied substantially for different activities in our study population, ranging from 2.1 MET hours/week for climbing flights of stairs to >40 MET hours/week for walking. We found non-linear associations between various types of physical activities and all cause mortality (P=0.02 for non-linearity for swimming and P<0.001 for all other activities; figure 3). The reduction in the risk of all cause mortality seemed to plateau at about 0.75 MET hours/week for climbing flights of stairs, 5 MET hours/week for tennis, squash, or racquetball, 7.5 MET hours/week for walking, and 7.5 MET hours/week for weight training or resistance exercises. For swimming, bicycling, and jogging, inverse associations with all cause mortality were apparent up to about 2.5, 7.5 and 9 MET hours/week, respectively; beyond these thresholds, the associations were no longer significant. For running and rowing or callisthenics, a sharp reduction in risk was seen up to about 3 and 2.5 MET hours/week, respectively, followed by a continued, although slower, decline with higher activity levels.

Although the range of physical activity levels was generally broader in the Health Professionals Follow-Up Study than in the Nurses' Health Study, the associations for total and most individual physical activities were consistent in both direction and shape for the two cohorts within comparable physical activity ranges (online supplemental table 6 and figure 2), supporting the validity of the pooled analysis approach. However, an inverse association at higher levels of bicycling (ie, 7.5 MET hours/week) was found in the Health Professionals Follow-Up Study but not in the Nurses' Health Study.

The associations between walking, jogging, bicycling, and rowing or callisthenics and cause specific mortality were similar to those observed for all cause mortality, whereas the associations for other activities differed depending on the cause of death (table 2 and online supplemental figure 3). For cardiovascular disease and respiratory disease mortality, running, tennis, squash, or racquetball, climbing flights of stairs, and weight training or resistance exercises were significantly associated with lower risks, with these associations appearing more linear (all P>0.05 for non-linearity and P<0.05 for linearity; exact P values are shown in online supplemental figure 3). Also, swimming had a significant non-linear association with respiratory disease mortality (P=0.003 for non-linearity) only but was not associated with mortality from other specific causes. Most physical activities, except swimming, showed significant non-linear associations with cancer mortality (all P<0.05 for non-linearity).

For specific activities in the Nurses' Health Study, lower intensity exercise was associated with lower all cause mortality, especially for cancer mortality (both P_trend=0.001; online supplemental table 7). Also, other vigorous activity was associated with lower mortality from causes other than cancer (all P_trend<0.001, except for cancer mortality (P_trend=0.12)). For specific activities in the Health Professionals Follow-Up Study, heavy outdoor work showed an inverse association with respiratory disease mortality (P_trend=0.003).

Physical activity variety and mortality

The physical activity variety score was inversely associated with all cause mortality, and this association remained significant after adjusting for total physical activity (P_trend<0.001; table 3 and online supplemental table 8). Compared with the lowest group (group 1), the highest group (group 5) for the physical activity variety score was associated with a 19% lower all cause mortality (95% CI 0.78 to 0.85) and 13-41% lower mortality from cardiovascular disease, cancer, respiratory disease, and other causes. The model fit was improved by adding physical activity variety to the model with total physical activity levels and other covariates (P<0.001 for likelihood ratio test in the pooled analysis).

Amount and variety of physical activity

Compared with the reference group, participants ranked highest for both total physical activity level and variety had 21% lower mortality (95% CI 0.76 to 0.82; figure 4). With further adjustment of total physical activity levels within each group, in the highest group the hazard ratio decreased to 0.75 (95% CI 0.72 to 0.79; online supplemental table 9). Higher physical activity variety scores within each group of total physical activity level were associated with lower all cause mortality, but this trend was less consistent for specific causes (online supplemental table 10). We found no interactions between total physical activity level and physical activity variety score on mortality.

Sensitivity analyses

The results were largely similar when we continuously updated the physical activity data (online supplemental tables 11–13) or used the simple updated physical activity levels (online supplemental tables 14–16).

Principal findings

In this study, we found that the total physical activity and most individual physical activities were associated with lower mortality in a non-linear manner, suggesting the presence of a potential threshold for the beneficial effects of physical activity. Also, a simple physical activity variety score was associated with lower mortality, independent of total physical activity levels, suggesting that long term engagement in multiple types of physical activity may have additional health benefits on longevity.

Comparison with other studies

Our finding that total physical activity was linked to lower mortality aligns with previous studies.3 The possibility of distinct physiological effects of different physical activities has received increasing attention. Bicycling and running have different ventilatory responses, blood flow, skeletal muscle oxidative capacity, and central and peripheral innervation.31 Also, several studies found significant differences between aerobic exercises and resistance training. In a 26 week randomised trial in 160 older adults with obesity,8 the aerobic group showed greater increases in peak oxygen consumption but fewer improvements in strength compared with the resistance group, whereas the resistance group had smaller decreases in lean mass and bone mineral density. Another 26 week randomised trial of 124 healthy participants showed that endurance and interval training, rather than resistance training, increased telomerase activity and telomere length.32 A recent study indicated the advantage of aerobic exercises over resistance training in improving the risk profiles for cardiovascular disease in adults with overweight or obesity.7 Research evaluating the health effects of engagement in multiple commonly practised physical activity types, however, is scarce.33

Although few studies have evaluated various physical activities simultaneously, existing data suggest that walking,9 12 14 15 climbing stairs,14 aerobic exercises,11–14 and racquet sports11–13 are associated with lower mortality after accounting for other physical activities. Most studies, however, only examined the influence of any participation in specific activities with one measurement at baseline. Our study showed the non-linear relation between long term engagement in these physical activities and mortality. A meta-analysis reported that running or jogging was associated with mortality, independent of their duration,34 although we found that running, rather than jogging, was linearly associated with mortality from cardiovascular and respiratory disease. This observation may support the notion that cardiorespiratory fitness is more responsive to the intensity rather than the volume of physical activity.35 Our findings confirmed the potential benefits of weight training on longevity, as reported previously.36 Evidence linking bicycling to mortality is controversial. A meta-analysis of 17 prospective studies reported a linear relation between bicycling and all cause mortality, with a 9% lower risk for each increase in bicycling of 5 MET hours/week.37 Updated evidence, however, indicated either a non-linear relation9 or no association.14 15 The inability to differentiate between the purpose of bicycling (active commuting or recreational) or to capture variations during follow-up38 may contribute to these inconsistencies. Our study focused on long term leisure time bicycling and found a reduction in all cause mortality before reaching 7.5 MET hours/week (ie, 64 min/week), with the lowest risk found at about 2.5 MET hours/week. Given the downward trend in risk in the Health Professionals Follow-Up Study and the suggestive upward trend in the Nurses' Health Study, more studies are needed to confirm the dose-response relation between bicycling and mortality in men and women, and to clarify the underlying biological mechanisms.

We found that higher levels of swimming activities were not associated with a lower all cause mortality, adding to the varied findings in this area.10–14 Self-reported swimming duration, even when specified as lap swimming, may correspond to a wide range of actual energy expenditures because of variations in exercise intensity.39 For example, individuals may report similar swimming durations regardless of whether they swim vigorously or casually. This potential misclassification of true energy expenditure in swimming, particularly among those reporting longer swimming durations, may bias the observed associations towards the null. Therefore, further research on both length of time spent swimming and swimming intensity is needed to clarify their relation with mortality.

Little evidence exists about the potential health effects of physical activity variety. Several studies found that engagement in both aerobic exercises and resistance training enhanced physical function more effectively than either type alone.8 40 Findings among male athletes suggested participation in multiple collegiate sports was associated with higher levels of physical activity in later life than participation in one sport.41 These data imply a potential extra advantage of engagement in multiple physical activities. Our study provides longitudinal evidence indicating a link between greater variety in physical activities and longevity. Consistent with previous studies,42 we found that the health benefits of most individual physical activities plateaued at certain levels. Individuals who engage in multiple physical activities may maximise the health benefits by engaging with each individual activity within the levels of the beneficial threshold. The finding that a greater variety of physical activities predicted lower all cause mortality among individuals within each category of total physical activity level indicated that engaging in multiple types of physical activities may exert additional health benefits independent of total physical activity levels. Notably, our approach, focusing on physical activity variety, did not specify fixed combinations of physical activity types because individuals may choose different types of physical activity over time according to their personal preferences and physical condition. Given the potential heterogeneous associations between specific types of physical activity and cause specific mortality, however, future studies exploring potential synergistic effects of different physical activities may help to refine the current guidelines on the health benefits of physical activity.

Strengths and limitations of this study

The strengths of our study included the large sample sizes and long follow-up periods with repeated physical activity measurements, allowing us to capture long term practices of multiple physical activities.

The study had several limitations. Firstly, the physical activity data were self-reported, inevitably introducing measurement errors. The prospective study design, however, may mean that the errors were largely non-differential because determining the outcomes was unrelated to the measurement errors. Secondly, not all types of physical activity were assessed in the baseline questionnaires for both cohorts. Some activities (eg, weight training or resistance exercises, or outdoor work) were only collected in later questionnaire cycles or in one cohort. Although we restricted analyses for these activities to participants and questionnaire cycles with available data, this approach may have limited the sample sizes and comparability between the cohorts. Nevertheless, we maximised the use of the available information by examining associations between each physical activity type and mortality within the relevant cohort, or in pooled analyses where applicable, providing new insights into the dose-response relations of specific activities that were previously under-explored. Thirdly, because MET scores are assigned assuming active engagement, the lack of information on intensity may cause misclassification of true energy expenditures, particularly for activities such as swimming, which could bias the observed associations towards the null. Fourthly, despite efforts, such as applying lag analyses, the possibility of reverse causation cannot be eliminated. This limitation is particularly relevant for respiratory disease, which can have longstanding insidious symptoms that may appear long before diagnosis. Nevertheless, we found similar associations between physical activity and mortality other than for respiratory disease (online supplemental table 17). Fifthly, because only women participated in the Nurses' Health Study and only men participated in the Health Professionals Follow-Up Study, our cohort specific analyses were equivalent to sex specific analyses. Potential differences should be confirmed and further explored in other studies that include both men and women. Lastly, although potential misclassification may arise from missing assessments of physical activity during follow-up, this effect should be minimal, given that 83% of participants had complete physical activity data or no more than two cycles with missing data, and we calculated cumulative average physical activity levels based on all available cycles. Also, our participants were mainly white health professionals, which might limit the generalisability of our findings to other populations.

Conclusions

Our findings suggest that an active lifestyle and engagement in multiple physical activities may facilitate long term survival.