Physical activity trajectories and mortality: population based cohort study
BMJ 2019; 365 doi: https://doi.org/10.1136/bmj.l2323 (Published 26 June 2019)
Cite this as: BMJ 2019;365:l2323
Article
Related content
Metrics
Responses
Peer review
Alexander Mok, PhD student1, Kay-Tee Khaw, professor2, Robert Luben, head of bioinformatics2, Nick Wareham, professor1, Soren Brage, research programme leader1
Author affiliations
Correspondence to: S Brage soren.brage@mrc-epid.cam.ac.uk
Accepted 25 April 2019
Abstract
Objective To assess the prospective associations of baseline and long term trajectories of physical activity on mortality from all causes, cardiovascular disease, and cancer.
Design Population based cohort study.
Setting Adults from the general population in the UK.
Participants 14 599 men and women (aged 40 to 79) from the European Prospective Investigation into Cancer and Nutrition-Norfolk cohort, assessed at baseline (1993 to 1997) up to 2004 for lifestyle and other risk factors; then followed to 2016 for mortality (median of 12.5 years of follow-up, after the last exposure assessment).
Main exposure Physical activity energy expenditure (PAEE) derived from questionnaires, calibrated against combined movement and heart rate monitoring.
Main outcome measures Mortality from all causes, cardiovascular disease, and cancer. Multivariable proportional hazards regression models were adjusted for age, sex, sociodemographics, and changes in medical history, overall diet quality, body mass index, blood pressure, triglycerides, and cholesterol levels.
Results During 171 277 person years of follow-up, 3148 deaths occurred. Long term increases in PAEE were inversely associated with mortality, independent of baseline PAEE. For each 1 kJ/kg/day per year increase in PAEE (equivalent to a trajectory of being inactive at baseline and gradually, over five years, meeting the World Health Organization minimum physical activity guidelines of 150 minutes/week of moderate-intensity physical activity), hazard ratios were: 0.76 (95% confidence interval 0.71 to 0.82) for all cause mortality, 0.71 (0.62 to 0.82) for cardiovascular disease mortality, and 0.89 (0.79 to 0.99) for cancer mortality, adjusted for baseline PAEE, and established risk factors. Similar results were observed when analyses were stratified by medical history of cardiovascular disease and cancer. Joint analyses with baseline and trajectories of physical activity show that, compared with consistently inactive individuals, those with increasing physical activity trajectories over time experienced lower risks of mortality from all causes, with hazard ratios of 0.76 (0.65 to 0.88), 0.62 (0.53 to 0.72), and 0.58 (0.43 to 0.78) at low, medium, and high baseline physical activity, respectively. At the population level, meeting and maintaining at least the minimum physical activity recommendations would potentially prevent 46% of deaths associated with physical inactivity.
Conclusions Middle aged and older adults, including those with cardiovascular disease and cancer, can gain substantial longevity benefits by becoming more physically active, irrespective of past physical activity levels and established risk factors. Considerable population health impacts can be attained with consistent engagement in physical activity during mid to late life.
Introduction
Physical activity is associated with lower risks of all cause mortality, cardiovascular disease, and certain cancers.123 However, much of the epidemiology arises from observational studies assessing physical activity at a single point in time (at baseline), on subsequent mortality and chronic disease outcomes. From 1975 to 2016, over 90% of these epidemiological investigations on physical activity and mortality have used a single assessment of physical activity at baseline.4 Relating mortality risks to baseline physical activity levels does not account for within-person variation over the long term, potentially diluting the epidemiological associations. As physical activity behaviours are complex and vary over the life course,5 assessing within-person trajectories of physical activity over time would better characterise the association between physical activity and mortality.
Fewer studies have assessed physical activity trajectories over time and subsequent risks of mortality.67891011 Some of these investigations have only included small samples of older adults, in either men or women. Importantly, most studies were limited by crude categorisations of physical activity patterns, without exposure calibration against objective measures with established validity. Many studies also do not adequately account for concurrent changes in other lifestyle risk factors—such as overall diet quality and body mass index—which might potentially confound the association between physical activity and mortality. This is important, as some studies have shown that associations between physical activity and weight gain are weak or inconsistent, suggesting that being overweight or obese might instead predict physical inactivity rather than the reverse.1213 Previous investigations have also not quantified the population impact of different physical activity trajectories over time on mortality. We examined associations of baseline and long term trajectories of within-person changes in physical activity on all cause, cardiovascular disease, and cancer mortality in a population based cohort study and quantified the number of preventable deaths from the observed physical activity trajectories.
Methods
Study population
The data for this investigation were from the European Prospective Investigation into Cancer and Nutrition-Norfolk (EPIC-Norfolk) study, comprising a baseline assessment and three follow-up assessments. The EPIC-Norfolk study is a population based cohort study of 25 639 men and women aged 40 to 79, resident in Norfolk, UK, and recruited between 1993 to 1997 from community general practices as previously described.14
After the baseline clinic assessment (1993 to 1997), the first follow-up (postal questionnaire) was conducted between 1995 and 1997 at a mean of 1.7 (SD 0.1) years after baseline, the second follow-up (clinic visit) took place 3.6 (0.7) years after baseline, and the third follow-up (postal questionnaire) was initiated 7.6 (0.9) years after the baseline clinic visit. All participants with repeated measures of physical activity (at least baseline and final follow-up assessments) were included, resulting in an analytical sample of 14 599 men and women.
Assessment of physical activity
Habitual physical activity was assessed with a validated questionnaire, with a reference time frame of the past year.1516 The first question inquired about occupational physical activity, classified as five categories: unemployed, sedentary (eg, desk job), standing (eg, shop assistant, security guard), physical work (eg, plumber, nurse), and heavy manual work (eg, construction worker, bricklayer). The second open ended question asked about time spent (hours/week) on cycling, recreational activities, sports, or physical exercise, separately for winter and summer.
The validity of this instrument has previously been examined in an independent validation study, by using individually-calibrated combined movement and heart rate monitoring as the criterion method; physical activity energy expenditure (PAEE) increased through each of four ordinal categories of self reported physical activity comprising both occupational and leisure time physical activity.15 In this study, we disaggregated the index of total physical activity into its original two variables that were domain specific and conducted a calibration to PAEE using the validation dataset, in which the exact same instrument had been used (n=1747, omitting one study centre that had used a different instrument). Specifically, quasi-continuous and marginalised values of PAEE in units of kJ/kg/day were derived from three levels of occupational activity (unemployed or sedentary occupation; standing occupation; and physical or heavy manual occupation) and four levels of leisure time physical activity (none; 0.1 to 3.5 hours; 3.6 to 7 hours; and >7 hours per week). This regression procedure allows the domain specific levels of occupational and leisure time physical activity to have independent PAEE coefficients, while assigning a value of 0 kJ/kg/day to individuals with a sedentary (or no) occupation and reporting no leisure time physical activity (LTPA). The resulting calibration equation was: PAEE (kJ/kg/day) =0 (sedentary or no job) + 5.61 (standing job) + 7.63 (manual job) + 0 (no LTPA) + 3.59 (LTPA of 0.1 to 3.5 hours per week) + 7.17 (LTPA of 3.6 to 7 hours per week) + 11.26 (LTPA >7 hours per week).
Assessment of covariates
Information about participants’ lifestyle and clinical risk factors were obtained at both clinic visits, carried out by trained nurses at baseline and 3.6 years later. Information collected during clinic visits included: age; height; weight; blood pressure; habitual diet; alcohol intake (units consumed per week); smoking status (never, former, and current smokers); physical activity; social class (unemployed, non-skilled workers, semiskilled workers, skilled workers, managers, and professionals); education level (none, General Certificate of Education (GCE) Ordinary Level, GCE Advanced Level, bachelor’s degree, and above); and medical history of heart disease, stroke, cancer, diabetes, fractures (wrist, vertebral, and hip), asthma, and other chronic respiratory conditions (bronchitis and emphysema). Additionally, updated information on heart disease, stroke, and cancer up to the final physical activity assessment (third follow-up) were also collected by using data from hospital episode statistics. This is a database containing details of all admissions, including emergency department attendances and outpatient appointments at National Health Service hospitals in England. Non-fasting blood samples were collected and refrigerated at 4°C until transported within a week of sampling to be assayed for serum triglycerides, total cholesterol, and high density lipoprotein cholesterol by using standard enzymatic techniques. We derived low density lipoprotein cholesterol by using the Friedewald equation.17
We assessed habitual dietary intake during the previous year by using validated 130 item food-frequency questionnaires administered at baseline and at the second clinic visit. The validity of this food-frequency questionnaire for major foods and nutrients was previously assessed against 16 day weighed diet records, 24 hour recall, and selected biomarkers in a subsample of this cohort.1819 We created a comprehensive diet quality score for each participant, separately for baseline and at follow-up, incorporating eight dietary components known to influence health and the risk of chronic disease.20 The composite diet quality score included: wholegrains, refined grains, sweetened confectionery and beverages, fish, red and processed meat, fruit and vegetables, sodium, and the ratio of unsaturated to saturated fatty acids from dietary intakes. We created tertiles for each dietary component and then scored these as −1, 0, or 1, with the directionality depending on whether the food or nutrient was associated with health risks or benefits.20 Scores from the eight dietary components were summed into an overall diet quality score which ranged from −8 to 8, with higher values representing a healthier dietary pattern. We also collected updated information on body weight and height from the two postal assessments (first and third follow-up).
Mortality ascertainment
All participants were followed-up for mortality by the Office of National Statistics until the most recent censor date of 31 March 2016. Causes of death were confirmed by death certificates which were coded by nosologists according to ICD-9 (international classification of diseases, ninth revision) and ICD-10 (international classification of diseases, 10th revision). We defined cancer mortality and cardiovascular disease mortality by using codes ICD-9 140-208 or ICD-10 C00-C97 and ICD-9 400-438 or ICD-10 I10-I79, respectively.
Statistical analysis
We used Cox proportional hazards regression models to derive hazard ratios and 95% confidence intervals. Individuals contributed person time from the date of the last physical activity assessment (third follow-up) until the date of death or censoring. We used all available assessments of physical activity to better represent long term habitual physical activity and used linear regression against elapsed time to derive an overall physical activity trajectory (ΔPAEE) for each individual. We used the resulting coefficient of the calibrated ΔPAEE values in kJ/kg/day/year, together with baseline PAEE, as mutually-adjusted exposure variables in the Cox regression models.
We created categories reflecting approximate tertiles of both baseline PAEE and ΔPAEE to investigate joint effects of baseline and long term trajectories of physical activity. We defined the categories of baseline PAEE as: low (PAEE=0 kJ/kg/day), medium (0<PAEE<8.4 kJ/kg/day), and high (PAEE≥8.4 kJ/kg/day). We defined the categories of ΔPAEE over time as: decreasers (ΔPAEE≤−0.20 kJ/kg/day/year), maintainers (−0.20<ΔPAEE<0.20 kJ/kg/day/year), and increasers (ΔPAEE≥0.20 kJ/kg/day/year). We then created joint exposure categories by cross-classifying the three baseline by the three trajectory categories, resulting in eight categories. The reference group was individuals with consistently low physical activity (by definition, there would be no exposure category comprising individuals declining from no baseline physical activity). We estimated the potential number of preventable deaths at the population level in each joint exposure category, using the absolute difference in adjusted mortality rates between the reference group (consistently inactive) and each joint exposure category, multiplied by the person years observed in the corresponding joint exposure category. We derived adjusted mortality rates by using multivariable exponential regression, with covariates used in the most comprehensively adjusted analytical model.
In model 1 we adjusted for: general demographics (age, sex, socioeconomic status, education level, and smoking status), dietary factors (total energy intake, overall diet quality, alcohol consumption), and medical history (asthma, chronic respiratory conditions, bone fractures, diabetes, heart disease, stroke, and cancer). Age, energy and alcohol intake, and diet quality were continuous variables. In model 2 we accounted for changes in the above covariates by further inclusion of updated variables at the second clinic visit (3.6 years later), as well as updated status of cardiovascular disease and cancer from hospital episode statistics up until the final physical activity assessment. In model 3 we further accounted for changes in body mass index by including continuous values of body mass index at baseline and at the final physical activity assessment. In model 4 we accounted for changes in blood pressure and lipids by further including continuous values of systolic and diastolic blood pressure, serum triglycerides, low density lipoprotein cholesterol, and high density lipoprotein cholesterol at baseline and at the second clinic visit.
We used height and weight measurements from the baseline and second clinic visit to calibrate self reported height and weight provided by the postal questionnaires. Self reported values were multiplied by the ratio of mean clinically-measured values and self-reported values. We imputed missing values of covariates at follow-up by using regression on their baseline values. A complete case analysis was conducted as a sensitivity analysis. Reverse causation owing to undiagnosed disease was mitigated by excluding participants who died within one year of the final physical activity assessment (beginning of follow-up for mortality) in all analyses. Predefined subgroups were age, sex, clinically-defined cut points of body mass index, and history of cardiovascular disease and cancer. We performed additional sensitivity analyses by excluding individuals with any period-prevalent chronic diseases (heart disease, stroke, and cancer) up to the final physical activity assessment, as well as excluding deaths occurring within two years of the final physical activity assessment. All analyses were performed by using Stata SE version 14.2.
Patient and public involvement
Patients and members of the public were not formally involved in the design, analysis or interpretation of this study. Nonetheless, the research question in this article is of broad public health interest. The results of this study will be disseminated to study participants and the general public through the study websites, participant engagement events, seminars, and conferences.
Results
Study population
Among 14 599 participants with a mean baseline age of 58.0 (SD 8.8), followed for a median of 12.5 (interquartile range 11.9-13.2) years after the final physical activity assessment, there were 3148 deaths (950 from cardiovascular disease and 1091 from cancer) during 171 277 person years of follow-up. Table 1 shows the study population characteristics at the four assessment time points. On average, dietary factors such as total energy intake, alcohol consumption, and overall diet quality were similar at baseline and at the second clinic visit. The prevalence of diabetes, cardiovascular disease, cancer, and respiratory conditions increased over time. From baseline to the final follow-up assessment, mean body mass index increased from 26.1 kg/m2 to 26.7 kg/m2, and mean PAEE declined by 17% from 5.9 kJ/kg/day to 4.9 kJ/kg/day. The Pearson correlation coefficients were r=0.57 between PAEE at baseline and 1.7 years later; and r=0.45 between PAEE at baseline and 7.6 years later (final physical activity assessment).
Table 1 Study population characteristics at baseline and follow-up assessments. Values are means (SD) unless stated otherwise
View popupView inline
Associations of baseline and trajectories of physical activity with mortality
Table 2 shows that for each 1 kJ/kg/day/year increase in PAEE over time (ΔPAEE), the hazard ratios were: 0.78 (95% confidence interval 0.73 to 0.84) for all cause mortality, 0.75 (0.66 to 0.86) for cardiovascular disease mortality, and 0.88 (0.79 to 0.98) for cancer mortality (model 1). Progressive adjustments for time-updated covariates (model 2), changes in body mass index (model 3), and changes in blood pressure and blood lipids (model 4) did not attenuate the strength of the associations. In all models, baseline PAEE was also independently associated with lower mortality; for each 10 kJ/kg/day difference between individuals, hazard ratios were 0.70 (95% confidence interval 0.64 to 0.78) for all cause mortality, 0.69 (0.57 to 0.83) for cardiovascular disease mortality, and 0.83 (0.70 to 0.98) for cancer mortality (table 2, model 4). There was no evidence of an interaction between baseline PAEE and ΔPAEE for all mortality outcomes (P>0.6 from likelihood-ratio tests). The effect of PAEE averaged across all assessments on overall mortality, was 0.70 (0.62 to 0.78) for each 10 kJ/kg/day difference between individuals. For single time point exposure assessments, the inverse association of PAEE with mortality at the most recent assessment was stronger than that for baseline PAEE; hazard ratios were 0.68 (0.62 to 0.75) and 0.87 (0.80 to 0.94) for each 10 kJ/kg/day difference, respectively.
Table 2 Associations of mutually-adjusted baseline physical activity energy expenditure (PAEE) and trajectories of physical activity (ΔPAEE) with mortality. Values are hazard ratios (95% confidence intervals) unless stated otherwise
View popupView inline
Sensitivity analyses excluding individuals with any period-prevalent heart disease, stroke, and cancer occurring up to the final physical activity assessment, as well as any deaths occurring within two years of this final assessment, showed similar associations with mortality for baseline PAEE and ΔPAEE, hazard ratios of 0.72 (95% confidence interval 0.63 to 0.81) for each 10 kJ/kg/day difference and 0.78 (0.71 to 0.86) for each 1 kJ/kg/day/year difference, respectively. Sensitivity analysis that used complete cases did not materially change the strength of associations (attenuation <5% for ΔPAEE estimates for all outcomes), but it attenuated the statistical significance for cancer mortality (supplementary table 1). Adjustments for occupational physical activity categories (sedentary, standing, physical, and heavy manual jobs) (supplementary table 2) slightly strengthened the association of ΔPAEE with cardiovascular disease mortality and attenuated the association with cancer mortality, whereas associations for baseline PAEE became stronger for both these outcomes.
Stratified analyses
Figure 1 shows that, based on the most comprehensively-adjusted analytical model (model 4), significant inverse associations for baseline PAEE and ΔPAEE with all cause mortality persisted in all subgroups of age, sex, adiposity, and chronic disease status. Although tests for interaction were not statistically significant for any subgroup, the benefit of baseline PAEE on all cause mortality tended to be stronger in women (hazard ratio 0.63, 95% confidence interval 0.53 to 0.74) than men (0.76, 0.66 to 0.87; P=0.08). Baseline PAEE and ΔPAEE were not associated with cardiovascular disease mortality in individuals with obesity. In stratified analyses for cancer mortality, the longevity benefits of both baseline PAEE and ΔPAEE were only significant in older adults.
Fig 1
Fig 1
Associations of baseline and long term trajectories of physical activity energy expenditure (PAEE) with all cause, cardiovascular disease, and cancer mortality, stratified by age group, sex, body mass index (BMI), and disease status. Hazard ratios are mutually adjusted for both baseline PAEE and ΔPAEE, and are based on the most comprehensively adjusted model for changes in covariates, including medical history, diet quality, body mass index, blood pressure, and lipids (model 4 from table 2).
Download figure Open in new tab Download powerpoint
Joint associations of baseline and trajectories of physical activity with mortality
Figure 2 shows that compared with individuals who were consistently inactive (low-maintainers), individuals with medium and high baseline physical activity who maintained these levels (medium-maintainers and high-maintainers) had significantly lower risks of all cause mortality, 28% and 33% respectively. Individuals with increasing physical activity trajectories experienced additional longevity benefits, including those with low baseline activity, as well as those with already high levels of baseline physical activity. Dose-response gradients were observed within and between strata of baseline physical activity levels. Within strata of low, medium, and high baseline physical activity, the risk of mortality was lower across ordinally increasing trajectories of: decreasers, maintainers, and increasers. Between strata of baseline physical activity, the risk of mortality decreased by 24% for low-increasers, 38% for medium-increasers, and 42% for high-increasers. Medium-decreasers and high-decreasers had 10% and 20% lower risks of mortality, respectively, compared with the reference group of low-maintainers.
Fig 2
Fig 2
Joint associations of baseline and trajectories of physical activity energy expenditure (PAEE) with all cause mortality. Hazard ratios (HR) are based on the most comprehensively adjusted model for age, sex, sociodemographics, and changes in medical history, diet quality, body mass index, blood pressure, and lipids (model 4 from table 2). Adjusted mortality rates are expressed per 100 000 person years. WHO=World Health Organization.
Download figure Open in new tab Download powerpoint
Estimation of population impact
Figure 2 shows that if the entire cohort remained inactive over time, an additional 24% of deaths (678 more than the observed 2840 deaths) would have potentially occurred. At the population level, the greatest number of potential deaths averted were in the medium-increasers and medium-maintainers, preventing 169 (25%) and 143 (21%) of the deaths associated with physical inactivity, respectively. All physical activity trajectories that culminated with meeting at least the minimum physical activity guidelines (equivalent to 5 kJ/kg/day) could potentially prevent 93% of the deaths associated with physical inactivity at the population level.
Discussion
In this prospective cohort study with repeated assessments, we found protective associations for increasing physical activity trajectories against mortality from all causes, cardiovascular disease, and cancer, irrespective of past physical activity levels. These associations were also independent of levels and changes in several established risk factors such as overall diet quality, body mass index, medical history, blood pressure, triglycerides, and cholesterol. Both higher physical activity levels at baseline and increasing trajectories over time were protective against mortality. Notably, the strength of associations was similar in individuals with and without pre-existing cardiovascular disease and cancer. These results are encouraging, not least for middle aged and older adults with cardiovascular disease and cancer, who can still gain substantial longevity benefits by becoming more active, lending further support to the broad public health benefits of physical activity.
Independent and joint effects of baseline and trajectories of physical activity
The absence of an interaction between baseline physical activity levels and long term trajectories of physical activity on the risk of mortality suggests that the relative longevity benefit of increasing physical activity is consistent, irrespective of baseline levels. Increasing PAEE by 1 kJ/kg/day per year—equivalent to a trajectory of being inactive at baseline and then subsequently increasing physical activity to 5 and 10 kJ/kg/day, five and 10 years later, respectively—was associated with a 24% lower risk of all cause mortality. This gain in longevity from increasing physical activity over time, is in addition to the benefits already accrued from baseline physical activity, such as a 30% lower risk of mortality for a between-individual difference of 10 kJ/kg/day. For reference, 5 kJ/kg/day corresponds to the World Health Organization minimum physical activity guidelines of 150 minutes per week of moderate-intensity physical activity, and 10 kJ/kg/day corresponds to the WHO recommendations of 300 minutes per week of moderate-intensity physical activity for additional health benefits. Figure 3 shows how these levels of physical activity can be achieved in any number of ways during leisure time and at work, with the required duration depending on relative intensities of the activities undertaken.
Fig 3
Fig 3
Physical activity energy expenditure (PAEE) of common activities performed during leisure time and at work. MET=metabolic equivalent of task. WHO=World Health Organization
Download figure Open in new tab Download powerpoint
The joint analyses of physical activity trajectories beginning from different baseline levels showed that adults who were already meeting at least the minimum physical activity recommendations (150 minutes per week of moderate physical activity), experience substantial longevity benefits by either maintaining or further increasing physical activity levels. This is evidenced by medium-maintainers and high-maintainers experiencing 28% and 33% lower risks of mortality, with an additional ~10% lower risk for increasers in both these baseline groups.
Adults already meeting the equivalent of the higher WHO physical activity recommendations (300 minutes per week of moderate physical activity) still gain further longevity benefits by increasing physical activity levels to over 14 kJ/kg/day. This energy expenditure corresponds to approximately three times the recommended minimum (equivalent to 450 minutes per week of moderate physical activity).
The joint analyses also revealed that some, but not all, of the longevity benefits from past physical activity levels are lost when previously active individuals decrease their activity levels. Compared with the consistently inactive, medium-decreasers and high-decreasers experienced 10% and 20% lower risks of mortality, respectively. However, these effects appear modest, compared with the 28% and 33% lower risks of mortality for medium-maintainers and high-maintainers, respectively. The low-increasers also experienced slightly lower risks of mortality than the high-decreasers (24% v 20% lower than the consistently-inactive reference group, respectively) but both had higher risks of mortality than the medium-maintainers (28% lower risk), despite all these three groups ending up at approximately the same physical activity level at the final exposure assessment. There might be several explanations for this, including: the relative importance of past versus more recent physical activity; differential factors that might have caused these specific physical activity trajectories in the first place, beyond differences in period-prevalent cardiovascular disease or cancer; and the degree to which physical activity levels were maintained or continued to change beyond the last exposure assessment until the date of final censoring.
Population impact
Although mortality benefits were greatest in the high-increasers (hazard ratio of 0.58), the fraction of potential deaths averted at the population level were greatest for the medium-increasers (25%) and the medium-maintainers (21%), in contrast with 10% for the high-increasers. This is owing to the combination of a moderately strong aetiological association (hazard ratio of 0.72) and a greater prevalence of medium-maintainers (23 032 person years, 15% of total person years), compared with high-increasers (6988 person years, 4% of total). All physical activity trajectories during middle to late adulthood that culminate with meeting at least the minimum physical activity guidelines could potentially prevent 93% of deaths attributable to physical inactivity. The last 7% were the 48 deaths potentially prevented by the medium-decreasers, who as a group did not meet the minimum physical activity guidelines at the final exposure assessment. Had this group maintained their baseline physical activity levels, an additional 136 deaths (nearly three times as many) may have been prevented. Comparatively fewer, yet still an extra 119 deaths (twice as many) could have been prevented if high-decreasers had maintained their baseline physical activity levels. These two groups with declining physical activity levels were also the most prevalent trajectories in the cohort. Thus, in addition to shifting the population towards meeting the minimum physical activity recommendations, public health efforts should also focus on the maintenance of physical activity levels, specifically preventing declines over middle and late life. The WHO minimum guidelines of 150 minutes per week of moderate-intensity physical activity appears to be a realistic public health target, given that these levels were observed to be broadly attainable at the population level. Individuals with existing chronic conditions such as cardiovascular disease and cancer—for whom our study has also shown to gain longevity benefits—might choose to engage in commensurably lower-intensity activities but for a longer duration (fig 3). Further research is, however, needed to specifically ascertain the health benefits of lower-intensity physical activity in both healthy individuals and those with major chronic diseases.21
Comparisons with existing studies
This study also showed that the longevity benefits of increasing physical activity are independent of intermediary changes in several established risk factors, including body mass index, blood pressure, triglycerides, and cholesterol. These results are interesting, relative to other studies that have shown considerable attenuation of the strength of associations after adjustment for similar cardiometabolic biomarkers.222 In our study, it is somewhat surprising that the inverse associations of physical activity with cardiovascular disease mortality were not attenuated (but rather strengthened) after adjusting for established cardiometabolic risk factors. These findings support research into other potential mechanisms, including vascular function,23 novel lipids,24 and autonomic nervous system activity,25 through which physical activity might protect against cardiovascular disease. In our study, the protective associations of physical activity were stronger for cardiovascular disease mortality than for cancer mortality, suggesting that longevity benefits were primarily driven through the prevention of cardiovascular-related deaths. Adjustment for occupational physical activity strengthened the inverse associations with cardiovascular disease mortality for both baseline PAEE and ΔPAEE; this was also the case for the association between baseline PAEE and cancer mortality, but the association between ΔPAEE and cancer mortality was attenuated. The existing body of evidence, from a meta-analysis of nine international cohort studies also reported stronger inverse associations for cardiovascular disease mortality, compared with cancer mortality.2627 The weaker associations with cancer mortality might reflect the notion that cancers are a collection of neoplastic diseases, which might be aetiologically diverse and characterised by separate pathophysiologies.1
Our results for the association of baseline physical activity with mortality were broadly similar to those reported in the literature, although our estimates have accounted for changes in physical activity over time, which to some degree would correct for regression dilution bias.28 In a pooled analysis of data from high income countries within the Prospective Urban Rural Epidemiologic (PURE) study,29 medium baseline physical activity (150 to 750 minutes per week of moderate-intensity physical activity) was associated with a 31% lower risk of mortality. This is similar to our estimates of a 30% lower risk of mortality for a 10 kJ/kg/day difference between individuals (equivalent to 300 minutes per week of moderate-intensity physical activity).
Another pooled analysis examining the dose-response relation between baseline leisure time physical activity and mortality also reported lower mortality risks of between 31% to 37% at a comparable volume of physical activity.30 Comparisons with previous studies, examining specifically the changes and patterns of physical activity over time on mortality, are difficult owing to methodological and analytical heterogeneity between studies, precluding the synthesis of published results using meta-analytic methods. There was considerable variation in the operational definitions of the “changes” in physical activity over time. Some classified “changes” as increases and decreases, compared with unchanged physical activity irrespective of baseline levels31; others grouped varying activity levels over time as “mixed patterns,”4 potentially obscuring the benefits for individuals who improved physical activity levels over time; yet others used a reference group of the “consistently-active.”69 Furthermore, the time periods for studying these physical activity trajectories were also variable, with some studies examining short term changes within one to two years,3233 and others examining changes over 10 years.61011 Nonetheless, the relative risks of our high-maintainer and medium-maintainer groups (with hazard ratios of 0.67 and 0.72, respectively) were broadly in the ranges of the consistently-active groups reported in previous studies.7811 Future work examining physical activity trajectories over time on health outcomes could consider pooling of individual-level harmonised data from compatible studies with repeated follow-up assessments, ideally combined with external calibration; this would enable standardisation of exposure definitions and analytical approaches.
Strengths and limitations of the study
On balance, we present a comprehensive analysis, examining longitudinal physical activity trajectories in a large cohort with long follow-up for mortality, and quantified the population health impact from different physical activity trajectories. To overcome limitations in the majority of studies which have predominantly examined mortality associations with physical activity assessed at a single time point, we incorporated repeated measures of physical activity calibrated against objective measurements of individually-calibrated combined movement and heart rate monitoring. The use of longitudinal, within-individual trajectories of physical activity over time also precludes any confounding by time-invariant factors such as genetics. Our approach offers a stronger operationalisation of physical activity exposures, representing a method which can be used in future longitudinal studies investigating the associations between physical activity and subsequent health outcomes. Our study showed robust protective associations between physical activity and mortality, even after controlling for established risk factors, such as overall diet quality, body mass index, blood pressure, triglycerides, and cholesterol.
Some limitations of our study are that the analytical sample comprised of individuals who were available for follow-up approximately a decade after initial recruitment. Thus, a healthy cohort effect cannot be excluded. This, however, would only serve to render our findings more conservative. As the study was observational, residual confounding owing to unmeasured factors might still be possible. However, it would be virtually impossible to study the effects of habitual physical activity on mortality in a randomised controlled trial, and the observational nature of this study broadly shows the attainable longevity benefits of physical activity trajectories observed in the real world.
Conclusion
We showed that middle aged and older adults, including those with cardiovascular disease and cancer, stand to gain substantial longevity benefits by becoming more physically active, irrespective of past physical activity levels and established risk factors—including overall diet quality, body mass index, blood pressure, triglycerides, and cholesterol. Maintaining or increasing physical activity levels from a baseline equivalent to meeting the minimum public health recommendations has the greatest population health impact, with these trajectories being responsible for preventing nearly one in two deaths associated with physical inactivity. In addition to shifting the population towards meeting the minimum physical activity recommendations, public health efforts should also focus on the maintenance of physical activity levels, specifically preventing declines over mid to late life.
What is already known on this topic
Physical activity assessed at a single time point is associated with lower risks of mortality from all causes, cardiovascular disease, and cancer
Fewer studies have examined long term changes in physical activity and quantified the population health impact of different activity trajectories
What this study adds
Middle aged and older adults, including those with cardiovascular disease and cancer, stand to gain substantial longevity benefits by becoming more physically active, regardless of past activity levels, and changes in established risk factors, including overall diet quality, bodyweight, blood pressure, triglycerides, and cholesterol
At the population level, meeting and maintaining at least the minimum public health recommendations (150 minutes per week of moderate-intensity physical activity) would potentially prevent 46% of deaths associated with physical inactivity
Public health strategies should shift the population towards meeting the minimum recommendations, and importantly, focus on preventing declines in physical activity during middle and late life
Acknowledgments
We are grateful to the EPIC-Norfolk study participants for their voluntary contribution towards public health research, and for the support of EPIC functional teams (study coordination, field epidemiology team, IT, and data management). We also thank Stephen Sharp (Senior Statistician, MRC Epidemiology Unit, University of Cambridge) and Charles Matthews (Senior Investigator, Division of Cancer Epidemiology and Genetics, Metabolic Epidemiology Branch, National Cancer Institute, USA) for helpful comments on statistical analyses and the contextualisation of results. The authors assume full responsibility for the analysis and interpretation of the data in this study.
Footnotes
Contributors: KTK, RL, and NW contributed to the design of the EPIC-Norfolk study. AM and SB conceptualised the design of the present analysis and analysed the data. AM wrote the first draft of the manuscript. All authors had full access to the data in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis. AM and SB and are the guarantors. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Funding: The EPIC-Norfolk study is supported by programme grants from the Medical Research Council and Cancer Research UK with additional support from the Stroke Association, British Heart Foundation, Department of Health, Food Standards Agency, and the Wellcome Trust. AM was supported by the National Science Scholarship from Singapore, A*STAR (Agency for Science, Technology and Research). The work of NW and SB was funded by the Medical Research Council UK (MC_UU_12015/1 and MC_UU_12015/3). The funders had no role in the study design; the collection, analysis, and interpretation of data; the writing of the manuscript; or the decision to submit the article for publication.
Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: no support from any additional organisations for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work.
Ethical approval: The study was approved by the Norfolk District Health Authority Ethics Committee and adhered to the World Medical Association’s Declaration of Helsinki.
Patient consent: All participants gave written informed consent before enrolment in the study.
Data sharing: No additional data are available.
The lead author (AM) affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.
This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.
References
↵Moore SC, Lee I-M, Weiderpass E, et al. Association of Leisure-Time Physical Activity With Risk of 26 Types of Cancer in 1.44 Million Adults. JAMA Intern Med2016;176:816-25. doi:10.1001/jamainternmed.2016.1548. pmid:27183032CrossRefPubMedGoogle Scholar
↵Mora S, Cook N, Buring JE, Ridker PM, Lee I-M. Physical activity and reduced risk of cardiovascular events: potential mediating mechanisms. Circulation2007;116:2110-8. doi:10.1161/CIRCULATIONAHA.107.729939. pmid:17967770Abstract/FREE Full TextGoogle Scholar
↵Lee I-M, Shiroma EJ, Lobelo F, Puska P, Blair SN, Katzmarzyk PT, Lancet Physical Activity Series Working Group. Effect of physical inactivity on major non--communicable diseases worldwide: an analysis of burden of disease and life expectancy. Lancet2012;380:219-29. doi:10.1016/S0140-6736(12)61031-9. pmid:22818936CrossRefPubMedWeb of ScienceGoogle Scholar
↵Bauman AE, Grunseit AC, Rangul V, Heitmann BL. Physical activity, obesity and mortality: does pattern of physical activity have stronger epidemiological associations?BMC Public Health2017;17:788. doi:10.1186/s12889-017-4806-6. pmid:28982371CrossRefPubMedGoogle Scholar
↵Barengo NC, Nissinen A, Tuomilehto J, Pekkarinen H. Twenty-five-year trends in physical activity of 30- to 59-year-old populations in eastern Finland. Med Sci Sports Exerc2002;34:1302-7. doi:10.1097/00005768-200208000-00011. pmid:12165685CrossRefPubMedWeb of ScienceGoogle Scholar
↵Byberg L, Melhus H, Gedeborg R, et al. Total mortality after changes in leisure time physical activity in 50 year old men: 35 year follow-up of population based cohort. BMJ2009;338:b688. doi:10.1136/bmj.b688. pmid:19264819Abstract/FREE Full TextGoogle Scholar
↵Gregg EW, Cauley JA, Stone K, et al., Study of Osteoporotic Fractures Research Group. Relationship of changes in physical activity and mortality among older women. JAMA2003;289:2379-86. doi:10.1001/jama.289.18.2379. pmid:12746361CrossRefPubMedWeb of ScienceGoogle Scholar
↵Schnohr P, Scharling H, Jensen JS. Changes in leisure-time physical activity and risk of death: an observational study of 7,000 men and women. Am J Epidemiol2003;158:639-44. doi:10.1093/aje/kwg207. pmid:14507599CrossRefPubMedWeb of ScienceGoogle Scholar
↵Bijnen FC, Feskens EJ, Caspersen CJ, Nagelkerke N, Mosterd WL, Kromhout D. Baseline and previous physical activity in relation to mortality in elderly men: the Zutphen Elderly Study. Am J Epidemiol1999;150:1289-96. doi:10.1093/oxfordjournals.aje.a009960. pmid:10604771CrossRefPubMedGoogle Scholar
↵Wannamethee SG, Shaper AG, Walker M. Changes in physical activity, mortality, and incidence of coronary heart disease in older men. Lancet1998;351:1603-8. doi:10.1016/S0140-6736(97)12355-8. pmid:9620713CrossRefPubMedWeb of ScienceGoogle Scholar
↵Paffenbarger RS Jr., Hyde RT, Wing AL, Lee IM, Jung DL, Kampert JB. The association of changes in physical-activity level and other lifestyle characteristics with mortality among men. N Engl J Med1993;328:538-45. doi:10.1056/NEJM199302253280804. pmid:8426621CrossRefPubMedWeb of ScienceGoogle Scholar
↵Pedisic Z, Grunseit A, Ding D, et al. High sitting time or obesity: Which came first? Bidirectional association in a longitudinal study of 31,787 Australian adults. Obesity (Silver Spring)2014;22:2126-30. doi:10.1002/oby.20817. pmid:24943057CrossRefPubMedGoogle Scholar
↵Golubic R, Ekelund U, Wijndaele K, et al. Rate of weight gain predicts change in physical activity levels: a longitudinal analysis of the EPIC-Norfolk cohort. Int J Obes (Lond)2013;37:404-9. doi:10.1038/ijo.2012.58. pmid:22531093CrossRefPubMedGoogle Scholar
↵Day N, Oakes S, Luben R, et al. EPIC-Norfolk: study design and characteristics of the cohort. European Prospective Investigation of Cancer. Br J Cancer1999;80(Suppl 1):95-103.pmid:10466767PubMedGoogle Scholar
↵Peters T, Brage S, Westgate K, et al., InterAct Consortium. Validity of a short questionnaire to assess physical activity in 10 European countries. Eur J Epidemiol2012;27:15-25. doi:10.1007/s10654-011-9625-y. pmid:22089423CrossRefPubMedGoogle Scholar
↵Khaw K-T, Jakes R, Bingham S, et al. Work and leisure time physical activity assessed using a simple, pragmatic, validated questionnaire and incident cardiovascular disease and all-cause mortality in men and women: The European Prospective Investigation into Cancer in Norfolk prospective population study. Int J Epidemiol2006;35:1034-43. doi:10.1093/ije/dyl079. pmid:16709620CrossRefPubMedWeb of ScienceGoogle Scholar
↵Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem1972;18:499-502.pmid:4337382Abstract/FREE Full TextGoogle Scholar
↵Bingham SA, Gill C, Welch A, et al. Validation of dietary assessment methods in the UK arm of EPIC using weighed records, and 24-hour urinary nitrogen and potassium and serum vitamin C and carotenoids as biomarkers. Int J Epidemiol1997;26(Suppl 1):S137-51. doi:10.1093/ije/26.suppl_1.S137. pmid:9126542CrossRefPubMedWeb of ScienceGoogle Scholar
↵Bingham SA, Welch AA, McTaggart A, et al. Nutritional methods in the European Prospective Investigation of Cancer in Norfolk. Public Health Nutr2001;4:847-58. doi:10.1079/PHN2000102. pmid:11415493CrossRefPubMedWeb of ScienceGoogle Scholar
↵Micha R, Peñalvo JL, Cudhea F, Imamura F, Rehm CD, Mozaffarian D. Association Between Dietary Factors and Mortality From Heart Disease, Stroke, and Type 2 Diabetes in the United States. JAMA2017;317:912-24. doi:10.1001/jama.2017.0947. pmid:28267855CrossRefPubMedGoogle Scholar
↵Chastin SFM, De Craemer M, De Cocker K, et al. How does light-intensity physical activity associate with adult cardiometabolic health and mortality? Systematic review with meta-analysis of experimental and observational studies. Br J Sports Med2019;53:370-6. doi:10.1136/bjsports-2017-097563Abstract/FREE Full TextGoogle Scholar
↵Sattelmair J, Pertman J, Ding EL, Kohl HW 3rd., Haskell W, Lee IM. Dose response between physical activity and risk of coronary heart disease: a meta-analysis. Circulation2011;124:789-95. doi:10.1161/CIRCULATIONAHA.110.010710. pmid:21810663Abstract/FREE Full TextGoogle Scholar
↵Hawkins M, Gabriel KP, Cooper J, Storti KL, Sutton-Tyrrell K, Kriska A. The impact of change in physical activity on change in arterial stiffness in overweight or obese sedentary young adults. Vasc Med2014;19:257-63. doi:10.1177/1358863X14536630. pmid:24879662CrossRefPubMedGoogle Scholar
↵Li S, Guo Y-L, Zhao X, et al. Novel and traditional lipid-related biomarkers and their combinations in predicting coronary severity. Sci Rep2017;7:360. doi:10.1038/s41598-017-00499-9. pmid:28336922CrossRefPubMedGoogle Scholar
↵Barthélémy J-C, Pichot V, Dauphinot V, et al. Autonomic nervous system activity and decline as prognostic indicators of cardiovascular and cerebrovascular events: the ‘PROOF’ Study. Study design and population sample. Associations with sleep-related breathing disorders: the ‘SYNAPSE’ Study. Neuroepidemiology2007;29:18-28. doi:10.1159/000108914. pmid:17898520CrossRefPubMedWeb of ScienceGoogle Scholar
↵Hupin D, Roche F, Gremeaux V, et al. Even a low-dose of moderate-to-vigorous physical activity reduces mortality by 22% in adults aged ≥60 years: a systematic review and meta-analysis. Br J Sports Med2015;49:1262-7. doi:10.1136/bjsports-2014-094306. pmid:26238869Abstract/FREE Full TextGoogle Scholar
↵Hupin D, Edouard P, Gremeaux V, et al. Physical activity to reduce mortality risk. Eur Heart J2017;38:1534-7. doi:10.1093/eurheartj/ehx236. pmid:29048470CrossRefPubMedGoogle Scholar
↵Clarke R, Shipley M, Lewington S, et al. Underestimation of risk associations due to regression dilution in long-term follow-up of prospective studies. Am J Epidemiol1999;150:341-53. doi:10.1093/oxfordjournals.aje.a010013. pmid:10453810CrossRefPubMedWeb of ScienceGoogle Scholar
↵Lear SA, Hu W, Rangarajan S, et al. The effect of physical activity on mortality and cardiovascular disease in 130 000 people from 17 high-income, middle-income, and low-income countries: the PURE study. Lancet2017;390:2643-54. doi:10.1016/S0140-6736(17)31634-3. pmid:28943267CrossRefPubMedGoogle Scholar
↵Arem H, Moore SC, Patel A, et al. Leisure time physical activity and mortality: a detailed pooled analysis of the dose-response relationship. JAMA Intern Med2015;175:959-67. doi:10.1001/jamainternmed.2015.0533. pmid:25844730CrossRefPubMedGoogle Scholar
↵Hulsegge G, Looman M, Smit HA, Daviglus ML, van der Schouw YT, Verschuren WM. Lifestyle changes in young adulthood and middle age and risk of cardiovascular disease and all-cause mortality: The doetinchem cohort study. J Am Heart Assoc2016;5:1-11. doi:10.1161/JAHA.115.002432. pmid:26764411CrossRefPubMedGoogle Scholar
↵Talbot LA, Morrell CH, Fleg JL, Metter EJ. Changes in leisure time physical activity and risk of all-cause mortality in men and women: the Baltimore Longitudinal Study of Aging. Prev Med2007;45:169-76. doi:10.1016/j.ypmed.2007.05.014. pmid:17631385CrossRefPubMedWeb of ScienceGoogle Scholar
↵Yates T, Haffner SM, Schulte PJ, et al. Association between change in daily ambulatory activity and cardiovascular events in people with impaired glucose tolerance (NAVIGATOR trial): a cohort analysis. Lancet2014;383:1059-66. doi:10.1016/S0140-6736(13)62061-9. pmid:24361242CrossRefPubMedGoogle Scholar
View Abstract
Article tools
PDF0 responses
Respond to this article
Data supplement
Print
Alerts & updates
Citation tools
Request permissions
Author citation
Add article to BMJ Portfolio
Email to a friend
UK jobs
International jobs
Cambridge University Hospitals NHS Foundation Trust: Locum Consultant in Palliative Care
Cleveland Clinic, Jersey: GP Partner required
Strathcarron Hospice: Consultant/Locum Consultant in Palliative Medicine
University of Glasgow: Clinical Lecturer in Geriatric Medicine
CARE Fertility Group: Consultant in Reproductive Medicine/Fertility Specialist
View more
Altmetric
Who is talking about this article?
Article has an altmetric score of 12
See more details
Picked up by 1 news outlets
Tweeted by 5
This week's poll
Is resilience training helpful?
Yes
No
Vote
View Results
Read related article
See previous polls
Other content recommended for you,
Cancer risk associated with chronic diseases and disease markers: prospective cohort study
Hua Zhao et al., The BMJ, 2018
Physical inactivity, cardiometabolic disease, and risk of dementia: an individual-participant meta-analysis
Mika Kivimäki et al., The BMJ, 2019
Associations of grip strength with cardiovascular, respiratory, and cancer outcomes and all cause mortality: prospective cohort study of half a million UK Biobank participants
Carlos A Celis-Morales et al., The BMJ, 2018
Association between muscular strength and mortality in men: prospective cohort study
Jonatan R Ruiz et al., The BMJ, 2008
Trajectory of body shape in early and middle life and all cause and cause specific mortality: results from two prospective US cohort studies
Mingyang Song et al., The BMJ, 2016
Exceptional longevity does not result in excessive levels of disability.
Kaare Christensen et al., Proc Natl Acad Sci U S A, 2008
Implement strategies that increase the frequency of guideline-compliant vaccination schedule completion for each IBD patient prior to initiating therapy
Gary R. Lichtenstein, MD, myCME, 2018
Study of influence factors on mortality risk of the elderly with mild cognitive impairment in China
LU Jiehua LI Yue, Population Journal
Powered by
I consent to the use of Google Analytics and related cookies across the TrendMD network (widget, website, blog). Learn more Yes No
Back to top
Investigating causal relations between sleep traits and risk of breast cancer in women: mendelian randomisation study
BMJ 2019; 365 doi: https://doi.org/10.1136/bmj.l2327 (Published 26 June 2019)
Cite this as: BMJ 2019;365:l2327
Linked Editorial
Larks, owls, and breast cancer
Article
Related content
Metrics
Responses
Peer review
Rebecca C Richmond, research fellow12, Emma L Anderson, research fellow12, Hassan S Dashti, postdoctoral researcher34, Samuel E Jones, research fellow5, Jacqueline M Lane, postdoctoral researcher34, Linn Beate Strand, associate professor6, Ben Brumpton, research fellow67, Martin K Rutter, senior lecturer89, Andrew R Wood, lecturer5, Kurt Straif, scientist10, Caroline L Relton, professor12, Marcus Munafò, professor111, Timothy M Frayling, professor5, Richard M Martin, professor1212, Richa Saxena, professor341314, Michael N Weedon, associate professor5, Debbie A Lawlor, professor1212, George Davey Smith, professor1212
Author affiliations
Correspondence to: R C Richmond, Office BS4, Oakfield House, Oakfield Grove, Bristol BS8 2BN, UK Rebecca.richmond@bristol.ac.uk (or @beckyrichmond90 on Twitter)
Accepted 26 April 2019
Abstract
Objective To examine whether sleep traits have a causal effect on risk of breast cancer.
Design Mendelian randomisation study.
Setting UK Biobank prospective cohort study and Breast Cancer Association Consortium (BCAC) case-control genome-wide association study.
Participants 156 848 women in the multivariable regression and one sample mendelian randomisation (MR) analysis in UK Biobank (7784 with a breast cancer diagnosis) and 122 977 breast cancer cases and 105 974 controls from BCAC in the two sample MR analysis.
Exposures Self reported chronotype (morning or evening preference), insomnia symptoms, and sleep duration in multivariable regression, and genetic variants robustly associated with these sleep traits.
Main outcome measure Breast cancer diagnosis.
Results In multivariable regression analysis using UK Biobank data on breast cancer incidence, morning preference was inversely associated with breast cancer (hazard ratio 0.95, 95% confidence interval 0.93 to 0.98 per category increase), whereas there was little evidence for an association between sleep duration and insomnia symptoms. Using 341 single nucleotide polymorphisms (SNPs) associated with chronotype, 91 SNPs associated with sleep duration, and 57 SNPs associated with insomnia symptoms, one sample MR analysis in UK Biobank provided some supportive evidence for a protective effect of morning preference on breast cancer risk (0.85, 0.70, 1.03 per category increase) but imprecise estimates for sleep duration and insomnia symptoms. Two sample MR using data from BCAC supported findings for a protective effect of morning preference (inverse variance weighted odds ratio 0.88, 95% confidence interval 0.82 to 0.93 per category increase) and adverse effect of increased sleep duration (1.19, 1.02 to 1.39 per hour increase) on breast cancer risk (both oestrogen receptor positive and oestrogen receptor negative), whereas evidence for insomnia symptoms was inconsistent. Results were largely robust to sensitivity analyses accounting for horizontal pleiotropy.
Conclusions Findings showed consistent evidence for a protective effect of morning preference and suggestive evidence for an adverse effect of increased sleep duration on breast cancer risk.
Introduction
In 2007 the World Health Organization’s International Agency for Research on Cancer classified shift work that involves circadian disruption as being probably carcinogenic to humans.1 Disturbed sleep, exposure to light at night, and exposure to other lifestyle factors have been proposed as possible underlying mechanisms.234 Although much of the literature on breast cancer risk has focused on the potentially adverse effects of night shift work and exposure to light at night, less investigation has been done into the potential adverse effects of sleep disruption and traits such as chronotype (morning or evening preference), sleep duration, and insomnia.5
In a meta-analysis of 28 studies, strong evidence suggested a positive association between circadian disruption and breast cancer risk (relative risk 1.14, 95% confidence interval 1.08 to 1.21). However, the association with short sleep duration (<7 hours a night) in seven contributing studies was much less conclusive (0.96, 0.86 to 1.06), and no dose-response association with sleep deficiency was observed.6 Findings from other meta-analyses have been conflicting, with two showing no conclusive evidence that sleep duration is associated with breast cancer risk78 and one showing evidence of an adverse effect of increased sleep duration (>7 hours a night).9 Most studies in the meta-analyses, however, have been case-control designs, vulnerable to reverse causation, or cohort studies with a small number of cases. Fewer studies have investigated associations between chronotype and insomnia with breast cancer risk. The Nurses’ Health Study cohort of 72 517 women (1834 breast cancer cases) found no strong evidence of an association with chronotype,10 and a prospective study of 33 332 women (862 incident breast cancer cases) in the Nord-Trøndelag Health Study (HUNT) found no strong evidence of an association with individual insomnia symptoms, although there was evidence of some excess risk among participants with multiple insomnia problems.11 Studies have tended to rely on self report of sleep exposures, meaning associations could be biased by measurement error and by residual or unmeasured confounding, making causal inference challenging.
Mendelian randomisation (MR) uses genetic variants that are robustly associated with potentially modifiable risk factors to explore causal effects on outcomes.121314 This method is less susceptible to measurement error, confounding, and reverse causation than conventional multivariable regression approaches, provided certain assumptions are met. These are that the genetic variants are robustly associated with the exposure of interest, are not associated with confounders of the exposure-outcome relation, and do not influence the outcome through pathways other than the exposure of interest. Genetic variants robustly associated with chronotype, sleep duration, and insomnia symptoms have recently been identified in large genome-wide association studies (GWAS) with sample sizes of around 50 000 to more than one million.151617181920212223 Findings from those GWAS have confirmed the role of several core circadian genes influencing sleep traits, and identified genetic variants with no previously known circadian role.24 These genetic variants have been used in two sample MR and provided some evidence that longer sleep has a causal effect on schizophrenia risk,16 whereas being a “morning person” is causally associated with a reduced risk of schizophrenia and depression,15 and insomnia is causally associated with an increased risk of type 2 diabetes, higher body mass index (BMI), coronary heart disease, and several psychiatric traits.1723 In our study we used MR to explore the causal effect of sleep traits on breast cancer risk.
We used genetic variants robustly associated with chronotype, sleep duration, and insomnia symptoms identified in three recent UK Biobank GWAS151617 to investigate whether these sleep traits have a causal effect on breast cancer risk. To do this, we performed a one sample MR analysis using data from UK Biobank, from which estimates were compared with conventional observational multivariable regression results in the same study, as well as a two sample MR analysis using data from the Breast Cancer Association Consortium (BCAC).25 Furthermore, we aimed to assess the extent to which findings were robust to potential pleiotropy and supported by genetic variants associated with accelerometer derived measures of chronotype (sleep midpoint timing of the least active five hours of the day), sleep duration, and sleep fragmentation (number of nocturnal sleep episodes).
Methods
Multivariable regression and one sample MR analysis
Study participants
We used data on women from the UK Biobank, which recruited more than 500 000 participants (55% women) out of 9.2 million eligible adults aged between 40 and 70 years in the UK who were invited to participate (5.5% response rate).26 The study protocol is available online (www.ukbiobank.ac.uk/wp-content/uploads/2011/11/UK-Biobank-Protocol.pdf) and more details are published elsewhere.27 At recruitment the participants gave informed consent to participate and be followed-up. Overall, 503 317 participants consented to join the study cohort and visited an assessment centre. Information on sleep traits (chronotype, sleep duration, and insomnia symptoms), breast cancer status (prevalent and incident cases with up to nine years of follow-up), relevant confounding factors, and genetic variants are available in UK Biobank.
Sleep traits
At baseline assessment, conducted in one of 22 UK Biobank assessment centres between 2006 and 2010, participants completed a touchscreen questionnaire, which included questions about sociodemographic status, lifestyle and environment, early life and family history, health and medical history, and psychosocial factors. Participants were asked about their chronotype (morning or evening preference), average sleep duration, and insomnia symptoms.
Chronotype (morning or evening preference) was assessed in the question “Do you consider yourself to be?” with one of six possible answers: “Definitely a ‘morning’ person,” “More a ‘morning’ than ‘evening’ person,” “More an ‘evening’ than a ‘morning’ person,” “Definitely an ‘evening’ person,” “Do not know,” or “Prefer not to answer.” We derived a five level ordinal variable for chronotype where “Definitely a ‘morning’ person,” “More a ‘morning’ than ‘evening’ person,” “More an ‘evening’ than a ‘morning’ person,” “Definitely an ‘evening’ person,” “Do not know,” or “Prefer not to answer” were coded as 2, 1, −1, −2, 0, and missing, respectively. Sleep duration was assessed by asking: “About how many hours sleep do you get in every 24 hours? (please include naps).” The answer could only contain integer values. Binary variables for short sleep duration (<7 hours v 7-8 hours) and long sleep duration (>8 hours v 7-8 hours) were also derived. To assess insomnia symptoms, participants were asked: “Do you have trouble falling asleep at night or do you wake up in the middle of the night?” with responses “Never/rarely,” “Sometimes,” “Usually,” or “Prefer not to answer.” Those who responded “Prefer not to answer” were set to missing. We derived a three level ordinal variable for insomnia symptoms where “Never/rarely,” “Sometimes,” and “Usually” were coded as 0, 1, and 2, respectively.
Breast cancer
Participants were followed through record linkage to the National Health Service central registers, which provide information on cancer registrations, using ICD-9 and ICD-10 (international classification of diseases, ninth and 10th revisions, respectively) codes and cancer deaths. The endpoints in these analyses were first diagnosis of invasive breast cancer (ICD-10 C50, ICD-9 174), or breast cancer listed as the underlying cause of death on the death certificate for women who died during follow-up but were not captured by the cancer registers. We excluded all women with any other cancer diagnosis from the analysis. At the time of analysis, mortality data were available up to February 2016 and cancer registry data up to April 2015. Prevalent cases were defined as women with a diagnosis of breast cancer before date of recruitment to the UK Biobank. Incident cases were defined as women with a diagnosis of breast cancer or dying from it during follow-up.
Confounders
We considered several factors to be potential confounders of the association between sleep traits and breast cancer risk: education, body mass index (BMI), alcohol intake, smoking, strenuous physical activity, family history of breast cancer, age at menarche, parity, use of oral contraceptives, menopause status, and hormone replacement therapy.
BMI was derived from weight and height measured when participants attended the initial assessment centre, whereas information on other potential confounders was obtained from questionnaire responses completed at baseline (see methods in supplementary file). Additional information extracted from the initial assessment visit included centre of initial assessment visit, age at recruitment derived from date of birth, and date of attending assessment centre. Participants who were employed were also asked whether their current job involved night shifts: never/rarely, sometimes, usually, or always.
Genetic variants
The full data release in UK Biobank contains the cohort of successfully genotyped people (n=488 377). A total of 49 979 people were genotyped using the UK BiLEVE genotyping chip and 438 398 using the UK Biobank axiom genotyping chip. Pre-imputation quality control, phasing, and imputation of the UK Biobank genetic data have been described elsewhere.28
In the MR analysis, we used a total of 341 single nucleotide polymorphisms (SNPs) associated with chronotype,15 91 SNPs associated with continuous sleep duration,16 and 57 SNPs associated with insomnia symptoms17 (see supplementary file, tables 1-3). These genetic variants were derived from self report and confirmed with objective sleep assessment and in independent cohorts.151617
Multivariable regression analysis
We carried out separate multivariable Cox regression between chronotype, insomnia symptoms, and sleep duration and incident breast cancer to investigate prospective associations between these sleep traits and to minimise the likelihood of reverse causality in observational associations. To minimise the role of confounding, we adjusted analyses for age, assessment centre, and the top 40 genetic principal components (obtained from principal components analysis (PCA) to detect and quantify the genetic structure of populations). A second model additionally adjusted for education, BMI, alcohol intake, smoking, strenuous physical activity, family history of breast cancer, age at menarche, parity, menopause status, use of oral contraceptives, and hormone replacement therapy.
One sample MR analysis
For one sample MR, the genetic variants used were extracted genotypes from the UK Biobank imputation dataset (imputed to the Haplotype Reference Consortium reference panel), which performed extensive quality control including exclusion of most third degree or closer relatives from a genetic kinship analysis, as well as those who were not classified as white British based on questionnaire and PCA29 (see methods in supplementary file). Unweighted allele scores were generated as the total number of sleep trait increasing alleles (morning preference alleles from chronotype) present in the genotype of each participant.
A two stage method was implemented to give a population average causal hazard ratio. The first stage model consisted of a regression of the sleep trait (chronotype, sleep duration, and insomnia symptoms) on the allele score and the second stage model consisted of a Cox regression of breast cancer status on the fitted values from the first stage regression, with adjustment for age at recruitment, assessment centre, 40 genetic principal components, and genotyping chip in both stages.
Sensitivity analyses
To check the proportional hazards assumption, we used Pearson correlations to test Schoenfeld residuals from both multivariable Cox regression and one sample MR Cox regression models for an association with follow-up time.
To assess the specificity of our findings to breast cancer, we performed multivariable regression and one sample MR analysis to assess the causal effect of the sleep traits on other cancer diagnoses and on all cause mortality.
We also performed MR analysis using only those genetic variants that replicated at Bonferroni significance in a large independent dataset for chronotype15 (242 variants in 23andMe, n=240 098, highlighted in supplementary file, table 1) to evaluate the potential impact of winner’s curse (ie, overestimation of genetic effects in the initial study), which can bias causal estimates in MR analysis. Given the relatively small sample size of replication datasets for sleep duration (CHARGE Consortium, n=47 180)16 and insomnia (HUNT, n=62 533),17 few SNPs independently replicated at Bonferroni significance to serve as sufficiently strong instruments for this sensitivity analysis.
To test the MR assumption that genetic variants should not be associated with confounders of the exposure-outcome relation, we investigated associations between the allele scores and potential confounders in UK Biobank. We then performed one sample MR analysis adjusted for any potential confounders found to be strongly associated with the allele scores (beyond a Bonferroni significance threshold of P<1.39×10−3) as a further sensitivity analysis.
We also conducted both multivariable regression and one sample MR using all breast cancer cases (incident and prevalent) in a logistic regression analysis in UK Biobank, and performed sensitivity analysis removing participants who reported currently working night shifts (sometimes, usually, or always).
Two sample MR analysis
We conducted a two sample MR analysis of sleep traits on breast cancer risk using female specific estimates of the associations between the genetic instruments and sleep traits identified in the respective GWAS151617 in UK Biobank (sample 1) (see supplementary file, tables 1-3), and estimates of the associations between the genetic instruments and breast cancer from a large scale GWAS of breast cancer (BCAC) (sample 2).
GWAS of chronotype (five level ordinal variable), sleep duration (continuous variable), and insomnia symptoms (three level ordinal variable) were performed among women of European ancestry (n=241 350 - 245 767) in the UK Biobank. This was done using BOLT-LMM30 linear mixed models and an additive genetic model adjusted for age, sex, 10 genetic principal components, genotyping array, and genetic correlation matrix, as was done previously.151617
The GWAS of breast cancer involved 122 977 women with the disease (oestrogen receptor positive and oestrogen receptor negative combined) and 105 974 controls of European ancestry from BCAC.25 BCAC summary data were based on imputation to the 1000 Genomes Project Phase 3 reference panel. To explore potential heterogeneity by breast cancer subtype, we also investigated the causal effect of the sleep traits on breast cancer stratified by oestrogen receptor status, using genetic association data from 69 501 oestrogen positive and 21 468 oestrogen negative cases within BCAC.25
Two sample MR analyses were conducted using “TwoSampleMR,” an R package for such analyses,31 which was first used to extract the SNPs being used to instrument the exposure (here the sleep trait of interest) from the outcome GWAS (here breast cancer in BCAC). If a SNP was unavailable in the breast cancer GWAS summary statistics, we identified proxy SNPs with a minimum linkage disequilibrium (LD) r2=0.8. We then performed harmonisation of the direction of effects between exposure and outcome associations, where palindromic SNPs were aligned when minor allele frequencies were less than 0.3, or they were otherwise excluded. We then used an inverse variance weighted method to meta-analyse the SNP specific Wald estimates (SNP outcome estimate divided by SNP exposure estimate) using random effects, to obtain an estimate for the causal effect of the sleep trait on breast cancer risk.
Sensitivity analyses
The inverse variance weighted random effects method will return an unbiased estimate in the absence of horizontal pleiotropy, or when horizontal pleiotropy is balanced.32 To account for directional pleiotropy, we compared results with three other MR methods, which each makes different assumptions about this: MR Egger,33 weighted median,34 and weighted mode,35 and therefore a consistent effect across multiple methods strengthens causal evidence.
To further detect and correct obtained causal estimates for potential violation of the MR assumptions,32 we performed RadialMR36 in the two sample analyses to identify outliers with the most weight in the MR analysis and the largest contribution to Cochran’s Q statistic for heterogeneity, which may then be removed and the data reanalysed. Radial MR analysis was conducted using modified second order weights and an α level of 0.05 divided by the number of SNPs being used to instrument the exposure. For the outliers identified, we also assessed their potential pleiotropic role by performing a phenome-wide association study (PheWAS) approach37 to investigate the associations between the SNPs and all available traits in the MR-Base PheWAS database (http://phewas.mrbase.org/).
To evaluate the potential impact of winner’s curse, we performed two sample MR analysis using 242 genetic variants that replicated at Bonferroni significance in a large independent dataset for chronotype15 (23andMe, n=240 098, highlighted in supplementary file, table 1). We also carried out further MR analysis using robust adjusted profile scores, which provide an unbiased causal estimate in the presence of weak instruments.38
Given potential non-linear associations between sleep duration and breast cancer risk,9 we also used data on 27 SNPs associated with short sleep (<7 hours v 7-8 hours) and eight SNPs associated with long sleep (>8 hours v 7-8 hours)16 in two sample MR analysis (see supplementary file, tables 4 and 5). Causal effect estimates (ie, odds ratios for breast cancer) were rescaled to be interpreted for each doubling of genetic liability for short or long sleep, as recommended elsewhere.39
Finally, we performed two sample MR using genetic variants robustly associated with accelerometer derived sleep traits in UK Biobank, to be compared with causal estimates obtained using genetic variants associated with self reported traits. For this we used genetic variants identified in GWAS in relation to three accelerometer based measures: timing of the least active five hours (L5 timing) (6 SNPs), nocturnal sleep duration (11 SNPs), and number of nocturnal sleep episodes (21 SNPs) in up to 85 205 participants, as previously described40 (see supplementary file, tables 6-8). Also see the methods section in the supplementary file for more details about how accelerometer sleep traits were derived. Effect estimates represented an hour earlier L5 timing (correlated positively with and to be compared with the self reported chronotype measure of increased morning preference), an hour increase of nocturnal sleep duration (to be compared with self reported sleep duration), and a unit increase in the number of nocturnal sleep episodes (to be compared with self reported insomnia symptoms).
All analyses were conducted using Stata (version 15) and R (version 3.4.1).
Patient and public involvement
The current research was not informed by patient and public involvement because it used secondary data. However, future research following on from our findings should be guided by patient and public opinions.
No patients were involved in setting the research question or the outcome measures, nor were they involved in developing plans for design or implementation of the study. No patients were asked to advise on interpretation or writing up of results. The results of the research will be disseminated to study participants on request, and to stakeholders and the broader public as relevant.
Results
Baseline characteristics
Of the 180 216 women in the UK Biobank who had been successfully genotyped and passed the genetic quality control, and after excluding 23 368 who had a diagnosis of other types of cancer, 7784/156 868 (4.9%) had received an exclusive diagnosis of breast cancer. Of these, 5036/156 868 (3.2%) were defined as prevalent cases and 2740/156 868 (1.7%) developed incident breast cancer over a median follow-up of 2.98 years.
Women with a breast cancer diagnosis (prevalent or incident) were more likely to be older, have a higher BMI, be less physically active, have had an earlier age at menarche, be postmenopausal, have ever used hormone replacement therapy, have a family history of breast cancer, and be nulliparous. They were less likely to be never smokers, work night shifts, and have ever used oral contraceptives (table 1) compared with women without a breast cancer diagnosis. No strong difference was found in education level between women with and without breast cancer, in line with previous findings,41 as well as no clear difference in relation to alcohol intake.
Table 1 Baseline characteristics of women who had and had not developed breast cancer by date of censoring in UK Biobank. Values are numbers (percentages) unless stated otherwise
View popupView inline
Multivariable analysis
In multivariable Cox regression analysis, an inverse association was observed between morning preference and risk of breast cancer, which remained similar in the fully adjusted model (hazard ratio 0.95, 95% confidence interval 0.93 to 0.98 per category increase) but there was no clear association between sleep duration and insomnia symptoms with risk of breast cancer (table 2). The proportional hazards assumption held for all the multivariable Cox regression analyses (see supplementary file, table 9). The inverse association with morning preference was not observed for other cancer diagnoses (hazard ratio 1.00, 95% confidence interval 0.99 to 1.02 per category increase) (see supplementary file, table 10), although it was evident in multivariable Cox regression analysis of all cause mortality (0.95, 0.93 to 0.97 per category increase) (see supplementary file, table 11). Associations with sleep duration and insomnia were also observed in relation to these other outcomes (see supplementary file, tables 10 and 11).
Table 2 Multivariable and mendelian randomisation Cox regression analysis for risk of breast cancer associated with sleep traits
View popupView inline
When incident and prevalent cases were combined and associations investigated in a logistic regression framework, evidence was consistent for an inverse association between morning preference and breast cancer risk (odds ratio 0.96, 95% confidence interval 0.94 to 0.98), as well a positive association between both sleep duration (1.02, 1.00 to 1.05 per hour increase) and insomnia symptoms (1.11, 1.07 to 1.15 per category increase) with breast cancer risk, potentially reflecting reverse causation (see supplementary file, table 12). Cox regression estimates were similar after excluding participants who reported working night shifts (see supplementary file, table 13).
One sample MR analysis
Among UK Biobank female participants, allele scores explained 2.3% of the variance in chronotype, 0.7% of the variance in sleep duration, and 0.4% of the variance in insomnia symptoms (table 3). Some evidence suggested a protective effect of morning preference on breast cancer risk (hazard ratio 0.85, 95% confidence interval 0.70 to 1.03 per category increase) and weaker evidence for an adverse effect of increased sleep duration (1.06, 0.70 to 1.59 per hour increase) and insomnia symptoms (1.37, 0.59 to 3.20 per category increase) (table 2), albeit imprecisely estimated (wide confidence intervals). The proportional hazards assumption held for all the one sample MR Cox regression analyses (see supplementary file, table 9). The protective effect of morning preference was not supported by MR analysis for other cancer diagnoses (1.05, 0.93 to 1.17 per category increase) (see supplementary file, table 10) or all cause mortality (1.15, 0.97 to 1.35 per category increase) (see supplementary file, table 11), although evidence suggested an adverse effect of insomnia on risk of other cancers (1.55, 0.94 to 2.55 per category increase) (see supplementary file, table 10).
Table 3 Allele scores for sleep traits in UK Biobank
View popupView inline
When using only the genetic variants that replicated in an independent dataset (242 variants in 23andMe) for chronotype, estimates of effect on breast cancer risk were similar (0.89, 0.71 to 1.12 per category increase); although with wider confidence intervals given that the replicated variants explained less of the variance in chronotype (1.6%) (see supplementary file, table 14).
Although most of the confounding factors were not associated with the sleep trait allele scores in UK Biobank, after accounting for multiple testing the chronotype allele score was associated with parity and vigorous activity; the sleep duration allele score was associated with age at menarche and BMI, and the insomnia allele score was associated with using hormone replacement therapy and age at menarche (see supplementary file, table 15). Further sensitivity analysis was undertaken adjusting for these potential confounders in the one sample MR analysis, and effect estimates were consistent (see supplementary file, table 16).
Findings of a protective effect of morning preference were supported in analysis of all breast cancer cases (incident and prevalent) in logistic regression. Evidence for sleep duration and insomnia symptoms was weaker, although both had effect estimates in the positive direction (see supplementary file, table 12). In analyses excluding women who reported working night shifts, findings were also consistent with the main results from Cox regression (see supplementary file, table 13).
Two sample MR analysis
After harmonisation of the SNP effects in the two summary datasets (UK Biobank and BCAC), 305 SNPs were used to instrument chronotype, 82 SNPs were used to instrument sleep duration, and 50 SNPs were used to instrument insomnia symptoms. This included three proxy SNPs (r2≥0.8) for chronotype (rs376957969 for rs111867612, rs1871516 for rs4550782, and rs6583802 for rs61875203). Two sample MR supported the findings of a protective effect of morning preference (inverse variance weighted odds ratio 0.88, 95% confidence interval 0.82 to 0.93 per category increase) (see supplementary file, table 17 and figure 1) as well as an adverse effect of increased sleep duration (1.19, 1.02, 1.39 per hour increase) on breast cancer risk (see supplementary file, table 17 and figure 2). Little evidence for a causal effect of insomnia symptoms was observed (0.93, 0.49, 1.76 per category increase) (see supplementary file, table 17 and figure 3). Figure 1 shows the inverse variance weighted estimates for chronotype, sleep duration, and insomnia symptoms from two sample MR compared with multivariable and one sample MR approaches in UK Biobank. Findings were similar when stratified by oestrogen receptor positive and oestrogen receptor negative breast cancer (see supplementary file, table 17).
Fig 1
Fig 1
Forest plot of multivariable and mendelian randomisation (MR) estimates for association between sleep traits and breast cancer risk. Odds ratios are per category increase in chronotype (from definite evening, intermediate evening, neither, intermediate morning, definite morning), per hour increase in sleep duration, and per category increase in insomnia risk (from no, some, and frequent insomnia symptoms). Odds ratios rather than hazard ratios for incident breast cancer are shown for multivariable and one sample MR analysis to compare estimates across methods
Download figure Open in new tab Download powerpoint
Effect estimates were broadly consistent between the inverse variance weighted method and the pleiotropy robust methods applied (MR Egger, weighted median, and weighted mode) in two sample MR (see supplementary file, table 17 and figures 1-3). Furthermore, the MR Egger test of directional pleiotropy was consistent with the null for all analyses (see supplementary file, table 18).
Evidence for heterogeneity in causal effects for most of the models (see supplementary file, table 19) could still indicate potential violations of the MR assumptions. We used radial plots to aid in the detection of outlying variants. Radial MR analysis identified six outliers for chronotype, three for sleep duration, and two for insomnia symptoms in both inverse variance weighted and MR Egger (see supplementary file, table 20 and figures 4-6). The pleiotropic effect of many of these outliers was indicated in a PheWAS of the SNPs on all existing traits in the MR-Base database (see supplementary file, figure 7). With removal of outliers, inverse variance weighted and MR Egger effect estimates were largely unchanged (see supplementary file, table 21).
Effect estimates for the causal effect of chronotype on breast cancer risk were consistent when using the 242 genetic variants associated with chronotype, which replicated at Bonferroni significance in 23andMe,15 indicating that winner’s curse is unlikely to have substantially biased effect estimates (see supplementary file, table 22). MR robust adjusted profile scores, which provide unbiased estimates in the presence of weak instruments, provided similar causal estimates to the main MR analysis (see supplementary file, table 23).
Findings of an adverse effect of increased sleep duration on breast cancer risk were supported using genetic variants specifically associated with short and long sleep duration, with evidence for a protective effect of short sleep duration on breast cancer (inverse variance weighted odds ratio 0.92, 95% confidence interval 0.86 to 0.99 per doubling of genetic liability for short sleep duration) and adverse effect of long sleep duration (1.24, 0.96 to 1.60 per doubling of genetic liability for long sleep duration) (see supplementary file, table 24).
Finally, we performed two sample MR using genetic variants robustly associated with accelerometer derived sleep traits in UK Biobank, to be compared with causal estimates obtained using genetic variants associated with self reported traits. Supplementary table 25 shows the genetic correlations between these traits. Using genetic variants robustly associated with accelerometer derived sleep traits in UK Biobank, we found no clear evidence of association with L5 timing measured objectively (1.04, 0.78 to 1.38 per hour decrease) (see supplementary file, table 26 and figure 8). However, an adverse effect of increased sleep duration was supported using estimates from objectively measured sleep duration (1.16, 1.02 to 1.32 per hour increase) (see supplementary file, table 26 and figure 9) and there was some evidence for a causal effect of increased fragmentation on breast cancer risk (1.14, 1.00 to 1.30 per sleep episode) (see supplementary file, table 26 and figure 10). Given the limited availability of SNPs being used to proxy for L5 timing to evaluate its causal role on breast cancer, and given the strong association found between chronotype and L5 timing (see supplementary file, table 25),15 we performed a further MR analysis using the 305 chronotype variants with SNP exposure effect estimates taken from the GWAS of L5 timing, to also evaluate the causal effect of L5 timing (see supplementary file, table 27). This analysis revealed some evidence for an association with L5 timing and risk of breast cancer in the inverse variance weighted analysis (0.86, 0.78 to 0.95), although this estimate was not consistent across the pleiotropy robust methods, which were more consistent with the null.
Discussion
Mendelian randomisation (MR) uses genetic variation to investigate causal relations between potentially modifiable risk factors and health outcomes. In this study we compared observational estimates from multivariable regression with those from MR analyses to make inferences about the likely causal effects of three sleep traits on breast cancer risk.
In multivariable regression analysis using data on breast cancer incidence in the UK Biobank study, morning preference was inversely associated with breast cancer, whereas there was little evidence for an association with sleep duration and insomnia. Using genetic variants associated with chronotype, sleep duration, and insomnia symptoms, one sample MR analysis in UK Biobank provided some evidence for a protective effect of morning preference but imprecise estimates for sleep duration and insomnia. Findings for a protective effect of morning preference and adverse effect of increased sleep duration on breast cancer (both oestrogen receptor positive and oestrogen receptor negative) were supported by two sample MR using data from the Breast Cancer Association Consortium (BCAC), whereas there was inconsistent evidence for insomnia symptoms. Results were largely robust to sensitivity analyses accounting for horizontal pleiotropy.
Comparison with other studies
Previous studies have found an enrichment of circadian pathway genetic variants in breast cancer.2542 Nonetheless, these studies did not directly implicate modifiable sleep traits by which risk of breast cancer could be minimised and did not attempt to separate the effects of the genetic variants on breast cancer risk through circadian disruption from pleiotropic pathways.
Findings of an adverse effect of evening preference on breast cancer risk in all analyses performed go some way to supporting hypotheses around carcinogenic light-at-night 4 and findings of increased risk among night shift workers who might be exposed to artificial light at night.1 In particular, the specificity of the causal effect of chronotype on breast cancer, which was not observed in relation to other cancers or all cause mortality, is consistent with the hormonal mechanisms implicated in the light-at-night hypothesis. However, findings when using an objective measure of chronotype (the least active five hours (L5 timing)) did not reveal the same adverse effect. Although this last analysis might be limited by the number and strength of the genetic variants used to instrument L5 timing, the lack of consistency in estimates draws to question the mechanisms by which morning or evening preference (rather than actual activity) influences breast cancer risk. Further analysis using the single nucleotide polymorphisms (SNPs) identified in relation to chronotype as instruments for L5 timing were consistent with a protective effect of morning preference, suggesting a protective effect of activity as well as reported preference, but as the pleiotropy robust tests were not consistent, more work is needed to distinguish the causal effect of morning preference from activity—for example, with the use of multivariable MR methods.43
Evidence for an adverse effect of increased sleep duration on breast cancer risk contrasts with the observational findings in UK Biobank as well as much of the literature on circadian disruption and breast cancer risk,6 and unlike our findings for chronotype are not aligned with the light at night hypothesis. However, recent studies implicate longer sleep duration as a risk factor for breast cancer.9 Given previous reports of a J-shaped relation between sleep duration and breast cancer risk,9 as well as investigating sleep duration as a continuous variable, we also investigated the causal effects of both short and long sleep duration to investigate non-linear effects. In line with our main findings, we found evidence for a protective effect of short sleep duration and adverse effect of long sleep duration on breast cancer risk. Furthermore, using genetic variants associated with accelerometer derived nocturnal sleep duration, we found evidence for an adverse effect of sleep duration with a similar magnitude of effect.
Overall, we found inconsistent evidence about the causal effect of insomnia symptoms on breast cancer risk in multivariable and MR analyses. A previous study of incident breast cancer in the Nord-Trøndelag Health Study (HUNT) revealed no strong evidence of an association with individual insomnia symptoms,11 although people with multiple insomnia problems were found to be at increased risk. In our analysis, insomnia was defined based on self report of either difficulty initiating sleep or waking in the night. Further work is therefore required to investigate individual symptoms of insomnia on breast cancer risk, and the potential cumulative effect. Interestingly, MR analysis provided some evidence for adverse causal effect of accelerometer derived number of nocturnal sleep episodes on breast cancer risk.
Strengths and limitations of this study
Key strengths of the study are the integration of multiple approaches to assess the causal effect of sleep traits on breast cancer, the inclusion of data from two large epidemiological resources—UK Biobank and BCAC—as well as use of data derived from both self reported and objectively assessed measures of sleep. Furthermore, for MR analysis we used the largest number of SNPs identified in the genome-wide association studies (GWAS) literature, with full summary statistics available to obtain strong genetic instruments for MR analysis and to explore potential pleiotropic pathways.
The approaches of multivariable Cox regression of incident cases, multivariable logistic regression of prevalent and incident cases, one and two sample MR, each have different strengths and limitations in terms of key sources of bias (see supplementary file, table 28). In multivariable analysis, attempts were made to mitigate key sources of bias, including confounding and reverse causation, with the use of multivariable Cox regression analysis of incident cases of breast cancer and adjustment for several hypothesised confounders. Nonetheless, residual or unmeasured confounding, selection bias, and measurement error could also have distorted effect estimates. We used MR analysis to minimise the likelihood of bias due to measurement error, confounding, and reverse causation. In addition, we conducted a series of sensitivity analyses to assess the core assumptions that the genetic instruments are strongly associated with the exposures of interest, are not influenced by confounding factors, and do not directly influence the outcome other than through the exposure.
One limitation of this study related to the self reported measures used in multivariable regression analyses and used to identify genetic variants for MR analysis. In particular, the measure of sleep duration might capture time spent napping and the any insomnia variable is really a measure of insomnia symptoms and not necessarily clinical insomnia. However, both these measures have been validated with the use of objective measures from accelerometer data in the UK Biobank and concordance is good, particularly for the effects of the genetic variants identified.151617
Another limitation relates to the selection of participants. Analysis in the two large epidemiological studies included here (UK Biobank and BCAC) was restricted to women of European ancestry. Further work is required to investigate whether these findings translate to women in other ancestry groups. Although the UK Biobank represents a large and well characterised epidemiological resource, it is not representative of the UK population owing to low participation.27 As well as influencing the generalisability of findings, selection into the study can lead to biased estimates of association through “collider bias.”44 To minimise the influence of this, we also used genetic data from a large case-control study of breast cancer (BCAC), and we compared MR effect estimates across these datasets.
In all MR analyses, SNP exposure estimates were obtained from the UK Biobank as this has formed a major component of the GWAS of sleep traits conducted to date.151617202123 This could lead to winner’s curse, when the magnitude of the effect sizes for genetic variants identified within a discovery sample are likely to be larger than in the overall population. In a one sample MR analysis, the impact of winner’s curse of the SNP exposure association can bias causal estimates towards the confounded observational estimate, whereas in two sample MR, winner’s curse can result in bias of the causal estimate towards the null. To minimise the impact of winner’s curse in one sample MR analysis we derived an additional allele score for chronotype composed of SNPs that replicated beyond a Bonferroni correction threshold in an independent study (23andMe).15 Similarly, for two sample MR analysis, we used SNP exposure estimates from this replication analysis in sensitivity analyses, and findings were consistent with the main analysis (see supplementary file, tables 14 and 22).
We were unable to apply the same approach to investigate the impact of winner’s curse in the sleep duration and insomnia analysis owing to the relatively small sample size of the replication datasets in those studies, meaning genetic associations could be imprecise. Although we are aware of a large GWAS for insomnia that was conducted using data from both UK Biobank and 23andMe, full summary data for the top SNPs in the replication analysis are not freely available.23 We used unweighted allele scores to minimise the contribution of potential weak instruments in the one sample MR analysis. We also applied a robust adjusted profile score method in the two sample MR analysis, which provides unbiased estimates in the presence of weak instruments, and this revealed similar causal estimates for chronotype, sleep duration, and insomnia as in the main analysis.
Although associations between the allele scores and confounders in UK Biobank imply violation of the MR assumption that genetic variants should not be associated with confounding factors, there are several explanations for these findings. Previous MR studies have identified causal effects of sleep traits on reproductive traits, body mass index, and activity levels,15161723 suggesting that these factors might be mediators of the association between sleep traits and breast cancer rather than confounders. Furthermore, some of the genetic variants associated with chronotype and insomnia have been found to be adiposity related loci,1516 implying potential pleiotropic pathways. Nonetheless, we also applied a series of pleiotropy robust MR methods and outlier detection to rigorously explore the possibility that findings of a causal effect of chronotype and sleep duration were not biased as a result of pleiotropy.
As well as attempting to mitigate key sources of bias for each epidemiological approach applied, we also assessed the consistency in estimates between the approaches to provide the best inference about the causal effect of sleep traits on breast cancer. This is aligned with the practice of triangulation, which aims to obtain more reliable answers to research questions through the integration of results from different approaches, where each approach has different sources of potential bias that are unrelated to each other.4546 We also compared estimates based on self reported sleep with the use of genetic variants associated with accelerometer derived measures of sleep,40 although we did not use female specific SNP estimates here given the smaller number of participants in UK Biobank with these data.
Implications of findings
Findings of a protective effect of morning preference on breast cancer risk add to other evidence from MR supporting a possible beneficial effect of morning preference on decreased risk of schizophrenia and depression.15 However, whether it is the actual behaviour that poses the health risk or the preference for morning versus evening requires further evaluation. Further work is also required to investigate the impact of circadian misalignment, which can be determined by genetic risk, self reported chronotype, and objectively measured L5 timing. In addition, suggestive evidence for a causal effect of increased sleep duration on breast cancer risk should be investigated further.
Conclusions
In this study, both multivariable regression and MR analysis were used to provide strong evidence for a causal effect of chronotype on breast cancer risk. Furthermore, some evidence suggested a causal effect of sleep duration on risk of breast cancer, although findings for these traits were less consistent across the different methods applied. However, the biological role of many of the genetic variants used to instrument these traits in MR and mechanistic pathways underlying the observed effects are not well understood. Previously reported pathways between sleep disruption and mammary oncogenesis include immunological, molecular, cellular, neuroendocrine, and metabolic processes.5 Further work to uncover these possible mediating processes is required. Nonetheless, these findings have potential implications for influencing sleep habits of the general population to improve health.
What is already known on this topic
The World Health Organization’s International Agency for Research on Cancer classifies shift work involving circadian disruption as probably carcinogenic to humans
Much of the literature on breast cancer risk has focused on the potentially adverse effects of night shift work and exposure to light at night, and less into the potential adverse effects of traits such as chronotype (morning or evening preference), sleep duration, and insomnia
Genetic variants robustly associated with chronotype, sleep duration, and insomnia symptoms have recently been identified in large genome-wide association studies
What this study adds
This study found consistent evidence for a protective effect of morning preference and suggestive evidence for an adverse effect of increased sleep duration on breast cancer risk
The evidence for insomnia symptoms was inconclusive
These findings have potential implications for influencing sleep habits of the general population to improve health
Acknowledgments
This research was conducted using the UK Biobank Resource under application numbers 9072, 6818, 15825, and 16391. We thank the participants and researchers from the UK Biobank who contributed or collected data; Ruth Mitchell, Gibran Hemani, Tom Dudding, and Lavinia Paternoster for conducting the quality control filtering of UK Biobank data; and Wes Spiller, Jie Zheng, Gibran Hemani, Philip Haycock, and Kaitlin Wade for help with data acquisition and statistical analysis. This study was made possible with the financial support of Jonathan de Pass and Georgina de Pass.
Footnotes
Contributors: RCR conceived the study and conducted the main analysis. HSD, SEJ, and JML conducted the female specific genome-wide association studies and assisted with sensitivity analyses. RCR, ELA, and GDS drafted the initial manuscript. All authors assisted with interpretation, commended on drafts of the manuscript, and approved the final version. RCR is the guarantor and attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Funding: The breast cancer genome-wide association analyses were supported by the Government of Canada through Genome Canada and the Canadian Institutes of Health Research, the Ministère de l’Économie, de la Science et de l’Innovation du Québec through Genome Québec and grant PSR-SIIRI-701, the National Institutes of Health (U19 CA148065, X01HG007492), Cancer Research UK (C1287/A10118, C1287/A16563, C1287/A10710), and the European Union (HEALTH-F2-2009-223175 and H2020 633784 and 634935). All studies and funders are listed in Michailidou et al.25 RCR, ELA, BMB, CLR, RMM, MM, DAL, and GDS are members of the MRC Integrative Epidemiology Unit at the University of Bristol funded by the Medical Research Council (grant Nos MM_UU_00011/1, MC_UU_00011/2, MC_UU_00011/5, MC_UU_00011/6, and MC_UU_00011/7). RCR is a de Pass VC research fellow at the University of Bristol. This study was supported by the NIHR Biomedical Research Centre at the University Hospitals Bristol NHS Foundation Trust and the University of Bristol. The views expressed in this publication are those of the authors and not necessarily those of the National Health Service, National Institute for Health Research, or Department of Health and Social Care. This work was also supported by Cancer Research UK (grant No C18281/A19169) and the Economic and Social Research Council (grant No ES/N000498/1). SEJ is funded by the Medical Research Council (grant No MR/M005070/1). TMF is supported by the European Research Council (grant No 323195:GLUCOSEGENES-FP7-IDEAS-ERC). MNW is supported by the Wellcome Trust Institutional Strategic Support Award (grant No WT097835MF).
Competing interests: All authors have completed the ICMJE uniform disclosure form at http://www.icmje.org/coi_disclosure.pdf. MKR reports receiving research funding from Novo Nordisk, consultancy fees from Novo Nordisk and Roche Diabetes Care, and modest owning of shares in GlaxoSmithKline, outside the submitted work. DAL reports receiving research support from Medtronic and Roche Diagnostics for research outside the submitted work. All other authors declare no support from any organisation for the submitted work, no financial relationships with any organisations that might have an interest in the submitted work in the previous three years, no other relationships or activities that could appear to have influenced the submitted work.
Ethical approval: UK Biobank has received ethical approval from the UK National Health Service’s National Research Ethics Service (ref 11/ NW/0382).
Data sharing: Scripts for the two sample mendelian randomisation analysis are available on GitHub at: https://github.com/rcrichmond/sleep_breastcancer_mr/. For statistical code relating to the individual level data analysis in UK Biobank, please contact the corresponding author at rebecca.richmond@bristol.ac.uk.
Transparency: The lead author (RCR) affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.
This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.
References
↵Straif K, Baan R, Grosse Y, et al., WHO International Agency For Research on Cancer Monograph Working Group. Carcinogenicity of shift-work, painting, and fire-fighting. Lancet Oncol2007;8:1065-6. doi:10.1016/S1470-2045(07)70373-X pmid:19271347CrossRefPubMedWeb of ScienceGoogle Scholar
↵Fritschi L, Glass DC, Heyworth JS, et al. Hypotheses for mechanisms linking shiftwork and cancer. Med Hypotheses2011;77:430-6. doi:10.1016/j.mehy.2011.06.002 pmid:21723672CrossRefPubMedWeb of ScienceGoogle Scholar
↵Haus EL, Smolensky MH. Shift work and cancer risk: potential mechanistic roles of circadian disruption, light at night, and sleep deprivation. Sleep Med Rev2013;17:273-84. doi:10.1016/j.smrv.2012.08.003 pmid:23137527CrossRefPubMedWeb of ScienceGoogle Scholar
↵Stevens RG. Light-at-night, circadian disruption and breast cancer: assessment of existing evidence. Int J Epidemiol2009;38:963-70. doi:10.1093/ije/dyp178 pmid:19380369CrossRefPubMedWeb of ScienceGoogle Scholar
↵Samuelsson LB, Bovbjerg DH, Roecklein KA, Hall MH. Sleep and circadian disruption and incident breast cancer risk: An evidence-based and theoretical review. Neurosci Biobehav Rev2018;84:35-48. doi:10.1016/j.neubiorev.2017.10.011 pmid:29032088CrossRefPubMedGoogle Scholar
↵He C, Anand ST, Ebell MH, Vena JE, Robb SW. Circadian disrupting exposures and breast cancer risk: a meta-analysis. Int Arch Occup Environ Health2015;88:533-47. doi:10.1007/s00420-014-0986-x pmid:25261318CrossRefPubMedGoogle Scholar
↵Yang WS, Deng Q, Fan WY, Wang WY, Wang X. Light exposure at night, sleep duration, melatonin, and breast cancer: a dose-response analysis of observational studies. Eur J Cancer Prev2014;23:269-76. doi:10.1097/CEJ.0000000000000030 pmid:24858716CrossRefPubMedGoogle Scholar
↵Qin Y, Zhou Y, Zhang X, Wei X, He J. Sleep duration and breast cancer risk: a meta-analysis of observational studies. Int J Cancer2014;134:1166-73. doi:10.1002/ijc.28452 pmid:24037973CrossRefPubMedGoogle Scholar
↵Lu C, Sun H, Huang J, et al. Long-Term Sleep Duration as a Risk Factor for Breast Cancer: Evidence from a Systematic Review and Dose-Response Meta-Analysis. Biomed Res Int2017;2017:4845059. doi:10.1155/2017/4845059 pmid:29130041CrossRefPubMedGoogle Scholar
↵Ramin C, Devore EE, Pierre-Paul J, Duffy JF, Hankinson SE, Schernhammer ES. Chronotype and breast cancer risk in a cohort of US nurses. Chronobiol Int2013;30:1181-6. doi:10.3109/07420528.2013.809359 pmid:23961712CrossRefPubMedGoogle Scholar
↵Sen A, Opdahl S, Strand LB, Vatten LJ, Laugsand LE, Janszky I. Insomnia and the Risk of Breast Cancer: The HUNT Study. Psychosom Med2017;79:461-8. doi:10.1097/PSY.0000000000000417 pmid:27763987CrossRefPubMedGoogle Scholar
↵Smith GD, Ebrahim S. ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease?Int J Epidemiol2003;32:1-22. doi:10.1093/ije/dyg070 pmid:12689998CrossRefPubMedWeb of ScienceGoogle Scholar
↵Davey Smith G, Hemani G. Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Hum Mol Genet2014;23(R1):R89-98. doi:10.1093/hmg/ddu328 pmid:25064373CrossRefPubMedWeb of ScienceGoogle Scholar
↵Davies NM, Holmes MV, Davey Smith G. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians. BMJ2018;362:k601. doi:10.1136/bmj.k601 pmid:30002074FREE Full TextGoogle Scholar
↵Jones SE, Lane JM, Wood AR, et al. Genome-wide association analyses of chronotype in 697,828 individuals provides insights into circadian rhythms. Nat Commun2019;10:343. doi:10.1038/s41467-018-08259-7 pmid:30696823CrossRefPubMedGoogle Scholar
↵Dashti H, Jones SE, Wood AR, et al. Genome-wide association study identifies genetic loci for self-reported habitual sleep duration supported by accelerometer-derived estimates. Nat Commun2019;10:1100. doi:10.1038/s41467-019-08917-4 pmid:30846698CrossRefPubMedGoogle Scholar
↵Lane JM, Jones SE, Dashti HS, et al., HUNT All In Sleep. Biological and clinical insights from genetics of insomnia symptoms. Nat Genet2019;51:387-93. doi:10.1038/s41588-019-0361-7 pmid:30804566CrossRefPubMedGoogle Scholar
↵Gottlieb DJ, Hek K, Chen TH, et al. Novel loci associated with usual sleep duration: the CHARGE Consortium Genome-Wide Association Study. Mol Psychiatry2015;20:1232-9. doi:10.1038/mp.2014.133 pmid:25469926CrossRefPubMedGoogle Scholar
↵Hu Y, Shmygelska A, Tran D, Eriksson N, Tung JY, Hinds DA. GWAS of 89,283 individuals identifies genetic variants associated with self-reporting of being a morning person. Nat Commun2016;7:10448. doi:10.1038/ncomms10448 pmid:26835600CrossRefPubMedGoogle Scholar
↵Jones SE, Tyrrell J, Wood AR, et al. Genome-Wide Association Analyses in 128,266 Individuals Identifies New Morningness and Sleep Duration Loci. PLoS Genet2016;12:e1006125. doi:10.1371/journal.pgen.1006125 pmid:27494321CrossRefPubMedGoogle Scholar
↵Lane JM, Liang J, Vlasac I, et al. Genome-wide association analyses of sleep disturbance traits identify new loci and highlight shared genetics with neuropsychiatric and metabolic traits. Nat Genet2017;49:274-81. doi:10.1038/ng.3749 pmid:27992416CrossRefPubMedGoogle Scholar
↵Hammerschlag AR, Stringer S, de Leeuw CA, et al. Genome-wide association analysis of insomnia complaints identifies risk genes and genetic overlap with psychiatric and metabolic traits. Nat Genet2017;49:1584-92. doi:10.1038/ng.3888 pmid:28604731CrossRefPubMedGoogle Scholar
↵Jansen PR, Watanabe K, Stringer S, et al. Genome-wide analysis of insomnia in 1 331 010 individuals identifies new risk loci and functional pathways. Nat Genet2019;51:394-403. doi:10.1038/s41588-018-0333-3 pmid:30804565CrossRefPubMedGoogle Scholar
↵Kalmbach DA, Schneider LD, Cheung J, et al. Genetic Basis of Chronotype in Humans: Insights From Three Landmark GWAS. Sleep2017;40. doi:10.1093/sleep/zsw048 pmid:28364486CrossRefPubMedGoogle Scholar
↵Michailidou K, Lindström S, Dennis J, et al., NBCS Collaborators, ABCTB Investigators, ConFab/AOCS Investigators. Association analysis identifies 65 new breast cancer risk loci. Nature2017;551:92-4. doi:10.1038/nature24284 pmid:29059683CrossRefPubMedGoogle Scholar
↵Fry A, Littlejohns TJ, Sudlow C, et al. Comparison of Sociodemographic and Health-Related Characteristics of UK Biobank Participants With Those of the General Population. Am J Epidemiol2017;186:1026-34. doi:10.1093/aje/kwx246 pmid:28641372CrossRefPubMedGoogle Scholar
↵Sudlow C, Gallacher J, Allen N, et al. UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med2015;12:e1001779. doi:10.1371/journal.pmed.1001779 pmid:25826379CrossRefPubMedGoogle Scholar
↵Bycroft C, Freeman C, Petkova D, et al. The UK Biobank resource with deep phenotyping and genomic data. Nature2018;562:203-9. doi:10.1038/s41586-018-0579-z pmid:30305743CrossRefPubMedGoogle Scholar
↵Mitchell R, Hemani G, Dudding T, Paternoster L. UK Biobank Genetic Data: MRC-IEU Quality Control, Version 1.University of Bristol, 2017.Google Scholar
↵Loh PR, Tucker G, Bulik-Sullivan BK, et al. Efficient Bayesian mixed-model analysis increases association power in large cohorts. Nat Genet2015;47:284-90. doi:10.1038/ng.3190 pmid:25642633CrossRefPubMedGoogle Scholar
↵Hemani G, Zheng J, Elsworth B, et al. The MR-Base platform supports systematic causal inference across the human phenome. Elife2018;7:7. doi:10.7554/eLife.34408 pmid:29846171CrossRefPubMedGoogle Scholar
↵Bowden J, Del Greco M F, Minelli C, Davey Smith G, Sheehan N, Thompson J. A framework for the investigation of pleiotropy in two-sample summary data Mendelian randomization. Stat Med2017;36:1783-802. doi:10.1002/sim.7221 pmid:28114746CrossRefPubMedGoogle Scholar
↵Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol2015;44:512-25. doi:10.1093/ije/dyv080 pmid:26050253CrossRefPubMedGoogle Scholar
↵Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent Estimation in Mendelian Randomization with Some Invalid Instruments Using a Weighted Median Estimator. Genet Epidemiol2016;40:304-14. doi:10.1002/gepi.21965 pmid:27061298CrossRefPubMedGoogle Scholar
↵Hartwig FP, Davey Smith G, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. Int J Epidemiol2017;46:1985-98. doi:10.1093/ije/dyx102 pmid:29040600CrossRefPubMedGoogle Scholar
↵Bowden J, Spiller W, Del Greco M F, et al. Improving the visualization, interpretation and analysis of two-sample summary data Mendelian randomization via the Radial plot and Radial regression. Int J Epidemiol2018;47:1264-78. doi:10.1093/ije/dyy101 pmid:29961852CrossRefPubMedGoogle Scholar
↵Millard LAC, Davies NM, Timpson NJ, Tilling K, Flach PA, Davey Smith G. MR-PheWAS: hypothesis prioritization among potential causal effects of body mass index on many outcomes, using Mendelian randomization.Scientific Reports, 2015: 5.Google Scholar
↵Zhao Q, Wang J, Hemani G, Bowden J, Small D. Statistical inference in two-sample summary data Mendelian randomization using a robust adjusted profile score. arXiv. 2018;https://arxiv.org/abs/1801.09652.Google Scholar
↵Burgess S, Labrecque JA. Mendelian randomization with a binary exposure variable: interpretation and presentation of causal estimates. arXiv. 2018;https://arxiv.org/abs/1804.05545.Google Scholar
↵Jones SE, van Hees VT, Mazzotti DR, et al. Genetic studies of accelerometer-based sleep measures yield new insights into human sleep behaviour. Nat Commun2019;10:1585. doi:10.1038/s41467-019-09576-1 pmid:30952852CrossRefPubMedGoogle Scholar
↵Braaten T, Weiderpass E, Kumle M, Adami HO, Lund E. Education and risk of breast cancer in the Norwegian-Swedish women’s lifestyle and health cohort study. Int J Cancer2004;110:579-83. doi:10.1002/ijc.20141 pmid:15122591CrossRefPubMedWeb of ScienceGoogle Scholar
↵Mocellin S, Tropea S, Benna C, Rossi CR. Circadian pathway genetic variation and cancer risk: evidence from genome-wide association studies. BMC Med2018;16:20. doi:10.1186/s12916-018-1010-1 pmid:29455641CrossRefPubMedGoogle Scholar
↵Burgess S, Thompson SG. Multivariable Mendelian randomization: the use of pleiotropic genetic variants to estimate causal effects. Am J Epidemiol2015;181:251-60. doi:10.1093/aje/kwu283 pmid:25632051CrossRefPubMedGoogle Scholar
↵Munafò MR, Tilling K, Taylor AE, Evans DM, Davey Smith G. Collider scope: when selection bias can substantially influence observed associations. Int J Epidemiol2018;47:226-35. doi:10.1093/ije/dyx206 pmid:29040562CrossRefPubMedGoogle Scholar
↵Lawlor DA, Tilling K, Davey Smith G. Triangulation in aetiological epidemiology. Int J Epidemiol2016;45:1866-86.pmid:28108528CrossRefPubMedGoogle Scholar
↵Munafò MR, Davey Smith G. Robust research needs many lines of evidence. Nature2018;553:399-401. doi:10.1038/d41586-018-01023-3 pmid:29368721CrossRefPubMedGoogle Scholar
View Abstract
Article tools
0 responses
Respond to this article
Data supplement
Print
Alerts & updates
Citation tools
Request permissions
Author citation
Add article to BMJ Portfolio
Email to a friend
UK jobs
International jobs
Cambridge University Hospitals NHS Foundation Trust: Locum Consultant in Palliative Care
Cleveland Clinic, Jersey: GP Partner required
Strathcarron Hospice: Consultant/Locum Consultant in Palliative Medicine
University of Glasgow: Clinical Lecturer in Geriatric Medicine
CARE Fertility Group: Consultant in Reproductive Medicine/Fertility Specialist
View more
Altmetric
Who is talking about this article?
Article has an altmetric score of 21
See more details
Picked up by 1 news outlets
Tweeted by 18
This week's poll
Is resilience training helpful?
Yes
No
Vote
View Results
Read related article
See previous polls
Other content recommended for you,
Evaluation of the causal effects between subjective wellbeing and cardiometabolic health: mendelian randomisation study
Robyn E Wootton et al., The BMJ, 2018
Height, body mass index, and socioeconomic status: mendelian randomisation study in UK Biobank
Katherine S Ruth et al., The BMJ, 2016
Height, body mass index, and socioeconomic status: mendelian randomisation study in UK Biobank
Jessica Tyrrell et al., The BMJ, 2016
Body mass index and all cause mortality in HUNT and UK Biobank studies: linear and non-linear mendelian randomisation analyses
Yi-Qian Sun et al., The BMJ, 2019
Role of obesity in smoking behaviour: Mendelian randomisation study in UK Biobank
Robert Carreras-Torres et al., The BMJ, 2018
Inference in Psychiatry via 2-Sample Mendelian Randomization—From Association to Causal Pathway?
Enda M. Byrne et al., JAMA Psychiatry, 2017
Genetic Association of Waist-to-Hip Ratio With Cardiometabolic Traits, Type 2 Diabetes, and Coronary Heart Disease
Connor A. Emdin et al., Journal of American Medical Association, 2017
Telomere Length Shortening and Alzheimer Disease—A Mendelian Randomization Study
Yiqiang Zhan et al., JAMA Neurology, 2015
Powered by
I consent to the use of Google Analytics and related cookies across the TrendMD network (widget, website, blog). Learn more Yes No
Back to top
BMJ 2019; 365 doi: https://doi.org/10.1136/bmj.l2323 (Published 26 June 2019)
Cite this as: BMJ 2019;365:l2323
Article
Related content
Metrics
Responses
Peer review
Alexander Mok, PhD student1, Kay-Tee Khaw, professor2, Robert Luben, head of bioinformatics2, Nick Wareham, professor1, Soren Brage, research programme leader1
Author affiliations
Correspondence to: S Brage soren.brage@mrc-epid.cam.ac.uk
Accepted 25 April 2019
Abstract
Objective To assess the prospective associations of baseline and long term trajectories of physical activity on mortality from all causes, cardiovascular disease, and cancer.
Design Population based cohort study.
Setting Adults from the general population in the UK.
Participants 14 599 men and women (aged 40 to 79) from the European Prospective Investigation into Cancer and Nutrition-Norfolk cohort, assessed at baseline (1993 to 1997) up to 2004 for lifestyle and other risk factors; then followed to 2016 for mortality (median of 12.5 years of follow-up, after the last exposure assessment).
Main exposure Physical activity energy expenditure (PAEE) derived from questionnaires, calibrated against combined movement and heart rate monitoring.
Main outcome measures Mortality from all causes, cardiovascular disease, and cancer. Multivariable proportional hazards regression models were adjusted for age, sex, sociodemographics, and changes in medical history, overall diet quality, body mass index, blood pressure, triglycerides, and cholesterol levels.
Results During 171 277 person years of follow-up, 3148 deaths occurred. Long term increases in PAEE were inversely associated with mortality, independent of baseline PAEE. For each 1 kJ/kg/day per year increase in PAEE (equivalent to a trajectory of being inactive at baseline and gradually, over five years, meeting the World Health Organization minimum physical activity guidelines of 150 minutes/week of moderate-intensity physical activity), hazard ratios were: 0.76 (95% confidence interval 0.71 to 0.82) for all cause mortality, 0.71 (0.62 to 0.82) for cardiovascular disease mortality, and 0.89 (0.79 to 0.99) for cancer mortality, adjusted for baseline PAEE, and established risk factors. Similar results were observed when analyses were stratified by medical history of cardiovascular disease and cancer. Joint analyses with baseline and trajectories of physical activity show that, compared with consistently inactive individuals, those with increasing physical activity trajectories over time experienced lower risks of mortality from all causes, with hazard ratios of 0.76 (0.65 to 0.88), 0.62 (0.53 to 0.72), and 0.58 (0.43 to 0.78) at low, medium, and high baseline physical activity, respectively. At the population level, meeting and maintaining at least the minimum physical activity recommendations would potentially prevent 46% of deaths associated with physical inactivity.
Conclusions Middle aged and older adults, including those with cardiovascular disease and cancer, can gain substantial longevity benefits by becoming more physically active, irrespective of past physical activity levels and established risk factors. Considerable population health impacts can be attained with consistent engagement in physical activity during mid to late life.
Introduction
Physical activity is associated with lower risks of all cause mortality, cardiovascular disease, and certain cancers.123 However, much of the epidemiology arises from observational studies assessing physical activity at a single point in time (at baseline), on subsequent mortality and chronic disease outcomes. From 1975 to 2016, over 90% of these epidemiological investigations on physical activity and mortality have used a single assessment of physical activity at baseline.4 Relating mortality risks to baseline physical activity levels does not account for within-person variation over the long term, potentially diluting the epidemiological associations. As physical activity behaviours are complex and vary over the life course,5 assessing within-person trajectories of physical activity over time would better characterise the association between physical activity and mortality.
Fewer studies have assessed physical activity trajectories over time and subsequent risks of mortality.67891011 Some of these investigations have only included small samples of older adults, in either men or women. Importantly, most studies were limited by crude categorisations of physical activity patterns, without exposure calibration against objective measures with established validity. Many studies also do not adequately account for concurrent changes in other lifestyle risk factors—such as overall diet quality and body mass index—which might potentially confound the association between physical activity and mortality. This is important, as some studies have shown that associations between physical activity and weight gain are weak or inconsistent, suggesting that being overweight or obese might instead predict physical inactivity rather than the reverse.1213 Previous investigations have also not quantified the population impact of different physical activity trajectories over time on mortality. We examined associations of baseline and long term trajectories of within-person changes in physical activity on all cause, cardiovascular disease, and cancer mortality in a population based cohort study and quantified the number of preventable deaths from the observed physical activity trajectories.
Methods
Study population
The data for this investigation were from the European Prospective Investigation into Cancer and Nutrition-Norfolk (EPIC-Norfolk) study, comprising a baseline assessment and three follow-up assessments. The EPIC-Norfolk study is a population based cohort study of 25 639 men and women aged 40 to 79, resident in Norfolk, UK, and recruited between 1993 to 1997 from community general practices as previously described.14
After the baseline clinic assessment (1993 to 1997), the first follow-up (postal questionnaire) was conducted between 1995 and 1997 at a mean of 1.7 (SD 0.1) years after baseline, the second follow-up (clinic visit) took place 3.6 (0.7) years after baseline, and the third follow-up (postal questionnaire) was initiated 7.6 (0.9) years after the baseline clinic visit. All participants with repeated measures of physical activity (at least baseline and final follow-up assessments) were included, resulting in an analytical sample of 14 599 men and women.
Assessment of physical activity
Habitual physical activity was assessed with a validated questionnaire, with a reference time frame of the past year.1516 The first question inquired about occupational physical activity, classified as five categories: unemployed, sedentary (eg, desk job), standing (eg, shop assistant, security guard), physical work (eg, plumber, nurse), and heavy manual work (eg, construction worker, bricklayer). The second open ended question asked about time spent (hours/week) on cycling, recreational activities, sports, or physical exercise, separately for winter and summer.
The validity of this instrument has previously been examined in an independent validation study, by using individually-calibrated combined movement and heart rate monitoring as the criterion method; physical activity energy expenditure (PAEE) increased through each of four ordinal categories of self reported physical activity comprising both occupational and leisure time physical activity.15 In this study, we disaggregated the index of total physical activity into its original two variables that were domain specific and conducted a calibration to PAEE using the validation dataset, in which the exact same instrument had been used (n=1747, omitting one study centre that had used a different instrument). Specifically, quasi-continuous and marginalised values of PAEE in units of kJ/kg/day were derived from three levels of occupational activity (unemployed or sedentary occupation; standing occupation; and physical or heavy manual occupation) and four levels of leisure time physical activity (none; 0.1 to 3.5 hours; 3.6 to 7 hours; and >7 hours per week). This regression procedure allows the domain specific levels of occupational and leisure time physical activity to have independent PAEE coefficients, while assigning a value of 0 kJ/kg/day to individuals with a sedentary (or no) occupation and reporting no leisure time physical activity (LTPA). The resulting calibration equation was: PAEE (kJ/kg/day) =0 (sedentary or no job) + 5.61 (standing job) + 7.63 (manual job) + 0 (no LTPA) + 3.59 (LTPA of 0.1 to 3.5 hours per week) + 7.17 (LTPA of 3.6 to 7 hours per week) + 11.26 (LTPA >7 hours per week).
Assessment of covariates
Information about participants’ lifestyle and clinical risk factors were obtained at both clinic visits, carried out by trained nurses at baseline and 3.6 years later. Information collected during clinic visits included: age; height; weight; blood pressure; habitual diet; alcohol intake (units consumed per week); smoking status (never, former, and current smokers); physical activity; social class (unemployed, non-skilled workers, semiskilled workers, skilled workers, managers, and professionals); education level (none, General Certificate of Education (GCE) Ordinary Level, GCE Advanced Level, bachelor’s degree, and above); and medical history of heart disease, stroke, cancer, diabetes, fractures (wrist, vertebral, and hip), asthma, and other chronic respiratory conditions (bronchitis and emphysema). Additionally, updated information on heart disease, stroke, and cancer up to the final physical activity assessment (third follow-up) were also collected by using data from hospital episode statistics. This is a database containing details of all admissions, including emergency department attendances and outpatient appointments at National Health Service hospitals in England. Non-fasting blood samples were collected and refrigerated at 4°C until transported within a week of sampling to be assayed for serum triglycerides, total cholesterol, and high density lipoprotein cholesterol by using standard enzymatic techniques. We derived low density lipoprotein cholesterol by using the Friedewald equation.17
We assessed habitual dietary intake during the previous year by using validated 130 item food-frequency questionnaires administered at baseline and at the second clinic visit. The validity of this food-frequency questionnaire for major foods and nutrients was previously assessed against 16 day weighed diet records, 24 hour recall, and selected biomarkers in a subsample of this cohort.1819 We created a comprehensive diet quality score for each participant, separately for baseline and at follow-up, incorporating eight dietary components known to influence health and the risk of chronic disease.20 The composite diet quality score included: wholegrains, refined grains, sweetened confectionery and beverages, fish, red and processed meat, fruit and vegetables, sodium, and the ratio of unsaturated to saturated fatty acids from dietary intakes. We created tertiles for each dietary component and then scored these as −1, 0, or 1, with the directionality depending on whether the food or nutrient was associated with health risks or benefits.20 Scores from the eight dietary components were summed into an overall diet quality score which ranged from −8 to 8, with higher values representing a healthier dietary pattern. We also collected updated information on body weight and height from the two postal assessments (first and third follow-up).
Mortality ascertainment
All participants were followed-up for mortality by the Office of National Statistics until the most recent censor date of 31 March 2016. Causes of death were confirmed by death certificates which were coded by nosologists according to ICD-9 (international classification of diseases, ninth revision) and ICD-10 (international classification of diseases, 10th revision). We defined cancer mortality and cardiovascular disease mortality by using codes ICD-9 140-208 or ICD-10 C00-C97 and ICD-9 400-438 or ICD-10 I10-I79, respectively.
Statistical analysis
We used Cox proportional hazards regression models to derive hazard ratios and 95% confidence intervals. Individuals contributed person time from the date of the last physical activity assessment (third follow-up) until the date of death or censoring. We used all available assessments of physical activity to better represent long term habitual physical activity and used linear regression against elapsed time to derive an overall physical activity trajectory (ΔPAEE) for each individual. We used the resulting coefficient of the calibrated ΔPAEE values in kJ/kg/day/year, together with baseline PAEE, as mutually-adjusted exposure variables in the Cox regression models.
We created categories reflecting approximate tertiles of both baseline PAEE and ΔPAEE to investigate joint effects of baseline and long term trajectories of physical activity. We defined the categories of baseline PAEE as: low (PAEE=0 kJ/kg/day), medium (0<PAEE<8.4 kJ/kg/day), and high (PAEE≥8.4 kJ/kg/day). We defined the categories of ΔPAEE over time as: decreasers (ΔPAEE≤−0.20 kJ/kg/day/year), maintainers (−0.20<ΔPAEE<0.20 kJ/kg/day/year), and increasers (ΔPAEE≥0.20 kJ/kg/day/year). We then created joint exposure categories by cross-classifying the three baseline by the three trajectory categories, resulting in eight categories. The reference group was individuals with consistently low physical activity (by definition, there would be no exposure category comprising individuals declining from no baseline physical activity). We estimated the potential number of preventable deaths at the population level in each joint exposure category, using the absolute difference in adjusted mortality rates between the reference group (consistently inactive) and each joint exposure category, multiplied by the person years observed in the corresponding joint exposure category. We derived adjusted mortality rates by using multivariable exponential regression, with covariates used in the most comprehensively adjusted analytical model.
In model 1 we adjusted for: general demographics (age, sex, socioeconomic status, education level, and smoking status), dietary factors (total energy intake, overall diet quality, alcohol consumption), and medical history (asthma, chronic respiratory conditions, bone fractures, diabetes, heart disease, stroke, and cancer). Age, energy and alcohol intake, and diet quality were continuous variables. In model 2 we accounted for changes in the above covariates by further inclusion of updated variables at the second clinic visit (3.6 years later), as well as updated status of cardiovascular disease and cancer from hospital episode statistics up until the final physical activity assessment. In model 3 we further accounted for changes in body mass index by including continuous values of body mass index at baseline and at the final physical activity assessment. In model 4 we accounted for changes in blood pressure and lipids by further including continuous values of systolic and diastolic blood pressure, serum triglycerides, low density lipoprotein cholesterol, and high density lipoprotein cholesterol at baseline and at the second clinic visit.
We used height and weight measurements from the baseline and second clinic visit to calibrate self reported height and weight provided by the postal questionnaires. Self reported values were multiplied by the ratio of mean clinically-measured values and self-reported values. We imputed missing values of covariates at follow-up by using regression on their baseline values. A complete case analysis was conducted as a sensitivity analysis. Reverse causation owing to undiagnosed disease was mitigated by excluding participants who died within one year of the final physical activity assessment (beginning of follow-up for mortality) in all analyses. Predefined subgroups were age, sex, clinically-defined cut points of body mass index, and history of cardiovascular disease and cancer. We performed additional sensitivity analyses by excluding individuals with any period-prevalent chronic diseases (heart disease, stroke, and cancer) up to the final physical activity assessment, as well as excluding deaths occurring within two years of the final physical activity assessment. All analyses were performed by using Stata SE version 14.2.
Patient and public involvement
Patients and members of the public were not formally involved in the design, analysis or interpretation of this study. Nonetheless, the research question in this article is of broad public health interest. The results of this study will be disseminated to study participants and the general public through the study websites, participant engagement events, seminars, and conferences.
Results
Study population
Among 14 599 participants with a mean baseline age of 58.0 (SD 8.8), followed for a median of 12.5 (interquartile range 11.9-13.2) years after the final physical activity assessment, there were 3148 deaths (950 from cardiovascular disease and 1091 from cancer) during 171 277 person years of follow-up. Table 1 shows the study population characteristics at the four assessment time points. On average, dietary factors such as total energy intake, alcohol consumption, and overall diet quality were similar at baseline and at the second clinic visit. The prevalence of diabetes, cardiovascular disease, cancer, and respiratory conditions increased over time. From baseline to the final follow-up assessment, mean body mass index increased from 26.1 kg/m2 to 26.7 kg/m2, and mean PAEE declined by 17% from 5.9 kJ/kg/day to 4.9 kJ/kg/day. The Pearson correlation coefficients were r=0.57 between PAEE at baseline and 1.7 years later; and r=0.45 between PAEE at baseline and 7.6 years later (final physical activity assessment).
Table 1 Study population characteristics at baseline and follow-up assessments. Values are means (SD) unless stated otherwise
View popupView inline
Associations of baseline and trajectories of physical activity with mortality
Table 2 shows that for each 1 kJ/kg/day/year increase in PAEE over time (ΔPAEE), the hazard ratios were: 0.78 (95% confidence interval 0.73 to 0.84) for all cause mortality, 0.75 (0.66 to 0.86) for cardiovascular disease mortality, and 0.88 (0.79 to 0.98) for cancer mortality (model 1). Progressive adjustments for time-updated covariates (model 2), changes in body mass index (model 3), and changes in blood pressure and blood lipids (model 4) did not attenuate the strength of the associations. In all models, baseline PAEE was also independently associated with lower mortality; for each 10 kJ/kg/day difference between individuals, hazard ratios were 0.70 (95% confidence interval 0.64 to 0.78) for all cause mortality, 0.69 (0.57 to 0.83) for cardiovascular disease mortality, and 0.83 (0.70 to 0.98) for cancer mortality (table 2, model 4). There was no evidence of an interaction between baseline PAEE and ΔPAEE for all mortality outcomes (P>0.6 from likelihood-ratio tests). The effect of PAEE averaged across all assessments on overall mortality, was 0.70 (0.62 to 0.78) for each 10 kJ/kg/day difference between individuals. For single time point exposure assessments, the inverse association of PAEE with mortality at the most recent assessment was stronger than that for baseline PAEE; hazard ratios were 0.68 (0.62 to 0.75) and 0.87 (0.80 to 0.94) for each 10 kJ/kg/day difference, respectively.
Table 2 Associations of mutually-adjusted baseline physical activity energy expenditure (PAEE) and trajectories of physical activity (ΔPAEE) with mortality. Values are hazard ratios (95% confidence intervals) unless stated otherwise
View popupView inline
Sensitivity analyses excluding individuals with any period-prevalent heart disease, stroke, and cancer occurring up to the final physical activity assessment, as well as any deaths occurring within two years of this final assessment, showed similar associations with mortality for baseline PAEE and ΔPAEE, hazard ratios of 0.72 (95% confidence interval 0.63 to 0.81) for each 10 kJ/kg/day difference and 0.78 (0.71 to 0.86) for each 1 kJ/kg/day/year difference, respectively. Sensitivity analysis that used complete cases did not materially change the strength of associations (attenuation <5% for ΔPAEE estimates for all outcomes), but it attenuated the statistical significance for cancer mortality (supplementary table 1). Adjustments for occupational physical activity categories (sedentary, standing, physical, and heavy manual jobs) (supplementary table 2) slightly strengthened the association of ΔPAEE with cardiovascular disease mortality and attenuated the association with cancer mortality, whereas associations for baseline PAEE became stronger for both these outcomes.
Stratified analyses
Figure 1 shows that, based on the most comprehensively-adjusted analytical model (model 4), significant inverse associations for baseline PAEE and ΔPAEE with all cause mortality persisted in all subgroups of age, sex, adiposity, and chronic disease status. Although tests for interaction were not statistically significant for any subgroup, the benefit of baseline PAEE on all cause mortality tended to be stronger in women (hazard ratio 0.63, 95% confidence interval 0.53 to 0.74) than men (0.76, 0.66 to 0.87; P=0.08). Baseline PAEE and ΔPAEE were not associated with cardiovascular disease mortality in individuals with obesity. In stratified analyses for cancer mortality, the longevity benefits of both baseline PAEE and ΔPAEE were only significant in older adults.
Fig 1
Fig 1
Associations of baseline and long term trajectories of physical activity energy expenditure (PAEE) with all cause, cardiovascular disease, and cancer mortality, stratified by age group, sex, body mass index (BMI), and disease status. Hazard ratios are mutually adjusted for both baseline PAEE and ΔPAEE, and are based on the most comprehensively adjusted model for changes in covariates, including medical history, diet quality, body mass index, blood pressure, and lipids (model 4 from table 2).
Download figure Open in new tab Download powerpoint
Joint associations of baseline and trajectories of physical activity with mortality
Figure 2 shows that compared with individuals who were consistently inactive (low-maintainers), individuals with medium and high baseline physical activity who maintained these levels (medium-maintainers and high-maintainers) had significantly lower risks of all cause mortality, 28% and 33% respectively. Individuals with increasing physical activity trajectories experienced additional longevity benefits, including those with low baseline activity, as well as those with already high levels of baseline physical activity. Dose-response gradients were observed within and between strata of baseline physical activity levels. Within strata of low, medium, and high baseline physical activity, the risk of mortality was lower across ordinally increasing trajectories of: decreasers, maintainers, and increasers. Between strata of baseline physical activity, the risk of mortality decreased by 24% for low-increasers, 38% for medium-increasers, and 42% for high-increasers. Medium-decreasers and high-decreasers had 10% and 20% lower risks of mortality, respectively, compared with the reference group of low-maintainers.
Fig 2
Fig 2
Joint associations of baseline and trajectories of physical activity energy expenditure (PAEE) with all cause mortality. Hazard ratios (HR) are based on the most comprehensively adjusted model for age, sex, sociodemographics, and changes in medical history, diet quality, body mass index, blood pressure, and lipids (model 4 from table 2). Adjusted mortality rates are expressed per 100 000 person years. WHO=World Health Organization.
Download figure Open in new tab Download powerpoint
Estimation of population impact
Figure 2 shows that if the entire cohort remained inactive over time, an additional 24% of deaths (678 more than the observed 2840 deaths) would have potentially occurred. At the population level, the greatest number of potential deaths averted were in the medium-increasers and medium-maintainers, preventing 169 (25%) and 143 (21%) of the deaths associated with physical inactivity, respectively. All physical activity trajectories that culminated with meeting at least the minimum physical activity guidelines (equivalent to 5 kJ/kg/day) could potentially prevent 93% of the deaths associated with physical inactivity at the population level.
Discussion
In this prospective cohort study with repeated assessments, we found protective associations for increasing physical activity trajectories against mortality from all causes, cardiovascular disease, and cancer, irrespective of past physical activity levels. These associations were also independent of levels and changes in several established risk factors such as overall diet quality, body mass index, medical history, blood pressure, triglycerides, and cholesterol. Both higher physical activity levels at baseline and increasing trajectories over time were protective against mortality. Notably, the strength of associations was similar in individuals with and without pre-existing cardiovascular disease and cancer. These results are encouraging, not least for middle aged and older adults with cardiovascular disease and cancer, who can still gain substantial longevity benefits by becoming more active, lending further support to the broad public health benefits of physical activity.
Independent and joint effects of baseline and trajectories of physical activity
The absence of an interaction between baseline physical activity levels and long term trajectories of physical activity on the risk of mortality suggests that the relative longevity benefit of increasing physical activity is consistent, irrespective of baseline levels. Increasing PAEE by 1 kJ/kg/day per year—equivalent to a trajectory of being inactive at baseline and then subsequently increasing physical activity to 5 and 10 kJ/kg/day, five and 10 years later, respectively—was associated with a 24% lower risk of all cause mortality. This gain in longevity from increasing physical activity over time, is in addition to the benefits already accrued from baseline physical activity, such as a 30% lower risk of mortality for a between-individual difference of 10 kJ/kg/day. For reference, 5 kJ/kg/day corresponds to the World Health Organization minimum physical activity guidelines of 150 minutes per week of moderate-intensity physical activity, and 10 kJ/kg/day corresponds to the WHO recommendations of 300 minutes per week of moderate-intensity physical activity for additional health benefits. Figure 3 shows how these levels of physical activity can be achieved in any number of ways during leisure time and at work, with the required duration depending on relative intensities of the activities undertaken.
Fig 3
Fig 3
Physical activity energy expenditure (PAEE) of common activities performed during leisure time and at work. MET=metabolic equivalent of task. WHO=World Health Organization
Download figure Open in new tab Download powerpoint
The joint analyses of physical activity trajectories beginning from different baseline levels showed that adults who were already meeting at least the minimum physical activity recommendations (150 minutes per week of moderate physical activity), experience substantial longevity benefits by either maintaining or further increasing physical activity levels. This is evidenced by medium-maintainers and high-maintainers experiencing 28% and 33% lower risks of mortality, with an additional ~10% lower risk for increasers in both these baseline groups.
Adults already meeting the equivalent of the higher WHO physical activity recommendations (300 minutes per week of moderate physical activity) still gain further longevity benefits by increasing physical activity levels to over 14 kJ/kg/day. This energy expenditure corresponds to approximately three times the recommended minimum (equivalent to 450 minutes per week of moderate physical activity).
The joint analyses also revealed that some, but not all, of the longevity benefits from past physical activity levels are lost when previously active individuals decrease their activity levels. Compared with the consistently inactive, medium-decreasers and high-decreasers experienced 10% and 20% lower risks of mortality, respectively. However, these effects appear modest, compared with the 28% and 33% lower risks of mortality for medium-maintainers and high-maintainers, respectively. The low-increasers also experienced slightly lower risks of mortality than the high-decreasers (24% v 20% lower than the consistently-inactive reference group, respectively) but both had higher risks of mortality than the medium-maintainers (28% lower risk), despite all these three groups ending up at approximately the same physical activity level at the final exposure assessment. There might be several explanations for this, including: the relative importance of past versus more recent physical activity; differential factors that might have caused these specific physical activity trajectories in the first place, beyond differences in period-prevalent cardiovascular disease or cancer; and the degree to which physical activity levels were maintained or continued to change beyond the last exposure assessment until the date of final censoring.
Population impact
Although mortality benefits were greatest in the high-increasers (hazard ratio of 0.58), the fraction of potential deaths averted at the population level were greatest for the medium-increasers (25%) and the medium-maintainers (21%), in contrast with 10% for the high-increasers. This is owing to the combination of a moderately strong aetiological association (hazard ratio of 0.72) and a greater prevalence of medium-maintainers (23 032 person years, 15% of total person years), compared with high-increasers (6988 person years, 4% of total). All physical activity trajectories during middle to late adulthood that culminate with meeting at least the minimum physical activity guidelines could potentially prevent 93% of deaths attributable to physical inactivity. The last 7% were the 48 deaths potentially prevented by the medium-decreasers, who as a group did not meet the minimum physical activity guidelines at the final exposure assessment. Had this group maintained their baseline physical activity levels, an additional 136 deaths (nearly three times as many) may have been prevented. Comparatively fewer, yet still an extra 119 deaths (twice as many) could have been prevented if high-decreasers had maintained their baseline physical activity levels. These two groups with declining physical activity levels were also the most prevalent trajectories in the cohort. Thus, in addition to shifting the population towards meeting the minimum physical activity recommendations, public health efforts should also focus on the maintenance of physical activity levels, specifically preventing declines over middle and late life. The WHO minimum guidelines of 150 minutes per week of moderate-intensity physical activity appears to be a realistic public health target, given that these levels were observed to be broadly attainable at the population level. Individuals with existing chronic conditions such as cardiovascular disease and cancer—for whom our study has also shown to gain longevity benefits—might choose to engage in commensurably lower-intensity activities but for a longer duration (fig 3). Further research is, however, needed to specifically ascertain the health benefits of lower-intensity physical activity in both healthy individuals and those with major chronic diseases.21
Comparisons with existing studies
This study also showed that the longevity benefits of increasing physical activity are independent of intermediary changes in several established risk factors, including body mass index, blood pressure, triglycerides, and cholesterol. These results are interesting, relative to other studies that have shown considerable attenuation of the strength of associations after adjustment for similar cardiometabolic biomarkers.222 In our study, it is somewhat surprising that the inverse associations of physical activity with cardiovascular disease mortality were not attenuated (but rather strengthened) after adjusting for established cardiometabolic risk factors. These findings support research into other potential mechanisms, including vascular function,23 novel lipids,24 and autonomic nervous system activity,25 through which physical activity might protect against cardiovascular disease. In our study, the protective associations of physical activity were stronger for cardiovascular disease mortality than for cancer mortality, suggesting that longevity benefits were primarily driven through the prevention of cardiovascular-related deaths. Adjustment for occupational physical activity strengthened the inverse associations with cardiovascular disease mortality for both baseline PAEE and ΔPAEE; this was also the case for the association between baseline PAEE and cancer mortality, but the association between ΔPAEE and cancer mortality was attenuated. The existing body of evidence, from a meta-analysis of nine international cohort studies also reported stronger inverse associations for cardiovascular disease mortality, compared with cancer mortality.2627 The weaker associations with cancer mortality might reflect the notion that cancers are a collection of neoplastic diseases, which might be aetiologically diverse and characterised by separate pathophysiologies.1
Our results for the association of baseline physical activity with mortality were broadly similar to those reported in the literature, although our estimates have accounted for changes in physical activity over time, which to some degree would correct for regression dilution bias.28 In a pooled analysis of data from high income countries within the Prospective Urban Rural Epidemiologic (PURE) study,29 medium baseline physical activity (150 to 750 minutes per week of moderate-intensity physical activity) was associated with a 31% lower risk of mortality. This is similar to our estimates of a 30% lower risk of mortality for a 10 kJ/kg/day difference between individuals (equivalent to 300 minutes per week of moderate-intensity physical activity).
Another pooled analysis examining the dose-response relation between baseline leisure time physical activity and mortality also reported lower mortality risks of between 31% to 37% at a comparable volume of physical activity.30 Comparisons with previous studies, examining specifically the changes and patterns of physical activity over time on mortality, are difficult owing to methodological and analytical heterogeneity between studies, precluding the synthesis of published results using meta-analytic methods. There was considerable variation in the operational definitions of the “changes” in physical activity over time. Some classified “changes” as increases and decreases, compared with unchanged physical activity irrespective of baseline levels31; others grouped varying activity levels over time as “mixed patterns,”4 potentially obscuring the benefits for individuals who improved physical activity levels over time; yet others used a reference group of the “consistently-active.”69 Furthermore, the time periods for studying these physical activity trajectories were also variable, with some studies examining short term changes within one to two years,3233 and others examining changes over 10 years.61011 Nonetheless, the relative risks of our high-maintainer and medium-maintainer groups (with hazard ratios of 0.67 and 0.72, respectively) were broadly in the ranges of the consistently-active groups reported in previous studies.7811 Future work examining physical activity trajectories over time on health outcomes could consider pooling of individual-level harmonised data from compatible studies with repeated follow-up assessments, ideally combined with external calibration; this would enable standardisation of exposure definitions and analytical approaches.
Strengths and limitations of the study
On balance, we present a comprehensive analysis, examining longitudinal physical activity trajectories in a large cohort with long follow-up for mortality, and quantified the population health impact from different physical activity trajectories. To overcome limitations in the majority of studies which have predominantly examined mortality associations with physical activity assessed at a single time point, we incorporated repeated measures of physical activity calibrated against objective measurements of individually-calibrated combined movement and heart rate monitoring. The use of longitudinal, within-individual trajectories of physical activity over time also precludes any confounding by time-invariant factors such as genetics. Our approach offers a stronger operationalisation of physical activity exposures, representing a method which can be used in future longitudinal studies investigating the associations between physical activity and subsequent health outcomes. Our study showed robust protective associations between physical activity and mortality, even after controlling for established risk factors, such as overall diet quality, body mass index, blood pressure, triglycerides, and cholesterol.
Some limitations of our study are that the analytical sample comprised of individuals who were available for follow-up approximately a decade after initial recruitment. Thus, a healthy cohort effect cannot be excluded. This, however, would only serve to render our findings more conservative. As the study was observational, residual confounding owing to unmeasured factors might still be possible. However, it would be virtually impossible to study the effects of habitual physical activity on mortality in a randomised controlled trial, and the observational nature of this study broadly shows the attainable longevity benefits of physical activity trajectories observed in the real world.
Conclusion
We showed that middle aged and older adults, including those with cardiovascular disease and cancer, stand to gain substantial longevity benefits by becoming more physically active, irrespective of past physical activity levels and established risk factors—including overall diet quality, body mass index, blood pressure, triglycerides, and cholesterol. Maintaining or increasing physical activity levels from a baseline equivalent to meeting the minimum public health recommendations has the greatest population health impact, with these trajectories being responsible for preventing nearly one in two deaths associated with physical inactivity. In addition to shifting the population towards meeting the minimum physical activity recommendations, public health efforts should also focus on the maintenance of physical activity levels, specifically preventing declines over mid to late life.
What is already known on this topic
Physical activity assessed at a single time point is associated with lower risks of mortality from all causes, cardiovascular disease, and cancer
Fewer studies have examined long term changes in physical activity and quantified the population health impact of different activity trajectories
What this study adds
Middle aged and older adults, including those with cardiovascular disease and cancer, stand to gain substantial longevity benefits by becoming more physically active, regardless of past activity levels, and changes in established risk factors, including overall diet quality, bodyweight, blood pressure, triglycerides, and cholesterol
At the population level, meeting and maintaining at least the minimum public health recommendations (150 minutes per week of moderate-intensity physical activity) would potentially prevent 46% of deaths associated with physical inactivity
Public health strategies should shift the population towards meeting the minimum recommendations, and importantly, focus on preventing declines in physical activity during middle and late life
Acknowledgments
We are grateful to the EPIC-Norfolk study participants for their voluntary contribution towards public health research, and for the support of EPIC functional teams (study coordination, field epidemiology team, IT, and data management). We also thank Stephen Sharp (Senior Statistician, MRC Epidemiology Unit, University of Cambridge) and Charles Matthews (Senior Investigator, Division of Cancer Epidemiology and Genetics, Metabolic Epidemiology Branch, National Cancer Institute, USA) for helpful comments on statistical analyses and the contextualisation of results. The authors assume full responsibility for the analysis and interpretation of the data in this study.
Footnotes
Contributors: KTK, RL, and NW contributed to the design of the EPIC-Norfolk study. AM and SB conceptualised the design of the present analysis and analysed the data. AM wrote the first draft of the manuscript. All authors had full access to the data in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis. AM and SB and are the guarantors. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Funding: The EPIC-Norfolk study is supported by programme grants from the Medical Research Council and Cancer Research UK with additional support from the Stroke Association, British Heart Foundation, Department of Health, Food Standards Agency, and the Wellcome Trust. AM was supported by the National Science Scholarship from Singapore, A*STAR (Agency for Science, Technology and Research). The work of NW and SB was funded by the Medical Research Council UK (MC_UU_12015/1 and MC_UU_12015/3). The funders had no role in the study design; the collection, analysis, and interpretation of data; the writing of the manuscript; or the decision to submit the article for publication.
Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: no support from any additional organisations for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work.
Ethical approval: The study was approved by the Norfolk District Health Authority Ethics Committee and adhered to the World Medical Association’s Declaration of Helsinki.
Patient consent: All participants gave written informed consent before enrolment in the study.
Data sharing: No additional data are available.
The lead author (AM) affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.
This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.
References
↵Moore SC, Lee I-M, Weiderpass E, et al. Association of Leisure-Time Physical Activity With Risk of 26 Types of Cancer in 1.44 Million Adults. JAMA Intern Med2016;176:816-25. doi:10.1001/jamainternmed.2016.1548. pmid:27183032CrossRefPubMedGoogle Scholar
↵Mora S, Cook N, Buring JE, Ridker PM, Lee I-M. Physical activity and reduced risk of cardiovascular events: potential mediating mechanisms. Circulation2007;116:2110-8. doi:10.1161/CIRCULATIONAHA.107.729939. pmid:17967770Abstract/FREE Full TextGoogle Scholar
↵Lee I-M, Shiroma EJ, Lobelo F, Puska P, Blair SN, Katzmarzyk PT, Lancet Physical Activity Series Working Group. Effect of physical inactivity on major non--communicable diseases worldwide: an analysis of burden of disease and life expectancy. Lancet2012;380:219-29. doi:10.1016/S0140-6736(12)61031-9. pmid:22818936CrossRefPubMedWeb of ScienceGoogle Scholar
↵Bauman AE, Grunseit AC, Rangul V, Heitmann BL. Physical activity, obesity and mortality: does pattern of physical activity have stronger epidemiological associations?BMC Public Health2017;17:788. doi:10.1186/s12889-017-4806-6. pmid:28982371CrossRefPubMedGoogle Scholar
↵Barengo NC, Nissinen A, Tuomilehto J, Pekkarinen H. Twenty-five-year trends in physical activity of 30- to 59-year-old populations in eastern Finland. Med Sci Sports Exerc2002;34:1302-7. doi:10.1097/00005768-200208000-00011. pmid:12165685CrossRefPubMedWeb of ScienceGoogle Scholar
↵Byberg L, Melhus H, Gedeborg R, et al. Total mortality after changes in leisure time physical activity in 50 year old men: 35 year follow-up of population based cohort. BMJ2009;338:b688. doi:10.1136/bmj.b688. pmid:19264819Abstract/FREE Full TextGoogle Scholar
↵Gregg EW, Cauley JA, Stone K, et al., Study of Osteoporotic Fractures Research Group. Relationship of changes in physical activity and mortality among older women. JAMA2003;289:2379-86. doi:10.1001/jama.289.18.2379. pmid:12746361CrossRefPubMedWeb of ScienceGoogle Scholar
↵Schnohr P, Scharling H, Jensen JS. Changes in leisure-time physical activity and risk of death: an observational study of 7,000 men and women. Am J Epidemiol2003;158:639-44. doi:10.1093/aje/kwg207. pmid:14507599CrossRefPubMedWeb of ScienceGoogle Scholar
↵Bijnen FC, Feskens EJ, Caspersen CJ, Nagelkerke N, Mosterd WL, Kromhout D. Baseline and previous physical activity in relation to mortality in elderly men: the Zutphen Elderly Study. Am J Epidemiol1999;150:1289-96. doi:10.1093/oxfordjournals.aje.a009960. pmid:10604771CrossRefPubMedGoogle Scholar
↵Wannamethee SG, Shaper AG, Walker M. Changes in physical activity, mortality, and incidence of coronary heart disease in older men. Lancet1998;351:1603-8. doi:10.1016/S0140-6736(97)12355-8. pmid:9620713CrossRefPubMedWeb of ScienceGoogle Scholar
↵Paffenbarger RS Jr., Hyde RT, Wing AL, Lee IM, Jung DL, Kampert JB. The association of changes in physical-activity level and other lifestyle characteristics with mortality among men. N Engl J Med1993;328:538-45. doi:10.1056/NEJM199302253280804. pmid:8426621CrossRefPubMedWeb of ScienceGoogle Scholar
↵Pedisic Z, Grunseit A, Ding D, et al. High sitting time or obesity: Which came first? Bidirectional association in a longitudinal study of 31,787 Australian adults. Obesity (Silver Spring)2014;22:2126-30. doi:10.1002/oby.20817. pmid:24943057CrossRefPubMedGoogle Scholar
↵Golubic R, Ekelund U, Wijndaele K, et al. Rate of weight gain predicts change in physical activity levels: a longitudinal analysis of the EPIC-Norfolk cohort. Int J Obes (Lond)2013;37:404-9. doi:10.1038/ijo.2012.58. pmid:22531093CrossRefPubMedGoogle Scholar
↵Day N, Oakes S, Luben R, et al. EPIC-Norfolk: study design and characteristics of the cohort. European Prospective Investigation of Cancer. Br J Cancer1999;80(Suppl 1):95-103.pmid:10466767PubMedGoogle Scholar
↵Peters T, Brage S, Westgate K, et al., InterAct Consortium. Validity of a short questionnaire to assess physical activity in 10 European countries. Eur J Epidemiol2012;27:15-25. doi:10.1007/s10654-011-9625-y. pmid:22089423CrossRefPubMedGoogle Scholar
↵Khaw K-T, Jakes R, Bingham S, et al. Work and leisure time physical activity assessed using a simple, pragmatic, validated questionnaire and incident cardiovascular disease and all-cause mortality in men and women: The European Prospective Investigation into Cancer in Norfolk prospective population study. Int J Epidemiol2006;35:1034-43. doi:10.1093/ije/dyl079. pmid:16709620CrossRefPubMedWeb of ScienceGoogle Scholar
↵Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem1972;18:499-502.pmid:4337382Abstract/FREE Full TextGoogle Scholar
↵Bingham SA, Gill C, Welch A, et al. Validation of dietary assessment methods in the UK arm of EPIC using weighed records, and 24-hour urinary nitrogen and potassium and serum vitamin C and carotenoids as biomarkers. Int J Epidemiol1997;26(Suppl 1):S137-51. doi:10.1093/ije/26.suppl_1.S137. pmid:9126542CrossRefPubMedWeb of ScienceGoogle Scholar
↵Bingham SA, Welch AA, McTaggart A, et al. Nutritional methods in the European Prospective Investigation of Cancer in Norfolk. Public Health Nutr2001;4:847-58. doi:10.1079/PHN2000102. pmid:11415493CrossRefPubMedWeb of ScienceGoogle Scholar
↵Micha R, Peñalvo JL, Cudhea F, Imamura F, Rehm CD, Mozaffarian D. Association Between Dietary Factors and Mortality From Heart Disease, Stroke, and Type 2 Diabetes in the United States. JAMA2017;317:912-24. doi:10.1001/jama.2017.0947. pmid:28267855CrossRefPubMedGoogle Scholar
↵Chastin SFM, De Craemer M, De Cocker K, et al. How does light-intensity physical activity associate with adult cardiometabolic health and mortality? Systematic review with meta-analysis of experimental and observational studies. Br J Sports Med2019;53:370-6. doi:10.1136/bjsports-2017-097563Abstract/FREE Full TextGoogle Scholar
↵Sattelmair J, Pertman J, Ding EL, Kohl HW 3rd., Haskell W, Lee IM. Dose response between physical activity and risk of coronary heart disease: a meta-analysis. Circulation2011;124:789-95. doi:10.1161/CIRCULATIONAHA.110.010710. pmid:21810663Abstract/FREE Full TextGoogle Scholar
↵Hawkins M, Gabriel KP, Cooper J, Storti KL, Sutton-Tyrrell K, Kriska A. The impact of change in physical activity on change in arterial stiffness in overweight or obese sedentary young adults. Vasc Med2014;19:257-63. doi:10.1177/1358863X14536630. pmid:24879662CrossRefPubMedGoogle Scholar
↵Li S, Guo Y-L, Zhao X, et al. Novel and traditional lipid-related biomarkers and their combinations in predicting coronary severity. Sci Rep2017;7:360. doi:10.1038/s41598-017-00499-9. pmid:28336922CrossRefPubMedGoogle Scholar
↵Barthélémy J-C, Pichot V, Dauphinot V, et al. Autonomic nervous system activity and decline as prognostic indicators of cardiovascular and cerebrovascular events: the ‘PROOF’ Study. Study design and population sample. Associations with sleep-related breathing disorders: the ‘SYNAPSE’ Study. Neuroepidemiology2007;29:18-28. doi:10.1159/000108914. pmid:17898520CrossRefPubMedWeb of ScienceGoogle Scholar
↵Hupin D, Roche F, Gremeaux V, et al. Even a low-dose of moderate-to-vigorous physical activity reduces mortality by 22% in adults aged ≥60 years: a systematic review and meta-analysis. Br J Sports Med2015;49:1262-7. doi:10.1136/bjsports-2014-094306. pmid:26238869Abstract/FREE Full TextGoogle Scholar
↵Hupin D, Edouard P, Gremeaux V, et al. Physical activity to reduce mortality risk. Eur Heart J2017;38:1534-7. doi:10.1093/eurheartj/ehx236. pmid:29048470CrossRefPubMedGoogle Scholar
↵Clarke R, Shipley M, Lewington S, et al. Underestimation of risk associations due to regression dilution in long-term follow-up of prospective studies. Am J Epidemiol1999;150:341-53. doi:10.1093/oxfordjournals.aje.a010013. pmid:10453810CrossRefPubMedWeb of ScienceGoogle Scholar
↵Lear SA, Hu W, Rangarajan S, et al. The effect of physical activity on mortality and cardiovascular disease in 130 000 people from 17 high-income, middle-income, and low-income countries: the PURE study. Lancet2017;390:2643-54. doi:10.1016/S0140-6736(17)31634-3. pmid:28943267CrossRefPubMedGoogle Scholar
↵Arem H, Moore SC, Patel A, et al. Leisure time physical activity and mortality: a detailed pooled analysis of the dose-response relationship. JAMA Intern Med2015;175:959-67. doi:10.1001/jamainternmed.2015.0533. pmid:25844730CrossRefPubMedGoogle Scholar
↵Hulsegge G, Looman M, Smit HA, Daviglus ML, van der Schouw YT, Verschuren WM. Lifestyle changes in young adulthood and middle age and risk of cardiovascular disease and all-cause mortality: The doetinchem cohort study. J Am Heart Assoc2016;5:1-11. doi:10.1161/JAHA.115.002432. pmid:26764411CrossRefPubMedGoogle Scholar
↵Talbot LA, Morrell CH, Fleg JL, Metter EJ. Changes in leisure time physical activity and risk of all-cause mortality in men and women: the Baltimore Longitudinal Study of Aging. Prev Med2007;45:169-76. doi:10.1016/j.ypmed.2007.05.014. pmid:17631385CrossRefPubMedWeb of ScienceGoogle Scholar
↵Yates T, Haffner SM, Schulte PJ, et al. Association between change in daily ambulatory activity and cardiovascular events in people with impaired glucose tolerance (NAVIGATOR trial): a cohort analysis. Lancet2014;383:1059-66. doi:10.1016/S0140-6736(13)62061-9. pmid:24361242CrossRefPubMedGoogle Scholar
View Abstract
Article tools
PDF0 responses
Respond to this article
Data supplement
Alerts & updates
Citation tools
Request permissions
Author citation
Add article to BMJ Portfolio
Email to a friend
UK jobs
International jobs
Cambridge University Hospitals NHS Foundation Trust: Locum Consultant in Palliative Care
Cleveland Clinic, Jersey: GP Partner required
Strathcarron Hospice: Consultant/Locum Consultant in Palliative Medicine
University of Glasgow: Clinical Lecturer in Geriatric Medicine
CARE Fertility Group: Consultant in Reproductive Medicine/Fertility Specialist
View more
Altmetric
Who is talking about this article?
Article has an altmetric score of 12
See more details
Picked up by 1 news outlets
Tweeted by 5
This week's poll
Is resilience training helpful?
Yes
No
Vote
View Results
Read related article
See previous polls
Other content recommended for you,
Cancer risk associated with chronic diseases and disease markers: prospective cohort study
Hua Zhao et al., The BMJ, 2018
Physical inactivity, cardiometabolic disease, and risk of dementia: an individual-participant meta-analysis
Mika Kivimäki et al., The BMJ, 2019
Associations of grip strength with cardiovascular, respiratory, and cancer outcomes and all cause mortality: prospective cohort study of half a million UK Biobank participants
Carlos A Celis-Morales et al., The BMJ, 2018
Association between muscular strength and mortality in men: prospective cohort study
Jonatan R Ruiz et al., The BMJ, 2008
Trajectory of body shape in early and middle life and all cause and cause specific mortality: results from two prospective US cohort studies
Mingyang Song et al., The BMJ, 2016
Exceptional longevity does not result in excessive levels of disability.
Kaare Christensen et al., Proc Natl Acad Sci U S A, 2008
Implement strategies that increase the frequency of guideline-compliant vaccination schedule completion for each IBD patient prior to initiating therapy
Gary R. Lichtenstein, MD, myCME, 2018
Study of influence factors on mortality risk of the elderly with mild cognitive impairment in China
LU Jiehua LI Yue, Population Journal
Powered by
I consent to the use of Google Analytics and related cookies across the TrendMD network (widget, website, blog). Learn more Yes No
Back to top
Δεν υπάρχουν σχόλια:
Δημοσίευση σχολίου