Article Text
PDF
Abstract
Objectives To examine independent and interactive associations of physical activity and diet with all-cause, cardiovascular disease (CVD) and physical activity, diet and adiposity-related (PDAR) cancer mortality.
Methods This population-based prospective cohort study (n=346 627) is based on the UK Biobank data with linkage to the National Health Service death records to 30 April 2020. A left-truncated Cox proportional hazards model was fitted to examine the associations between exposures (self-reported total moderate-to-vigorous intensity physical activity (MVPA), vigorous-intensity physical activity (VPA) and a diet quality index (score ranged 0–3)) and outcomes (all-cause, CVD and PDAR cancer mortality).
Results During a median follow-up of 11.2 years, 13 869 participants died from all causes, 2650 from CVD and 4522 from PDAR cancers. Compared with quartile 1 (Q1, 0–210 min/week), Q2–Q4 of MVPA were associated with lower risks of all-cause (HR ranged from 0.87 (95% CI: 0.83 to 0.91) to 0.91 (95% CI: 0.87 to 0.96)), CVD (HR ranged from 0.85 (95% CI: 0.76 to 0.95) to 0.90 (95% CI: 0.81 to 1.00)) and PDAR cancer mortality (HR ranged from 0.86 (95% CI: 0.79 to 0.93) to 0.94 (95% CI: 0.86 to 1.02)). Compared with no VPA, any VPA was associated with lower risk for all-cause and CVD mortality (HR ranged from 0.85 (95% CI: 0.80 to 0.89) to 0.88 (95% CI: 0.84 to 0.93) and from 0.75 (95% CI: 0.68 to 0.83) to 0.90 (95% CI: 0.80 to 1.02), respectively). Although not reaching statistical significance for all-cause and CVD mortality, being in the best dietary category (diet quality index=2–3) was associated with a reduction in PDAR cancer mortality (HR=0.86, 95% CI: 0.78 to 0.93). No additive or multiplicative interactions between physical activity categories and dietary quality was found. When comparing across physical activity and diet combinations, the lowest risk combinations consistently included the higher levels of physical activity and the highest diet quality score.
Conclusions Adhering to both quality diet and sufficient physical activity is important for optimally reducing the risk of mortality from all causes, CVD and PDAR cancers.
- epidemiology
- preventive medicine
Data availability statement
Data may be obtained from a third party upon approval and payment. UK Biobank data could be obtained on application from https://www.ukbiobank.ac.uk/enable-your-research/apply-for-access.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
Both regular physical activity and a healthy diet play an important role in promoting health and longevity.
A small number of overfeeding studies found that high-intensity exercise may counteract detrimental physiological responses to overfeeding; however, the long-term interactive effects between diet and physical activity remained less explored.
WHAT THIS STUDY ADDS
The combinations with the lowest risk of mortality consistently included the highest diet quality score and higher levels of physical activity.
High levels of physical activity did not counteract the detrimental effects of a poor diet on mortality risk.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
This study reinforces the importance of both physical activity and diet quality for achieving the greatest reduction in mortality risk.
Public health messages and clinical advice should focus on promoting both physical activity and dietary guidelines.
Introduction
Regular physical activity and a healthy diet play a major role in the prevention of a range of chronic diseases, and are recommended by health authorities around the world.1–5 Although both behaviours are critical components of a healthy lifestyle, their interlinked effects on energy balance and metabolic health have inspired popular beliefs, media headlines6 and scientific discussion7 8 on whether one can ‘outrun a bad diet’ through high levels of physical activity.
The public’s interest in this topic may be further fuelled by a small number of overfeeding studies, which provided tantalising evidence that high-intensity exercise may counteract inflammation and the deterioration of insulin sensitivity and the cardiometabolic profile associated with overfeeding, despite energy surplus.9 10 However, these studies were designed to study the acute effects over a short period of time.
To date, a very small number of studies have explored the interaction between diet and physical activity in relation to long-term outcomes, such as mortality, and found no significant attenuation of risk associated with poor diet in those who were the most active.11–13 These studies did not examine both additive and multiplicative interactions and were primarily focused on the overall volume/pattern of physical activity, without specifically examining the role vigorous-intensity physical activity (VPA) plays. Considering the promising evidence on the additional benefits of VPA independent of the overall volume of physical activity,14 15 it is yet to be examined whether VPA offers particular risk reduction against a poor diet.
To address this research gap, we examine independent and joint associations of diet and physical activity (measured as both MVPA and VPA) with all-cause, cardiovascular disease (CVD) and cancer mortality using a large population-based sample of British adults from the UK Biobank study. 14 15We hypothesise that physical activity and diet quality are independently associated with lower mortality risk, and that high levels of physical activity, either in total MVPA or VPA, cannot offset detrimental effects of poor-quality diet.
Methods
Sampling and procedures
Details about the UK Biobank cohort have been previously published.16 Briefly, between April 2007 and December 2010, 502 632 participants (aged 40–69 years) were assessed at 22 assessment centres throughout the UK (5.5% response rate). During the baseline assessment visit, participants provided electronically signed consent, a self-completed touch-screen questionnaire, a brief computer-assisted interview and had their physical measures and biological samples taken. In the current study, main exposures (ie, diet and physical activity) and sociodemographic covariates were collected using the baseline touch-screen questionnaire and all-cause and cause-specific mortality data are based on linked death register data.
Measurement
Exposure variables
MVPA was measured using the short version of the International Physical Activity Questionnaire (IPAQ), which has established reliability and validity.17 IPAQ asked separate questions regarding total minutes of walking, moderate physical activity (MPA) and VPA that lasted for at least 10 min at a time, in a typical week. First, we summed the total minutes of walking, MPA and VPA (weighted by two) as the total MVPA and analysed it as quartiles (0–210, 211–450, 451–920, 921–5040 min/week).18 Second, considering that existing overfeeding studies that postulated ‘outrunning a bad diet’ focused on VPA,9 10 and that VPA is found to convey potential additional benefits independent of the overall volume of MVPA,14 15 we further used unweighted VPA minutes, instead of MVPA, as an exposure of interest. VPA was categorised as 0, 10–74 (1–9 min long bouts were not captured by the IPAQ), 75–149 and 150+ min/week based on data distribution and congruent with recommended amount by physical activity guidelines.2
Diet quality was assessed at recruitment using a self-reported questionnaire regarding the frequency of consumption of common food items (https://biobank.ndph.ox.ac.uk/showcase/ukb/docs/TouchscreenQuestionsMainFinal.pdf). We adopted the dietary index developed by Rutten-Jacobs et al 19 based on adherence to selected dietary targets recommended by the American Heart Association.20 Briefly, the index included the following categories of food items: (1) fruit and vegetables; (2) fish; and (3) red and processed meat. Meeting the target for each category (ie, ≥4.5 cups/day of fruit and vegetables, ≥2 serves/week of fish, and ≤2 times/week intake of processed meat and ≤5 times/week of red meat intake) is dichotomised as yes (1) and no (0). An overall diet quality score is derived by summing the three categories of food items (0–3) and recoded into: 0 (lowest diet quality), 1 (medium diet quality) and 2–3 (high diet quality).
Outcome variables
Date of death was ascertained from death certificates held by the National Health Service. All participants from England, Scotland and Wales were censored on 30 April 2020. All-cause mortality was defined as a death record from any causes. CVD mortality was ascertained when the primary cause of death was CVD, defined by International Classification of Diseases (ICD)-10 I00-I99. Physical activity, diet and adiposity-related cancer (PDAR cancer hereafter) mortality was determined when the primary cause of death was a type of cancer that is evidentially linked to physical inactivity and/or diet and/or adiposity. Specifically, we selected the types of cancer for which there is strong/convincing evidence based on the 2018 US Physical Activity Guidelines Advisory Committee Scientific Report21 and the World Cancer Research Fund Strong Evidence Matrix.5 Cancer in the following sites was included for PDAR cancer mortality: bladder (C67), breast (postmenopausal; ascertained by C50 combined with self-related menopausal status), colorectal (C18-21), endometrium (C54.1), stomach (C16), kidney (C64), liver (C22), oesophagus (C15) and pancreas (C25). Based on the current sample, deaths from these PDAR cancers represented 59% of all cancer mortality.
Covariates
We used directed acyclic graphs (DAG) developed in Dagitty22 to identify confounders. DAG is a conceptual representation for mapping assumptions about a related set of variables involved in a causal research question.22 Confounding paths and minimal sufficient adjustment variables were identified through the DAG (online supplemental figures 2 and 3). We accordingly adjusted for age group category with categories based on inflection points from natural splines (online supplemental figure 4), sex (male vs female), educational attainment (categorised based on the British educational system as: university, A/AS levels or equivalent (11–12 years), O levels/ secondary or equivlanet (7-10 years; including General Certificate of Secondary Education, Certificate of Secondary Education), vocational (including National Vocational Qualifications, Higher National Diploma, Higher National Certificates) or other), Townsend Index (quartiles, an indicator for area-level material deprivation,23 measured immediately prior to participants joining the study), living arrangement (living with vs without partner), country of birth (UK vs non-UK), type of job (categorised into: manual and physically demanding, mainly walking or standing, sedentary, and not employed, based on self-reported frequencies of jobs involving mainly heavy manual and physical work and walking or standing), smoking status (never, former, current), weekly alcohol consumption (0, 1–14 drinks, 14+ drinks) and physician diagnosis of hypertension and diabetes at baseline (yes vs no). It is unclear whether body mass index (BMI) is a confounder or a mediator, we therefore did not adjust for BMI in the model, but explored the effects of adjusting for BMI in a sensitivity analysis.
Supplemental material
Statistical analysis
The relationship between physical activity (MVPA or VPA), diet and mortality outcomes (all-cause, CVD and PDAR cancer mortality) was examined using Cox proportional hazards model among non-pregnant, apparently healthy participants with complete data on exposure measures and covariates. To minimise the risk of ‘reverse causation’ (ie, lifestyle change as a result of health conditions that increase risk of mortality), those with a baseline diagnosis of CVD, cancer, degenerative neurological disorder, late-stage renal disease, chronic obstructive pulmonary disease (COPD) and emphysema (based on self-report and a range of administrative linkage data, such as hospital admissions and primary care data) and those who were underweight (BMI <18.5) or reported chronic widespread pain were excluded.24 A left-truncated Cox proportional hazards model was fitted using age as the underlying time scale. This model is recommended where a wide range of ages is being assessed over a relatively long period, especially when investigating effects in older adults.25 For all analyses, HRs and 95% CIs were presented adjusted for all covariates described under ‘covariates’ above. Analyses were further adjusted for MPA and walking time when VPA was modelled as the exposure variable. Models were assessed for any violation of the proportional hazards assumption by testing for any significant correlation between Schoenfield residuals and time, and covariates were stratified if necessary until the assumption was met.
We calculated effective power using the number of events per variable (EPV), by dividing the number of events by the number of variables in our primary analysis model, and compared against a critical value of 10.26 The EPV in this study was at least 115 for all analyses, suggesting the study was adequately powered.
To explore the independent and joint associations of physical activity and diet quality, two sets of models were fitted. First, both diet and physical activity were included in the same model to assess the independent effects of an exposure variable when the other exposure and all covariates were adjusted for. Multiplicative interactions were tested by including a diet×physical activity interaction term to the adjusted model and using a Wald’s test to assess the differences between the models with and without the interaction term. Additive interactions were tested by refitting all models using non-parametric Aalen additive regression models, allowing for an estimation of hazard difference on a linear scale, and we tested for interactions in these models. Where no evidence was found for multiplicative interactions between exposures of interest, the combined effects of diet and physical activity were estimated using linear combinations of HRs with the theoretically highest-risk category (lowest physical activity: MVPA Q1 or no VPA and worst diet quality: diet quality index=0) as the reference group. All analysis was performed in R (V.4.1.2).
We conducted several sensitivity analyses. First, considering the inconsistent evidence on occupational physical activity and mortality,27 28 we conducted a sensitivity analysis excluding those whose job mainly involved manual and physically demanding work (8%), based on the assumption that for these participants, heavy occupational physical activity constituted most of their reported total MVPA. Second, to account for reverse causation caused by occult disease (ie, people with low physical activity and poor diet were more likely to die within the first 2 years as a result of undiagnosed conditions), we reran the analysis by excluding the first 2 years of follow-up time. Third, as mentioned above, we explored models with BMI as an additional confounder. Fourth, considering the competing risk between CVD and cancer mortality, a Fine-Gray model of competing risks29 was used accounting for the marginal probability of each ‘subdistribution’ (eg, those who died from CVD vs those dying from non-CVD-related causes). Finally, we explored the associations of physical activity and diet with all cancer, in comparison with PDAR cancer. The reporting of the study follows the STrengthening the Reporting of OBservational studies in Epidemiology checklist (online supplemental file 1).
Supplemental material
Results
The final analytical sample included 346 627 participants with complete data on physical activity, diet and all covariates, without diagnosed CVD, cancer, degenerative neurological disorder, late stage renal disease, chronic widespread pain, COPD and emphysema, or being underweight at baseline (online supplemental figure 1). During a median of 11.2 (IQR= 1.4) years and a total of 3 841 534 person-years of follow-up, 13 869 participants died. Among those, 2650 died from CVD and 4522 from PDAR cancers.
Baseline characteristics
Baseline characteristics of participants are described in table 1. Overall, participants reported an average of 686 min/week (SD=679) (median=672, IQR=84) of weighted total MVPA (walking+MPA+VPA×2) with Quartile 1(Q1) reporting 0–210 min/week (mean=117; SD=60; median=117, IQR=50), Q2 reporting 211–450 min/week (mean=331; SD=69; median=330, IQR=150), Q3 reporting 451–920 min/week (mean=658; SD=134, median=635, IQR=135) and Q4 reporting 921–5040 min/week (mean=1644; SD=686; median=1440, IQR=780). In terms of VPA, 40.8% reported doing no VPA at all, 26.5% <75 min/week, 15.4% reported 76–150 min/week and 17.4% reported performing >150 min/week. Regarding diet quality, 22.1% of participants had a dietary index score of 0 (the poorest), 53.4% of participants had a dietary index score of 1 and 24.5% of participants had a dietary index score of 2–3 (the best). Overall, males, participants without university education, with manual jobs or jobs involving walking and standing, and those of normal weight reported higher levels of MVPA. Meanwhile, participants who were slightly older, female, university educated, not living with a partner, born overseas, not working, and those not currently smoking or drinking excessively, and of normal weight were more likely to report higher diet quality index (table 1).
Descriptive statistics of the study sample (n=346 627, 2007–2010)
Physical activity and diet as individual risk factors
MVPA, VPA and diet quality index were significantly associated with all-cause, CVD and PDAR cancer mortality outcomes in unadjusted models (table 2). The magnitude of associations was attenuated in the adjusted models. In adjusted models, higher levels of MVPA (Q2–Q4) were significantly and inversely associated with all three mortality outcomes with the strongest association observed in Q4 for all-cause (HR=0.87, 95% CI: 0.83 to 0.91) and PDAR cancer mortality (HR=0.86, 95% CI: 0.79 to 0.93) and in Q2 for CVD mortality (HR=0.85, 95% CI: 0.76 to 0.95). VPA was inversely associated with all-cause and CVD mortality but not significantly associated with PDAR cancer mortality. The association between VPA and CVD mortality appeared stronger than that between MVPA and CVD mortality. However, the associations between VPA and CVD mortality were not dose-dependent with the lowest risk observed for the two middle VPA categories (10–75 min/week: HR=0.75, 95% CI: 0.68 to 0.83; 76–150 min/week: HR=0.76, 95% CI: 0.67 to 0.86), not the highest category (Q4 HR=0.90, 95% CI: 0.80 to 1.01). The inverse associations of diet quality index with all-cause and CVD mortality were non-significant in most models, except that being in the best category of diet quality was associated with 14% lower risk in cancer mortality (HR=0.86, 95% CI: 0.78 to 0.93).
Associations of moderate-to-vigorous physical activity (MVPA), vigorous-intensity physical activity (VPA) and diet with all-cause, cardiovascular disease and cancer mortality (n=346 627)
Physical activity and diet as interactive risk factors
We did not find evidence for statistically significant multiplicative or additive interactions in either MVPA or VPA models for any outcome (online supplemental tables 4–6). When exploring the linear combinations of joint effects between physical activity (MVPA and VPA) and diet quality index, we found that compared with the highest risk combination (MVPA Q1 or no VPA and diet quality index=0), most other combinations had a significantly lower HR for all-cause, CVD and PDAR cancer mortality. For all-cause and PDAR cancer mortality, the incremental reductions of HR by both physical activity and diet quality index were clear, and the combinations with the lowest HR involved being in MVPA Q3 and Q4 and a diet quality index of 2–3. For CVD mortality, the lowest HR was observed in Q2 and Q3 and the highest diet quality index (figure 1). The analysis of VPA and diet quality index combinations showed generally similar patterns (figure 2). Overall, there is no evidence for high levels of physical activity, measured as MVPA or VPA, fully offsetting low diet quality in any of the analyses; neither was there evidence for a higher diet quality index fully offsetting a lack of physical activity.
Combinations of moderate-to-vigorous physical activity (MVPA) quartiles and diet quality index and all-cause, cardiovascular disease (CVD) and physical activity, diet and adiposity-related (PDAR) cancer mortality outcomes (n=346 627). Quartiles: Q1: 0–210 min/week; Q2: 211–450 min/week; Q3: 451–920 min/week; Q4: 921–5040 min/week. An overall score ranged from 0 to 3 and was recoded into: 0 (poorest), 1 (medium diet quality) and 2–3 (best). All models adjusted for age, sex, educational attainment, marital status, country of birth, type of occupation, neighbourhood deprivation, smoking status, alcohol intake, hypertension and diabetes .
Combinations of vigorous-intensity physical activity (VPA) and diet quality index and all-cause, cardiovascular disease (CVD) and physical activity, diet and adiposity-related (PDAR) cancer mortality outcomes (n=346 627). An overall diet quality score ranged from 0 to 3 and was recoded into: 0 (poorest), 1 (medium diet quality) and 2–3 (best). All models adjusted for age, sex, educational attainment, marital status, country of birth, type of occupation, neighbourhood deprivation, smoking status, alcohol intake, hypertension, diabetes, moderate-intensity physical activity and walking.
Sensitivity analysis
All sensitivity analyses, including restricting to participants without a physically demanding job (online supplemental table 7, figures 5-7), those excluding the first 2 years of follow-up (online supplemental table 8, figures 8–10), those additionally adjusted for BMI as a covariate (online supplemental table 9, figures 11-13) and those accounting for the marginal probability of subdistributions (online supplemental figures 14–17) revealed no substantial changes in the associations identified from the main analyses, suggesting robustness of the findings. Finally, compared with the associations between exposures (MVPA, VPA and diet quality) and PDAR cancer, all exposures had a weaker association with all cancers (online supplemental figure 18), perhaps because this outcome lacked specificity (including less relevant cancers to physical activity and diet), but the CI was narrower, due to more statistical power.
Discussion
The current study examined independent and joint effects of physical activity and diet quality on long-term mortality outcomes. Findings from this study confirm the importance of both physical activity and quality diet in all-cause and cause-specific mortality, suggesting that one should adhere to both sufficient physical activity and quality diet to minimise mortality risk.
Both healthy diet and regular physical activity have long been considered critical protective lifestyle factors against chronic disease and premature mortality.30 Because of their complementary and interactive effects on energy, lipid, glucose and metabolic homeostatic processes,31 and their shared physiological mechanisms for chronic disease prevention,32 it has been suggested that diet and physical activity be jointly considered in epidemiological studies.30 31 However, more than two decades after such call for consideration,30 epidemiological studies examining their interactive effects on health and mortality are still rare.11–13 33–35
Findings of this study
Our study found an inverse association between physical activity and mortality from any causes, CVD and cancer. Particularly, associations of MVPA with all-cause and PDAR cancer mortality showed a dose-response pattern, with 13%–14% lower hazards observed in the most active quartile compared with the least active one. Such magnitude of risk reduction is similar to that identified by previous analyses.24 36 37 When we particularly examined the role of VPA, we found a larger magnitude of risk reduction in CVD mortality, particularly in the middle categories (10–150 min/week), while accounting for MPA and walking. Although the evidence on whether VPA and MPA protect against mortality risk to the same extent remains equivocal,38 several large cohort studies have found that doing some VPA may offer additional benefits than doing none, holding constant the total volume of MVPA.14 15 39 It has been argued that VPA may lead to more physiological adaptations14 and elicit more insulin sensitising and anti-inflammatory effects40 than lower-intensity physical activity, which may explain why the stronger association with VPA are particularly pronounced for CVD mortality in our study.
Despite the association between diet and all-cause and CVD mortality not reaching statistical significance, we found that those in the best diet quality category had 14% lower risk of PDAR cancer mortality compared with those in the worst category. It is important to emphasise that diet quality remains an important aspect of non-communicable disease prevention. Diet consists of countless health influencing components, and different measures capture different aspects of dietary health.41 For example, studies based on the UK Biobank data found dietary indicators to be differentially associated with CVD events, all-cause and cause-specific mortality,41 42 suggesting that effects of diet on health could be specific to the dietary indicators. The current diet quality index is based on meeting the recommendations for fruit and vegetables, fish, and red and processed meat, all of which are important dietary components for chronic disease prevention.43 These food groups were selected as markers for overall diet quality, because other important dietary components and/or nutrient groups, such as whole grains and dairy, were not measured during baseline assessment. Considering that errors in reported dietary measures are likely to attenuate diet-disease associations towards the null,44 one may expect the ‘true’ association between diet quality and mortality outcomes to be stronger.
Consistently across the three outcomes, nearly all risk combinations involving higher physical activity categories had significantly reduced mortality risk compared with the highest-risk reference groups. In fact, across outcomes, the lowest-risk combinations almost always involved the highest diet quality index and the highest or second-highest physical activity categories.
Comparison with other studies
The current investigation was motivated by several animal and human studies with promising findings suggesting that exercise may counteract short-term adverse effects of overfeeding. For example, Gioscia-Ryan et al randomised mice into normal versus Western diet (characterised as high fat, high sugar, low fibre) and the absence versus presence of aerobic exercise (running wheel); their study found that a Western diet exacerbates age-related endothelial dysfunction, aortic stiffening, vascular oxidative stress and aortic inflammation, but all these effects were attenuated or completely prevented by exercise.45 A small number of experimental studies conducted in human participants generally found that high-intensity exercise protects against detrimental metabolic effects of overfeeding,46 such as insulinemic responses,9 cardiometabolic profiles10 and genes expression in adipose tissue,9 despite energy surplus. While these studies focused on short-term outcomes, long-term cumulation of adverse physiological outcomes could lead to chronic disease and ill-health. A small number of epidemiological studies have examined the joint associations between physical activity and diet on mortality outcomes. Alvarez-Alvarez et al examined the joint association between physical activity and Mediterranean diet score, in the Seguimiento Universidad de Navarra cohort, and found the largest all-cause mortality risk reductions in the groups that adhered to both high physical activity and healthier dietary scores.13 Two studies based on the National Health and Nutrition Examination Survey found that the participants with higher levels of physical activity and better quality diet had the lowest risk of all-cause mortality in both general adults11 and prefrail and frail older adults.12 However, an analysis of a cohort of postmenopausal black women in the USA found that neither diet nor physical activity, independently or combined, was associated with site-specific obesity-related cancers.35 A study based on healthy middle-aged participants from the Melbourne Collaborative Cohort Study found a stronger association between physical activity and mortality than that between diet and mortality, and that adherence to a Mediterranean diet contributed to little gain in addition to physical activity.33 These mixed findings imply that more research with valid measures are needed to fully understand the long-term joint effects of diet and physical activity on health and longevity.
Strengths and limitations
This study is the first to our knowledge to examine the joint association of physical activity and diet quality with all-cause and cause-specific mortality. Comprehensive analysis was conducted in a large sample with >10 years of follow-up and both multiplicative and additive interactions were tested. Various sensitivity analyses led to similar results, suggesting robustness of the findings. However, findings should be interpreted in light of limitations. First, despite using validated measures,17 19 both physical activity and diet quality measures are subject to measurement bias. Particularly, biases associated with the dietary instrument are likely to explain the non-significant associations between diet and all-cause and CVD mortality outcomes. UK Biobank did introduce accelerometer and 24-hour dietary recall, both considered more comprehensive and valid measures, during later follow-ups. However, because accelerometer and dietary recall were measured in different years and the sample with both measures is small, we therefore chose to use the large baseline sample with diet and physical activity measured at the same time. Second, both physical activity and diet were measured at one time point only, therefore, could not capture long-term behavioural patterns. Furthermore, the lack of time-varying data could influence the relationships between exposures and covariates, and consequentially, lead to biases in model specification, such as collider stratification bias.47 For example, BMI could be a ‘cause’ or ‘effect’ of physical activity, but because BMI was measured at the same time as physical activity, it is impossible to determine the temporal order to inform covariate selection. The same applies to intermediate health outcomes, such as hypertension. In this case, if BMI was adjusted for, the models could be subject to overadjustment, but if not adjusted for, they could be subject to confounding. We built DAGs with several unmeasured time-varying variables to inform model specification. However, with the presence of unmeasured confounding and selection bias, we had to make various assumptions in model specification. An example of such assumptions is that hypertension and diabetes were ‘pre-existing’ conditions and BMI was an outcome of baseline exposures. Although we compared models with and without BMI as a covariate in a sensitivity analysis and found minimal difference in the exposure-outcome associations, the current model is still far from the ‘ideal correct’ model where all relevant variables were measured, at different time points, and the causal relationships among variables could be clearly mapped. It is important to interpret findings from the current models within the context of the possibility of residual confounding from both unmeasured confounders and measurement biases associated with existing covariates (eg, socioeconomic status, alcohol consumption). In addition, because confounding variables were only measured at baseline, the models do not account for any possible time-varying confounding. Third, missing data (around 13% of participants omitted in complete case analysis) could result in selection bias. Fourth, in analysing physical activity as a categorical variable, there is an assumption of equal risk within each category. Although this is a common practice, measurement error may have been introduced via this process. Fifth, HRs assume a consistent exposure-outcome association over time. Although all final models satisfied the proportional hazards assumption, selection bias in these estimates is likely, due to the inherent time-varying nature of mortality risk.48 Sixth, the competing risk between PDAR cancer and CVD should ideally be modelled using cause-specific models, which were not computationally feasible in the present study. Fine-Gray models treat those who died from competing events not of interest to the research question (eg, non-CVD causes in the CVD mortality model and non-PDAR cancer in the PDAR cancer model) as still within the set at risk, which is difficult to interpret and may potentially lead to biases. Finally, the UK Biobank study had a participation rate of <6%,16 and the participants appeared much more active than average, posing concerns regarding the representativeness of the sample and the generalisability of the findings. A recent study comparing unweighted risk factor-outcome associations in the UK Biobank with those using poststratification to match a population representative sample found that the lack of population representativeness could distort risk factor-outcome associations, but the association of physical activity and a dietary indicator with all-cause and CVD mortality remained similar.49 Future studies could improve the current evidence by combining accelerometer-based physical activity measures with higher quality dietary measures, such as those derived from 24-hour dietary recall, and ideally, assessing behavioural patterns repeatedly across the lifespan.
Implications
This study reinforces the importance of physical activity and diet quality. Adherence to both health behaviours is endorsed by health professionals and disseminated through clinical and public health recommendations worldwide. However, sensationalised headlines50 and misleading advertisement for exercise regimens to lure consumers into the idea of ‘working out to eat whatever they want’51 have fuelled the circulation of the myth about ‘exercise outrunning a bad diet’. Our study provided important evidence for health professionals that exercise does not fully compensate for a poor diet and that we should recommend and advocate for both an active lifestyle and a healthy diet.
Conclusion
Findings from the UK Biobank study suggest that both physical activity and diet are important for reducing mortality risk, and participants with both higher levels of physical activity and high-quality diet had the lowest mortality risk. Furthermore, our data provide evidence that high levels of physical activity did not fully counteract the detrimental effects of a poor diet on mortality risk.
Data availability statement
Data may be obtained from a third party upon approval and payment. UK Biobank data could be obtained on application from https://www.ukbiobank.ac.uk/enable-your-research/apply-for-access.
Ethics statements
Patient consent for publication
Ethics approval
The UK Biobank is a large population-based cohort study, which received ethical approval from the North-West Research Ethics Committee. Participants gave informed consent to participate in the study before taking part.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
IL, UE and LF are joint senior authors.
Twitter @DrMelodyDing, @M_Stamatakis, @pjclare, @LuigiFontana22
Contributors DD and LF conceptualised the research question. DD led the investigation and drafted the paper. JvB conducted data analysis, PC supported revision of data analysis, BN helped with literature searches, ME with data management and NV with dietary measures. All authors provided critical feedback on the paper at various stages and approved the final version of the paper. DD is the guarantor of the project and accepts full responsibility for the work and/or the conduct of the study, had access to the data and controlled the decision to publish.
Funding DD ES abd LF were each funded by an Investigator Grant from the National Health and Medical Research Council of Australia (APP 2009254, APP1194510, and APP1177797, respectively). DD was additionally funded by Heart Foundation Australia, and LF additionally funded by the Australian Youth and Health Foundation and Philip Bushell Foundation.
Disclaimer The funder played no role in any aspects of the research project.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.