Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Several studies have suggested that a ‘physical activity (PA) paradox’ may exist—that is, the well documented beneficial health effects of leisure-time physical activity (LTPA) are not found for occupational physical activity (OPA). A BJSM editorial has outlined potential explanations for such a paradox.1 A recent systematic review has provided some empirical evidence that men, but not women, engaging in high (compared with low) level OPA have an 18% (95% CI 5% to 34%) increased risk of all-cause mortality.2 Nonetheless, the existence of this paradox has been questioned in a discussion paper by Professor Roy Shephard.3 Below we expand on three limitations this discussion paper raised and suggest a research agenda for generating more conclusive evidence regarding this paradox.
Cohort study origin
The current evidence originated from certain regions including Scandinavia, Spain, Iran and Israel. It was suggested that this restricted geographical representation weakens the evidence. However, relatively good working conditions in western European and Scandinavian countries could attenuate the negative health effects of OPA. Studies from countries with higher work demands and/or higher environmental temperatures are likely to provide evidence for even larger negative health effects. Apart from physical work demands, other factors (eg, LTPA or socioeconomical gradients) may also vary between countries and may have an impact on the association between OPA and health. The restricted origin of evidence also raises the possibility of similar (potentially flawed) analytical methods and even overlap in study participants. While similarity in methods is present in some of the reported studies, cohorts vary considerably in terms of recruitment (eg, general population vs workers only, and national representative samples vs samples from specific regions), assessment of variables, and analytical approach. Thus, the potential for common flaws across studies to account for the findings is unlikely.
In recent years, researchers around the globe have dealt with disentangling the differential health effects of OPA and LTPA. After a systematic search we identified studies from Germany, Switzerland, Finland, Norway, Japan and China, published after our review (table 1; online supplementary file 1), showing evidence for and against the PA paradox. Nevertheless, we agree that there is a need to uncover evidence from other parts of the world. If existing cohorts have data available to study longitudinally the health effects of OPA and LTPA, we urge researchers to do so.
Physical activity measurement
In the reviewed studies, OPA and LTPA were typically assessed by questionnaires with relatively low validity, providing exposure estimates prone to random error and effect estimates biased towards the null. Moreover, rather crude categories for PA created for analyses may have led to misclassification, which will likely bias study effect sizes.
We recommend that future research addressing the PA paradox should be based on more objective PA assessment methods consisting of, for example, accelerometry, inclinometry and other methods that can provide detailed information on PA characteristics, including duration, intensity, breaks, recovery time, postures, biomechanical loading, relative aerobic loads taking absolute fitness levels into account, cumulative loads and the composition of PA.
There is a need for evidence from better PA measures in longitudinal cohorts. Although some cohort studies have measured accelerometer-based PA,4 most of these studies failed to distinguish OPA from LTPA (which could be done by combining device-based PA measurements with, for example, diary information). Some cohorts that distinguished PA domains used only proxy health measures, for example, heart rate and blood pressure, but currently lack sufficient follow-up duration required for adequate statistical power in linking cohort data with register-based health outcomes. Such data may, however, become available in the future.5
Like in most observational studies, the possibility of residual confounding exists in the reviewed studies. The way confounding factors were measured, often using self-reports, could possibly result in biased effect sizes due to (differential) misclassification.6 Moreover, the distribution of confounding factors (eg, smoking prevalence) and OPA in the population may have changed over time. Therefore, generalisation to current day settings and conditions should be done with caution. Additionally, there may be evidence for effect modification, for example, by gender and cardiovascular disease status.2 7
We argue that it is unlikely that residual confounding (from socioeconomic or other factors) could fully explain the effects reported in our review.2 Evidence supporting a PA paradox was even found among samples adjusting for or stratifying on socioeconomic position,8 and with participants from lower socioeconomic position only.9
Nevertheless, more adequate ways to address confounding and effect modification should be pursued. Re-analysis of existing data, for example, in the context of individual participant data meta-analyses,7 can be done to adjust for confounders properly (beyond what has been done in earlier articles on the same study) and to harmonise these adjustments across studies. Although harmonising individual data from different cohorts is challenging, combined databases can help to identify (socioeconomic) confounding factors, or to address these factors by conducting stratified analyses. Alternative research designs (other than from observational studies) or analytical techniques could be used to address residual confounding (table 2).10
In conclusion, while we acknowledge methodological limitations of the studies shown in our review,2 existing evidence from prospective cohort studies on all-cause mortality provides some support for a PA paradox, at least among men. However, considering the inconsistent findings and limitations, we propose a research agenda to examine this paradox further (table 2). After conducting the proposed research, triangulation of evidence from different populations and research methods should be performed to provide a clearer understanding of any paradox and its mechanisms.
Twitter @coenen_pieter, @profHoltermann, @WvanMechelen, @Leon_Straker
Correction notice This article has been corrected since it published Online First. The funding statement has been corrected.
Contributors All authors (PC, MH, AH, NK, WvM, LS and AvdB) reviewed the manuscript for important intellectual content. AvdB is the study guarantor.
Funding This study has been funded by The Netherlands Organisation for Health Research and Development; ZonMw (grant #: 531-00141-3). The funding body played no role in the development of this protocol.
Competing interests For the avoidance of doubt, WvM wishes to declare that he is a non-executive board member of Arbo Unie B.V. WvM and AvdB are director-shareholders of Vrije University Medical Center (VUmc) spin-off company Evalua Nederland B.V. Both Arbo Unie and Evalua operate in the Dutch occupational healthcare market. There are no conflicts of interest reported by the other authors.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.