From London buses to activity trackers: A reflection of 70 years of physical activity research

In 1953, Morris et al.¹ published their ground-breaking findings among London bus drivers and conductors, linking physical activity to a major chronic disease for the first time. This seminal paper marked the birth of a new field—physical activity epidemiology.¹ Since then, the field has flourished, evidenced by mounting scientific publications.² Our knowledge about the role of physical activity in disease prevention has expanded from coronary heart disease in Morris et al.’s¹ initial study to dozens of additional diseases and conditions, such as diabetes, cancers, and dementia.³ More than 7 decades and thousands of publications later, we critically reflect on several aspects of the progress in physical activity epidemiology.

1. How much progress have we made?

A paradigm-shifting event in the field of physical activity and health occurred in 1995 when the Centers for Disease Control and Prevention released their physical activity guidelines. In contrast to previous guidelines that focused on exercise training to enhance cardiorespiratory fitness, the 1995 guidelines moved the field into the public health realm by recognizing the importance of moderate-intensity physical activity, such as brisk walking. These recommendations have been systematically and regularly updated. However, to date, the evidence underpinning physical activity guidelines development is primarily based on self-reported data on physical activity.

In the last few years, device-measured physical activity has become increasingly common in large cohort studies. These measures are less prone to information biases and can assess physical activity across the entire intensity spectrum at a high resolution (seconds). The availability of accelerometer data has led to advances in epidemiological evidence. For example, emerging studies suggest a greater magnitude of association between physical activity and risk for morbidity and mortality, as well as a lower risk reduction threshold compared with self-reported physical activity.⁴ Further, the total amounts of physical activity, assessed either as energy expended or steps per day, are strongly associated with a lower risk of morbidity and mortality.⁵ Additionally, some data suggest that high-intensity physical activity performed for a few minutes per day is associated with health benefits.⁶ Still, most evidence on device-measured physical activity and health outcomes is derived from a single study, the UK Biobank study. Such data are subject to healthy volunteer selection bias, which may be accentuated in the subgroup with accelerometer measures.

The possibility of allowing researchers from around the world to access data in studies like the UK Biobank is an important step forward in scientific advancement. However, there is also an apparent risk that easily accessible data preclude us from taking a step back to thoroughly consider what the pertinent research questions are. In a “publish or perish” environment, researchers are often more motivated by the outcomes of publishing than the actual purpose of the publication. Unless pre-registration of analytical protocols for observational studies becomes a common practice,⁷ data dredging or fishing will remain tempting and undetectable. Further, it is nearly impossible to avoid partial or complete duplications across studies, either unintentionally, due to not knowing what others are working on, or intentionally, by producing least-publishable units.⁸ “Quick and dirty” research outputs may help advance academic careers, but their usefulness for advancing science and informing guidelines and practice remains unknown.

In conclusion, physical activity research has made important progress despite the apparent issue of over-publication. Future research should prioritize important yet under-studied areas and the areas most needed for informing guidelines and practice.

2. Are we overestimating the health benefits of physical activity?

There is a fundamental challenge in examining the causal relationship between physical activity and health: the outcomes of physical activity may take decades to manifest. Therefore, using randomized control trials, often considered the gold standard for causality, to examine the health benefits of physical activity, poses practical, ethical, and budgetary challenges. Still, some randomized control trials have been conducted, but their findings have been inconclusive,⁹ contrasting with the relatively consistent and much larger effect sizes from observational studies. Although evidence from observational studies and randomized control trials are not exactly comparable,⁵ such a counterintuitive observation has made some question whether we, as a field, have been overestimating the causal effects of physical activity.¹⁰

There are reasons to believe that the association between physical activity and health outcomes may still be underestimated by most studies because physical activity tends to be measured at one point in time, but evidence suggests that the association may be stronger based on repeated measures of physical activity.¹¹ Moreover, improvements in device-based measures may further strengthen the association due to fewer measurement errors, which tend to bias the association towards the null.

However, there are fundamental reasons why the observed association could be overestimated. While physical activity is very plausibly a cause of various health outcomes, being able to perform physical activity is also an important marker and predictor of good health, especially at an older age. This poses the challenge of “time-varying confounding” where the confounder (e.g., health or vitality) for the association between physical activity (exposure) and a long-term endpoint (outcome), such as mortality, is affected by previous physical activity (exposure).¹² Unfortunately, time-varying confounding is rarely addressed by research in physical activity epidemiology and it is likely to bias the association, in most cases, away from the null. The field of causal inference epidemiology could potentially offer tools for addressing such complex challenges.¹³ There are promising signs that such methods have started to appear in physical activity research, such as using directed acyclic graphs¹⁴ for guiding model building and covariate selection and applying Mendelian randomization¹⁵ and marginal structural model¹⁶ for testing the causal associations between physical activity and health-related outcomes.

In conclusion, whether observational studies have underestimated or overestimated the associations between physical activity and long-term health outcomes remains inconclusive and complex. As a field, we should be serious about reducing bias and improving methodologies.

3. Have we sufficiently considered equity in our research?

Physical activity epidemiology research has been traditionally conducted in apparently healthy Western cohorts, who are, on average, better educated and healthier than average due to healthy volunteer bias.¹⁷ Despite the repeated calls for better diversity in study populations,¹⁸ generalizability remains an Achilles heel for the field.

Perhaps a more concerning issue is the perspectives from which we conduct research on physical activity—there is often an implicit assumption that physical activity is a “free choice” and one should always aim to do more, not less. As a result, physical activity has been glorified as a panacea and sitting has been vilified. It was not until recently that researchers challenged this assumption by questioning whether all physical activity is health enhancing. For example, physical activity may be performed out of necessity, not choice, in hazardous and unenjoyable settings¹⁹ or in a demanding occupational context without sufficient recovery periods.²⁰ Furthermore, more physical activity may be unobtainable when individuals are battling against fatigue as a result of a major or debilitating condition. Future research should incorporate and consider such nuances. Messages that are not deemed inclusive, such as “move more, sit less”, should be reconsidered.

In the past decade, device-based measures have become increasingly common in physical activity research. This advancement has offered unprecedented opportunities to study physical activity in granularity with much better accuracy. However, an unintended consequence is that device-based evidence may particularly over-represent the privileged populations. While wearables are changing the paradigm of physical activity research, we must be mindful of who the data represent and who can benefit from the research based on such data, to ensure that research does not exacerbate existing inequalities.

In conclusion, as a field, we have not sufficiently considered or addressed equity or equality. Diversity, inclusion, equity, and equality should be important points of consideration for future research.

4. What are the missed opportunities in physical activity research?

One under-studied and under-emphasized area is the potential co-benefits of physical activity beyond apparent physical health outcomes. For example, physical activity is consistently linked to better mental health outcomes²¹ and emotional well-being of individuals,²² as well as the economic outcomes of societies.²³ Physical activity has also been favorably associated with other social outcomes, such as academic performance²⁴ and social functioning.²⁵ Finally, specific types of physical activity, such as active travel, may offer environmental benefits.²⁶ Improving and capitalizing on the evidence base for the co-benefits of physical activity may foster collaboration across sectors, strengthen the return-on-investment case for physical activity promotion, and offer unique pathways for physical activity interventions. Most important, in a world faced with multiple major global challenges but finite resources, physical activity has the potential to concurrently address some of these challenges, such as climate change and social isolation, along with improving physical health. However, any decisions on resource allocation should be based on solid evidence and the field of physical activity epidemiology has a golden opportunity to build such an evidence base that is urgently needed in the age of syndemics.

In conclusion, methods and approaches from physical activity epidemiology can also be used to explore a broad range of “co-benefits” through interdisciplinary collaboration. There are many pathways through which physical activity benefits our society, quantifying and recognizing the non-health benefits helps us to create more avenues for positive changes.