Anticipating key innovations in physical activity epidemiology for the coming decades

Main text

In a previously published article in this journal, a research team from the China Kadoorie Biobank (CKB) (www.ckbiobank.org) used a novel phenome-wide association analysis to examine the prospective associations between physical activity (PA), 425 different types of diseases, and 53 specific causes of death in more than 500,000 participants.¹ Throughout the duration of this study, there were 722,183 incident events and 39,320 mortality events recorded over a period of 12 years. PA was associated with a lower risk for 65 different diseases and lowered mortality risk of 19 causes. This study emphasizes that the benefits of PA go beyond those traditionally recognized, such as all-cause mortality, cardiovascular disease, certain types of cancer, and diabetes. Strengths of the study include the large sample size, a long follow-up period, and linkage to multiple health registries. Thus, this study provides comprehensive insights into the role of PA to improve long-term health.

This study also has important limitations acknowledged by the authors. One of them is that PA information was obtained by self-reports, which is a method known to include important measurement errors. Measurement error in self-reports can result, for example, from recall bias, misclassification of activity intensity, and individual and contextual interpretation differences when individuals are asked to estimate their PA levels. Research in PA epidemiology has increasingly incorporated more device-based measures as a result of recent technological advancements. Wearable devices, ranging from simple pedometers to more sophisticated tools, such as heart rate monitors and accelerometers, are being used to assess PA more accurately and have gained widespread use in large cohorts. To date, the UK Biobank (UKB; www.ukbiobank.ac.uk) is the largest cohort with device-measured PA. The UKB includes a population-based sample of approximately 100,000 adults who agreed to participate in a study of PA using wearable devices. One example of the use of the UKB wearable device data includes a phenome-wide association analysis performed recently, examining the associations of device-measured moderate to vigorous PA and the incidence of 697 diseases over a median of 6.3 years of follow-up.² For comparative purposes, the authors also examined the associations between self-reported PA and the incidence of diseases in the total sample (n = 500,000), and, interestingly, they found similar benefits (e.g., cardiovascular and metabolic diseases) or harms (e.g., musculoskeletal conditions and injuries) of PA compared with those found in the CKB cohort. However, there were substantially stronger effect sizes, and harms were not so evident when using device-measured data as opposed to self-reported PA. These two phenome-wide association analyses in the CKB and UKB exemplify how research in PA epidemiology continues to evolve.³ The question then arises: what key technological and methodological innovations should PA epidemiologists leverage to further advance our knowledge about the health benefits of PA?

Despite the UKB being the largest cohort with device-measured PA information to date, it also includes some limitations. The UKB cohort is subject to the healthy volunteer selection bias, and this selection bias is potentially more pronounced in the ancillary UKB study with wearable data. Harmonized meta-analyses, where exposure data and confounders are harmonized across cohorts and meta-analyzed thereafter, may also partially overcome additional limitations associated with single cohorts, such as limited generalizability.⁴ Individual participant data analyses pooling data at the individual level from multiple cohorts allows examining exposure-outcome associations with higher certainty, preserving continuous PA data, and can minimize the loss of information while increasing statistical power.⁵ One example of an ongoing pooled initiative is the ProPASS Consortium (www.propassconsortium.org), an international collaboration platform working in harmonizing thigh- and wrist-worn accelerometer data to promote collaborative individual participant and prospective meta-analyses on PA, posture, sleep, and health outcomes (Figure 1).

Figure 1.

Summary of some future innovations in PA epidemiology for the next decades

(A) The ProPASS Consortium (www.propassconsortium.org).

(B) Repeated measures of accelerometer data in the Seniors-ENRICA-2 cohort.

(C) The IMPaCT cohort (https://cohorte-impact.es) protocols for device-based measurements.

(D) Skin patches for glucose control.

Another limitation of numerous cohort studies using both self-reporting and wearable devices to examine the associations between PA and health outcomes is the reliance on a single baseline measure of PA, assuming that this estimate is the participant’s PA level throughout the follow-up. Considering that PA is a moderately variable behavior, repeated administrations can minimize measurement error and potential PA-health outcome attenuations that result from within-individual variability in PA estimates. Repeated measurements of the exposure variable in a subsample of the cohort can be used to correct for regression dilution, assuming all error in the exposure measurement is random. This work in the CKB and another recent one in the UKB studies have shown how the associations between PA and health outcomes can be corrected with this approach.^1,6 If data collection in cohorts using wearable devices are already more costly and complex than those using self-reporting, then the logistics of collecting repeated measurements can be even more challenging. The Seniors-ENRICA-2 cohort is a study of community-dwelling older adults that has successfully conducted three follow-up assessments using wearable devices, with a possible fourth wave coming up in the near future.⁷ This longitudinal design enables the detailed characterization of changes in PA as participants age and would allow for detailed examination of associations between PA and health outcomes, likely yielding more robust effect estimates (Figure 1).

Most studies examining device-measured PA associations with health outcomes have used wearable devices with accelerometer sensors. Integrating heart rate and other physiological sensors may improve accuracy and allow a more precise measure of PA than using only accelerometer sensors, helping to differentiate between activities with similar movements but different intensities as well as daily activities with minimal movement. The integration of biomechanical and physiological data in research-grade wearable devices may be essential for more precise PA assessments in future cohort studies. The IMPaCT cohort (https://cohorte-impact.es), an ongoing population-based study in Spain, aims to recruit a representative sample of 200,000 participants aged 16–79 years. This cohort exemplifies future innovation in PA assessments, as all participants will be monitored using the multisensor matrix wearable device that includes an accelerometer, gyroscope, LED-based photoplethysmography, skin temperature, and ambient temperature. Moreover, to enhance PA characterization, the wearable device will be worn on two body locations (i.e., the non-dominant wrist and the opposite thigh), which likely increases the possibility for activity type recognition and posture differentiation (Figure 1).

Furthermore, ongoing advancements in bioengineering will likely serve as the next foundational pillar in PA epidemiology. An example to illustrate this future innovation is skin patches. The best example is the skin patch that has been commercialized by several pharmaceutical companies for continuous glucose monitoring, becoming a game changer in diabetes control. Using skin patches, several sensors can be fitted to assess biomechanical and physiological data. Incorporating chemical sensors can detect biomarkers in the sweat of the epidermis and in the interstitial fluid of the dermis. Beyond enabling a more precise assessment of PA by integrating accelerometer data from multiple sensors, this emerging technology represents a significant advancement toward continuous, non-invasive health monitoring. By capturing early physiological signals associated with the onset and progression of diseases, it offers a novel opportunity to investigate the real-time PA-health outcome effects using an all-in-one wearable patch.⁸ Such innovation may serve as a precursor to personalized exercise medicine (Figure 1).

Although key innovations in this scientific field will likely involve new technologies, we must not overlook inherent challenges. Thus, it would be important to consider the limitations of the devices, which, for example, cannot capture the context in which PA (i.e., leisure time, travel, domestic, and occupational) or sedentary time (e.g., TV viewing, computer, and reading) occur.³ Therefore, it is suggested to complement device-based measurements with self-reporting. Additionally, the placement of the device (i.e., dominant vs. non-dominant wrist, hip, thigh, or ankle) may influence the accuracy with which certain variables of interest are assessed. Second, a multidisciplinary approach will be necessary, involving exercise specialists, epidemiologists, mathematicians, physicists, and engineers, with all of them working together to address how device-measured data must be processed, including handling large volumes of information and missing data, combining data from multiple sensors, developing AI algorithms, interpreting the output, etc. Finally, it is also important to acknowledge that these devices collect a large amount of personal data, which warrants a discussion of ethical issues, including privacy, data governance, potential uses, and who is granted access. One example is the All of Us study, in which some participants agreed to share data from their own consumer-grade devices worn for a median of 4 years.⁹ These data are not publicly available but can be accessed by approved researchers following registration, completion of ethics training, and agreement to a data use policy through the All of Us Research Workbench platform.

While randomized clinical trials are considered to provide estimates of the causal effect of PA on disease incidence and mortality, the use of randomized trials for assessing long-term disease and survival is not feasible and impractical.¹⁰ It is also difficult to understand the dose-response relationships with health outcomes in a randomized clinical design, as this would require multiple trial arms and adequately powered designs. Thus, investing in new well-characterized cohort studies that include novel technological and methodological innovations is likely the best bet for advancing the field of PA epidemiology and our understanding of the health benefits associated with different characteristics of PA.

Funding and acknowledgments

The authors express their gratitude to all volunteers who have dedicated their time to participating in cohort studies, which have contributed to the advancement of numerous significant studies in the field of PA epidemiology.

Declaration of interests

The authors declare no competing interests.

Published Online: July 23, 2025