Approaches to Confounder Selection in Observational Studies Investigating Associations Between Physical Activity and Health-Related Outcomes: A Review.

BACKGROUND: There are many different confounder selection approaches. It is considered best practice, and recommended by reporting guidelines, that the choice of confounders is described and justified. We aimed to describe practices in confounder selection in epidemiological studies using observational data to investigate the effect of physical activity-related exposures on health-related outcomes. METHODS: A structured search strategy was used to identify studies estimating the effect of a physical activity-related exposure on a health-related outcome using individual-level observational data, published in 2021-2022. A total of 1353 articles were identified in the screening process, 300 of which were randomly selected for full-text screening and data extraction. Data extracted included the approach(es) used to select confounders and study characteristics. RESULTS: Of the 300 selected articles, 279 were eligible. Most (61%) did not describe their approach to choosing confounders. Around a quarter (27%) of studies used a confounder selection approach based on prior knowledge or theory, and a further 3% used a directed acyclic graph. The remaining studies (9%) used only a data-driven approach. CONCLUSIONS: Most published studies aiming to estimate the effect of physical activity-related exposures on health-related outcomes do not report the method used to choose confounders. This goes against reporting guidelines and negatively impacts transparency and reproducibility. These results suggest that more awareness and professional development of confounder selection methods are required among people conducting observational studies in physical activity epidemiology and that greater adherence to reporting guidelines is required by study authors, reviewers, and journal editors.