Background

Longitudinal physical activity can be modelled using various approaches, but it remains unclear which approaches better describe long-term activity and how they influence study findings. Therefore, this study examined eight commonly used modelling approaches for physical activity over 12 years and their associations with all-cause, cardiovascular disease (CVD), and cancer mortality in women.

Methods

We used data collected at 3-year intervals from participants in the 1946-51 cohort of the Australian Longitudinal Study on Women’s Health from age 47-52 to 59-64 years (N = 10,536). All-cause, CVD, and cancer mortality were ascertained through December 2022 (age 71-76). Physical activity was modelled using four volumes (baseline, most recent, cumulative average, and time-varying) and four profile types (count of meeting guidelines, first-last change, group-based trajectory modelling, and growth curve parameters). Cox regressions and Fine-Gray models were used to examine their associations with all-cause, CVD, and cancer mortality outcomes.

Results

Between age 59-64 and age 71-76, 791 deaths occurred (279 from CVD and 422 from cancer). Physical activity volumes modelled using cumulative average showed smoother distributions and clearer patterns of mortality risk reduction than baseline or most recent activity models. When examining activity profiles, subgroups derived from group-based trajectory modelling provided more accurate descriptions of activity over time and more consistent risk reductions than simpler approaches (e.g., count of meeting guidelines, first-last change). Similar results were observed for CVD and cancer mortality.

Conclusions

Different modelling approaches for longitudinal physical activity yield substantially different estimates in associations with mortality outcomes, even within the same study sample. Future studies should incorporate repeated activity measurements and apply advanced modelling approaches to capture long-term activity.