Objectives We aimed to investigate the dose–response associations of long-term leisure-time physical activity (LTPA) obtained from repeated measures with all-cause and cardiovascular disease (CVD) mortality outcomes in Taiwanese adults.

Methods We included 210 327 participants with self-reported LTPA at least in two medical examinations (867 968 data points) for up to 20 years (median, IQR: 4.8 years, 2.3–9.0). Dose–response relationships were modelled with restricted cubic spline functions and Cox regressions HRs (95% CIs) adjusted for main covariates.

Results During up to 23 years of follow-up (3 655 734 person-years), 10 539 participants died, of which 1919 of CVD. We observed an inverse, non-linear dose–response association between long-term LTPA and all-cause and CVD mortality. Compared with the referent (0 metabolic equivalent of task (MET) hours/week), insufficient (0.01–7.49 MET hours/week), recommended (7.50–15.00 MET hours/week) and additional (>15 MET hours/week) amounts of LTPA had a lower mortality risk of 0.74 (0.69–0.80), 0.64 (0.60–0.70) and 0.59 (0.54–0.64) for all-cause mortality and 0.68 (0.60–0.84), 0.56 (0.47–0.67) and 0.56 (0.47–0.68) for CVD mortality. When using only baseline measures of LTPA, the corresponding mortality risk was 0.88 (0.84–0.93), 0.83 (0.78–0.88) and 0.78 (0.73–0.83) for all-cause and 0.91 (0.81–1.02), 0.78 (0.68–0.89) and 0.80 (0.70–0.92) for CVD mortality.

Conclusion Long-term LTPA was associated with lower risks of all-cause and CVD mortality. The magnitude of risk reductions was larger when modelling repeated measures of LTPA compared with one measure of LTPA at baseline.