Response to “Physical activity has no significant association with stroke?”

We thank Du et al. for their comments on our paper.¹ Du et al. mentioned that stroke is not significantly associated with most of the subgroups stratified by the duration of exercise per week (Supplementary Table 5)² and thus the results do not suggest a definite association between physical activity (PA) and reduced risk of stroke. However, the stroke analysis was a secondary analysis of cardiovascular disease by subtypes and thus we had relatively limited number of cases across 9 finely categorized groups of vigorous and moderate PA, which we utilized to make results comparable to the main analysis. Because of the lower number of cases in the sub-group analyses, the hazard ratio (HR) and 95% confidence interval (CI) for the individual categories appear less stable. When we collapsed them into smaller groups to increase statistical stability, we tended to see a clearer pattern that PA is associated with reduced risk of stroke. For example, HRs (95% CI) of stroke across 6 categories of vigorous PA were 1 (ref) for 0 min/week, 0.99 (0.85–1.16) for 1–74 min/week, 0.86 (0.73–1.01) for 75–149 min/week, 0.89 (0.75–1.06) for 150–224 min/week, 0.92 (0.76–1.12) for 225–299 min/week, and 0.80 (0.67–0.97) for ≥300 min/week. For moderate PA, HRs (95% CI) of stroke across 6 categories were 1 (ref) for 0–19 min/week, 0.96 (0.88–1.04) for 20–149 min/week, 0.99 (0.91–1.08) for 150–299 min/week, 1.04 (0.94–1.15) for 300–449 min/week, 0.88 (0.77–1.01) for 450–599 min/week, and 0.75 (0.64–0.89) for ≥600 min/week.

As the authors pointed out, our multivariable models included various potential confounders. More specifically, our models mutually adjusted for vigorous and moderate PA and additionally adjusted for body mass index. This statistical approach can help to examine the independent association of vigorous and moderate PA and better control for confounding by adiposity. However, because vigorous and moderate PA are correlated and body mass index could be an intermediate in the relationship between PA and stroke, the estimates are likely to be conservative and may underestimate the true associations. In multivariable models without mutual adjustment for vigorous and moderate PA and for body mass index, we observed stronger inverse associations between PA and stroke (P-trend<.001 for vigorous PA and 0.002 for moderate PA). For vigorous PA, HRs (95% CI) of stroke across categories used in the above paragraph were 1 (ref), 0.97 (0.84–1.14), 0.83 (0.71–0.98), 0.85 (0.72–1.01), 0.88 (0.73–1.07), and 0.75 (0.63–0.91). For moderate PA, the HRs (95% CI) of stroke across categories were 1 (ref), 0.93 (0.86–1.01), 0.93 (0.85–1.01), 0.95 (0.86–1.05), 0.81 (0.71–0.93), and 0.70 (0.59–0.82). By not controlling for body mass index, these results assume that the benefit of PA is in part mediated by controlling body weight. Therefore, the overall results for stroke are generally in line with previous studies supporting the benefits of PA on stroke. Nevertheless, we agree with the authors that more studies are needed to better understand the underlying factors in the relationship between long-term PA intensity and stroke as well as its potential differences by types of stroke (ischemic vs. hemorrhagic).