Effect of non-leisure physical activity on mortality in U.S. adults: does propensity score matching make a difference?

PurposeTo reanalyze results reported in 2008, using propensity score matching, to test the treatment effect of non-leisure physical activity on survival.MethodsMcCullagh's ordinal logit model was used to estimate propensity scores, separately for adults aged 35–59 and 60–74 years at baseline in the National Health and Nutrition Examination Survey I Epidemiologic Followup Study, for three levels of non-leisure activity. Each untreated adult (low non-leisure activity) was matched to two treated adults from the same age group, one with moderate and one with high non-leisure activity. Multivariable Cox proportional hazard regressions were fitted to the matched samples and compared with regressions fitted to unmatched samples.ResultsExcept for moderate leisure-time activity and age, observed covariates were well balanced after matching. For adults aged 35–59 years, matched regressions showed smaller mortality reductions: 23% for moderate and 32% for high non-leisure activity, compared with low non-leisure activity. For adults aged 60–74 years, matched regressions showed a smaller mortality reduction for moderate non-leisure activity (29%) but a larger reduction for high non-leisure activity (47%).ConclusionsIn line with published reviews of conventional multivariable analysis versus propensity score matching, Cox regressions on matched and unmatched samples produced similar estimates of treatment effects for non-leisure activity, but the estimates from matched regressions were usually smaller.