Residual cholesterol and inflammatory risk in statin-treated patients undergoing percutaneous coronary intervention: proportional hazards assumption in context

This commentary refers to ‘Residual cholesterol and inflammatory risk in statin-treated patients undergoing percutaneous coronary intervention’, by B. Bay et al., https://doi.org/10.1093/eurheartj/ehaf196 and the discussion piece ‘Residual cholesterol and inflammatory risk in statin-treated patients undergoing percutaneous coronary intervention: challenges with the proportional hazards assumption and coping strategies’, by Z. Huang, https://doi.org/10.1093/eurheartj/ehag325. We thank Dr Huang for the interest in our study and for the opportunity to clarify selected methodological aspects of our analysis on residual cholesterol and inflammatory risk in statin-treated patients undergoing percutaneous coronary intervention, published in a recent issue of the European Heart Journal.1 In the original manuscript, we did not report the results of the formal assessment of the proportional hazards assumption, an omission that we recognize would have benefited readers. As part of the routine diagnostic evaluation of Cox regression models, this assumption was nonetheless formally examined. Specifically, the global test based on Schoenfeld residuals did not provide evidence of violation of the proportional hazards assumption for the primary analysis reported, yielding P = .162 for the univariable Cox model for major adverse cardiovascular events (MACEs) and P = .219 for the multivariable model (similarly, no individual covariate showed evidence of departure from proportional hazards in covariate-specific Schoenfeld tests in either model). This result should be interpreted in the context of our large study population, which included 15 494 patients. In settings of this magnitude, formal tests of model assumptions typically have adequate sensitivity to detect even modest departures from proportional hazards. Therefore, the absence of statistical evidence against the assumption is unlikely to reflect limited power. Overall, the results of these diagnostic evaluations support the use of the Cox regression framework as an appropriate and statistically robust approach for estimating relative effects in the present analysis.