Censoring, Competing Events, and Multistate Models: Comment on Beyersmann et al. "Hazards Constitute Key Quantities for Analyzing, Interpreting and Understanding Time-to-Event Data"

Beyersmann et al. propose a functional interpretation of hazards, viewing them as evolving quantities describing the entire event process rather than as pointwise causal contrasts. In this commentary, we elaborate on the implications of this view for causal inference in modern clinical trials with survival outcomes. We emphasize how censoring, competing events, and multistate structures shape not only identifiability but also the definition and transportability of hazard-based estimands. We highlight that, even within a functional framework, censoring mechanisms may implicitly determine the statistical estimand through time-dependent weighting, with direct implications for generalizability across studies and populations. We further discuss how these issues are amplified in competing-risks and multistate settings, where causal interpretation requires careful consideration of intercurrent events and selection induced by post-randomization state occupancy.