Cause-specific Cox proportional hazards regression does not exhibit an intrinsic statistical bias when estimating hazard ratios in the presence of competing risks, according to a simulation-based investigation published in Population Health Metrics. Instead, the researchers found that numerical instability and unreliable finite-sample estimates arise primarily when a frequent competing event substantially depletes the risk set, leading to a loss and imbalance of observable primary events.
In epidemiological survival analysis, evaluating hazard ratios under competing risks is a fundamental component of time-to-event research. While the cause-specific Cox model remains a standard analytical framework, the authors sought to identify the specific epidemiological configurations and study design parameters under which resulting hazard ratio estimates become unstable, particularly in settings where competing events actively deplete the population remaining at risk.
To evaluate these dynamics, the investigators implemented a simulation algorithm to construct synthetic cohorts subject to two competing events governed by Weibull cause-specific hazards. Within this framework, a binary exposure variable was introduced to evaluate its statistical association with the hazard of the primary outcome event. Each distinct parametric configuration was evaluated across 1,000 independent simulation runs.
The research team systematically varied several core epidemiological and study design factors across the generated cohorts. These parameters included the incidence densities of both competing causes, exposure prevalence across the study groups, overall sample size, follow-up duration, and the underlying generating cause-specific hazard ratio. Cause-specific log-hazard ratios were estimated using Cox proportional hazards models in which competing events were treated as censored observations.
In four selected simulation settings, the authors compared the competing-event scenarios with single-event simulations that incorporated calibrated external censoring. This external censoring was calibrated to reproduce approximately the numbers of subjects removed by competing events in the primary analyses. The comparisons demonstrated closely comparable confidence intervals, bias, and root mean squared error across both models, indicating that the immediate statistical mechanism driving instability is the depletion and numerical imbalance of information regarding primary events rather than an intrinsic model failure.
The simulations revealed specific combinations of primary and competing event incidence, follow-up duration, sample size, and exposure prevalence where the absolute number or distribution of primary events was insufficient for stable estimation. When observable primary events became too scarce or markedly unbalanced between exposure groups, reliable finite-sample estimation of the generating cause-specific hazard ratio became unattainable.
Although calibrated external censoring produced statistically similar consequences in the simulations, the authors emphasized that competing events carry a fundamentally distinct epidemiological meaning because they permanently preclude the occurrence of the primary event. Consequently, the resulting loss of information cannot be remedied by improving outcome ascertainment or diagnostic surveillance. The investigators concluded that applied epidemiological studies must carefully evaluate whether their planned sample size, follow-up duration, disease incidence patterns, and event distributions across exposure categories provide sufficient observable information to ensure reliable estimation and valid interpretation of cause-specific hazard ratios.
Subject of Research: Finite-sample hazard ratio estimation under competing risks in survival analysis
Article Title: A simulation analysis of hazard ratio estimation with competing risks
Article References: Rosso, E., Berchialla, P., Ricceri, F., & Giraudo, M. T. (2026). A simulation analysis of hazard ratio estimation with competing risks. Population Health Metrics. https://doi.org/10.1186/s12963-026-00529-7
Image Credits: AI Generated
DOI: 10.1186/s12963-026-00529-7
Keywords: cause-specific hazard ratios, competing risks, Cox proportional hazards model, survival analysis, finite-sample estimation, Weibull hazards, exposure prevalence, sample size, follow-up duration, incidence density, external censoring, bias, root mean squared error, confidence intervals, risk set depletion, epidemiological methods, synthetic cohorts
News Source: Ophelia Keating. (November 16, 2026). Simulations Link Hazard Ratio Estimation Instability to Lost Primary Event Information. Scienmag.



