A simulation analysis of hazard ratio estimation with competing risks

In epidemiological research, the estimation of hazard ratios (HRs) under competing risks is an important aspect of survival analysis. The cause-specific Cox model remains a standard analytical tool, however it is important to recognize the situations in which the resulting HR estimates become unstable or unreliable. This structured simulation-based investigation aims to examine finite-sample estimation of cause-specific hazard ratios in epidemiological settings where a frequent competing event substantially depletes the risk set, particularly focusing on the influence of selected epidemiological and design parameters. We used a simulation algorithm to generate synthetic cohorts with two competing events governed by Weibull cause-specific hazards. A binary exposure variable was introduced to evaluate its association with the hazard of the primary event. We varied the incidence densities of the two causes, exposure prevalence, sample size, follow-up duration, and the generating cause-specific HR. Cause-specific log-HRs were estimated using Cox proportional hazards models, treating competing events as censored. Each configuration was simulated 1,000 times. In four selected settings, results were compared with those from single-event simulations incorporating external censoring calibrated to reproduce approximately the numbers of removed subjects as reported for the competing events in the original analyses. We identified some combinations of primary and competing event incidence, exposure prevalence, follow-up duration, and sample size for which the number or allocation of primary events was insufficient for stable estimation. In four additional comparisons, competing-event scenarios and calibrated external-censoring scenarios produced closely comparable confidence intervals, bias, and root mean squared error. In the four additional comparisons, competing-event and calibrated external-censoring scenarios produced broadly similar confidence intervals, bias, and root mean squared error. This finding indicates that the immediate statistical mechanism was primarily the loss and imbalance of information on the primary event. The simulations do not indicate an intrinsic bias of cause-specific Cox regression in the presence of competing events. They identify settings in which the generating cause-specific HR is difficult to estimate reliably because too few or markedly unbalanced primary events remain observable. Although calibrated external censoring produced similar statistical consequences, competing events have a distinct epidemiological meaning: they permanently preclude the primary event, and the resulting loss of information cannot be remedied by improving outcome ascertainment. Applied studies should therefore assess whether the follow-up duration, sample size, incidence patterns, and distribution of primary events across exposure groups provide sufficient information for reliable estimation and interpretation of the cause-specific HR.

Authors

Institutions

Publication Details

Journal
Population Health Metrics
Published
2026-11-14
DOI
https://doi.org/10.1186/s12963-026-00529-7
Primary Topic
Statistical Methods in Epidemiology
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A simulation analysis of hazard ratio estimation with competing risks

Fulvio Ricceri, Paola Berchialla, Marı́a Teresa Giraudo, Elena Rosso
1 links
Population Health Metrics
Statistical Methods in Epidemiology
article

A simulation analysis of hazard ratio estimation with competing risks

Fulvio Ricceri, Paola Berchialla, Marı́a Teresa Giraudo, Elena Rosso
article en

Abstract

In epidemiological research, the estimation of hazard ratios (HRs) under competing risks is an important aspect of survival analysis. The cause-specific Cox model remains a standard analytical tool, however it is important to recognize the situations in which the resulting HR estimates become unstable or unreliable. This structured simulation-based investigation aims to examine finite-sample estimation of cause-specific hazard ratios in epidemiological settings where a frequent competing event substantially depletes the risk set, particularly focusing on the influence of selected epidemiological and design parameters. We used a simulation algorithm to generate synthetic cohorts with two competing events governed by Weibull cause-specific hazards. A binary exposure variable was introduced to evaluate its association with the hazard of the primary event. We varied the incidence densities of the two causes, exposure prevalence, sample size, follow-up duration, and the generating cause-specific HR. Cause-specific log-HRs were estimated using Cox proportional hazards models, treating competing events as censored. Each configuration was simulated 1,000 times. In four selected settings, results were compared with those from single-event simulations incorporating external censoring calibrated to reproduce approximately the numbers of removed subjects as reported for the competing events in the original analyses. We identified some combinations of primary and competing event incidence, exposure prevalence, follow-up duration, and sample size for which the number or allocation of primary events was insufficient for stable estimation. In four additional comparisons, competing-event scenarios and calibrated external-censoring scenarios produced closely comparable confidence intervals, bias, and root mean squared error. In the four additional comparisons, competing-event and calibrated external-censoring scenarios produced broadly similar confidence intervals, bias, and root mean squared error. This finding indicates that the immediate statistical mechanism was primarily the loss and imbalance of information on the primary event. The simulations do not indicate an intrinsic bias of cause-specific Cox regression in the presence of competing events. They identify settings in which the generating cause-specific HR is difficult to estimate reliably because too few or markedly unbalanced primary events remain observable. Although calibrated external censoring produced similar statistical consequences, competing events have a distinct epidemiological meaning: they permanently preclude the primary event, and the resulting loss of information cannot be remedied by improving outcome ascertainment. Applied studies should therefore assess whether the follow-up duration, sample size, incidence patterns, and distribution of primary events across exposure groups provide sufficient information for reliable estimation and interpretation of the cause-specific HR.

Population Health Metrics
University of Turin (IT)
Good health and well-being
Statistical Methods in Epidemiology
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon