[Application of multi-stage competing risk model to survival data]

Area of Science:

• Medical Statistics
• Oncology
• Survival Analysis

Background:

• Traditional proportional hazard models are standard for analyzing main outcomes and predictors.
• Medical studies frequently feature non-unique endpoints, posing challenges for standard analysis.
• Treating competing outcomes as censored data in traditional models can bias main outcome risk estimates.

Purpose of the Study:

• To compare the clinical application value of traditional proportional hazard, traditional competing risk, and multi-state competing risk models.
• To evaluate the suitability of the multi-state competing risk model for survival data with multiple outcomes.
• To address the limitations of traditional models in handling non-unique endpoints and comparing risks of different outcomes.

Main Methods:

• Development of traditional proportional hazard model, traditional competing risk model, and multi-state competing risk model.
• Application of the three models to a previously published follow-up data set of prostate cancer patients.
• Comparative analysis of the advantages and disadvantages of each model using identical survival data.

Main Results:

• The traditional proportional hazard model may yield biased estimates when competing outcomes exist.
• The traditional competing risk model adjusts for competing outcomes but does not directly compare their risks.
• The multi-state competing risk model effectively handles multiple outcomes and allows for direct risk comparisons, demonstrating significant clinical utility.

Conclusions:

• The multi-state competing risk model offers a more suitable solution for survival data with multiple outcomes compared to traditional models.
• This model enhances the understanding of clinical outcomes in complex scenarios like prostate cancer.
• The study highlights the importance of selecting appropriate statistical models for accurate survival data analysis in medical research.