Statistical methods for clustered competing risk data when the event types are only available in a training dataset

Area of Science:

• Biostatistics
• Epidemiology
• Health Data Science

Background:

• Analyzing clustered competing risks data is challenging when event types are missing in the main study.
• Existing methods may not adequately address missing event type information within clusters.

Purpose of the Study:

• To develop and evaluate novel statistical methods for analyzing clustered competing risks data with missing event types.
• To estimate exposure effects accurately in the presence of incomplete event type information.

Main Methods:

• Cause-specific proportional hazards frailty model incorporating random effects for within-cluster correlation.
• Weighted penalized partial likelihood method using event type probabilities derived from a classification model.
• Imputation approach for missing event types based on classification model predictions.

Main Results:

• Analytical variances were derived for the proposed methods.
• Extensive simulation studies demonstrated the finite sample properties of the methods.
• The methods were applied to assess associations between tinnitus and hearing loss in a real-world study.

Conclusions:

• The proposed methods provide a robust framework for analyzing clustered competing risks data with missing event types.
• These techniques enhance the ability to estimate exposure effects in epidemiological and clinical research.
• The application to tinnitus and hearing loss highlights the practical utility of the developed statistical approaches.