Boosting for Mixture Cure Models Using Survival Data Under Competing Risks for High-Dimensional Data: Application to Second Primary Lung Cancer Data

• Activity Number: 83 - Recent Methodological Developments and Applications in Statistical and Machine Learning Approaches for Predictive Modeling Using Competing Risk Data
• Type: Topic-Contributed
• Date/Time: Monday, August 9, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Section on Risk Analysis
• Author(s): Nilotpal Sanyal* and Summer S. Han
• Companies: Stanford University School of Medicine and Stanford University School of Medicine
• Keywords: competing risks; cure model; boosting; survival analysis; high-dimensional data; machine learning

Abstract

Mixture cure models have been widely used to analyze survival data for cancer, where many patients can be cured, manifesting in the data as long-term survivors of their disease. A mixture cure model is a survival model that incorporates a cure rate with the assumption that the population contains both uncured and cured individuals. Recently, several statistical methods have been proposed to develop a mixture cure model for survival data under competing risks. However, approaches for selecting features for predicting time-to-event outcome under competing risks using high-dimensional sparse data have been lacking for mixture cure models. In this study, we propose a boosting approach for fitting mixture cure models for high-dimensional sparse data under competing risks. We consider a component-wise boosting algorithm that minimizes squared error loss and iteratively and automatically selects features. We conduct a simulation study to compare the performance of the proposed method to existing approaches without a boosting. We apply our method to lung cancer data to identify features associated with risk of developing second primary lung cancer among lung cancer survivors.