Penalized landmark supermodels (penLM) for dynamic prediction for time-to-event outcomes in high-dimensional data: application to lung cancer mortality prediction integrating multiple data sources

Oregon Convention Center 

To effectively monitor long-term patient outcomes, it is critical to assess the dynamic risk of prognosis. This often involves utilizing multiple data sources (e.g., tumor registries, treatment histories, and patient-reported outcomes). However, challenges arise in selecting predictive features for patient outcomes from high-dimensional data, aligning longitudinal measurements from multiple sources, and summarizing model performance. We develop the penalized landmark supermodel (penLM) for dynamic risk prediction with high-dimensional, potentially multi-source data and novel metrics that summarize model performance (AUC or Brier Score) across several time points by incorporating temporal correlations. Through simulations, we assessed the coverage of the novel metrics' confidence intervals and the tests' power and type I error. We applied penLM to predict the updated 5-year risk of lung cancer mortality at diagnosis and for subsequent years by combining data from SEER registries, Medicare insurance claims, Medicare Health Outcome Survey, and the U.S. Census, revealing superior predictive accuracy compared to single-source models. The framework is available in our R package, dynamicLM.