DnnRMST

DnnRMST predicts restricted mean survival time (RMST) from baseline covariates in time-to-event data using deep neural networks.

Key Features:

• Direct Relationship Modeling: Establishes a direct relationship between RMST and baseline covariates and predicts RMST at multiple time points simultaneously.
• Transformation to Pseudo Observations: Transforms each subject's survival time into jackknife pseudo observations that serve as quantitative response variables for model training.
• Information Sharing Across Times: Jointly models RMST at multiple times to enable information sharing across time points and improve prediction accuracy.
• Model Evaluation and Validation: Evaluated using extensive simulation studies and applied to three real datasets.
• Interpretability Features: Identifies subject-specific predictors and assesses their importance in risk prediction.

Scientific Applications:

• Clinical Research: Provides RMST estimates and covariate effects for analysis of time-to-event outcomes in clinical studies.
• Personalized Medicine: Highlights subject-specific predictors to support individualized risk profiling and decision-making.

Methodology:

Survival times are transformed into jackknife pseudo observations used as quantitative responses to train a deep neural network that jointly models RMST at multiple time points, with an architecture designed to facilitate information sharing across times.