Optimal Subsampling for the Proportional Mean Residual Life Model With Massive Survival Data

The proportional mean residual life (PMRL) model enables flexible survival analysis but can be computationally intensive for massive datasets. To address this, we introduce an optimal subsampling method that selects a small, representative subset to efficiently estimate model parameters. We derive subsample-based estimating equations, establish the asymptotic properties of the resulting estimator, and develop optimal sampling probabilities under an L-criterion. A practical two-step algorithm is presented for implementation. Simulation studies and a real-data application demonstrate that our method substantially reduces computational costs while maintaining high statistical efficiency.