Renewable estimation for partly interval-censored data streams

In the analysis of large-scale, censored survival data streams, traditional estimation techniques often become computationally infeasible and fail to achieve real-time processing capabilities. To address this challenge, we propose a renewable estimation approach specifically designed for partly interval-censored data streams. This method leverages only the most recent data batch and summary statistics derived from historical data, thereby substantially reducing computational and storage requirements. Furthermore, it is designed to flexibly accommodate various types of censoring and employs approximated information criteria techniques for variable selection, ensuring stable results and improving overall performance. The asymptotic properties of the proposed method are rigorously established, and extensive numerical simulations are conducted to assess its efficiency and accuracy. The simulation results demonstrate that, even with a sample size as large as 100,000, the proposed method offers up to a 20-fold improvement in computational efficiency compared to conventional single-pass estimation using the full dataset. Additionally, the practical applicability of the method is further validated through its implementation on the Alzheimer’s Disease Neuroimaging Initiative database.