New Framework Estimates Individual Treatment Benefit with Dose Variation

This research introduces Dose-AIPTB, a novel framework designed to estimate the Individual Probability of Treatment Benefit (IPTB) in clinical settings where treatment doses can vary. Unlike previous methods limited to binary treatments, Dose-AIPTB can handle multiple discrete dose levels and ordinal outcomes. The approach reframes the complex problem of predicting individual treatment effects into a simpler binary classification task. It achieves this by generating "pseudo-labels" from comparisons between patients with similar characteristics and then aggregating these labels using attention mechanisms or kernel regression. Experiments on both real-world and synthetic data demonstrated that the attention-based aggregation consistently outperformed kernel-based alternatives, especially under conditions like covariate shift and varying sample sizes. This framework provides a robust foundation for personalized medicine, enabling more precise dose selection based on individual patient benefit probabilities.