Surv-IPTB Uses Attention to Estimate Per-Patient Probability That a Treatment Helps, Not Just Average Effect

arXiv 2608.06288·low signalXBlueskyLinkedInCopy link

Submitted 2026-08-06 (2608.06288, Utkin, Kogan, and Konstantinov), Surv-IPTB is an attention-based framework for the Individual Probability of Treatment Benefit in survival analysis — directly quantifying the probability that a specific patient experiences extended survival under treatment, rather than reporting a population-average hazard ratio. The shift from average effects to per-individual benefit probabilities is the methodological point, and it applies well beyond medicine to any uplift-modeling or personalized-intervention problem where you need a calibrated per-unit answer. Single-source preprint with no reported benchmark comparison in the abstract.