CACSurv: Concordance-Aligned Comparative Learning with Large Language Models for Cancer Survival Prediction

2026-08-17

AI Summary

This study addresses the misalignment between time-to-event regression formulations and supervision signals in cancer survival prediction by proposing CACSurv. This framework reconceptualizes survival modeling as small-cohort comparative reasoning, leveraging consistency-aligned rewards and Monte Carlo reference aggregation to enable large language models to perform relative prognostic ranking based on pathology reports. Additionally, we introduce the TCGA-SurvReport benchmark. Experimental results across six TCGA cohorts demonstrate that CACSurv achieves a C-index of 0.722, outperforming state-of-the-art methods by 6.5 percentage points. By effectively resolving ranking supervision challenges under right-censored data, this approach significantly advances text-driven survival prediction performance.

Abstract

Cancer survival prediction supports treatment planning, risk stratification, and follow-up management. Existing methods use structured clinical variables, whole-slide images, genomic profiles, or multimodal inputs, while patient reports remain underexplored. We study report-centric survival prediction using reports that organize pathological, clinical, and molecular evidence. Large language models (LLMs) can reason over such reports, but case-wise time regression introduces two mismatches. First, a formulation mismatch arises because survival evaluation depends on ordering comparable patients, whereas independent time predictions do not enforce ranking consistency. Second, a supervision mismatch arises because a censored patient's observed time indicates survival beyond that point and cannot serve as an exact regression target, although it still implies orderings relative to patients who died earlier. To address these mismatches, we propose CACSurv, a Concordance-Aligned Comparative framework for report-centric survival prediction. CACSurv reformulates survival modeling as mini-cohort comparative reasoning, where an LLM predicts relative prognostic orderings. We introduce concordance-aligned rewards derived from comparable relations under right censoring, enabling censored outcomes to provide ranking supervision without exact event-time targets. At inference, Monte Carlo Reference Aggregation compares each patient with sampled references and aggregates positions into a cohort-level ranking. We establish TCGA-SurvReport, a benchmark covering six TCGA cancer cohorts. CACSurv achieves the highest C-index on all six cohorts and an average C-index of 0.722, outperforming the strongest published survival model by 6.5 percentage points and the strongest LLM time-regression baseline by 4.2 percentage points. Our code, models, and dataset will be available at https://github.com/xmed-lab/CACSurv.

Problem

• Cancer Survival Prediction
• Large Language Models
• Formulation Mismatch
• Supervision Mismatch
• Right Censoring

Innovation

• Concordance-Aligned Comparative Learning
• Large Language Models
• Cancer Survival Prediction
• Monte Carlo Reference Aggregation
• Right Censoring