An explainable transformer model learning from entire treatment timelines for pan-cancer risk profiling across healthcare systems

作者：Philipp Keyl, Niklas Kiermeyer, Jonah Bosserhoff, Tim Lenfers, Thibault Niederhauser, Bowen Fan, Thomas Schnake, Simon Schallenberg, Fabio Aubele, Solveig Kuss, Mina Jamshidi Idaji, Philipp Jurmeister, Moon Kim, Sebastian Bauer, Nikolaos Bechrakis, Michael Forsting, Dagmar Fuehrer-Sakel, Martin Glas, Viktor Gruenwald, Boris Hadaschik, Ken Herrmann, Stefan Kasper, Rainer Kimmig, Stephan Lang, Ina Pretzell, Tienush Rassaf, Alexander Roesch, Jens T. Siveke, Maja Guberina, Ulrich Sure, Marc Wichert, Michael Ingrisch, K Unger, Juergen Behr, Daniel Teupser, Christian G. Stief, J Mayerle, Nadia Harbeck, Amanda Tufman, Jens Ricke, Lars H. Lindner, Siegfried Priglinger, Guenter Hoeglinger, Sven Mahner, Martin Canis, Lucie Heinzerling, Christine Spitzweg, Alpaslan Tasdogan, Matthias Totzeck, Anja Welt, Marcel Wiesweg, C. Benedikt Westphalen, Reinhard Thasler, Fady Albashiti, Grégoire Montavon, Nicola Miglino, Zsolt Balázs, Michael von Bergwelt‐Baildon, Volker Heinemann, Claus Belka, Sylvia Hartmann, A. Wicki, Felix Nensa, Dirk Schadendorf, Daniel Teupser, Klaus-Robert Mueller, Martin Schüler, Frederick Klauschen, Jens Kleesiek, Julius Keyl · 发表于：medRxiv · 年份：2026 · DOI：10.64898/2026.07.24.26358838 · 研究领域：Machine Learning in Healthcare、Artificial Intelligence in Healthcare and Education、Radiomics and Machine Learning in Medical Imaging

Abstract Cancer outcomes vary widely between individual patients, each accumulating an irregular record of treatments, diagnoses, measurements, and complications. Current prognostic models reduce this complexity into a single snapshot, focus on narrow clinical settings, and rarely generalize across hospitals. Here we introduce Chronicle , an explainable transformer that learns from entire patient trajectories to predict diverse clinical outcomes throughout the disease course while capturing both short- and long-term temporal dependencies. Trained on 53.7 million longitudinal data points from 51,711 patients spanning 67 cancer types, Chronicle operates natively on irregular data without imputation and jointly predicts eight endpoints within a flexible framework adaptable to additional outcomes. Chronicle outperformed cross-sectional models for overall survival prediction (C-index 0.84 vs 0.76-0.79), stratified patients more accurately than established prognostic systems, including TNM stage, and predicted seven adverse event and transfusion endpoints (AUC 0.80-0.92). Applied without retraining to 69,341 patients in Germany, Switzerland, and the United States, Chronicle generalized across healthcare systems and improved further with local fine-tuning. Integrated explainability traced each risk update to patient-specific clinical factors, revealing distinct temporal persistence of prognostic information, with relevance half-lives ranging from weeks for therapies to nearly one ye...