Skip to main navigation Skip to search Skip to main content

Machine Learning Models in Pan-Cancer Research: A Review of Validation, Interpretability, and Clinical Usefulness

Research output: Chapter in Book/Report/Conference proceeding › Chapter › peer-review

Abstract

Pan-cancer machine learning now extends well beyond tumour-type
classification. Recent studies use these methods for survival prediction, immuno-
therapy response, treatment resistance, early detection, and image-based biomarker prediction to mention few. The breadth of these applications makes evaluation difficult because studies differ in their targets, datasets, and intended clinical roles. This review examined 17 studies published between January 2022 and June 2026, focusing on whether model evaluation supported the claims being made. Each study was assessed across seven domains; external validation, subgroup reporting, leakage safeguards, meaningful comparators, calibration, interpretability support and decision readiness. The evidence was uneven. Independent validation was reported in several studies, but its scope and relevance varied, and cancer-specific performance was not always clear. Calibration and clinically justified thresholds were uncommon, while interpretability ranged from simple feature ranking to support from ablation, external datasets and experimental analyses. Few studies evaluated whether model outputs could improve clinical decisions or patient outcomes. Overall, the strongest studies aligned prediction target, validation design and comparator with the intended use. Most reviewed models are currently better suited to discovery, retrospective risk stratification or early decision-support development than direct clinical deployment. Pan-cancer models should therefore be judged not only by discrimination or accuracy, but by how well their validation, interpretation and intended application fit together.
Original languageEnglish
Title of host publication International Conference on Artificial Intelligence in Healthcare 2026 (AIiH 2026), London.
PublisherZenodo
DOIs
Publication statusPublished - 26 Aug 2026

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

  1. SDG 3 - Good Health and Well-being
    SDG 3 Good Health and Well-being

Keywords

  • Cancer research
  • Machine learning
  • Artificial intelligence (AI)

Fingerprint

Dive into the research topics of 'Machine Learning Models in Pan-Cancer Research: A Review of Validation, Interpretability, and Clinical Usefulness'. Together they form a unique fingerprint.

Cite this