ASCO AI Study Develops Machine Learning Models for Multicancer Risk Stratification Using Routine Clinical Data
An international study presented at the ESMO Congress 2026 by ASCO AI has developed and validated approximately 500 machine learning models designed to stratify risk for the 20 most common cancer types. These models utilize routinely collected clinical data, including demographic information, lifestyle factors, ICD-10 coding history, blood biomarkers, genetic data, and metabolomic data. The research, led by Dr. Paul-Henry Koop of the Else Kröner Fresenius Center for Digital Health, TU Dresden, Germany, aims to move cancer screening eligibility from broad criteria to individual risk assessment. The models were trained on a cohort of 463,620 individuals from the U.K. Biobank, with follow-up extending up to 18 years, and included 57,055 patients with incident cancer. External validation was conducted using data from 42,036 individuals in the U.S. All of Us program and over 3 million individuals from the Japan Medical Data Center database, alongside data from Scotland and Wales.