Radiation Therapy Outcomes Models in the Era of Radiomics and Radiogenomics: Uncertainties and Validation

Models by their nature are mathematical approximations of reality, as conveyed by the statement that “all models are wrong but some are useful ( 1 Box G.E.P. Science and statistics. J Am Stat Assoc. 1976; 71: 791-799 Crossref Scopus (1109) Google Scholar ).” Their usefulness in radiation therapy (RT) is highlighted by the roles that outcome models play in improving the quality and efficacy of radiation treatment of tumors by predicting response, individualizing prescriptions, and optimizing and ranking planning options. These models are generally categorized into those for tumor response prediction by tumor control probability (TCP) and those for predicting radiation-induced toxicities by normal-tissue complication probability (NTCP). Traditionally, these models were simplistic and included a few dose and volume metrics summarizing the delivered treatment and baseline patient-specific prognostic risk factors, rendering them less powerful but also less prone to overfitting pitfalls ( 2 El Naqa I. A Guide to Outcome Modeling in Radiotherapy and Oncology: Listening to the Data. CRC Press: Taylor & Francis Group, Boca Raton, FL2018 Google Scholar ).