Predicting Cancer Risk: Practical Considerations in Developing and Validating a Cancer Risk Prediction Model

Abstract

As the interest in absolute cancer risk prediction models that estimate the probability of developing cancer has grown steadily, so does the debate on the usefulness of cancer risk prediction models. A newly developed risk prediction model should demonstrate its accuracy, reliability, and generalizability through a validation, preferably using external data. However, a limited number of cancer risk prediction models were independently validated and showed good calibration but weak discriminatory accuracy. Several methods and practical considerations in evaluating and improving cancer risk prediction models’ performance are discussed. To increase the utility of cancer risk prediction models in cancer prevention at the population as well as individual level, more powerful and efficient approaches for model development and validation are needed. A collective effort to pool data from cohort studies may provide an excellent opportunity to improve existing models and develop new models, especially for rare cancers and minority populations.