Prognostic model research

Abstract

Prognostic models combine multiple prognostic factors to estimate the risk of future outcomes in individuals with a particular disease or health condition. A useful model provides accurate predictions to support decision making by individuals and caregivers. This chapter describes the three phases of prognostic model research development (including internal validation), external validation (including model updating), and impact on decision making and individual health outcomes. Methodology is detailed for each phase, including the need for large representative datasets, methods to avoid or reduce overfitting and optimism, and the use of both discrimination and calibration to assess a model’s predictive performance. TRIPOD reporting guidelines are introduced. Emphasis is also given to the application of models in practice, including linking the model to clinical decisions using risk thresholds, and evaluating this using measures of net benefit, decision curves, cost-effectiveness analyses, and impact studies (such as randomized trials) to evaluate the effectiveness of models in improving outcomes.