BACKGROUND: In clinical practice, accurately estimating a patient’s prognosis is essential for clinical decision-making. Prognostic prediction models can support this process by combining multiple patients’ characteristics into individualized predictions. However, prognostic models require careful methodological evaluation before being applied in practice. CLINICAL QUESTION: How can clinicians determine whether a prognostic prediction model is ready to implement in routine practice? KEY RESULTS: Prognostic prediction models are algorithms that combine multiple variables to estimate the likelihood of an outcome. They can be developed using regression-based methods or machine learning approaches. They are usually developed in a specific dataset. However, good performance in the dataset does not guarantee that the model will perform well when applied to new patient data in practice. Prognostic models must undergo internal validation, external validation, and clinical impact evaluation before being implemented in practice. The performance of the model (at development and validation) should be evaluated across three key aspects: discrimination, calibration, and clinical utility. The risk of bias of a model should also be assessed, as weaknesses in study design or analysis can undermine the reliability of results. An effectiveness study should finally evaluate whether a model improves patients’ outcomes. CLINICAL APPLICATION: Clinicians should follow a structured process to review prognostic models before adopting them in practice. Clinicians should verify that the model has been validated in the target population, evaluate its performance metrics, and appraise its risk of bias. Only models that meet minimum standards across these domains should be considered suitable for clinical use.
Feller et al. (Thu,) studied this question.