バイオメトリクスと生物統計学ジャーナル

Evaluating Reproducibility and Generalizability of a Medical Predictive Model in Clinical Research

Abstract

Fengnan Li* and Shein Chung Chow

In clinical research, a medical predictive model is intended to provide insight into the impact of risk factors (predictors) such as demographics and patient characteristics on clinical outcomes. A validated medical predictive model informs disease status and treatment effects under study. More importantly, it can be used for disease management. However, a gap in the development process of these models is often observed. That is, most studies only focus on the internal validation for the model's reproducibility but overlook the external validation needed for evaluating generalizability. To solve this issue, this article proposes several methods for assessing both the reproducibility and generalizability of a developed/ validated medical predictive model. The generalizability estimation approaches allow for sensitivity analysis in situations where data on new populations is not available, which provides valuable insights into the model's applicability to patients from a different population.