Testing Clinical Prediction Models

To the Editor Mr Adibi and colleagues argued that practitioners are developing too many clinical prediction models (CPMs) and not testing model performance enough. The authors suggested placing all models in a cloud-based environment and allowing for a natural-selection approach to determine the best models. We agree that the current process, in which investigators develop models based on convenience cohorts without clear clinical use cases for their application and with little validation, needs a major overhaul. We also agree that placing models —with appropriate metadata—in a cloud-based environment is a necessary step to enable rapid model evaluation and monitoring of performance. However, we believe that a cloud-based environment is only the first step and the paradigm should be developed even further.
Source: JAMA - Category: General Medicine Source Type: research