Significance of external validation in clinical machine learning: let loose too early?

With great interest, we recently read the article by Karnuta et al. [1], in which the authors describe the development and internal validation of a prediction model for patient resource utilization. They involved 38,070 patients and trained a na ïve Bayes classifier in repeated 10-fold cross validation. Resampled training areas-under-the-curves were 0.880, 0.941, and 0.906. We commend the authors for their generally excellent machine learning (ML) methods and clean execution with sufficient training data.
Source: The Spine Journal - Category: Orthopaedics Authors: Tags: Letters to the editor Source Type: research