Machine learning: it takes more than select models to draw general conclusions

We read with great interest the article by Truchot et  al., in which the authors compared the performance of a select subset of machine learning (ML) models with their previously developed and commercially promoted Cox-based prognostication system (iBox). They concluded that, in general, ML models do not outperform the Cox-based prognostication system measured by Harrell's concordance index, discrimination, and calibration.1 It is important to realize that this conclusion is only valid for the limited number of ML models tested, and thus does not necessarily apply to the broader field of ML (Figure 1).

Source: Kidney International - October 18, 2023 Category: Urology & Nephrology Authors: Tamas Szili-Torok, Uwe J.F. Tietge, Max J. Verbeek, Stephan J.L. Bakker, Martin H. de Borst Tags: Letter to the Editor Source Type: research