Rice P. et al., 2021: Machine Learning Models for Predicting Stone-Free Status after Shockwave Lithotripsy: A Systematic Review and Meta-Analysis.
Rice P, Pugh M, Geraghty R, Hameed BZ, Shah M, Somani BK.
Newcastle Medical School, Newcastle-upon-Tyne, England.
Department of Urology, Freeman Hospital, Newcastle-upon-Tyne, England.
Translational and Clinical Research Institute, Faculty of Medical Sciences, Newcastle University, International Centre for Life, Newcastle upon Tyne, England.
Department of Urology, Churchill Hospital, Oxford, England.
Nuffield Department of Surgery, Oxford University, Oxford, England.
Department of Urology, University Hospital Southampton, Southampton, England.
Abstract
We performed a systematic review and meta-analysis to investigate the use of machine learning techniques for predicting stone-free rates following Shockwave Lithotripsy (SWL). Eight papers (3264 patients) were included. Two studies used decision-tree approaches, five studies utilised Artificial Neural Networks (ANN), and one study combined a variety of approaches. The summary true positive rate was 79%, summary false positive rate was 14%, and Receiver Operator Characteristic (ROC) was 0.90 for machine learning approaches. Machine learning algorithms were at least as good as standard approaches. Further prospective evidence is needed to routinely apply machine learning algorithms in clinical practice.
Urology. 2021 Apr 21:S0090-4295(21)00331-9. doi: 10.1016/j.urology.2021.04.006. Online ahead of print. PMID: 33894229.
Comments 1
This report is a systematic review and meta-analysis of the application of machine learning methods for predicting the outcome of SWL in terms of stone-free rates. Three different methods were scrutinized: Artificial neuron network, decision tree algorithms and random forests.
It is emphasized that selection of method for stone removal should be based on patients’ choice, expected stone-free rates, QoL and cost.
A brief introduction explains how the three methods work, but without a deep understanding of the details of the underlying mathematics it is difficult to evaluate the clinical value of the methodology. The authors emphasize the heterogeneity between included studies and the conclusion was that “there is no clear benefit” of using machine learning for selecting patients for SWL.
Back to basics: The best way to predict outcome of SWL is clinical experience!
Hans-Göran Tiselius