Renewing Focus on Methodological Rigour in Clinical Machine Learning

Clinical machine learning is increasingly used for prediction, diagnosis, prognosis, risk stratification, and treatment-related decision support. These applications are particularly relevant to precision medicine, where machine learning models may help provide individualised risk estimates to help guide personalised medical decision-making. However, the rapid growth of the clinical machine learning literature has been accompanied by persistent concerns regarding methodological quality, reproducibility, and clinical relevance. Common problems include inadequate sample sizes, information leakage, inappropriate data splitting, incomplete handling or reporting of missing data, over-interpretation of internal validation results, and insufficient description of analytical methods. In many cases, machine learning is applied as a data-mining exercise to readily available public datasets without a clearly defined clinical question or meaningful outcome. These problems limit reproducibility and make it difficult to determine whether reported models are likely to be useful in clinical practice.This Research Topic aims to promote rigorous, transparent, and reproducible supervised machine learning research in clinical medicine, including work relevant to precision medicine. We welcome machine learning modelling studies using original clinical datasets, appropriate study designs, clinically meaningful outcomes, adequate sample sizes, and robust model-development and validation methods. Modelling studies should be reported in accordance with established guidelines, such as TRIPOD+AI. External validation, shadow testing, pilot implementation, and clinical trials are highly encouraged, however well-conducted preliminary studies using rigorous internal validation approaches are also acceptable. Negative or inconclusive findings will be considered if authors critically examine the causes of failure, limitations of the available data, and opportunities for improving future models. Methodological articles, tutorials, reviews, mini-reviews, perspectives, and commentaries addressing supervised-learning research methods are also within scope. Studies based primarily on exploratory data mining of single public datasets, surface-level evaluation of general-purpose large language model or chatbot outputs, or systematic review and meta-analysis of predictive performance metrics without critical commentaries on the methodology and reporting quality of included articles, will generally be out of scope.We are additionally interested in articles covering (but not limited to) the following themes:• Designing clinical machine learning studies around clearly defined clinical questions, intended uses, and meaningful prediction targets• A priori and post hoc determination of sample size requirements in machine learning modelling• Strategies and best practices for applying advanced internal validation methods when independent validation datasets are unavailable, including resampling and nested validation approaches• Strategies and best practices for external, temporal, and multicentre validation• Evaluation of clinical prediction models beyond headline discrimination metrics, including calibration, clinically relevant thresholds, and decision utility• Methodological approaches for identifying and understanding model failure, including negative, inconclusive, or unexpectedly poor performance• Reproducibility and transparent specification of clinical machine learning analyses• Strategies and best practices for prospective evaluation of machine learning models, including shadow testing, pilot implementation, and clinical trials• Methodological research, reviews, and critical analyses aimed at improving the design, evaluation, and interpretation of supervised machine learning studies in medicineWe welcome original research, methods articles, reviews, mini-reviews, perspectives, and commentaries relevant to the Research Topic. Topic editors: Jiawen Deng, Kiyan Heybati, Fangwen Zhou Source: https://www.frontiersin.org/research-topics/86361/renewing-focus-on-methodological-rigour-in-clinical-machine-learning