Resampling methods for class imbalance in clinical prediction models: A scoping review protocol

INTRODUCTION: Class imbalance-where clinically important "positive" cases make up less than 30% of the dataset-systematically reduces the sensitivity and fairness of medical prediction models. Although data-level techniques, such as random oversampling, random undersampling, SMOTE, and algorithm-level approaches like cost-sensitive learning, are widely used, the empirical evidence on when these corrections improve model performance remains scattered across different diseases and modelling frameworks. This protocol outlines a scoping systematic review with meta-regression that will map and quantitatively summarise 15 years of research on resampling strategies in imbalanced clinical datasets, addressing a key methodological gap in reliable medical AI. METHODS AND ANALYSIS: We will search MEDLINE, EMBASE, Scopus, Web of Science Core Collection, and IEEE Xplore, along with grey literature sources (medRxiv, arXiv, bioRxiv) for primary studies (2009-31 Dec 2024) that apply at least one resampling or cost-sensitive strategy to binary clinical prediction tasks with a minority-class prevalence of less than 30%. There will be no language restrictions. Two reviewers will screen records, extract data using a piloted form, and document the process in a PRISMA flow diagram. A descriptive synthesis will catalogue the clinical domain, sample size, imbalance ratio, resampling strategy, model type, and performance metrics where 10 or more studies report compatible AUCs. A random-effects mixed-effects meta-regression (logit-transformed AUC) will be used to examine the effect of moderators, including imbalance ratio, resampling strategy, model family, and sample size. Small-study effects will be assessed with funnel plots, Egger's test, trim-and-fill, and weight-function models; influence diagnostics and leave-one-out analyses will evaluate robustness. Since this is a methodological review, formal clinical risk-of-bias tools are optional; instead, design-level screening, influence diagnostics, and sensitivity analyses will enhance transparency. DISCUSSION: By combining a comprehensive conceptual framework with quantitative estimates, this review aims to determine when data-level versus algorithm-level balancing leads to genuine improvements in discrimination, calibration, and cost-sensitive metrics across various medical fields. The findings will help researchers select concise, evidence-based methods for addressing imbalance, inform journal and regulatory reporting standards, and identify research gaps such as the under-reporting of calibration and misclassification costs, which must be addressed before balanced models can be reliably trusted in clinical practice. SYSTEMATIC REVIEW REGISTRATION: INPLASY202550026.