Machine learning-based prediction models for recurrent stroke after an index stroke: a scoping review protocol

Introduction Recurrent stroke remains a major threat after an index stroke, but individual risk is difficult to estimate because recurrence reflects interacting clinical, imaging, treatment, behavioural and health-system factors. Machine learning (ML)-based prediction models may capture complex prognostic patterns, yet the evidence has not been comprehensively mapped across model development, validation, reporting and translational evaluation. This scoping review will map ML-based prognostic models for recurrent stroke among adult stroke survivors. Methods and analysis This protocol follows JBI methodological guidance for scoping reviews and the scoping review-specific protocol reporting guidance proposed by Peters et al. The completed review will be reported in accordance with Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR), and its searches will be documented using PRISMA-S. The Population-Concept-Context framework includes adults aged 18 years or older with a previous stroke, ML-based prognostic models predicting a subsequent stroke and any healthcare, registry, rehabilitation, community, administrative or population-based setting. Two reviewers will independently screen records. Evidence will be organised at report or study, cohort, model and model-evaluation levels to prevent double counting. Data charting will cover model time zero, prediction horizon, outcome events, development, validation, performance, explainability, equity and translational evaluation. Full Prediction model Risk Of Bias ASsessment Tool + Artificial Intelligence assessment and a separate Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis + Artificial Intelligence reporting map are planned. Findings will be summarised descriptively and narratively; no pooled performance estimate or algorithm ranking is planned. Ethics and dissemination Ethical approval is not required because the review will use publicly available literature. Findings will be disseminated through a peer-reviewed publication, conference presentation and where feasible, open supplementary materials containing the final search strategies and data-charting template. Trial registration number zdy7m; https://osf.io/zdy7m