Abstract
Background Hip fracture is one of the most severe musculoskeletal injuries in older adults, associated with high mortality, functional decline, and substantial healthcare burden. Globally, the incidence of hip fracture is projected to increase dramatically with population aging, with estimates suggesting over 6 million cases annually by 2050. Despite surgical intervention, post-operative rehabilitation outcomes remain highly variable, with approximately 50% of patients failing to recover pre-fracture functional levels. Accurate prognostic prediction of rehabilitation outcomes is critical for individualized treatment planning, resource allocation, and shared decision-making between clinicians, patients, and families. In recent years, machine learning (ML) methods — including random forest, gradient boosting (XGBoost/LightGBM), support vector machines, artificial neural networks, and deep learning — have been increasingly applied to clinical prognostic prediction, demonstrating advantages over traditional statistical models in handling high-dimensional, nonlinear data. However, the application of ML-based prognostic models in elderly hip fracture rehabilitation remains fragmented, with considerable heterogeneity in algorithms used, outcomes predicted, and methodological quality reported. No comprehensive evidence synthesis has systematically mapped this body of literature. Objectives This study aims to conduct a scoping review and evidence map to systematically identify, describe, and visualize the existing evidence on ML-based prognostic prediction models for post-operative rehabilitation outcomes in elderly hip fracture patients. The specific objectives are: To identify the volume, scope, and characteristics of published studies in this field; To describe the distribution of ML algorithms, predicted outcomes, model performance metrics, and key predictive features; To assess the risk of bias of included studies using the Prediction model Risk Of Bias ASsessment Tool (PROBAST) and reporting completeness using the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) checklist; To generate evidence maps (bubble chart and heat map) visualizing the density and gaps of evidence across ML algorithm types and rehabilitation outcome categories; To identify research gaps and priorities to guide future high-quality ML prediction model development in this domain. Methods This scoping review and evidence map will be conducted in accordance with the JBI methodology for evidence mapping and reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) checklist. Eligibility criteria: Studies will be included if they: (1) enrolled elderly patients (aged ≥60 years) with hip fracture (femoral neck fracture, intertrochanteric fracture, or subtrochanteric fracture) undergoing surgical intervention; (2) developed or validated a prognostic prediction model using ML methods (including but not limited to logistic regression as baseline comparator, decision tree, random forest, support vector machine, XGBoost/LightGBM, artificial neural network, or deep learning); (3) predicted at least one post-operative rehabilitation-related outcome, including functional recovery (e.g., Barthel Index, FIM score, ADL), mortality, readmission, complications, length of hospital stay, discharge destination, or quality of life. All study designs will be considered, including model development studies, validation studies, and systematic reviews. No restrictions will be applied to publication date or language (English and Chinese). Search strategy: A systematic search has been conducted across six electronic databases: PubMed, Web of Science (Core Collection), Scopus, Embase, Cochrane Library, and CINAHL. The search was completed on December 12, 2025. Study selection: Two independent reviewers will screen titles and abstracts, followed by full-text assessment of potentially eligible studies. Discrepancies will be resolved through discussion or arbitration by a third reviewer. Data extraction: Standardized data extraction will be performed independently by two reviewers, capturing study characteristics (author, year, country, sample size), population characteristics (age, fracture type, surgical approach), ML model features (algorithm type, number and category of predictors, feature selection method, missing data handling), predicted outcomes (outcome type, definition, follow-up time point), model performance metrics (AUC/C-index, sensitivity, specificity, calibration), validation strategy (internal/external), model interpretability (SHAP values, feature importance), and clinical translation (decision curve analysis, web calculator). Quality assessment: Risk of bias will be assessed using the PROBAST tool across four domains (participants, predictors, outcome, analysis), yielding domain-specific and overall risk of bias ratings (low/high/unclear). Reporting completeness will be evaluated using the TRIPOD checklist. Evidence mapping: A bubble chart (ML algorithm type × outcome category, with bubble size representing study count) and a heat map (evidence density matrix) will be generated to visualize the distribution of evidence and identify research gaps. Registration Note Database searches were completed on December 12, 2025. Title and abstract screening has been completed prior to this registration. This registration is submitted during the full-text screening phase, before data extraction, quality assessment, or any form of evidence synthesis has been initiated. No outcomes or patterns within the data have been observed or analyzed at the time of registration. Expected Outcomes This study is expected to produce the first comprehensive evidence map of ML-based prognostic prediction models for post-operative rehabilitation in elderly hip fracture patients. Key outputs will include: a characterization of the current evidence landscape (study volume, geographic distribution, algorithm types, outcome coverage); an assessment of overall methodological quality and reporting standards in this field; visual evidence maps identifying areas of concentrated research and critical evidence gaps; and a research agenda highlighting priorities for future model development, external validation, and clinical implementation studies. Findings will be submitted for publication in a peer-reviewed journal focused on evidence synthesis or clinical prediction modeling.