Prediction-time information leakage in a published machine-learning model of extended length of stay after hip fracture surgery

Companion deposit for the article <em>Prediction-time information leakage in a published machine-learning model of extended length of stay after hip fracture surgery: a temporal validation and deployment simulation</em>.</p><p>Contains the a priori time-stamped protocol and its dated amendment log, the de-identified development cohort (n = 1,137) and 2025 evaluation cohort (n = 100), the machine-readable prediction-time audit of all 43 candidate predictors with the evidence for each classification, the frozen feature-set definitions, all analysis code with a pinned Python environment, the analysis reports, and the three completed reporting and appraisal instruments (TRIPOD+AI, TRIPOD+AI for Abstracts, PROBAST+AI).</p><p>Data and documents are released under CC BY 4.0; code under the MIT licence. The raw hospital information system and DRG settlement extracts are not included, because they contain direct identifiers and the admissions remain re-identifiable within the institution; they are available to named researchers under a data-use agreement with the approval of the Ethics Committee of the People's Hospital of Chongqing Hechuan (approval HX-2025-009).