Incremental predictive value of cystatin C for next-visit care needs in older adults: reproducibility code, models, and supporting materials from CHARLS

Abstract

This record provides reproducibility materials for the study “Incremental predictive value of cystatin C for next-visit care needs in older adults: a longitudinal study using CHARLS data.” The study evaluates prediction of three next-visit states: alive without basic activities of daily living (BADL) assistance, alive needing BADL assistance, and death. This updated release accompanies the manuscript revision dated 11 October 2026. It includes the original analysis workflow and additional analyses incorporating grip strength, C-reactive protein, and hemoglobin; comparisons under common regularization settings and alternative calibration approaches; sensitivity analyses; assessment-capacity comparisons; and community-based outer-fold evaluation. The materials include data-construction and analysis code, 27 saved final care-prediction models and one outcome-observation model, aggregate results and bootstrap metric draws, figure-generation code, software-environment documentation, validation records, and SHA-256 checksums. The revised manuscript and supplementary materials are provided in Word and PDF formats, together with Figures 1–5 in TIFF, PNG, PDF, and SVG formats. The release is distributed across five ZIP archives:01 — Readme and verification;02 — Original analysis and models;03 — Extended analysis and models;05 — Publication figures. Download all five archives and extract them into the same destination directory, preserving their internal paths. Together they reconstruct the CHARLS_CysC_20261011 folder. Start with README.md and run “python reproduce.py --stage verify” to check file integrity. The README and module documentation describe dependencies, required inputs, and execution instructions. CHARLS participant-level records, derived individual-level panels, individual predictions, and participant-specific fold assignments are not distributed. Full analysis reproduction requires appropriately authorized CHARLS data. The package provides code for data preparation, model fitting, calibration, evaluation, resampling, and figure generation; saved-model application is also supported. Validation records distinguish executed data reconstruction, model replay, aggregate-result checks, and figure reproduction from complete model refitting. The release includes the full fitting and resampling code, but packaging validation did not repeat every historical model fit and bootstrap analysis. Historical components retained for provenance are identified in the README; the current publication files are in the manuscript and figures directory.