Data and Code for a 24-Hour Landmark Model of 28-Day Mortality in Sepsis-Associated Acute Respiratory Distress Syndrome

This record accompanies the manuscript "Development and Internal Validation of a 24-Hour Landmark Model for 28-Day Mortality in Sepsis-Associated Acute Respiratory Distress Syndrome." The archive contains the de-identified final analytic dataset used in the revised analysis. It includes 955 records, an analytic row identifier, 26 candidate predictors, and a binary 28-day all-cause mortality outcome. The outcome variable, death_28d, is coded as 0 for survival (481 records) and 1 for death (474 records). The materials include the CSV dataset, a data dictionary, Python analysis scripts, software-version information, model coefficients, derived result tables, and figures. The analysis compares SOFA-only logistic regression, conventional multivariable logistic regression, LASSO logistic regression, random forest, and XGBoost. Internal validation uses five repetitions of five-fold outer cross-validation with three-fold inner tuning. The code also includes calibration assessment, paired AUC comparisons, decision-curve analysis, sensitivity analyses, and held-out XGBoost TreeSHAP analysis. This dataset is the final analytic extract, not the complete original screening database or source clinical records. Source-cohort exclusion records, original missingness records, and time-stamped clinical measurements are not included. The binary outcome file does not independently establish the follow-up time origin. Source-verification notes in the accompanying documentation remain relevant to the interpretation of time-dependent clinical fields. These materials support research verification and secondary methodological research. The models have not been externally validated and are not intended for direct clinical decision-making.