Code for: Parsimonious mortality prediction in critically ill ICD-coded COPD and respiratory failure with external validation
Year: 2026
Type: software
Abstract: Code and derived supplementary materials accompanying the manuscript "Parsimonious mortality prediction in critically ill ICD-coded COPD and respiratory failure with external validation" submitted to Scientific Reports (Article). Archive 1 — T2_code_zenodo_v1.1.3.zip: cohort extraction and counting scripts; seed-locked main analysis (repeated 10-fold cross-validation with the entire selection and preprocessing chain re-run inside every fold); internal and no-refit external validation; parsimony chain (univariate screening, LASSO with the 1-SE rule, backward AIC, Wald trimming); competing-score comparisons (SOFA, OASIS, SAPS-II, age+sex+SOFA) with paired DeLong and nested-bootstrap inference for the prespecified SAPS-II equivalence analysis; calibration, decision-curve and clinical-impact analyses; sensitivity analyses S5/S7/S8; the supplementary analyses underlying Supplementary Tables S5-S11; figure scripts including the canonical R version of the LASSO figure; locked software versions and a three-layer reproduction guide with QC anchors. Archive 2 — T2_derived_supplementary_materials_v1.1.3.zip: the consolidated Supplementary Information (Tables S1-S11, including panels E-F of Table S10 and the new Table S11 on 24-h cohort identifiability, component-complete score comparison and calibration by stratum), all aggregate data files referenced by it, the Supplementary Figures S1-S6, the individual tables in Markdown and Word format, and the supplementary index. Archive 3 — T2_additional_analyses_v1.1.3.zip: scripts and aggregate results for the temporal-validation (MIMIC-IV 2008-2016 to 2017-2022), outcome-ascertainment (eICU-CRD missing discharge status and inter-hospital transfer), 24-h cohort-identifiability (eICU-CRD diagnosis time stamps), and component-complete comparison analyses; includes the exact metric values quoted in the manuscript. Data availability: patient-level data from MIMIC-IV and eICU-CRD are NOT included and cannot be redistributed under the PhysioNet Data Use Agreements. Researchers with PhysioNet credentialed access can obtain the databases from https://physionet.org/content/mimiciv/ and https://physionet.org/content/eicu-crd/. All database-facing scripts are environment-variable driven (set T2_MIMIC_DB to a DuckDB catalog containing the MIMIC-IV and eICU-CRD schemas). Reproducibility: random seed 20261018 throughout; the reproduction guide lists cohort counts and headline statistics as QC anchors for drift checking. All files contain aggregate results only. Related identifiersRelated identifier = the accompanying article in Scientific Reports (add the article DOI after acceptance)Related identifier = https://doi.org/10.5281/zenodo.23824112 (earlier, superseded record, v1.0.0) Keywordschronic obstructive pulmonary disease; respiratory failure; intensive care; prediction model; external validation; MIMIC-IV; eICU-CRD; TRIPOD+AI
Author: Tiantian Zhang
Language: English
Subfield: Pulmonary and Respiratory Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7226026792
