Predicted acute kidney injury risk and early postoperative nonsteroidal anti-inflammatory drug administration in two noncardiac surgery cohorts: Aggregate results and reproducibility code

This package accompanies the study of predicted preoperative acute kidney injury (AKI) risk and early postoperative nonsteroidal anti-inflammatory drug (NSAID) administration in two noncardiac surgery cohorts. INSPIRE and MIMIC-IV are the main cohorts; eICU provides a distinct exploratory ICU-landmark, new-order setting. Version 2026.09.30.1 contains 10 aggregate CSV tables, a data dictionary, analysis and display code, shared upstream sources, computational settings, provenance and checksums. The 2026-09-30 numerical-threshold correction was followed by refitting the original risk models with retained patient folds and rerunning the original analyses. The adjusted highest-versus-lowest risk-decile administration differences are -22.2 percentage points in INSPIRE and -25.4 in MIMIC-IV. These are descriptive associations, not estimates of NSAID effects on AKI. Licenses by component: the aggregate CSV files and data dictionary use CC BY 4.0; the authors' original code and technical documentation use MIT. The eight upstream files preserve their original MIT notice and attribution to PostopAKI Transport, DOI 10.5281/zenodo.23015065. LICENSE_CODE.txt, LICENSE_AGGREGATE_DATA.txt and THIRD_PARTY_NOTICES.md define these scopes. No license to the source databases is granted. Restricted-data boundary: this archive contains no patient-level records, identifiers, individual risk scores, fold maps, fitted models, database files, credentials or internal logs. Readers must obtain their own authorized access to INSPIRE v1.4.2, MIMIC-IV v3.1 and eICU-CRD v2.0 through PhysioNet and comply with the respective data-use requirements. Reproducibility scope: aggregate checks and all five figures can be reproduced from this archive. Patient-level analyses require authorized harmonized inputs. Exact replication of the correction additionally requires private retained cohort/fold and feature checkpoints; fresh extraction, laboratory ties, thread settings and row order can change exact outputs. Risk-score estimation uncertainty was not propagated. See README.md, analysis/INPUTS.md and REPRODUCIBILITY.md for the supported routes and limitations.