Data readiness and cross database transportability of creatinine based acute kidney injury prediction in intensive care

Clinical deployment of acute kidney injury (AKI) AI models depends on whether labels, predictors, missingness, and calibration remain stable across electronic health record (EHR) systems. To develop an AI-ready data passport and transportability stress-test framework for creatinine-based AKI prediction using MIMIC-IV v3.1 and eICU-CRD v2.0. We performed a retrospective secondary analysis of adult ICU stays. Predictions were generated at approximately 24 h after ICU admission using information observed during the first 24 h; this was not an admission-time prediction protocol. The primary outcome was a post-24-hour day-7 serum-creatinine-based operational endpoint, requiring first-24-hour creatinine and post-24-hour follow-up creatinine through day 7. Predictors included age, sex, first-24-hour renal and electrolyte laboratories, missingness indicators, and first-24-hour creatinine count. The endpoint was not intended to represent full KDIGO staging or incident AKI at ICU admission. We assessed endpoint computability and laboratory coverage, then trained L2-regularized logistic regression models in each database and externally validated them bidirectionally. Evaluation emphasized discrimination, Brier score, calibration, and measurement-density failure modes. The feasibility audit included 94,458 MIMIC-IV and 200,234 eICU adult ICU stays. First-24-hour creatinine coverage was 96.8% and 84.6%, respectively; the primary post-24-hour day-7 endpoint was computable in 89.2% and 65.6%. The modeling cohorts included 84,237 stays from 58,782 MIMIC-IV patients and 131,429 stays from 102,013 eICU patients, with primary endpoint event rates of 26.9% and 18.5%. Internal AUROC was 0.732 in MIMIC-IV and 0.723 in eICU. External AUROC was 0.703 for MIMIC-IV to eICU and 0.722 for eICU to MIMIC-IV, with confidence intervals reported in the main table. Calibration shifted directionally: MIMIC-IV to eICU overpredicted risk, whereas eICU to MIMIC-IV underpredicted higher-risk strata. Prediction of a post-24-hour day-7 creatinine-rise operational endpoint was feasible across MIMIC-IV and eICU, but endpoint computability, laboratory coverage, measurement density, and calibration shift differed substantially. The endpoint was not full KDIGO staging or proof of incident AKI at ICU entry, and the retrospective results do not establish clinical deployment readiness. Renal-risk AI should include an EHR data passport and calibration-focused algorithmovigilance before transport or deployment claims.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-14
DOI
https://doi.org/10.1038/s41598-026-77681-5
Primary Topic
Acute Kidney Injury Research
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Data readiness and cross database transportability of creatinine based acute kidney injury prediction in intensive care

Lixin Yin, Xiaoxi Zhang, Shaonan Wang, Giselle Chan
3 links
Scientific Reports
Acute Kidney Injury Research
article

Data readiness and cross database transportability of creatinine based acute kidney injury prediction in intensive care

Lixin Yin, Xiaoxi Zhang, Shaonan Wang, Giselle Chan
article en

Abstract

Clinical deployment of acute kidney injury (AKI) AI models depends on whether labels, predictors, missingness, and calibration remain stable across electronic health record (EHR) systems. To develop an AI-ready data passport and transportability stress-test framework for creatinine-based AKI prediction using MIMIC-IV v3.1 and eICU-CRD v2.0. We performed a retrospective secondary analysis of adult ICU stays. Predictions were generated at approximately 24 h after ICU admission using information observed during the first 24 h; this was not an admission-time prediction protocol. The primary outcome was a post-24-hour day-7 serum-creatinine-based operational endpoint, requiring first-24-hour creatinine and post-24-hour follow-up creatinine through day 7. Predictors included age, sex, first-24-hour renal and electrolyte laboratories, missingness indicators, and first-24-hour creatinine count. The endpoint was not intended to represent full KDIGO staging or incident AKI at ICU admission. We assessed endpoint computability and laboratory coverage, then trained L2-regularized logistic regression models in each database and externally validated them bidirectionally. Evaluation emphasized discrimination, Brier score, calibration, and measurement-density failure modes. The feasibility audit included 94,458 MIMIC-IV and 200,234 eICU adult ICU stays. First-24-hour creatinine coverage was 96.8% and 84.6%, respectively; the primary post-24-hour day-7 endpoint was computable in 89.2% and 65.6%. The modeling cohorts included 84,237 stays from 58,782 MIMIC-IV patients and 131,429 stays from 102,013 eICU patients, with primary endpoint event rates of 26.9% and 18.5%. Internal AUROC was 0.732 in MIMIC-IV and 0.723 in eICU. External AUROC was 0.703 for MIMIC-IV to eICU and 0.722 for eICU to MIMIC-IV, with confidence intervals reported in the main table. Calibration shifted directionally: MIMIC-IV to eICU overpredicted risk, whereas eICU to MIMIC-IV underpredicted higher-risk strata. Prediction of a post-24-hour day-7 creatinine-rise operational endpoint was feasible across MIMIC-IV and eICU, but endpoint computability, laboratory coverage, measurement density, and calibration shift differed substantially. The endpoint was not full KDIGO staging or proof of incident AKI at ICU entry, and the retrospective results do not establish clinical deployment readiness. Renal-risk AI should include an EHR data passport and calibration-focused algorithmovigilance before transport or deployment claims.

Scientific Reports
Cornell University (US), Ansteel (China) (CN), Jiangyin People's Hospital (CN), The Affiliated Taizhou People's Hospital of Nanjing Medical University (CN), Nanjing Medical University (CN)
Good health and well-being
Acute Kidney Injury Research
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon