A Population-Based Imputation Approach for Missing Baseline Serum Creatinine Values

Background: Diagnosing acute kidney injury (AKI) requires a baseline serum creatinine (S-Cre), which is often missing in the real world. We aimed to develop and validate a new method for estimating baseline S-Cre and compare it to existing methods Methods: From all S-Cre measurements collected from adult outpatients at CMUH in 19 years, we excluded those were 1) <0.3 or >30 mg/dL, 2) from patients with end-stage kidney disease or nephrectomy, 3) within 72 hours following dialysis or cardiopulmonary resuscitation, or 4) within 24 hours after massive transfusion, resulting in 3403528 measurements for baseline S-Cre reference. We developed the “median” imputation methods that utilized the median S-Cre within subgroups stratified by age, gender, and chronic kidney disease (median A), or in combination with body mass index (median B). A sepsis cohort meeting Sepsis-3 was used to assess the performance of 4 imputation methods (median A, median B, eGFR 75, and multiple imputation [MI]) in discriminating AKI according to KDIGO criteria. Bland-Altman plot was used to visualize bias between measured and imputed S-Cre. Results: In the sepsis cohort, median A and B method had a smaller mean of difference, compared to eGFR 75 and MI method. Median A method demonstrated the highest agreement for AKI (84%; Kappa 0.68), compared with eGFR 75 (82%; 0.63) and MI method (81%; 0.62). eGFR 75 had superior sensitivity (85% vs 82%), while MI showed higher specificity compared to median A method (92% vs 86%). Conclusion: The median methods showed satisfactory agreement in AKI detection comparable to eGFR 75 and MI approaches, offering a simple and intuitive approach for real-world application for baseline S-Cre estimation. External validation in the US population is underway.Performance of baseline S-Cre imputation methods using sepsis AKI cohort.Bland-Altman plots of true and imputed baseline S-Cre using different imputation methods.