Admission versus historical serum creatinine alters AKI classification in cirrhosis: a retrospective study of MIMIC-IV and eICU-CRD

The serum creatinine (SCr) value used to define and stage acute kidney injury (AKI) can alter case identification and severity classification. The terminology used to identify kidney replacement therapy (KRT) also affects clinically important outcome counts. We quantified both effects in adults with cirrhosis. Adults with coded cirrhosis and AKI were identified in MIMIC-IV and eICU-CRD using prior measured values or, when these were unavailable, the lower of admission and back-calculated SCr. Within fixed MIMIC-IV admissions, AKI status and stage were re-derived using admission SCr. The SCr measured when the creatinine criteria were first met was also divided into 3 categories chosen to approximate the stage sizes obtained using historical SCr. The outcome was in-hospital KRT or death within 28 days. Paired comparisons used cluster-bootstrap 95% confidence intervals. Among 889 MIMIC-IV admissions meeting AKI criteria under both denominators, stage changed in 136 (15.3%), including 99 downgrades. Admissions downgraded to stage 1 had a 59.6% event rate versus 26.9% in those remaining stage 1; the 23 stage 3-to-1 admissions had the highest rate among downgraded cells (78.3%). Stage based on historical SCr exceeded the 3-category SCr grouping by 0.041 in AUROC (95% CI, 0.013–0.071), whereas stage based on admission SCr performed similarly to that grouping (difference, − 0.002; 95% CI, − 0.033 to 0.031). Admission SCr caused 396 of 1,285 admissions (30.8%) to no longer meet AKI criteria at the same time point; 206 met no criterion anywhere in 7 days (16.0%). In the same 785 eICU admissions evaluated using both KRT term lists, the expanded KRT term list raised recorded events from 53 to 94. Using admission rather than historical SCr changed AKI case identification and shifted admissions from higher to lower stages. Using an expanded KRT term list increased recorded events in eICU-CRD. Clinical interpretation and multicenter database studies should explicitly document which SCr value was used to define and stage AKI and the terms used to identify KRT.