Assessing the Performance of the Current KDIGO Stage I Criteria to Predict AKI Progression

Background: Kidney Disease: Improving Global Outcomes (KDIGO) are a set of clinical practice guidelines that provide a framework to diagnose and treat Acute Kidney Injury (AKI). Patients often recover from KDIGO Stage I AKI without further intervention. This study assessed the performance of the current KDIGO Stage I criteria to predict progression to Stage II AKI. Methods: We examined de-identified inpatient data obtained from the electronic health record (EHR) from 6 hospitals in a New England hospital system. The exposure of interest were patients with Stage I-defined AKI, and the primary outcome was the progression to Stage II or higher within 48 hours. Variables were created using statistical software to flag AKI using the current KDIGO criteria of a 0.3 mg/dL increase within 48 hours or 50% increase over 7 days from baseline creatinine. Model accuracy was confirmed using the AKIFlagger library, and model discrimination was assessed using AUROC. Results: Of 20,274 patients, 6,352 (31.3%) had KDIGO-Stage I AKI, 1,134 (5.6%) had Stage II, and 120 (0.59%) had Stage III; 1,125 (5.5%) progressed to Stage II within 48 hours; those diagnosed as Stage I by the 50% (7-day) guideline, had a higher rate of progression (15.3% vs. 12.4%); 2,431 (12.0%) were diagnosed with both criteria for Stage I, and 392 (16.1%) progressed to Stage II within 48 hours. The 0.3 (48hr) guideline had lower sensitivity (0.46 vs. 0.63) and higher specificity (0.81 vs. 0.80). These criteria yielded an AUC of 0.66; secondary models, which individually utilize the 0.3 (48hr) and 50% (7-day) criteria, yielded AUCs of 0.64 and 0.60 respectively. Conclusion: The current KDIGO definitions for AKI are not accurate in predicting higher-stage disease progression.