Fig. 4

Comparison of performance of AKIpredictor and physicians for prediction of AKI-23 by SCr and UO. The black dot represents the classification threshold from the physicians. a At ICU admission (n = 120), AUROCs [95% CI] were 0.80 [0.69–0.92] and 0.75 [0.62–0.88] (P = 0.25), net benefit in ranges 0–26% and 0–74% for clinicians, and AKIpredictor respectively. Physicians’ classification threshold achieved 55% sensitivity, 82% specificity, 33% positive predictive value, and 94% negative predictive value. b On the first morning of ICU stay (n = 187), AUROCs [95% CI] were 0.94 [0.89–0.98] and 0.89 [0.82–0.97] (P = 0.27), net benefit in ranges (0–10% + 90–96%) and (0–48%) for clinicians and AKIpredictor respectively. Physicians’ classification threshold achieved 85% sensitivity, 86% specificity, 31% positive predictive value, and 99% negative predictive value. c After 24 h (n = 89), AUROCs [95% CI] were 0.95 [0.89–1.00] and 0.89 [0.79–0.99] (P = 0.09), with net benefit in ranges (0–36% + 40–48% + 50–67% + 80–100%) and (0–58%) for clinicians and AKIpredictor respectively. Clinicians’ classification threshold achieved 75% sensitivity, 90% specificity, 43% positive predictive value, and 97% negative predictive value. The wide confidence interval for high risk thresholds on the decision curve is amenable to the low number of patients. Therefore, findings should be interpreted with caution