The following is a summary of “Accurate and interpretable prediction of ICU-acquired AKI ,” published in the February 2023 issue of Critical Care by Schwager et al.

Two parsimonious algorithms were formulated and validated to forecast the time of diagnosis of any stage of acute kidney injury (any AKI) or moderate-to-severe AKI within clinically actionable prediction windows. Two gradient-boosting models were developed in this retrospective cohort study of adult ICU admissions. The first model predicted the risk of any AKI at least 6 hours before diagnosis for 50,342 admissions. The second model predicted the risk of moderate-to-severe AKI at least 12 hours before diagnosis for 39,087 admissions. The performance evaluation was conducted before the ailment diagnosis and subsequently confirmed through prospective validation.

The obtained AUROC values for the models were 0.756 and 0.721 at six and twelve hours, respectively, for diagnosing any AKI and moderate-to-severe AKI. In the study, both models exhibited high positive predictive values for any-AKI and moderate-to-severe AKI models, with values of 0.796 and 0.546, respectively. Additionally, the models were more frequently activated in patients who developed AKI compared to those who did not, with a median of 1.82 (IQR 0-4.71) vs. 0 (IQR 0-0.73) triggers per 8 hours for any AKI and 2.35 (IQR 0.14-4.96) vs. 0 (IQR 0-0.8) triggers per 8 hours for moderate-to-severe AKI models. The two acute kidney injury (AKI) prediction models exhibit favorable discriminative performance using conventional features. This can facilitate precise and informative surveillance of AKI susceptibility among patients in the intensive care unit (ICU).

Source: sciencedirect.com/science/article/abs/pii/S0883944123000278