Severance Hospital researchers develop AI to predict surgical kidney injury

A research team at Yonsei University Severance Hospital has developed an artificial intelligence model capable of predicting acute kidney injury (AKI) in patients undergoing general anesthesia using only eight preoperative clinical data points. The study, published in the ‘International Journal of Surgery’, utilized data from 184,560 adult patients treated at Severance and Gangnam Severance hospitals between 2006 and 2022.

Unlike previous models that required complex intraoperative or postoperative data, this simplified model relies on routine preoperative information such as kidney function, general physical condition, and nutritional or anemia status. This allows for risk assessment without additional specialized testing. In external validation involving 52,480 patients from other hospitals, the model achieved an AUROC of 0.809 for all stages of AKI and 0.908 for moderate to severe stages (stages 2–3).

The researchers proposed categorizing patients into high-risk (over 20%), medium-risk (5–20%), and low-risk (under 5%) groups. This classification aims to assist clinicians in preemptively managing high-risk patients by monitoring blood pressure and fluid status more closely or avoiding medications that could strain the kidneys.