Risk Prediction Model for Postoperative Acute Kidney Injury in a Broad Surgical Population.

BACKGROUND

Postoperative acute kidney injury (AKI) significantly impacts patient recovery, increasing morbidity, mortality, and healthcare costs. This study aims to develop and validate a supervised learning model to predict the risk of AKI in patients undergoing surgical procedures, using data from the American College of Surgeons NSQIP (ACS NSQIP).

STUDY

DESIGN

ACS NSQIP data were collected from 12 hospitals in the Tennessee Surgical Quality Collaborative from 2020 to 2023. The dataset was divided into a training set (2020 to 2022) and a temporal validation set (2023), with an external validation set from the 2023 ACS NSQIP Participant Use Data File. Patients with a history of dialysis and an American Society of Anesthesiologists classification of 5 were excluded. AKI was defined on the basis of postoperative renal insufficiency or dialysis within 30 days of surgery.

RESULTS

The study included 59,706 surgical cases in the training set and 980,323 cases in the external validation set, with AKI incidences of 1.8% and 2.4%, respectively. An additive logistic model was selected as the most parsimonious, achieving area under the receiver operating characteristic curve values of 0.87 to 0.88 in both validation sets. Key predictors included inpatient status, ascites, renal failure, preoperative creatinine, sepsis, American Society of Anesthesiologists classification, and age.

CONCLUSIONS

The model demonstrated excellent discrimination and calibration, particularly in the temporal validation set. The model has the potential to help identify high-risk patients preoperatively, guiding perioperative interventions to reduce AKI incidence and the associated healthcare costs. Further studies are needed to validate, refine, and study real-world applications of the model in diverse clinical settings.