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

Area of Science:

• Nephrology
• Surgical Outcomes Research
• Health Informatics

Background:

• Postoperative acute kidney injury (AKI) is a significant complication impacting patient recovery, increasing morbidity, mortality, and healthcare expenses.
• Predictive modeling offers a promising approach to mitigate AKI risk in surgical patients.

Purpose of the Study:

• To develop and validate a supervised learning model for predicting the risk of AKI in patients undergoing surgery.
• To utilize the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database for model development and validation.

Main Methods:

• Data from 12 hospitals in the Tennessee Surgical Quality Collaborative (TSQC) (2020-2023) were used for training and temporal validation.
• An external validation set was derived from the 2023 ACS-NSQIP Participant Use Data File (PUF).
• An additive logistic model was selected, excluding patients with prior dialysis or ASA classification of 5; AKI was defined by postoperative renal dysfunction or dialysis within 30 days.

Main Results:

• The study analyzed 59,706 training cases and 980,323 external validation cases, with AKI rates of 1.8% and 2.4%, respectively.
• The additive logistic model achieved an Area Under the Curve (AUC) of 0.87-0.88 in both validation sets.
• Key predictors for AKI included inpatient status, ascites, renal failure, preoperative creatinine, sepsis, American Society of Anesthesiologists (ASA) classification, and patient age.

Conclusions:

• The developed model demonstrates strong predictive performance (discrimination and calibration) for postoperative AKI.
• This tool can identify high-risk patients preoperatively, facilitating targeted perioperative interventions.
• Further research is warranted to validate and implement this model in diverse clinical settings to reduce AKI incidence and associated costs.