Risk Prediction Model Reliably Forecasts Postoperative AKI

Shotwell MS, Hennessy C, Martin BJ, et al. Risk Prediction Model for Postoperative Acute Kidney Injury in a Broad Surgical Population. J Am Coll Surg.  February 2026.

Using ACS-NSQIP data from 12 hospitals in the Tennessee Surgical Quality Collaborative (2020–2023), a supervised learning model was developed to predict 30-day postoperative acute kidney injury (AKI), defined as renal insufficiency or new dialysis within 30 days. Patients on preoperative dialysis and those with ASA class V were excluded. The training cohort included 59,706 cases (2020–2022), with temporal validation performed on 2023 cases and external validation using the 2023 ACS-NSQIP Participant Use Data File (n = 980,323). 

AKI incidence was 1.8% in the training cohort and 2.4% in the external validation cohort. An additive logistic regression model was selected as the most parsimonious approach and demonstrated excellent discrimination, with an AUC of 0.87–0.88 in both temporal and external validation sets, along with strong calibration performance. The most influential predictors included inpatient status, ascites, preoperative renal failure, elevated preoperative creatinine, sepsis, higher ASA classification, and advancing age.

With a relatively low overall incidence (~2%), postoperative AKI remains a predictable and clinically significant complication across a broad surgical population. The model relies entirely on readily available preoperative variables, making it feasible for real-world implementation without intraoperative data inputs. High-risk phenotypes, elderly inpatients with physiological derangement (sepsis, ascites), impaired baseline renal function, and elevated ASA class, can be identified before incision. 

With discrimination approaching an AUC of 0.88 across nearly 1 million externally validated cases, this tool supports structured preoperative risk stratification, informed consent discussions, ICU triage planning, and targeted perioperative renal-protective strategies aimed at reducing morbidity and resource utilization.