대한신장학회

논문분류	춘계학술대회 초록집
제목	Development and Validation of a Simplified Prediction Model for Postoperative Acute Kidney Injury in 180,000 surgical patients
저자	Dong Hoon Kang
출판정보	2026; 2026(1):
키워드	Acute Kidney Injury, Postoperative, Maching learning
초록	Objectives: Acute kidney injury (AKI) is a common postoperative complication associated with substantial morbidity and mortality. Most existing prediction models rely on extensive perioperative data, limiting their feasibility across diverse surgical settings. This study aimed to develop and externally validate a simplified preoperative machine learning model for predicting postoperative AKI in cardiac and non-cardiac surgical populations. Methods: This retrospective multicenter cohort study included 184,560 patients who underwent surgery at two tertiary hospitals in Korea between 2006 and 2022. Patients receiving preoperative kidney replacement therapy or with baseline estimated glomerular filtration rate <15 mL/min/1.73 m² were excluded. The internal cohort comprised 132,080 patients, and the external validation cohort comprised 52,480 patients. Three prediction models were developed using eXtreme Gradient Boosting with five-fold cross-validation: (1) a full model using all pre-, intra-, and postoperative variables, (2) a preoperative model, and (3) a reduced preoperative model with 8 key variables. Model performance was evaluated using area under the receiver operating characteristic curve (AUROC), calibration plots, and decision curve analysis. Results: Of 184,560 patients (41.7% aged ≥65 years; 53.5% male), postoperative AKI occurred in 11.2% of the internal cohort and 7.9% of the external cohort. The reduced preoperative model demonstrated robust discrimination, with internal AUROC 0.826 (95% CI, 0.818–0.834) for any-stage AKI and 0.918 (0.909–0.926) for severe AKI, and external AUROC 0.809 (0.803–0.817) and 0.908 (0.900–0.916), respectively. Without recalibration, calibration slopes were 0.890 for any-stage and 1.003 for severe AKI (expected calibration error ≤0.010). . Calibration and decision curve analyses confirmed consistent clinical utility across surgical contexts. Conclusion: A simplified, preoperative machine learning model for predicting postoperative AKI was developed and externally validated. The model demonstrated strong performance in diverse surgical populations, supporting its potential use for preoperative risk stratification and informed surgical decision-making.
원문(PDF)	PDF 원문보기

간행물 검색

대한신장학회 회원관리 규정