Cancer - Kidney(구연) (E-085) 설명가능한 AI 기법을 이용한 신장암 수술 후 수술후 급성 신손상 발생 예측 모델 개발 연구 1. 서울대학교병원 비뇨의학과 고용석1, 서준교1,유상현1, 정규환1, 윤현식1, 박대형1, 육형동1, 정창욱1, 구자현1, 김현회1, 곽철1 Purpose: To develop and validate a post-operative acute kidney injury (AKI) prediction model after kidney cancer surgery using explainable AI (XAI) Materials and methods: We used a retrospective database renal cell carcinoma patients who underwent radical or partial nephrectomy from Feb. 1988 to Apri. 2020 at Seoul National University Hospital. We removed predisposing end-stage renal disease, missing creatinine data from this database. The definition of AKI following Kidney Disease Improving Global Outcomes (KDIGO) guideline and immediate post-OP AKI were measured until postoperative 1 Day. After missing value imputation, we divided the data into development and validation sets. An extreme gradient‐boosting algorithm was applied to the development calculator using five‐fold cross‐validation with hyperparameter tuning following feature selection in the development set. The model feature importance was determined based on the Shapley value. The area under the curve (AUC) of the receiver operating characteristic curve was analyzed for determining model performance. Results: From 3010 available data, immediate post-operative AKI occurred in 29.8% (899/3010 patients) of patients. AKI stage II was 1.7% (15/3010 patients) and AKI stage III was 0.3% (9/3010). The data of 2408 patients were used for development, whereas the data of 602 patients were used as a test set. We selected the variables for the model according to the least absolute shrinkage and selection operator regression. The AUC of the final PC model was 0.801 (95% confidence interval (CI); 0.760-0.862). According to the feature of importance mapping by average Shapley Additive explains suggested Radical nephrectomy, Female, mass size highly impacted on immediate postoperative AKI. Conclusion: We successfully developed and internally validated an immediate post-operative AKI prediction model after kidney cancer surgery using XAI. keywords : Post operative complication, explainable artificial intelligence, prediction model