A web-based machine-learning algorithm predicting postoperative acute kidney injury after total knee arthroplasty

• Published in: Knee surgery, sports traumatology, arthroscopy : official journal of the ESSKA Vol. 30; no. 2; pp. 545 - 554
• Main Authors: Ko, Sunho, Jo, Changwung, Chang, Chong Bum, Lee, Yong Seuk, Moon, Young-Wan, Youm, Jae woo, Han, Hyuk-Soo, Lee, Myung Chul, Lee, Hajeong, Ro, Du Hyun
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.02.2022; John Wiley & Sons, Inc
• Subjects: Acute Kidney Injury - diagnosis; Acute Kidney Injury - epidemiology; Acute Kidney Injury - etiology; Aldosterone; Algorithms; Anesthesia; Angiotensin; Arthroplasty (knee); Arthroplasty, Replacement, Knee - adverse effects; Calibration; Confidence intervals; Creatinine; End-stage renal disease; Hospitals; Humans; Internet; Joint replacement surgery; Joint surgery; Kidney diseases; Kidneys; Knee; Learning algorithms; Machine Learning; Male; Medicine; Medicine & Public Health; Orthopedics; Patients; Performance prediction; Postoperative Complications - diagnosis; Postoperative Complications - epidemiology; Postoperative Complications - etiology; Postoperative period; Prediction models; Renin; Retrospective Studies; Risk analysis; Risk Assessment; Risk Factors; Sports Medicine; Statistical analysis; End-stage renal disease; Total knee arthroplasty; Total knee replacement; Acute kidney injury; Machine learning; Prediction
• ISSN: 0942-2056; 1433-7347; 1433-7347
• DOI: 10.1007/s00167-020-06258-0

Purpose Acute kidney injury (AKI) is a deleterious complication after total knee arthroplasty (TKA). The purposes of this study were to identify preoperative risk factors and develop a web-based prediction model for postoperative AKI, and assess how AKI affected the progression to ESRD. Method The study included 5757 patients treated in three tertiary teaching hospitals. The model was developed using data on 5302 patients from two hospitals and externally validated in 455 patients from the third hospital. Eighteen preoperative variables were collected and feature selection was performed. A gradient boosting machine (GBM) was used to predict AKI. A tenfold-stratified area under the curve (AUC) served as the metric for internal validation. Calibration was performed via isotonic regression and evaluated using a calibration plot. End-stage renal disease (ESRD) was followed up for an average of 41.7 months. Results AKI develops in up to 10% of patients undergoing TKA, increasing the risk of progression to ESRD. The ESRD odds ratio of AKI patients (compared to non-AKI patients) was 9.8 (95% confidence interval 4.3–22.4). Six key predictors of postoperative AKI were selected: higher preoperative levels of creatinine in serum, the use of general anesthesia, male sex, a higher ASA class (> 3), use of a renin–angiotensin–aldosterone system inhibitor, and no use of tranexamic acid (all p  < 0.001). The predictive performance of our model was good (area under the curve 0.78 [95% CI 0.74–0.81] in the developmental cohort and improved in the external validation cohort (0.89). Our model can be accessed at https://safetka.net . Conclusions A web-based predictive model for AKI after TKA was developed using a machine-learning algorithm featuring six preoperative variables. The model is simple and has been validated to improve both short- and long-term prognoses of TKA patients. Postoperative AKI may lead to ESRD, which surgeons should strive to avoid. Level of evidence Diagnostic level II.