Abstract: FR-PO055
Machine Learning Algorithm to Predict Mortality in Patients Undergoing Continuous Renal Replacement Therapy
Session Information
- AKI: Clinical Outcomes, Trials
November 08, 2019 | Location: Exhibit Hall, Walter E. Washington Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Acute Kidney Injury
- 102 AKI: Clinical, Outcomes, and Trials
Authors
- Kang, Min woo, Seoul National University Hospital, Seoul, Korea (the Republic of)
- Kim, Dong Ki, Seoul National University Hospital, Seoul, Korea (the Republic of)
- Oh, Kook-Hwan, Seoul National University College of Medicine, Seoul, Korea (the Republic of)
- Joo, Kwon Wook, Seoul National University Hospital, Seoul, Korea (the Republic of)
- Kim, Yon Su, Seoul National University College of Medicine, Seoul, Korea (the Republic of)
- Han, Seung Seok, Seoul National University Hospital, Seoul, Korea (the Republic of)
Background
Many scoring systems such as the Acute Physiologic Assessment and Chronic Health Evaluation II (APACHE II) and the Sequential Organ Failure Assessment (SOFA) have been used in predicting outcomes in patients admitted to the intensive care unit (ICU), but these original systems show poor predictability in the subset of patients undergoing continuous renal replacement therapy (CRRT). Accordingly, this study developed the machine learning model to improve the predictability in this subset.
Methods
1,571 adult patients undergoing CRRT were reviewed from 2010 to 2016 years: 70% and 30% of patients were randomly assigned into training and testing set. The primary outcome was mortality in the ICU or hospital admission. To develop the machine learning model, several algorithms were used. (logistic regression, linear discriminant analysis, κ-nearest neighbors, support vector machine, multivariate adaptive regression spline, random forest, extreme gradient boosting and neural networks model) Area under the receiver operating characteristic curves (AUCs) from original scoring systems and the machine learning models were compared using the DeLong test.
Results
Among the machine learning models for ICU mortality, the linear support vector machine showed the highest AUC (0.733), and logistic regression and linear discriminant analysis were the second (0.730 in both). The AUCs of APACHE II and SOFA scores were 0.611 and 0.677, respectively. The support vector machine showed greater predictability than the original systems (Ps<0.05). The machine learning models for in-hospital mortality had a similar trend.
Conclusion
Machine learning models show a better performance in predicting mortality of CRRT patients than the original scoring systems. Accordingly, incorporating the machine learning-based mortality-prediction may be needed when starting CRRT.