Abstract: TH-PO014
The Renal Prognosis Prediction Model of Diabetes Nephropathy Based on Machine Learning
Session Information
- AI, Digital Health, Data Science - I
November 02, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Wang, Chuanpeng, Department of Nephrology, State Key Laboratory of Complex Severe and Rare Diseases, Peking Union Medical College Hospital, Chinese Academy of Medical Science and Peking Union Medical College, Beijing, Beijing, China
- Xia, Peng, Department of Nephrology, State Key Laboratory of Complex Severe and Rare Diseases, Peking Union Medical College Hospital, Chinese Academy of Medical Science and Peking Union Medical College, Beijing, Beijing, China
- Zhao, Ze, Institute of Computing Technology, Chinese Academy of Sciences, Beijing, Beijing, China
- Wen, Yubing, Department of Nephrology, State Key Laboratory of Complex Severe and Rare Diseases, Peking Union Medical College Hospital, Chinese Academy of Medical Science and Peking Union Medical College, Beijing, Beijing, China
- Chen, Limeng, Department of Nephrology, State Key Laboratory of Complex Severe and Rare Diseases, Peking Union Medical College Hospital, Chinese Academy of Medical Science and Peking Union Medical College, Beijing, Beijing, China
Background
Diabetes nephropathy (DN) has become one of the most common causes of chronic kidney disease (CKD). This study is to establish a prognosis prediction model of DN using machine learning methods.
Methods
This retrospective cohort study enrolled 247 DN patients diagnosed by renal biopsy at Peking Union Medical College Hospital from December 2012 to September 2021 and collected clinical data from the EMR system. The primary endpoint was all-cause mortality, end-stage renal diseases (ESRD) requiring dialysis and kidney transplantation. We used maximum likelihood estimation to complete the missing data and principal component analysis to standardize the data. K-means, hierarchical, and SOM clustering were compiled in Python to classify the data set. The weight of each variable in the clustering model was measured by several model misjudgments after removing the variable. The weight Analysis was to find the potential risk factors for poor prognosis.
Results
1. About 57.5% of patients had renal insufficiency, and 62% with massive proteinuria. A total of 100 of them reached the primary endpoint with a median renal survival time of 2 years. Multivariate Cox regression showed that the independent risk factors for renal survival included proteinuria (OR = 1.13, 95%CI (1.07, 1.20), P＜0.001), grade 3 hypertension (OR = 2.55，95%CI (1.09, 5.99), P＜0.031) and low eGFR (OR = 1.02, 95%CI (1.00, 1.03), P=0.043).
2. By cluster analysis, two groups of patients had significant differences in renal survival at six months (OR=3.06, 95%CI (1.05, 8.92)), 12 months (OR=4.00, 95%CI (1.65, 9.70)), and 24 months (OR=2.46, 95%CI (1.78, 3.40)), as well as 24hUP (p＜0.001), urinary red blood cells (p＜0.001), hemoglobin (p=0.003), and albumin (p＜0.001).
3. The machine learning models using LR, xgboost, and AdaBoost have the highest accuracy of prediction results in the test set, with an accuracy of up to 87.8% and auc = 0.87; The ten characteristics with the highest weight in this model were blood chlorine, hypertension grade, creatinine, EGFR, gender, diabetes retinopathy, age since the onset of diabetes, free triiodothyronine, urinary red cells, and platelet count.
Conclusion
Machine learning models combined with patient history and laboratory examination are a potentially powerful method for predicting the DN prognosis.
Funding
- Government Support – Non-U.S.