Using Machine Learning to Identify Risk Factors and Establishing a Clinical Prediction Model to Predict Atherosclerosis Complications in Idiopathic Membranous Nephropathy

Abstract

Background: Clinically, it has been observed that patients with idiopathic membranous nephropathy (IMN) have a higher probability of coronary heart disease. We aim to investigate the risk factors associated with coronary heart disease in IMN patients using a mechanomics approach and establish a clinical diagnosis model.

Methods: We collected sixty-nine clinical data points from patients undergoing phospholipase A2 receptor (anti-PLA2R) tests at the Affiliated Hospital of Qingdao University between July 9, 2019 and March 15, 2021. We excluded patients with cancer, hepatitis B, recent injuries or surgeries, and those under 18. Finally, 162 patients were considered for our study, which included 73 patients with coronary heart disease. The patients were split into test and validation groups at a 7:3 ratio. We utilized the Mann-Whitney U test for initial factor screening and the least absolute shrinkage and selection operator (LASSO) regression for further index screening. Eventually, the effectiveness of the clinical model was evaluated through visual statistical methods.

Results: Age, lymphocyte count, the sum of high-density lipoprotein (HDL) and low-density lipoprotein (LDL), serum creatinine, and antithrombin III were risk factors for coronary heart disease in patients with idiopathic membranous nephropathy in a multivariate regression (p < 0.1). In the training group, 14 clinical features were finally screened by the LASSO regression, and the area under the curve (AUC) of the training group was 0.90 (95% CI 0.877-0.959), accuracy (ACC) was 0.85, sensitivity was 0.76, specificity was 0.91, and precision was 0.85. F1 scored 0.80. In the verification group, AUC was 0.84 (0.743-0.927), ACC was 0.80, sensitivity was 0.67, specificity was 0.87, precision was 0.75, and F1 scored 0.71. We then visualized them using a nomogram based on multivariate regression. The C index and clinical decision curve evaluated them. The C index was 83.8%, and the clinical decision curve was also excellent.

Conclusions: We've established an effective clinical prediction model for patients with IMN who also have coronary heart disease. This model holds significant potential for enhancing clinical decision-making.

Keywords: atherosclerosis (AS); idiopathic membranous nephropathy (IMN); machine learning; prediction model.