Development of machine learning-driven non-invasive diagnostic models for idiopathic membranous nephropathy in Chinese patients.

BACKGROUND

Idiopathic membranous nephropathy (IMN) is a major cause of nephrotic syndrome and end-stage renal disease, but the gold-standard diagnostic method is invasive. This study aims to develop a non-invasive diagnostic model for IMN, focus on the diagnostic value of anti-phospholipase A2 receptor antibody (anti-PLA2R-Ab).

PATIENTS AND

METHODS

In this single-center retrospective study,we included 9524 patients with chronic kidney disease patients who received renal biopsies, extracted 139 clinicopathological data from their records, and divided them into two groups based on pathological results.Renal biopsy cases were collected to form an independent external validation cohort.Seven machine learning methods were used to develop and verify models, and anti-PLA2R-Ab data were used to optimize and evaluate these models. Seventy percent of the patients were used for training, and the other 30% for verification. The area under the receiver operating characteristic curve, F1-score, accuracy, and confusion matrix were used to evaluate the diagnostic performance of the models.

RESULTS

We analyzed 8840 patients and 10 indicators, excluding anti-PLA2R-Ab, to develop and validate diagnostic models, and then analyzed 2457 patients and 6 indicators, including anti-PLA2R-Ab, to develop and validate optimized models. With or without anti-PLA2R-Ab, the CatBoost model provided more accurate diagnosis of IMN (internal vs. external verification AUC:0.921 vs.0.901 and 0.950 vs.0.904, respectively) than anti-PLA2R-Ab alone (AUC: 0.867).

CONCLUSION

The CatBoost model was an accurate and non-invasive method that provided better diagnosis of IMN than anti-PLA2R-Ab in Chinese patients. This model is especially when anti-PLA2R-Ab testing and kidney biopsy are difficult or impossible.