Abstract: TH-PO1040
A Novel Approach in Differential Diagnosis of Primary Glomerulonephritis Using the Decision Tree Algorithm Model Based on Biomarkers Panel
Session Information
- Glomerular Diseases: Epidemiology, Mechanisms, Complications, Outcomes
November 07, 2019 | Location: Exhibit Hall, Walter E. Washington Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Glomerular Diseases
- 1203 Glomerular Diseases: Clinical, Outcomes, and Trials
Authors
- Saganova, Elena, Pavlov First Saint-Petersburg State Medical University, Saint-Petersburg, Russian Federation
- Galkina, Olga, Pavlov First Saint-Petersburg State Medical University, Saint-Petersburg, Russian Federation
- Sipovskii, Vasiliy, Pavlov First Saint-Petersburg State Medical University, Saint-Petersburg, Russian Federation
- Smirnov, Alexey, Pavlov First Saint-Petersburg State Medical University, Saint-Petersburg, Russian Federation
Background
Recent studies showed that measurement of various biomarkers (BM) can be useful in differential diagnosis of some primary glomerulonephritis (GN) forms. Our aim was to develop an algorithm for differential diagnosis of primary GN based on BM panel using decision tree learning approach.
Methods
74 patients [39 male, age Me (min 18; max 83) – 37,5 (25; 54) years] with biopsy proven primary GN and without AKI, infectious diseases, severe heart failure, respiratory insufficiency, cancer, abnormal thyroid status, treatment with prednisolone more than 10 mg/per day were included in the study. Based on the results of kidney biopsy (KB) in 7% of cases minimal change disease (MCD) was diagnosed, in 27% – FSGS, in 27% – membranous nephropathy (MN), in 39% – IgA-nephropathy. BM were measured in the morning on the day of KB: serum creatinine(sCr), albumin(sAlb), CysC(sCysC), 24-hour total protein(24hTP), urinary (24-hour collection) cystatin C(uCysC), transferrin(uTr), IgG(uIgG), α1-microglobulin(uα1-mg), β2-microglobulin(uβ2-mg), serum/urine magnesium. The Classification and Regression Trees (CART) learning algorithm with FACT-style direct stopping as pruning criteria was used to create a model for differential diagnosis. Complete machine learning, statistical analysis were performed with Statistica v.12.
Results
A decision tree algorithm was developed including ten predictor variables: age, sAlb, uTrans, t24hTP, sCysC, uIgG, uβ2-mg, sCr, uα1-mg, EFMg. This algorithm accurately classified patients with MCD in 100% cases (5 out of 5 cases), FSGS - 80% (16/20 cases), MN - 85% (17/20 cases), IgA-nephropathy - 96,6% (28/29 cases) (Figure 1).
Conclusion
A “decision tree” algorithm based on age and few urinary, serum BM can be a powerful diagnostic tool in differential diagnosis of primary GN. Application of this algorithm allows to evaluate patients with high risk progression of CKD, identify treatment targets before or instead of KB.
Funding
- Government Support - Non-U.S.