Abstract: FR-PO468
A Prediction Model for Progression to ESKD, Based on the Neural Network (ANN) Analysis, in IgA Nephropathy Patients
Session Information
- CKD: Risk Factors for Incidence and Progression - II
November 03, 2017 | Location: Hall H, Morial Convention Center
Abstract Time: 10:00 AM - 10:00 AM
Category: Chronic Kidney Disease (Non-Dialysis)
- 301 CKD: Risk Factors for Incidence and Progression
Authors
- Schena, Francesco Paolo, University of Bari, Bari, Italy
- Anelli, Vito walter, Polytechnic University of Bari , Bari, Italy
- Tomeo, Paolo, Polytechnic University of Bari, Italy, Bari, Italy
- Di noia, Tommaso, Polytechnic University of bari, Bari, Italy
- Russo, Maria luisa, University of Turin, Turin, Italy
- D'arrigo, Graziella, Clin. Epid. and Physiopath. of Renal Dis. and Hypertens., CNR-IBIM, Reggio Calabria, Italy
- Tripepi, Giovanni, CNR-IBIM, Reggio Calabria, Italy
- Tesar, Vladimir, None, Prague, Prague, Czechia
- Zoccali, Carmine, Nephrology, Transplantation and Hypertension, Reggio Calabria, Italy
- Coppo, Rosanna, None, Prague, Prague, Czechia
Group or Team Name
- On behalf of VALIGA Study
Background
IgA nephropathy(IgAN) is the most common biopsy-proven glomerulonephritis in the world characterized by progressive deterioration of the renal function. Many statistical appoaches have been used to predict ESKD in IgN patients. The ANN is a non-linear statistical approach for pattern recognition in order to weight all the relationships between input and output variables. We have updated and used the clinical decision support system, previously published (NDT 2016), to estimate the ESKD risk.
Methods
A cohort of 680 adult patients from the European VALIGA study with complete dataset was used for this study. Of these 178 reached ESKD. Five indicators (age,sex,serum creatinine,daily proteinuria,hypertension) and MEST-C (mesangial and endocapillary hypercellularity, segmental glomerulosclerosis, tubular atrophy/interstitial fibrosis and presence of crescents) classification at time of kidney biopsy were used for the training and validation process of the ANNs. Then, we included follow-up (years) and therapy.Two indipendent ANNs were used for predicting, first, ESKD and, then, the time occurring to ESKD. For every ANN, 80% of IgAN patients were included in the training set and 20% in the validation set. Four well-known classification metrics were used for the first ANN and two error-based metrics were used to evaluate the regression algorithm for the second ANN.
Results
The first model to predict ESKD by 15 years of clinical outcome had accuracy 98%, precision 93%, recall 87% and F1-measure 87%.The second ANN to predict the number of years to achieve ESKD showed a RMSE (root mean squared error) of 2.68 years and a MAE(mean absolute error) of 2.12 years.
Conclusion
We have developed a new clinical decision support system to estimate the risk of ESKD and its timing in IgAN patients. This tool showed an excellent performance. Interestingly, the VALIGA cohort included many patients who received or not ACEi or ARBs or immunosuppressive therapy that may influence the clinical course of the disease. Thus, the training and validation phase of our ANNs considered therapy exposure.
Funding
- Government Support - Non-U.S.