A prognostic model of immunoglobulin A nephropathy using artificial neural network: a retrospective study based on integrated Chinese and Western Medicine.

Objective: To establish and evaluate a prognostic model of immunoglobulin A nephropathy (IgAN) based on integrated Chinese and Western Medicine.

Methods: Retrospective analysis from 1/1/2013 to 12/31/2015 was performed on 735 patients who were diagnosed with IgAN. In addition, 105 external data sets from 1/1/2016 to 12/31/2018 were used to verify the constructed model. The end point was entry into end-stage renal disease or a doubling of serum creatinine (Scr) level from baseline. Kaplan-Meier curve survival analysis and multivariable Cox regression analysis were used to find independent prognostic factors. MATLAB 2018b and artificial neural network (ANN) were used to construct prognostic risk factor prediction models each for Traditional Chinese Medicine (TCM), Western Medicine, and integrated TCM and Western Medicine. The ANN model incorporated WANG Yongjun's new five-type syndrome differentiation for IgAN. The prediction efficiencies of the three models were compared using the confusion matrix and the area under thecurve (AUC).

Results: Patients from 1/1/2013 to 12/31/2015 were followed for a mean of (46 ± 19) months. The 5-year median overall renal survival time was 58.6 months, and a total of 40 patients (5.4%) entered the endpoint. Ratio of males to females was 1.48:1. Median age of patients undergoing renal puncture was 35 years. Median 24-hour urinary protein was 0.55 g and 37 patients (5.0%) had pronounced proteinuria (24-hour urine protein ≥ 3.5 g). Median serum creatinine was 76 μmol/L and mean estimated glomerular filtration rate was (90 ± 33) mL/min per 1.73 m². Oxford classification of renal pathology suggested a high rate of focal segmental glomeru-losclerosis (80.3%). Use of immunosuppressants was the most common (71.3%) treatment after renal puncture and improved clinical outcomes of IgAN. TCM differentiation of kidney deficiency was the most common syndrome (69.5%). Independent risk factors for the endpoint were male, anemia, high urinary protein, and an Oxford classification of segmental sclerosis (S). AUCs of the Western Medicine, TCM, and integrated Chinese and Western Medicine models were 0.89, 0.87, and 0.92, respectively. In external data (1/1/2016 to 12/31/2018), the performance of the three models was 0.88, 0.80, and 0.94, respectively.

Conclusions: ANN can be used to successfully construct a 5-year prediction model of IgAN after renal puncture. The efficiency of this model, which combines TCM and Western Medicine factors based on Wang's new five-type syndrome differentiation, exceeds that of Western Medicine factors or TCM factors alone in data from this single-center retrospective study.