Application of artificial neural networks in estimating predictive factors and therapeutic efficacy in idiopathic membranous nephropathy

Abstract

Idiopathic membranous nephropathy (IMN) is one of the most frequent causes of the nephrotic syndrome in adults and one of the most common cause of chronic renal failure among primary glomerular diseases. The aim of this study was to develop artificial neural networks (ANN) to investigate factors of poor outcome for IMN and to evaluate the efficacy of different therapeutic protocols. Data were collected retrospectively for 124 patients with IMN (82 males, mean based on the received therapy patients were divided into three groups: corticosteroids only (group 1), cyclophsophamide with corticosteroids (group 2), and so called Ponticelli protocol (group 3). After achieving satisfactory truthfulness of the transcription function of ANN through clustering, we have applied the efficacy analysis to all patients and then compared them to each group separately, and evaluated the influence of initial characteristics on disease outcome as well as the therapy efficacy. The greatest therapy inefficiency was recorded for isolated corticosteroid therapy (29.41%) and the smallest inefficiency for Ponticelli protocol, for which the greatest accuracy of prognosis was recorded (82.09%). The greatest negative prognostic influence had kidney insufficiency (22%), quantitative proteinuria (15%) and index of interstitial infiltration (14%). Based on our results, we can recommend that patients diagnosed with IMN with renal insufficiency, nephrotic syndrome or a high degree of interstitial infiltration at the time of diagnosis should be treated concomitantly with cytotoxic drugs and corticosteroids, particularly with the Ponticelli protocol.