AN ARTIFICIAL NEURAL-NETWORK CAN SELECT PATIENTS AT HIGH-RISK OF DEVELOPING PROGRESSIVE IGA NEPHROPATHY MORE ACCURATELY THAN EXPERIENCED NEPHROLOGISTS

Background. The object of the study was to develop an artificial neura l network (ANN) to identify patients with IgA nephropathy (IgAN) with a poor prognosis and to compare the predictions of the ANN with the pr edictions of six experienced nephrologists. Methods. The following dat a from the time of renal biopsy were retrieved from the records of 54 patients with IgAN: age, sex, systolic and diastolic blood pressure, n umber of prescribed antihypertensive drugs, 24-h urine protein excreti on, and serum creatinine. Patients aged less than 14 years, or who had serum creatinine >350 mu mol/l at presentation, liver disease or conc omitant kidney disease were excluded. Outcome was assigned as 'stable' if serum creatinine was <150 mu mol/l after 7 years and 'non-stable' if serum creatinine was greater than or equal to 150 mu mol/l. The ANN was trained and tested using a 'jack-knife' sampling technique and pe rformance evaluated in terms of number of correct predictions, sensiti vity and specificity. The six nephrologists were asked to predict outc ome at 7 years for each patient using the same data as the ANN and the ir performance was assessed in the same manner. Results. The ANN assig ned the correct outcome to 47/54 (87.0%) patients: sensitivity 19/22 ( 86.4%), specificity 28/32 (87.5%). The mean score for nephrologists wa s 37.5/54 (69.4%, range 35-40), mean sensitivity 72% and mean specific ity 66%. Conclusions. An ANN trained using routine clinical informatio n obtained at the time of diagnosis can potentially predict 7-year out come for renal function in IgAN more accurately than experienced nephr ologists, and can therefore identify a group of high-risk patients req uiring close follow-up.