Predicting chronic renal insufficiency in idiopathic membranous glomerulonephritis. We developed an approach in quantifying the risk of developing chronic renal insufficiency (CRI) based on a cohort of 184 patients with idiopathic membranous glomerulonephritis (IMGN), prospectively followed by the Toronto Glomerulonephritis Registry between 1974 and 1988. After a mean follow-up period of 5.8 years, 26% of patients developed CRI (defined as persistent reduction of creatinine clearance (C_Cr) ≤ 60 ml/min/1.73 m² for ≥12 months). We found that when compared to the baseline probability of the unselected patients, the severity of proteinuria at kidney biopsy added only marginally to the prediction of CRI. We introduced a special test condition: persistent proteinuria (PP) (that is, duration of proteinuria, g/day, above different cut-off levels). We examined the positive predictive value (PPV) and sensitivity (SEN) of 15 arbitrarily chosen levels of PP (that is, proteinuria ≥4, 6 or 8 g/day persisting for ≥6, 9, 12, 18 or 24 months) to select levels with optimal predictive characteristics. We found that PP ≥ 8 g/day for ≥six months was a simple and useful predictor of CRI with a PPV and SEN of 66%. To further improve our prediction, we tested the following parameters: age, sex, initial S_Cr and C_Cr, proteinuria, serum albumin, hypertension, rate of change of C_Cr over time, and therapy (steroids ± immunosuppressive drugs) in a multivariate analysis. Proteinuria, initial C_Cr, and rate of change of C_Cr were most important in predicting CRI. Fifteen models were then developed by including each patient's C_Cr at the start of PP and its rate of change during the time period selected. Two models based on PP ≥ 4 g/day for ≥18 months, or ≥6 g/day for ≥9 months significantly improved the PPV's for CRI from those based on the same levels of PP alone. Using these test conditions, we can improve the prediction of CRI from a baseline probability of 26% in unselected patients to a range of 55 to 86% in the “high-risk” patients (with SEN > 60%). Application of these predictive strategies in IMGN will be useful in managing the individual patients and in selecting patients for clinical trials by limiting the exposure of potentially toxic therapy to the “high-risk” patients.