P0713DIFFERENTIAL PREDICTION OF DECLINE IN GFR IN PROTEINURIC VS NON-PROTEINURIC DIABETIC PATIENTS

Abstract

Abstract Background and Aims Diabetes mellitus (DM) is known to cause a progressive loss in renal function in at leaset 2 types of clinical presentation: classical diabetic nephropathy with overt proteinuria and non-proteinuric kidney diseases. The mechanisms of the progression should be multifactorial and are considered different at least in some ways in each patient. Now it is important to predict whether each diabetic patient has progressive renal disease or not, hopefully before overt proteinuria appears. Method A hospital-wide study with all the laboratory data for a period of 4 years and 2 months was conducted. A total of 2,804 non-dialysis patients with an age 18 or more in whom the eGFR slope was calculated over 731 days or more with HbA1c measured at least twice were retrieved. Medians were used for each laboratory datum, including dipstick proteinuria and haematuria (both changed to numeric variables, e.g. 1+ into 1.0), blood cell count, serum chemistry, and HbA1c. Medication class counts were defined as a total number of groups (DPP4, SGLT2, SU, .. ) where insulin was classified into 3 groups (rapid, mixed, and basal); the required prescription period was 3 months or more for each class. Statistical analysis was performed with R 3.6.0 on Ubuntu. For prediction, missing values were imputed by an R package {mice}. Results A total of 2,804 patients (M:F = 1375:1429, age 66.9+-13.6 (18--101, median 68) years) were analyzed whose median HbA1c was 6.14+-0.79 (3.6-11.6, median 5.95)%. This population was further divided into a proteinuric group (the median of dipstick proteinuria of 1+ or more, n=296) and a non-proteinuric group (the median of dipstick proteinuria of +- or less, n=2508). In the multivariate analysis, the proteinuric DM patients had the eGFR slope associated with initial eGFR, proteinuria, uric acid, and the number of classes of the anti-diabetic medications; higher uric acid and less number of DM medication were significantly associated with faster decline in eGFR. The non-proteinuric DM patients had eGFR slope associated with age, initial GFR, uric acid, chloride, and haemoglobin, but not with proteinuria; the patients with higher uric acid with lower haemoglobin had significantly faster decline in eGFR (-1.86 vs -1.34, P = 0.013). Prediction of faster eGFR decline was analyzed with machine learning technologies, ie, {ranger} and {randomForest}, where precision was 68.9%/67.9% (proteinuric group, ranger/randomForest) vs 76.8%/75.9% (non-proteinuric group, ranger/randomForest). Conclusion Propensity to lose kidney function is different in each diabetic patient, which should to be ideally predicted in individual basis.