Features Analysis And New Model Establishment For Diabetic Nephropathy And Nondiabetic Renal Disease

Background:Diabetic nephropathy(DN)has become the main reason for end-stage renal disease.Aging plays an important role in diabetes mellitus(DM),chronic kidney disease(CKD).What's more,the Kidney Disease Outcomes Quality Initiative(KDOQI)guideline offered the clinical diagnosis standards for DN and non-diabetic renal disease(NDRD),which is widely accepted.But its differential diagnosis efficacy for Chinese patients is still unknown.And machine learning behaved well in differential diagnosis for many diseases,but wether it could have good performance in differential diagnosis in DN and NDRD is still unknown.Thus,our study is sought to analysis the clinical and pathological features for NDRD patients in different age goups and validate the differential diagnosis performance of KDOQI guideline in Chinses patients.At the same time,we would establish new diagnosis model by machine learning.Methods:There are three parts in our study.First,we included all DM patients with renal biopsy from 1997 to 2017,which were divided into three group according to age to analysis the clinical and pathological features among them.Second,we included all the DM patients with renal biopsy in our center from 2007 to 2016.All the patients were divided into three groups according to pathological patterns-DN,NDRD,DN combined with NDRD,to verify the differential diagnosis performance of KDOQI guideline.Third,we included all DM patients with renal biopsy form 2005-2017 to establish differential diagnosis model by support vector machine and random forest.Results:The first part enrolled 982 patients.IgA nephropathy was the most common pathological pattern in young patients with NDRD.Membranous nephropathy was the most common pathological pattern in elder patients with NDRD.And the odds ratios of risk factors for NDRD also have something to do with age.The second part enrolled 773 patients in the end.KDOQI guideline showed good sensitivity but poor specificity.Rapidly decreasing eGFR,systemic disease,refractory hypertension and the exitence of grey area patients all contribute to the poor performance of the guideline.The third part enrolled 942 patients.The areas of ROC curves for SVM and RF models are all around 0.95.Conclusions:The clinical and pathological features for NDRD patients are all related to their ages.Good specificity makes the guideline more suitable as the screening criteria.Machine learning showed good performance in differential diagnosis between DN and NDRD.