BACKGROUND

Diabetic kidney disease (DKD) refers to chronic kidney disease in diabetes, but not a specific diagnosis. This report describes a patient cohort with electronic health record (EHR) data linked to manual data abstraction for kidney histopathology. METHODS

Patients were selected from the Center for Kidney Disease Research, Education, and Hope (CURE-CKD) registry which contains curated EHR clinical and administrative data from two large healthcare systems. Inclusion criteria consisted of a native kidney biopsy, diagnoses of diabetes and CKD but not on dialysis. Clinical investigators manually abstracted health history, laboratory data, and histological features from kidney biopsy reports. DKD was classified as: diabetic nephropathy (DN), DN mixed with nondiabetic lesions (Mixed), and nondiabetic lesions only (Other). RESULTS

In 523 patients with diabetes who underwent kidney biopsy in the years 2015-2017 (Table), diagnostic frequencies were DN 39.8% (n=208), Mixed 36.9% (n=193), Other 23.3% (n=122). Patients with DN were younger, displayed higher albuminuria, increased nodular glomerulosclerosis and arteriolar hyalinosis than the Mixed group. Those with DN more commonly had diabetes duration >10 years and higher albuminuria compared to the Other group, while lesions characteristic of DN (mesangial expansion, nodular glomerulosclerosis, GBM thickening, arteriolar hyalinosis, tubular basement membrane thickening) were uncommon in Other. CONCLUSION

Higher levels of albuminuria, nodular glomerulosclerosis and arteriolar hyalinosis were distinctly more common in DN compared to Mixed and Other groups, and nodular glomerulosclerosis was rarely observed in the Other group. Future work will use machine learning models of the EHR data to predict DN and select precision therapies.