Development and internal validation of a machine-learning-based risk identification model for diabetic kidney disease in patients with type 2 diabetes

Description

Jianquan Luo, Guoxin Chen, and Jiacheng Yang build a machine-learning tool that flags diabetic kidney disease early in people with type 2 diabetes. Kidney damage is the leading route to end-stage renal failure in China, yet awareness and screening lag. Their 2026 study draws on 602 patients at Sihui People's Hospital, 145 of them with kidney disease, using routine laboratory indicators rather than invasive tests.

Six algorithms competed on the data, and every one scored validation accuracy above 0.975 on the area-under-curve scale. Plain logistic regression won, reaching 0.979 on the held-out test set with sensitivity near 89.7 percent and specificity near 92.4 percent. Explanation analysis ranked glycated hemoglobin, serum albumin, urine protein, filtration rate, and beta-2 microglobulin as the five strongest signals. The authors pitch the model as cheap early triage for primary care, while noting that hospitals elsewhere must validate it before wider use.

Catalog record

Source: https://europepmc.org/article/MED/PMC13457287