#4519 CAN WE REPLACE URINE TESTING AND DELIVER A STEP CHANGE IN ACCESS TO RENOPROTECTIVE MEDICATIONS?

Abstract Background and Aims Albuminuria is a medical condition characterized by leakage of albumin into the urine due to kidney damage. The diagnosed heightened Albumin Creatinine ratio in (ACR) typically categorized microalbuminuria (ACR ≥ 30, < 300 mg/g mg/g) macroalbuminuria mg/g). Currently, despite compelling data, only minority patients with diabetes, rarely individuals without are screened for albuminuria systematic way. ACR tests also present high intra-person variability, making it harder identify meaningful biologic changes increasing complexity clinical trials. In this study we develop method using machine learning predict level from Electronic Health Records, excluding tests, validated model identifying albuminuria. Identifying undiagnosed could help slow progression disease be used speed up recruitment to, reduce screen failures in, Method We developed Quantile Regression US Limited IBM MarketScan Explorys Claims-EMR Data Set (LCED). included subjects who had an test (0 ≤ 5000 was at least 18 years age. Patient demographics (age, sex), vital signs (BMI, Blood Pressure) 8 common blood (Albumin, Bilirubin, Creatinine, HbA1C, Triglyceride, Glucose, White Cell count & ALT) were as covariates. A tree based gradient boosting framework (LightGBM) train quantile regression models, viz., 25-, 50-, 75-percentiles conditioned on then all qualified Optum's de-identified Clinformatics® Mart Database (2007-2021). evaluated performance metrics Area Under Curve (AUC), precision (PPV), specificity (TNR), sensitivity (TPR). Finally, use Kaplan-Meier estimates compare risk failure identified ICD codes (Chronic disease, stage 5; End renal disease; Dependence dialysis; Unspecified failure; Kidney transplant) both predicted measured values. Results final cohort 63,459 matched inclusion exclusion criteria LCED 5,857,385 Optum. Using 25% risk, consistently reaches PPV greater than 0.8 (TNR) 0.99 (Table 1). increases increased (Figure Conclusion results show that models have discriminative power datasets. It predicts micro- macro-albuminuria above 80% 25-quartile. However, classification lacking sensitivity, i.e., suffering may not classified such. By median prediction patient subpopulations par true subpopulation targeted. This means can confidently at-risk individuals. Therefore, our advantageous applications such pre-screening trials, where more important sensitivity. intend validate further outcomes upcoming CKD trial.