OBJECTIVES: Renal impairment is common co-morbidity in patients with diverse main underlying diseases and might be a modifier of treatment effects. Renal dysfunction can be measured by the estimated glomerular filtration rate eGFR level from lab tests. Population-based administrative claims databases are increasingly used in large-scale comparative outcomes studies of drug treatments. However, claims databases often lack information on laboratory tests results. This study aims to develop an approach to predict eGFR level based on longitudinal patient data in the absence of lab measures and identify predictive items (demographic characteristics, diagnoses and procedure codes, therapies) using data-driven analytic techniques. METHODS: Individual level data from 3,951,271 adult patients with recorded eGFR values in OPTUM claims database for years 2007 - 2016 were used to develop a classification system for prediction of eGFR [ml/min/1.73m] classes 0-15 (end-stage renal disease), 15-50 (renal impairment), 50+ (normal renal function). The classifier was built utilizing all available information, aggregated in specific categories, from inpatient, outpatient and pharma facilities. RESULTS: The developed data-driven classification system identified the following main predictors for eGFR-classes: (1) Diagnoses for chronic kidney diseases, ICD-10-CM codes: N02, N03, N11, N18; (2) Disorders resulting from impaired renal tubular function, disorders of kidney and ureter: N25, N28; (3) Phosphorus removing agents from the therapeutic group “Electrolytic, Caloric, Water”; (4) Diagnoses of acute kidney conditions, nephrotic and nephritic syndromes, glomerular disorders and kidney failure: N00, N01, N04, N05, N07, N08, N14, N17, N19; (5) Age. Out of sample validation step showed the sensitivity of the system of 70%, 80%, 82% and specificity of 99%, 82%, 84% to predict eGFR classes 0-15, 15-50, 50+. CONCLUSIONS: A new approach to assess patients’ renal function in administrative claims databases in the absence of lab measures was developed. The approach is applicable to analyses of claims databases of general structure.