Development and Validation of a Machine Learning Model for Reduced Kidney Health Risk Assessment

Abstract Background Reduced kidney function is associated with increased risk of all-cause and cardiovascular mortality, cardiovascular disease, end-stage renal disease, yet early detection is hampered by the lack of symptoms and accessible screening tools. Standard definitions may not adequately capture early risk across all age and sex groups. We aimed to develop and validate a machine learning (ML) model using available health parameters to predict risk of reduced kidney health, enabling earlier intervention. Methods Data from adults aged 50 or older from 30 years of National Health and Nutrition Examination Survey (US-NHANES) data (1988-2018) were used. The outcome, reduced kidney health, used age- and sex-specific thresholds based on healthy population percentiles for low estimated Glomerular Filtration Rate (eGFR) or elevated urinary Albumin-to-Creatinine Ratio (uACR). Models were developed on US-NHANES and externally validated in Korea NHANES 2021-2023. The Boruta algorithm identified key predictors. Three ML algorithms were compared; XGBoost was selected as the best model for the Machineborne Early Renal Warning And Control System (MERWACS). Results Among 20,073 adults, 9,534 were used for model development, 4,085 US for internal validation, and 6,454 Korean for external validation. Ten key parameters were identified: age, history of diabetes, hypertension, heart failure, systolic/diastolic blood pressure, ethnicity, arm circumference, pulse, and weight. MERWACS had ROCAUC of 0.716 (95% CI, 0.698-0.734) in internal validation. External validation in South Korea confirmed strong generalizability (ROCAUC 0.728 [0.710-0.744]). The system showed robust calibration. An online interface is being developed. Conclusions The ML-derived system provides a non-invasive early warning tool for identifying risk of reduced kidney function, aiding individuals and physicians in timely interventions to reduce the public health burden. Key messages • MERWACS, an ML model, accurately predicts reduced kidney health risk using 10 accessible health inputs. Strong, generalizable performance validated in US/Korean cohorts. • Validated MERWACS system offers non-invasive early warning for reduced kidney health risk using accessible data, enabling timely assessment & intervention to mitigate risks.