Abstract
In this study, we attempted to identify threshold values for kidney function measures that maximally discriminate short-term mortality, to identify major population segments in which these thresholds apply, and to classify the hierarchical rank of the thresholds when other classic risk factors are also considered. To do this we retrospectively identified estimated glomerular filtration rate (eGFR) and urinary albumin-creatinine ratio (ACR) thresholds to maximize sensitivity and specificity predictions for death in non-institutionalized NHANES III participants, representative of the United States population from 1988 to 1994 and followed through 2000. In a classification tree excluding dichotomizing variables, age 57 years was initially selected; ACR appeared in the second round and eGFR in the third. The prognostic discrimination of optimum eGFR and ACR thresholds exceeded those of commonly advocated public health screening measures, such as LDL cholesterol and fasting blood glucose, with body mass index appearing in the third round, and smoking and LDL cholesterol in the fourth. In a tree permitting dichotomizing variables, the ACR, systolic blood pressure, and glucose first appeared in the third round, with eGFR, smoking, and LDL in the fourth. Thus, the albumin-creatinine ratio and eGFR may be at least as efficient for survival-based clinical triage as most other classic risk factors.