Characteristics of the non-institutionalized US population between 1988 and 1994, based on the Third National Health and Nutrition Examination Survey (NHANES III), are shown in . Mean age was 44.9 years; 53.2% were women, 9.4% were African American, and 5.0% were Hispanic. Mean eGFR was 99.4 ml/min per 1.73 m2 and the median urinary ACR value was 5.7 mg/g. Older age was associated with lower eGFR (r = −0.76) and higher serum creatinine (r = 0.24) and ACR (r = 0.09) levels. Other positive correlations included female sex; self-reported hypertension, diabetes, and cardiovascular disease; systolic and diastolic blood pressure; body mass index; waist–hip ratio; low-density lipoprotein (LDL) cholesterol; C-reactive protein; and glucose. Negative correlations of age included African American and Hispanic race-ethnicity and smoking.
Characteristics of US adults, 1988–1994
The death rate was 10.0 per 1000 person years; 8.7% of the study population died during a mean follow-up interval of 8.7 years. Sensitivity and specificity values for predicting death at different eGFR and ACR thresholds are shown in . Sensitivity (Sn) and specificity (Sp) values for a threshold eGFR of 60 ml/min per 1.73 m2 were 0.25 and 0.98, respectively, and for an ACR threshold of 30 mg/g, 0.27 and 0.94, respectively. MaxSn + Sp thresholds were 94 ml/min per 1.73 m2 for eGFR (Sn/Sp, 0.80/0.67) and 9 mg/g (0.62/0.75) for ACR. When the study population was randomly split into five similarly sized subgroups, maximum MaxSn + Sp levels for eGFR were 85 (0.70/0.79), 94 (0.81/0.73), 95 (0.83/0.63), 93 (0.78/0.68), and 87 (0.68/0.78) ml/min per 1.73 m2. Corresponding values for ACR were 12 (0.63/0.82), 12 (0.57/0.81), 10 (0.52/0.76), 9 (0.64/0.76), and 9 (0.64/0.79) mg/g. In the subgroup with diabetes, MaxSn + Sp thresholds were 76 ml/min per 1.73 m2 for eGFR (0.49/0.82) and 12 mg/g (0.73/0.57) for ACR; corresponding values for the subgroup with hypertension were 83 ml/min per 1.73 m2 (0.69/0.69) and 12 mg/g (0.64/0.70).
Sensitivity (exposure among those who died) and specificity (non-exposure among those who survived) values for death on the y-axis at different estimated glomerular filtration rate (eGFR) and albumin-creatinine ratio (ACR) thresholds
MaxSn + Sp thresholds for other intrinsically continuous variables are shown in , as are mortality odds ratios. Ranked by MaxSn + Sp, age > 57 years was the best discriminator between survival and death, followed by eGFR ≤ 94 ml/min per 1.73 m2, systolic blood pressure > 127 mm Hg, ACR > 9 mg/g, waist–hip ratio > 0.91, glucose > 101 mg/dl, self-reported cardiovascular disease, self-reported hypertension, standardized serum creatinine > 0.97 mg/dl, C-reactive protein > 0.3 mg/l, LDL cholesterol > 148 mg/dl, self-reported diabetes, body mass index > 26 kg/m2, and male sex. In a similar analysis in the subgroup with self-reported diabetes, age > 62 years was the first-ranked discriminator (Sn/Sp 0.84/0.67), followed by systolic blood pressure > 136 mm Hg (0.59/0.75), eGFR ≤ 76 ml/min per 1.73 m2 (0.49/0.82), ACR > 12 mg/g (0.73/0.57), self-reported cardiovascular disease (0.42/0.88), LDL cholesterol > 165 mg/dl (0.32/0.87), HDL cholesterol ≤ 41 mg/dl (0.59/0.6), and C-reactive protein > 0.3mg/l (0.52/0.63). Among participants with self-reported hypertension, age > 65 years was the first-ranked discriminator (0.68/0.79), followed by eGFR ≤ 83 ml/min per 1.73 m2 (0.69/0.69), systolic blood pressure > 136 mm Hg (0.72/0.63), ACR > 12 mg/g (0.64/0.70), self-reported cardiovascular disease (0.33/0.92), serum creatinine > 0.97 mg/dl (0.40/0.83), glucose > 109 mg/dl (0.35/0.79), waist–hip ratio > 0.91 (0.78/0.36), C-reactive protein > 1.4 mg/l (0.14/0.95), and self-reported diabetes (0.15/0.92).
Threshold values for mortality discrimination, ranked by maximum values of sensitivity plus specificitya
Because its MaxSp + Sn for predicting death or survival was highest, age > 57 years was the first item chosen in the classification tree, and it formed the basis for the first two subgroups chosen, ages ≤ 57 and > 57 years. When an otherwise identical approach was used in the subgroup aged ≤ 57 years and age was not considered, systolic blood pressure > 120 mm Hg was the best discriminator; ACR > 12 mg/g was selected for the subgroup aged > 57 years (, ). eGFR first appeared in the third round in two branches of the classification tree: age ≤ 57 years and systolic blood pressure > 120 mm Hg, eGFR ≤ 105 ml/min per 1.73 m2; age > 57 years and ACR ≤ 12 mg/g, eGFR ≤ 63 ml/min per 1.73 m2. ACR also appeared in the fourth round: age ≤ 57 years, systolic blood pressure > 120 mm Hg and body mass index > 26 kg/m2, ACR > 7 mg/g. eGFR also appeared in the fourth round: age > 57 years, ACR > 12 mg/g and no cardiovascular disease, eGFR ≤ 62 ml/min per 1.73 m2; age > 57 years, ACR > 12 mg/g and cardiovascular disease, eGFR ≤ 49 ml/min per 1.73 m2. Regarding other classic risk factors, body mass index first appeared in the third round of the classification tree, and smoking and LDL cholesterol in the fourth round.
When age was not excluded as a candidate variable, age > 44 years and age > 74 years, respectively, were the best mortality discriminators in the subgroups aged ≤ 57 and > 57 years (). shows the corresponding four-level mortality classification tree, in which dichotomizing variables were not excluded. ACR first appeared in the third round (ACR > 12 mg/g, in the subset aged 58–74 years); ACR also appeared in the fourth round, in two subsets (ACR > 11 mg/g in the subset aged ≤ 44 years with glucose > 96 mg/dl; ACR > 7 mg/g in the subset aged 45–57 years with systolic blood pressure ≤ 124 mm Hg. eGFR first appeared in the fourth round (eGFR ≤ 60 ml/min per 1.73 m2 in the subset aged > 80 years). Regarding other classic risk factors, glucose and systolic blood pressure first appeared in the third round of the classification tree, and smoking and LDL cholesterol in the fourth round.
shows mean ages, death rates, and mortality odds ratios when the terminal nodes of the two classification trees were used to classify the study population. Overall, both classification systems appeared to exhibit satisfactory mortality discrimination characteristics, with or without adjustment for age, as C-statistics were ≥ 0.85 in each model.
Mortality risk estimates from categories derived from classification tree analysisa