Figure 1. Blood pressure among normotensive and untreated individuals. The solid red horizontal line represents the diagnostic threshold of 140 mm Hg and the solid blue horizontal line represents the diagnostic threshold of 90 mm Hg. We next investigated seven distributions, six parametric, and one nonparametric, for the baseline hazard function in a proportional hazards model. Generalized gamma, loglogistic, and exponential Figure 2. Time-to-event curves. Figure 3. Parametric hazard functions. Proportional hazards vs. logistic models Using the HUFS data, we found that logistic regression yielded a better fit than the proportional haz- ards model based on the DIC, provided that both age and age” were included as covariates in the logistic regression model (Table 2). Using logistic regression, the addition of age to the reduced (intercept-only) model resulted in a substantially lower DIC (Table 2), with a linear effect of age explaining 27.8% of the variance at the cost of one additional parameter. The addition of age? fur- ther decreased the DIC (Table 2), explaining an additional 0.8% of the variance at the cost of one additional parameter. With smoothing, the effective dimensionality of the proportional hazards model was 3.3, comparable to the dimensionality of 3.0 for the logistic model adjusted for age and age’ (Table 2). We also found that inclusion of a permanent stayer fraction increased the DIC of the proportional hazards model, indicating that inclusion of a permanent stayer fraction was not sup- ported (Table 2). Table 3. Replication of covariates from model selection in African Americans. Table 4. Covariates from model selection in European Americans and Mexican Americans. Discovery study Discovery study The Howard University Family Study (HUFS) is a population-based observational study of African American families and unrelated individuals from Washington, D.C. (Adeyemo et al., 2009). Ethics approval for the Howard University Family Study (HUFS) was obtained from the Howard University nstitutional Review Board (protocol number IRB-06-GSAS-32-A) and written informed consent was sbtained from each participant. All clinical investigation was conducted according to the principles 2xpressed in the Declaration of Helsinki. Families and individuals were not ascertained based on any »henotype. Weight was measured on an electronic scale to the nearest 0.1 kg. Height was measured nN a stadiometer to the nearest 0.1 cm. Body mass index (BMI) was calculated as weight divided by he square of height (kg/m*). Waist circumference was measured to the nearest 0.1 cm at the nar- owest part of the torso. Hip circumference was measured to the nearest 0.1 cm at the widest part of the buttocks. The waist-hip ratio was calculated as waist circumference in cm divided by hip cir- sumference in cm. Fat mass and fat-free mass were estimated using bioelectrical impedance analysis vith a validated population-specific equation as previously described (Luke et al., 1997). Percent fat nass was defined as fat mass divided by weight x100. Blood pressure was measured while seatec ising an oscillometric device (Omron Healthcare, Inc, Bannockburn, Illinois). Three readings were aken at 10 min intervals. Reported readings were the averages of the second and third readings. 4ypertension was defined as SBP >140 mm Hg, DBP >90 mm Hg, or treatment with anti-hyperten- sive medication. Blood was drawn after an overnight fast of at least 8 hr and all collected samples vere stored at —80°C pending biochemical assay. Creatinine, total cholesterol, high-density lipopro- ein cholesterol, low-density lipoprotein cholesterol, triglycerides, fructosamine, glucose, alkaline ohosphatase, alanine aminotransferase, total bilirubin, sodium, potassium, chloride, calcium, uric acid, urea, C-reactive protein, albumin, bicarbonate, and total protein were measured using COBAS NTEGRA tests (Roche Diagnostics, Indianapolis, Indiana). Cortisol and insulin were measured using -lecsys assays (Roche Diagnostics). Creatinine clearance was calculated using the Cockcroft-Gault 2>quation and the estimated glomerular filtration rate (eGFR) was calculated using the four-variable Modification of Diet in Renal Disease Study equation (National Kidney Foundation, 2002; Levey et al., 1999). T2D case status was defined as fasting plasma glucose level >126 mg/dL or reatment with anti-diabetic medication. T2D control status was defined as fasting plasma ylucose <100 mg/dL and no treatment with anti-diabetic medication. The last completed grade ot 2>ducation and income were self-reported on a questionnaire. The proportions of African and Euro- ean ancestry were estimated as described previously (Shriner et al., 2011). We extracted a subset of 1014 unrelated individuals. Materials and methods In summary, by Bayesian modeling of the baseline hazard function in time-to-event analysis of hypertension, we found that logistic regression, if adjusted for both linear and quadratic effects of age, yielded a fit comparable to proportional hazards regression. We found no evidence to support the existence of true controls, suggesting that if an individual lives long enough, hypertension is inevitable. Finally, we found that the combination of chloride, low-density lipoprotein cholesterol, uric acid, and weight, in addition to age and age”, accounted for 40.9% of the variance of hyperten- sion in African Americans, 34.8% in European Americans, and 28.3% in Mexican Americans, simulta- neously consistent with common risk factors among the groups and heterogeneity across the groups. obtained from each participant. All clinical investigation was conducted according to the principles sion in African Americans, 34.8% in European Americans, and 28.3% in Mexican Americans, simulta- uric acid, and weight, in addition to age and age“, accounted for 40.9% of the variance of hyperten-
