To create a propensity score for each subject, we constructed a nonparsimonious multiple logistic regression model with nondipping as the dependent variable. We incorporated the following clinical variables into the model as independent variables: age, sex, active smoking, body mass index, race (white vs nonwhite), gout, DM, sleep apnea, obstructive lung disease, congestive heart failure, history of hypertension (defined before the ambulatory BP recording), use of any antihypertensive medication, specific use of various antihypertensive medications (including diuretics, calcium channel antagonists, β-adrenergic receptor antagonists, angiotensin-converting enzyme [ACE] inhibitors, or angiotensin II receptor blockers), baseline laboratory test values (GFR and fasting glucose, log triglyceride, uric acid, total cholesterol, HDL cholesterol, low-density lipoprotein cholesterol, glycosylated hemoglobin A1c, log high-sensitivity C-reactive protein, and baseline serum urea nitrogen and creatinine levels), left ventricular ejection fraction, echocardiographic left ventricular hypertrophy (present, absent, or unavailable), diastolic dysfunction (present, absent, or unavailable), 24-hour mean arterial pressure, and 24-hour mean SBP, diastolic BP, and HR. For the propensity score model, missing values were assigned the mean value for the given variable. Because urinary microalbumin data were available for only 45 patients and because this variable could not be assumed to be missing entirely at random (because this test is primarily ordered among patients at risk for renal disease), it was not included in the propensity score model. Variables reflecting diurnal hemodynamic variation (daytime and nocturnal HR and BP) were excluded from the propensity score model. Propensity scores were divided into deciles, assigning each patient a score between 1 and 10, reflecting the likelihood of being a nondipper.