We used a prospective cohort design to investigate predictors of persistent normotension 12 months after withdrawal of antihypertensive drug treatment. We used participating general practitioners' databases to identify suitable participants—patients aged 65-84 years with a history of treated hypertension. We drew participants from among patients volunteering for inclusion in the second Australian national blood pressure study. This was a large randomised controlled trial comparing angiotensin converting enzyme inhibitor and diuretic based antihypertensive treatment for major cardiovascular outcomes and all cause mortality.2
Patients taking antihypertensive drugs at screening were offered withdrawal of drugs as part of the run-in phase. Pretreatment blood pressure could not be identified for all participants, so hypertensive status relied on self reporting. Patients who returned to hypertension were eligible for enrolment in the second Australian national blood pressure study.
Patients who agreed to participate had their previous antihypertensive drug treatment withdrawn gradually under the supervision of a research nurse. During the drug withdrawal phase participants were seen weekly for blood pressure monitoring until a minimum of two weeks after cessation of all antihypertensive drugs. Only those patients whose blood pressure remained within the normotensive range at the two week post-withdrawal visit entered the present study. We defined “normotension” as a sitting systolic blood pressure below 160 mm Hg and a diastolic pressure below 90 mm Hg. These criteria are now historical as they were established before the first patient entered the second Australian national blood pressure study in early 1995. However, in previous studies the level of defined hypertension did not alter the success of drug withdrawal.1
Candidate predictors of maintenance of normotension included body mass index, waist:hip ratio, blood pressure (on-treatment diastolic and systolic), heavy or higher weekend (binge) alcohol intake, recent exercise (walking or other vigorous activity), number of antihypertensive drugs taken, sex, and age. We selected these potential predictors on the basis of previous studies and ready availability to a general practitioner.
After a minimum of two visits to the nurse after cessation of all antihypertensive drugs, participants were followed up by their general practitioner. Typically, general practitioners reviewed each participant 10 times during the subsequent 12 month period (range 1-56 reviews) and recorded blood pressure on four or five occasions (range 0-29 recordings). We reviewed the clinical notes of all participants six and 12 months after withdrawal of treatment and extracted data on blood pressure, drugs, and adverse cardiovascular events. A research nurse measured participants' sitting blood pressure with a standard sphygmomanometer at a 12 month visit.
Twelve months after their entry into the study we classified patients into three groups: (1) Remained off antihypertensive treatment and were normotensive at the 12 month visit (“maintain normotension”). (2) Met study criteria for return to hypertension according to measurement by the study nurse (seated systolic blood pressure
160 mm Hg or diastolic blood pressure
90 mm Hg where systolic blood pressure
140 mm Hg) or had restarted antihypertensive treatment because of a blood pressure level that the general practitioner considered to justify reinstitution of treatment at or before the 12 month visit (“return to hypertension”). We also analysed this group as “return to hypertension early” (<70 days) and “return to hypertension late” (
70 days). (3) Restarted antihypertensive treatment for reasons unrelated to blood pressure or died before classification—this group is referred to as “other.”
We assessed the relation between potential predictors and normotensive status at 12 months by using Cox's proportional hazards regression in order to estimate relative risks, using a constant follow up time of one year with robust estimation of variance to account for clustering within doctor.3,4
We used a multivariate model to determine independent predictors, after standardising continuous predictors to account for differences in scale. We used SAS version 8.2 for all analyses.