Participants and interventions
Patients were eligible for inclusion in the study if they were aged 18–80 years, of South Asian origin (defined as both parents originating from the Indian subcontinent), and were known to have mild to moderate hypertension (defined as having a sitting diastolic blood pressure of 90–114 mmHg and systolic blood pressure of <180 mmHg). Patients were excluded if they had type 1 or type 2 diabetes mellitus, hypercholesterolaemia or established CHD (that is, multiple transient ischaemic attacks, angina, treated heart failure, myocardial infarction within the previous 3 months or stroke in the previous year).
Before entering the study, all eligible patients gave written, witnessed informed consent, having been provided with information sheets in English, Bengali, Gujerati, Hindi, Punjabi or Urdu, as appropriate.
Before starting active treatment, patients underwent the following examinations: demographic details, medical history (including previous and concurrent treatments), height, weight, waist/hip ratio, sitting blood pressure, heart rate, 2-hour oral glucose tolerance test, fasting, 30-minute and 2-hour insulin and proinsulin, fasting lipids, fasting, 30-minute and 2-hour non-esterified fatty acid (NEFA) lipid particles, urate, haematology, electrolytes and liver function tests. Blood pressure was measured locally using a mercury sphygmomanometer, following standard practice procedures. All blood and urine tests were performed by a central laboratory.
After the initial assessment, blood pressure, heart rate and adverse events were recorded at each subsequent visit throughout the study (that is, at weeks 0, 2, 4, 6, 8, 14, 21 and 34). Weight, waist:hip ratio, glucose, insulin, proinsulin, lipids, NEFA, urate, haematology, electrolytes and liver function tests were recorded again at weeks 21 and 34.
Patients entered a 2-week placebo run-in (weeks −2 to 0) to ensure that they were moderately hypertensive. At week 0, eligible patients were randomised to take one capsule of doxazosin (1 mg) or bendrofluazide (2.5 mg) each morning. The doxazosin dose could be titrated to the next highest dose (2, 4 or 8 mg once daily) at weeks 2, 4 and 6 if the patient's diastolic blood pressure was more than 90 mmHg or had not fallen by at least 5 mmHg since the previous visit. Patients receiving either treatment could be co-prescribed open-label amlodipine (5 mg) from week 8 onwards if diastolic blood pressure was at or above 90 mmHg.
Blinding was maintained by having four packs (A, B, C, D) for each drug. For doxazosin, packs A, B, C and D contained doxazosin 1, 2, 4 and 8 mg, respectively. For bendrofluazide, packs A, B, C and D were identical, with each containing bendrofluazide (2.5 mg). Placebo, doxazosin and bendrofluazide were presented as identical-sized dark-grey capsules. Amlodipine tablets were provided from commercial stocks.
Efficacy and safety variables
The primary efficacy variable in assessing the effects of doxazosin and bendrofluazide on glucose metabolism was the change in glucose concentrations from baseline (week −2) to week 21, measured by a 2-hour oral glucose tolerance test. The primary variable in assessing the effects on lipid metabolism was the change in total cholesterol concentrations from baseline (week −2) to week 21.
Secondary efficacy variables were the changes in insulin, proinsulin, LDL-C, HDL-C, triglycerides, NEFA and sitting blood pressure from baseline to weeks 21 and 34. Baseline was week −2 for all variables apart from blood pressure, when it was week 0.
Safety variables included clinical and laboratory adverse events (defined as any untoward medical occurrence, regardless of cause, in a patient administered a pharmaceutical product). If an adverse event was reported the investigator recorded the intensity (mild, moderate or severe, defined on subjective criteria), the relationship to the study drug and the outcome.
Power calculations and statistical methods
Sample size was calculated assuming a 5% improvement in LDL-C in patients treated with doxazosin compared with those treated with bendrofluazide. This was based on a baseline LDL-C of 4.5 mmol/l with a standard deviation of 0.614 mmol/l. The correlation between LDL-C concentrations at baseline and after 12 months was 0.8 in a previous, unpublished study of doxazosin (Pfizer Inc, data on file). A sample of 100 patients (50 per treatment group) was sufficient to result in a power of 80% at a significance level of 0.05.
Randomisation was performed centrally, so at week 0, the site telephoned the randomisation centre, with a minimisation code based on age (18–59 years or 60–80 years) and sex for each eligible patient. This code was used to assign a treatment number from the batch that had been issued to the site so that the overall stratification for age and sex was balanced between treatment groups.
The primary population for analysis was the intention-to-treat (ITT) population, which included all randomised patients who received at least one dose of the study drug and had baseline efficacy measurements. Missing measurements in the ITT population were treated as missing, as was pre-specified for the primary analysis. The alternative ‘last-observation carried forward’ analysis (when patients withdrew from treatment or had missing measurements, the last observation was carried forward for the analysis) is not presented here, but produced similar results and does not alter conclusions.
The ITT population was analysed for treatment differences (doxazosin–bendrofluazide). The general linear model ‘mean change in parameters = a+b (centre)+f(treatment)+j(sex)+k(age)+w(baseline)' was fitted. Age was considered as a categorical variable and patients were classified as under 60 years or 60 years and over. The model was fitted using type II sums of squares, and least squares means were computed. Data transformations were used on non-normally distributed data. No adjustments were made for multiplicity, because there was only one primary efficacy variable each for glucose metabolism and lipid metabolism. Version 6.12 of SAS for Windows was used for the production of all data summaries and analyses.
Statistical analyses were performed using analysis of covariance, which has less power to detect interaction effects than main effects; hence a higher significance level was used to test for interaction effects. The treatment effect was assessed for statistical significance at the 5% level, and, if the covariates were significant, additional tabular data summaries were generated displaying the primary parameter means by covariate categories. If the treatment effect was not found to be significant, no further models were fitted in the main analysis. If the treatment effect was significant, the model was refitted with terms for treatment by covariate interaction to assess for robustness of the treatment effect across the covariates. The interactions were tested for significance at the 10% level and least squares means computed.
