Background: Results of the Anglo-Scandinavian Cardiac Outcomes Trial-Blood Pressure Lowering Arm (ASCOT-BPLA) show significantly lower rates of coronary and stroke events in individuals allocated an amlodipine-based combination drug regimen than in those allocated an atenolol-based combination drug regimen (HR 0.86 and 0.77, respectively). Our aim was to assess to what extent these differences were due to significant differences in blood pressures and in other variables noted after randomisation.
Methods: We used data from ASCOT-BPLA (n=19 257) and compared differences in accumulated mean blood pressure levels at sequential times in the trial with sequential differences in coronary and stroke events. Serial mean matching for differences in systolic blood pressure was used to adjust HRs for differences in these events. We used an updated Cox-regression model to assess the effects of differences in accumulated mean levels of various measures of blood pressure, serum HDL-cholesterol, triglycerides and potassium, fasting blood glucose, heart rate, and bodyweight on differences in event rates.
Findings: We noted no temporal link between size of differences in blood pressure and different event rates. Serial mean matching for differences in systolic blood-pressure attenuated HRs for coronary and stroke events to a similar extent as did adjustments for systolic blood-pressure differences in Cox-regression analyses. HRs for coronary events and stroke adjusted for blood pressure rose from 0.86 (0.77-0.96) to 0.88 (0.79-0.98) and from 0.77 (0.66-0.89) to 0.83 (0.72-0.96), respectively. Multivariate adjustment gave HRs of 0.94 (0.81-1.08) for coronary events (HDL cholesterol being the largest contributor) and 0.87 (0.73-1.05) for stroke events.
Interpretation: Multivariate adjustment accounted for about half of the differences in coronary events and for about 40% of the differences in stroke events between the treatment regimens tested in ASCOT-BPLA, but residual differences were no longer significant. These residual differences could indicate inadequate statistical adjustment, but it remains possible that differential effects of the two treatment regimens on other variables also contributed to the different rates noted, particularly for stroke.