Are the estimates of blood pressure (BP) lowering effect the same in parallel and cross-over trials?

Tags: Oral
Wong GW1, Wright JM1
1Cochrane Hypertension Group, Canada

Background: Parallel and cross-over design are two possible designs for randomized control trials assessing the blood pressure (BP) lowering effect of antihypertensive drugs. Each design has strengths and limitations. Both study designs are valid scientific tools to measure the real effect of these interventions. One would expect the effect estimate from parallel studies to be similar to that from cross-over studies.

We conducted four systematic reviews on the BP lowering efficacy of beta-blockers. Both parallel and cross-over studies were included in our review so it was possible for us to conduct an indirect comparison of the two types of designs.

Method: We compared the BP lowering efficacy from parallel and cross-over studies within the same subclass of beta-blockers. The data were synthesized using RevMan 5.2 and the mean estimates from the two designs were compared using Student’s t tests.

Results: The analysis included 97 trials that examined the BP lowering efficacy of 20 different beta-blockers. To be included the studies had to be randomized, double-blind, placebo-controlled trials in primary hypertensive patients taking fixed dose beta-blockers for 3 to 12 weeks.

The BP lowering estimates (systolic/diastolic) from cross-over trials exceeded those from parallel trials for all subclasses: nonselective beta-blockers -12/-9 mmHg vs -5/-4 mmHg, alpha and beta receptor blockers -11/-7 mmHg vs -4/-3 mmHg, partial agonists -9/-5 mmHg vs -7/-3 mmHg and beta-1 selective blockers, -13/-11 mmHg vs -7/-6 mmHg. The overall weighted mean was -13/-9 mmHg and -7/-5 mmHg respectively. All the comparisons were statistically significant except for the partial agonists.

Conclusion: The findings suggest that the BP lowering effect of beta-blockers is greater by 6/4 mmHg in cross-over trials than parallel trials. This cannot be a real effect and creates a serious dilemma for reviewers who are pooling results from cross-over and parallel trials. In this case we suspect that the estimate from the parallel trials is closer to the real effect. These findings need to be replicated in other reviews. It is possible that results from cross-over trials are not reliable.