Instrumental variable meta‐analysis of the effect of systolic blood pressure reduction on cognitive change: Evidence from randomized trials of antihypertensive therapies

Abstract

AbstractBackgroundPrevious meta‐analyses of trials of antihypertensive drugs have only evaluated the effect of randomization to treatment on cognition, without accounting for duration of treatment or degree to which treatment decreased blood pressure. We assumed that any cognitive benefit of antihypertensive treatment would be proportional to the magnitude of change in systolic blood pressure (SBP) achieved and the duration of follow‐up during which a lower SBP was sustained.MethodWe aggregated data from randomized trials of antihypertensive drugs which reported data on change in both SBP and the Mini Mental Status Exam (MMSE). We estimated the effect of decreasing SBP on annual cognitive change for each trial using an instrumental‐variable approach. This was determined for each trial separately using maximum‐likelihood estimation, assuming differences in cognitive change between treatment arms was proportional to the difference in SBP between the groups and the duration of follow‐up. For interpretability, we expressed effects as difference in MMSE per 10mmHg change in SBP per year. Results were combined using a fixed‐effects meta‐analysis.ResultOut of 8 trials (N=70,734; follow‐up: 2.3 to 6 years), 7 suggested benefit of SBP reduction on cognitive change, with 2 of these 8 trials reaching statistical significance. No trials showed statistically significant cognitive harm. The pooled estimate suggested that a 10mmHg decrease in blood pressure slowed MMSE decline by 0.033 (95% CI: 0.023 to 0.044) points per year.ConclusionRandomized‐controlled trials of antihypertensives indicate that decreases in SBP slow cognitive decline. Even with a clinically substantial decrease in blood pressure, the average long‐term effect appears likely to be modest. Instrumental‐variable meta‐analysis can be used to provide an estimated annual benefit per change in SBP, allowing results from trials of heterogeneous treatments to be put on the same scale. Such estimates have more clinical relevance than standard intent‐to‐treat estimates.