A meta-analysis of the long-term effects of antihypertensive therapy on the risk of major cardiovascular disease across 51 randomized trials

Figures

• 

  Fig. 1. Standardized effects of BP-lowering treatment on primary and secondary outcomes across follow-up intervals. HRs and 95% CIs for MACE, ischemic heart disease, stroke, heart failure, cardiovascular death and all-cause death are shown across successive follow-up intervals (0−1, 1−2, 2−3, 3−4, 4−5 and >5 years), estimated using stratified Cox proportional hazards models. Effect estimates were standardized to a 5-mmHg reduction in systolic BP. Participants contributed data to each interval until the occurrence of the outcome, death or censoring. Squares indicate HR point estimates, with square size proportional to the number of events. P values for trend were calculated using two-sided Wald tests to assess linear changes in treatment log-HR across ordered follow-up intervals; for heart failure, P = 0.0004. No adjustment was made for multiple comparisons. SBP, systolic blood pressure.

• 

  Fig. 2. Continuous modeling of BP-lowering treatment effects on primary and secondary outcomes over follow-up time. Estimated HRs and 95% CIs are shown for MACE (a), ischemic heart disease (b), stroke (c), heart failure (d), cardiovascular death (e) and all-cause death (f), standardized to a 5-mmHg reduction in systolic BP, plotted as continuous functions of time since randomization. Solid lines indicate HR estimates; shaded areas represent 95% CIs. The horizontal dashed line denotes no effect (HR = 1). Numbers at risk by treatment group are shown below. For heart failure, the narrow CIs observed early after randomization reflect the mathematical constraints of the Royston−Parmar flexible parametric survival model near the baseline.

• 

  Fig. 3. Effects of BP-lowering treatment on MACE, stratified by follow-up duration and baseline age groups. HRs and 95% CIs for MACE are shown across successive follow-up intervals (0−1, 1−2, 2−3, 3−4, 4−5 and >5 years) and overall within baseline age groups <55, 55−64, 65−74, 75−84 and ≥85 years. Estimates were derived from trial-stratified Cox proportional hazards models and standardized to a 5-mmHg reduction in SBP. Squares indicate HR point estimates; horizontal lines indicate 95% CIs. P values for trend were derived within each age group using two-sided Wald tests to assess linear changes in the treatment log-HR across ordered follow-up intervals. SBP, systolic blood pressure.

• 

  Fig. 4. Effects of antihypertensive drug classes on MACE, stratified by follow-up duration. HRs and 95% CIs for MACE are shown for the major antihypertensive drug classes relative to placebo across successive follow-up intervals (0−1, 1−2, 2−3, 3−4, 4−5 and >5 years). Estimates were derived from network meta-analyses combining two-arm and multi-arm randomized trials, using log-HRs and corresponding standard errors from trial-level Cox models. Squares represent point estimates; horizontal lines indicate 95% CIs. P values for trend were calculated using meta-regression of interval-specific log-HRs to assess evidence for a linear change in treatment effects across ordered follow-up intervals within each drug class, with two-sided Knapp−Hartung tests for linear trend across ordered follow-up intervals. ACEI, ACE inhibitor.

• 

  Extended Data Fig. 1. Standardised effects of systolic blood pressure-lowering treatment on outcomes, stratified by follow-up duration, adjusted for selection bias using inverse probability of censoring weighting. Squares indicate hazard ratio (HR) point estimates and horizontal lines indicate 95% confidence intervals. The vertical line indicates HR = 1. n/N indicates the number of participants with events divided by the number of participants included in the corresponding analysis interval. For each follow-up interval, inverse probability of censoring weights (IPCW) were estimated from logistic models for the probability of remaining event-free beyond the corresponding landmark. Stabilised weights used a numerator model including randomised treatment allocation only and a denominator model additionally adjusted for baseline covariates. Weights were estimated among participants surviving beyond each landmark and truncated at the 1st and 99th percentiles. Interval-specific HRs were estimated using trial-stratified weighted Cox models with robust variance estimation and standardised to a 5 mm Hg reduction in systolic blood pressure. P values for trend were calculated using two-sided Wald tests from…

• 

  Extended Data Fig. 2. Standardised effects of systolic blood pressure-lowering treatment on outcomes, stratified by follow-up duration, using inverse probability of censoring weighting, incorporating time-varying adherence. Nine trials (AASK, ACCORD, ALLHAT, ASCOT, HIJ-CREATE, IDNT, PROGRESS, SHEP and SPRINT) with longitudinal adherence data were included. Squares indicate hazard ratio (HR) point estimates and horizontal lines indicate 95% confidence intervals. The vertical line indicates HR = 1. n/N indicates the number of participants with events divided by the number of participants included in the corresponding analysis interval. Treatment effects were estimated using trial-stratified weighted Cox models, standardised to a 5 mm Hg reduction in systolic blood pressure. Stabilised inverse probability of censoring weights incorporated baseline covariates and the most recently observed adherence status (high adherence was defined as taking ≥80% of prescribed medication) at each landmark. P values for trend were calculated using two-sided Wald tests from weighted trial-stratified Cox proportional hazards models with robust variance estimation to assess linear changes in the treatment log-hazard ratio…

• 

  Extended Data Fig. 3. Standardised effects of systolic blood pressure-lowering treatment on outcomes among participants with high adherence, stratified by follow-up duration. Twelve trials collected and shared adherence data have been included in this analysis. For nine trials (AASK, ACCORD, ALLHAT, ASCOT, HIJ-CREATE, IDNT, PROGRESS, SHEP, and SPRINT) with longitudinal adherence data, a counting process model was employed. Only follow-up periods where participants maintained high adherence status (defined as taking ≥80% of prescribed medication) were included in the analysis. For three trials (ANBP, PREVENT, and PROFESS) lacking longitudinal data, the entire follow-up period of participants who demonstrated overall high adherence (taking ≥80% of prescribed medication) was included. Squares indicate hazard ratio (HR) point estimates and horizontal lines indicate 95% confidence intervals. The vertical line indicates HR = 1. n/N indicates the number of participants with events divided by the number of participants included in the corresponding analysis interval. Treatment effects were estimated using trial-stratified Cox models and standardised to a 5 mm Hg reduction in systolic blood pressure. P values…