Alternative clinical trial designs to find a clinically important effect size of blood pressure-lowering on cognitive impairment and dementia

Abstract

Background

Blood pressure (BP) lowering might reduce the risk of cognitive impairment and dementia (CID), but more information is needed to design trials optimally.

Objective

To estimate trial sample sizes and durations for detecting BP-lowering effects on CID.

Methods

We estimated trial sample sizes and durations for finding an effect of a Systolic Blood PRessure INtervention Trial (SPRINT)-based BP-lowering strategy versus usual care on CID and global cognition using our MIchigan ChROnic Disease SIMulation Model (MICROSIM), which simulated adults with the US population's demographics and vascular risk factors and assumed BP-lowering's effect on CID and global cognition from our pooled cardiovascular cohort study.

Results

Over >139 million simulated trials across 78,000 parameters, the SPRINT-based BP-lowering strategy (versus usual care) resulted in a 2% CID relative risk reduction (RRR), much smaller than the CID RRR (15%) found in SPRINT Memory and Cognition IN Decreased Hypertension (MIND). Detecting a 2% CID RRR with >80% power would require 10-year trials with samples >50,000. Identifying the BP-lowering strategy's effect on mean global cognition would require trials of ≥10 years with >30,000 participants. However, trials of 5 years with 10,000 participants would have the power to detect the strategy's effect assuming a 15% CID RRR in SPRINT MIND.

Conclusions

Keywords

Alzheimer's disease blood pressure cognition dementia

Introduction

Cognitive impairment and dementia (CID) affect millions of people worldwide.^1,2 Interventions to prevent CID are urgently needed. Modifiable vascular risk factors contribute substantially to CID risk,³ and vascular risk factor interventions are likely to reduce these risks. However, studying potentially effective strategies to prevent CID is challenging because the optimal trial designs for vascular health interventions are often unclear.

High blood pressure (BP) is a common risk factor for CID, and BP-lowering interventions might prevent CID. The Systolic Blood PRessure INtervention Trial Memory and Cognition IN Decreased Hypertension (SPRINT MIND) trial found that treating BP to a goal of <120/80 mmHg using up to 4 BP medications reduced mild cognitive impairment risk in adults aged ≥50 with high atherosclerotic cardiovascular disease (ASCVD) risk.⁴ However, SPRINT MIND was underpowered to examine clinically significant reductions in dementia as a stand-alone outcome. Moreover, the effect size of BP lowering on CID in SPRINT MIND was larger than the effect size of BP lowering on CID observed in observational studies,³ raising concerns that SPRINT MIND's effect size overestimates the true effect. Whether lowering BP reduces dementia risk remains unclear, especially for those without elevated ASCVD risk. Understanding the influence of ASCVD and dementia risks on trial sample size and duration will also help identify participants to target for trials.⁵

Using our dementia and ASCVD microsimulation analysis platform and data from our BP-COG consortium of pooled cohort studies,^6–9 we determined the sample size and duration of a trial that would be adequately powered to find an effect size of BP-lowering on CID and global cognition that is clinically important and examined how participants’ ASCVD and CID risks influence trial design. Specifically, we examined whether the cognitive effects of BP-lowering, assuming the associations found in the best available observational studies are causal,⁷ would be detectable in a large clinical trial, and what trial parameters are required to identify these effects.

Methods

Overview of MICROSIM

Following reporting guidelines,¹⁰ we conducted a microsimulation analysis using an open-source dementia and ASCVD microsimulation framework we developed, the Michigan Chronic Disease Simulation Model (MICROSIM). MICROSIM has been described elsewhere.⁶ MICROSIM is an agent- and regression-based Monte Carlo microsimulation model that estimates annual transitions in risk factors and outcomes, including all-cause dementia and ASCVD events, in US adults. The MICROSIM population includes adults aged ≥18 without dementia or ASCVD and is nationally representative of the demographic and ASCVD risk factors of US adults, based on the National Health and Nutrition Examination Survey (NHANES).¹¹ We used simulation methods instead of traditional methods to estimate the sample size, duration, and statistical power of the trials because simulation methods are well-suited to exploring power across varying baseline population parameters, such as varying baseline ASCVD and CID risks, and also can account for the known complex interplay between factors that may influence trial outcomes such as competing risks between cardiovascular and cognitive outcomes.¹²

MICROSIM's model inputs were derived from the best available evidence, largely modeling state transitions with regression models derived from individual participant data pooled analyses of 5 US observational cardiovascular cohort studies in our BP-COG project.^7,13 Specifically, the BP-COG project (The Effect of Lower Blood Pressure over the Life Course on Late-life Cognition in Black, Hispanic, and White Individuals; R01 NS102715; Levine DA, PI) pooled and harmonized individual participant data from 5 US observational cardiovascular cohort studies with longitudinal cognitive data: Atherosclerosis Risk in Communities Study (ARIC),¹⁴ Coronary Artery Risk Development in Young Adults Study (CARDIA),¹⁵ the Cardiovascular Health Study (CHS),¹⁶ Framingham Offspring Study (FOS),¹⁷ and Northern Manhattan Study (NOMAS).¹⁸ MICROSIM's code and input parameters are publicly available (https://github.com/jburke5/microsim). The University of Michigan Institutional Review Board approved this study. Participating institutions’ review boards approved the cohort studies. All participants provided written informed consent.

MICROSIM updates individuals’ statuses annually as all-cause dementia, fatal or non-fatal ASCVD, non-ASCVD death, or remaining free of all events, using Monte Carlo methods. MICROSIM has been validated to reproduce population-level all-cause dementia, stroke, and myocardial infarction (MI) incidence.⁶ Estimates of dementia risk, non-ASCVD death, changes in global cognition scores, and changes in vascular risk factors were estimated using predictive models derived from individual participant data pooled analyses of BP-COG's cardiovascular cohort studies.⁷ Dementia risk was defined from a model based on cognition, change in cognition, race, and education, while cognition was defined in a model that included demographics, education, and vascular risk factors.^6,7 The effect sizes of BP-lowering on CID, global cognition decline, and cardiovascular outcomes were derived from BP-COG because its cardiovascular cohort studies represent the best available US observational evidence for these effect sizes. We aimed to compare the sample sizes and durations of trials to find the smaller effect size of BP lowering on CID observed in observational studies to those needed to detect the larger effect size of BP lowering on CID in SPRINT MIND, since the true effect is unclear. No explicit effect of BP medications on either BP levels or on ASCVD events was built into the usual care strategy.

BP treatment strategies

We explored the primary BP treatment strategy based on the intensive BP treatment algorithm from SPRINT's intensive arm, with strong safety and efficacy evidence.¹⁹ Simulated patients were randomized to this SPRINT-based BP-lowering strategy versus usual care. In usual care, BP medications were assigned using a regression model based on real-world BP management in BP-COG's pooled cardiovascular cohort studies.

In the SPRINT-based BP-lowering strategy,¹⁹ all adults were treated to a systolic BP (SBP) target of <120 mm Hg. However, SPRINT's BP measurement method averages BP values about 6 mmHg lower than those generated using the BP measurement protocol applied in NHANES.²⁰ So, we assumed SBP measures derived from NHANES and the BP-COG pooled cohort studies were 6 mmHg higher than they would have been in SPRINT; thus, our effective SBP treatment target was 126 mmHg.^19,20 For each individual, BP medications were added annually, representing the average change in clinical care over a year implemented at the year's beginning. This protocol could add up to the SPRINT protocol's maximum of 4 BP medications, assuming an average BP-lowering effect of 5.5/3.1 mmHg per medication.²¹ Like SPRINT, we capped treatment at 4 medications because the benefits of adding a fifth medication are uncertain,²² and guidelines recommend evaluating for secondary causes of hypertension at this threshold.²³ As these medications were added on top of the usual care strategy, strategies were designed to capture the marginal effect of changing BP treatment strategies compared to current US BP treatment.

Simulated trials

Using the MICROSIM structure, we executed many simulated trials to identify trial parameters that would have sufficient power to detect a clinically significant effect of a SPRINT-based BP-lowering strategy (versus usual care) on CID and global cognition. We included populations of various sizes with experimental subjects randomly selected from risk strata, randomized them to either a SPRINT-based BP-lowering strategy or usual care, and ran the trial for specific durations. We simulated trials with different combinations of parameters: sample size, duration, baseline ASCVD risk, baseline dementia risk (based on a population-level model),⁵ primary outcome (CID versus global cognition), and analysis type. We varied trial sample sizes from 500 to 50,000 and trial durations from 3 to 20 years. We calculated ASCVD and dementia risks for all individuals in the baseline MICROSIM population eligible for these trials using published models.⁶ For each trial, the initialized MICROSIM population randomized 50% of the subjects to usual care and 50% to the SPRINT-based BP treatment algorithm. We executed at least 100 trials for each unique set of trial parameters because exploratory simulations showed that approximately 50–75 trial runs typically produced stable mean effects of BP-lowering on the study's outcomes.

Outcomes

The first trial outcome was CID. Dementia risks were assigned using pooled cohort-derived Cox proportional hazard models of incident dementia based on demographics, baseline global cognition score, and global cognition trajectories, and data from 3 BP-COG pooled cohort studies (ARIC, CHS, and FOS).⁶ These cohort studies measured incident dementia by physician adjudication using standard diagnostic criteria, study-specific protocols, and all available data, including in-person neuropsychological and neurologic assessments, telephone interviews (participant or informant), brain imaging, and medical record review. Global cognition was estimated using a published linear mixed-effects model and data from the 5 BP-COG pooled cohort studies.⁷ We published the methods for harmonizing global cognition in the BP-COG project.¹³ Global cognition is scaled as a t-score, with a mean [SD] of 50 [10] at the first cognitive assessment.^7,13 Higher cognitive scores indicate better performance.

We defined cognitive impairment as a decline of more than 0.5 standard deviations from the baseline global cognition score, as some BP-COG cohort studies did not measure mild cognitive impairment (MCI).^24–26 The criterion of a decline of 0.5 or more standard deviations from a baseline cognitive score is used in clinical trials, is considered a clinically meaningful decline by experts, and has been validated against a clinically meaningful decline in global cognition in a cohort of cognitively normal adults aged ≥50 and established cognitive decline measures in older adults with dementia.^24–26 We also ran simulated trials with other outcomes: mean global cognition throughout the trial, final global cognition at the end of the trial, and a composite ASCVD outcome of stroke or MI. Stroke and MI risks were assigned using pooled cohort-derived Cox proportional hazard models of incident stroke and MI based on demographics, baseline vascular risk factor levels, vascular risk factor trajectories, and data from the 5 BP-COG pooled cohort studies.⁶ These cohort studies measured incident stroke and MI by physician adjudication using standard diagnostic criteria, study-specific protocols, and all available data.

Analysis

For binary outcomes like dementia incidence, logistic regression and Cox proportional hazards models were used. The Cox models specifically accounted for time-to-event data, with censoring occurring upon participants’ deaths. Continuous outcomes like global cognition were analyzed using linear regression. We calculated the power to detect an effect of a SPRINT-based BP-lowering strategy on the outcomes for each set of trial parameter combinations by calculating the proportion of positive trials (alpha=0.05) given the true effect specified by the underlying MICROSIM structure. We summarized how power varied for each outcome across sample sizes and trial durations using heat maps. Among trial parameter combinations with ≥80% power, we calculated absolute effect sizes for each trial parameter combination and plotted histograms.

To compare the results between our simulation study and SPRINT MIND, we performed power calculations exploring how assumptions differed between the two studies and influenced power. Specifically, we focused on CID risk separate from ASCVD risk. We assigned the risk of CID in a usual care group, assuming the approximate rate of CID in the standard BP treatment arm (targeting SBP of <140 mmHg) of SPRINT MIND (2%/year) and an average follow-up duration of 5 years, resulting in a control event rate of 10%. We then estimated the CID rate in an intensive BP treatment arm by applying a series of relative risks (RRs) from 0.99 to 0.80. The RR of CID in SPRINT MIND was 0.85. We varied the sample size from 500 to 50,000 subjects. We calculated the power for a z-test for each set of parameters, testing whether the two independent proportions were equal using the Python statsmodel.power package.²⁷ Statistical power across the range of sample sizes and RRs of the CID outcome was summarized using contour plots.

Results

We ran over 139 million simulated trials across 78,000 unique parameter combinations of outcomes, analysis type, sample size, duration, and baseline ASCVD and dementia risks. The 10-year ASCVD risk ranged from <1% in the lowest decile of trials to 16% in the highest decile. Conversely, 10-year dementia risk ranged from <1% in the lowest decile to 5% in the highest decile. Assuming an effect size of the SPRINT-based BP-lowering strategy on CID equivalent to that found in our individual participant data pooled analysis of BP-COG observational cohort studies, the simulated trials of the SPRINT-based BP-lowering strategy, compared to usual care, averaged an RR reduction of 2% (RR, 0.98) for the CID outcome, much smaller than the 25% (RR, 0.75) for the stroke-MI outcome.

We explored 6,000 unique trial parameter combinations (sample size, duration, and baseline ASCVD and dementia risks) to determine the sample sizes and durations of trials adequately powered to find the 2% RR reduction in CID from BP-lowering with the SPRINT-based BP-lowering strategy suggested by our individual participant data pooled analysis of observational studies (Figure 1). Adding longer follow-up to a trial beyond 5–10 years is inefficient in gaining more statistical power than adding more sample size. Power increases substantially by extending trial follow-up from 5 to 10 years. Modest power is gained from extending trial follow-up to 15 years. No power is gained by extending trial follow-up to 20 years. Thus, increasing the sample size and not extending the follow-up beyond 10 years is probably the most efficient design.

Figure 1.

Power to identify any effect of a Systolic Blood PRessure INtervention Trial (SPRINT)-based blood pressure-lowering strategy on cognitive impairment and dementia across all trials. This heat map summarizes the mean power (cell value) across all studied sampled sizes and durations among the clinical trial parameters to detect an effect of the SPRINT-based BP-lowering strategy, compared to usual care, on the outcome of cognitive impairment and dementia.

A trial detecting the relative risk reduction (2%) in CID from the SPRINT-based BP-lowering strategy with >80% power reliably would require trials with samples larger than 50,000 population and/or durations greater than 20 years (Figure 1). Only trials with 50,000 patients and at least 10 years of follow-up averaged greater than 70% power, and even a trial of 50,000 patients for 20 years did not average 80% power. When summarizing mean power across all CID analyses by sample size and duration, no combination of sample size and duration resulted in more than 80% power (Figure 1). Mean statistical power ranged from 5% to 70%, with most trials having <55% power given their sample sizes and durations (Figure 1).

The optimal trial to detect the 2% RR reduction in CID with very close to ∼80% power has approximately 60,000 patients with a 10-year follow-up. If the true RR reduction and absolute risk reduction in CID of the SPRINT-based BP-lowering strategy are closer to those observed in SPRINT MIND, a trial of approximately 10,000 patients with a 5-year follow-up would have ∼80% power to detect this effect size. The BP-lowering effects of a SPRINT-based BP-lowering strategy were generally small—with a mean effect of 0.1% per year reduction in absolute CID risk among all positive trial combinations (number needed to treat of 1,000) and a maximum effect of 0.26% per year reduction in CID (number needed to treat of ∼400/year).

Using the change in global cognition as the primary endpoint yields significantly more power than using an endpoint of CID. The statistical power to detect the effects of BP-lowering with a SPRINT-based BP-lowering strategy on the mean global cognition outcome (continuous score) was considerably greater than that for the CID outcome with >80% power for trials with large sample size (e.g., ≥30,000 persons) and long durations (e.g., ≥10 years) (Figure 2). Approximately 20% of parameter combinations (n = 1,222) yielded power >80% of 6,000 total combinations testing global cognition. Mean statistical power ranged from 6% to 93%, with more trials having >80% power to detect a significant effect of the SPRINT-based BP-lowering strategy on the mean global cognition score than for the CID outcome (Figure 2).

Figure 2.

Power to identify any effect of a Systolic Blood PRessure INtervention Trial (SPRINT)-based blood pressure-lowering strategy on mean global cognition across all trials. This heat map summarizes the mean power (cell value) across all studied sample sizes and durations among the clinical trial parameters to detect a SPRINT-based blood pressure-lowering strategy's effect on mean global cognition scores compared to usual care.

However, the effect sizes in the trials with 80% or greater power were small. Among trial parameter combinations with >80% power to detect any effect of BP-lowering on global cognition, the mean difference in global cognition scores at trial end between patients treated with a SPRINT-based BP-lowering strategy compared to usual care was 0.02 points, and the maximum difference was 0.08 points. The mean global cognition score in the simulation population is ∼50, and the standard deviation is ∼14. Thus, even the trial parameter combinations with the largest observed effects of BP-lowering on global cognition score (0.08 points over trials with durations of 5 or more years) are unlikely to identify clinically significant effects.

To explore how our findings of limited power could be reconciled with the SPRINT MIND trial identifying a significant effect of BP-lowering on CID, we next examined sample sizes needed to detect different RR reductions of CID, assuming baseline ASCVD and CID risk rates similar to those in the SPRINT MIND trial^19,28 and a trial duration of 5 years (Figure 3). Assuming a sample size of 9,500 and an RR of CID of 0.85, similar to the SPRINT MIND trial, we projected approximately 67% power to identify a true effect of BP-lowering with the SPRINT-based BP-lowering strategy on CID. Conversely, with the same baseline ASCVD and dementia risks and sample size based on the SPRINT MIND trial, but an RR of CID of 0.98 (approximating the RR reduction of the SPRINT-based BP-lowering strategy on CID estimated by MICROSIM and the individual participant data pooled analysis of the BP-COG cohort studies),⁷ power would decrease to 7% and only increase to 19% at a sample size of 50,000.

Figure 3.

Power to identify a significant difference between the Systolic Blood PRessure INtervention Trial (SPRINT)-based blood pressure-lowering strategy and usual care on cognitive impairment and dementia by relative risk levels and sample size. The contour plot illustrates how estimated power varies across trials with the untreated risk profile of the SPRINT MIND trial participants while varying relative risk (RR) levels of the blood pressure (BP)-lowering strategy's effect on cognitive impairment and dementia (CID) and sample size. The SPRINT MIND trial observed that the SPRINT-based BP-lowering strategy reduced the risk of CID by 15% (RR, 0.85). MICROSIM observed that the SPRINT-based BP-lowering strategy reduced the risk of CID by 2% (RR, 0.98).

Discussion

MICROSIM found that assuming the effect size of BP lowering on CID estimated from our individual participant data pooled analysis of observational studies, the SPRINT-based BP-lowering strategy, compared to usual care, had an average RR reduction of 2% (RR, 0.98) for CID, much smaller than the effect size found in SPRINT MIND (RR reduction of 15%). Increasing the sample size rather than extending follow-up beyond 10 years is the most efficient trial design because a 15-year follow-up increases power modestly, and a longer follow-up offers little gains in power. A trial large enough to find the 2% RR reduction in CID from the SPRINT-based BP-lowering strategy suggested by our individual participant data pooled analysis of observational studies would be very large and expensive with sample sizes of 60,000 patients with a 10-year follow-up. Using the change in global cognition as the primary endpoint yields significantly more power than using a CID endpoint. It is likely not important to power for the RR reduction in CID from the SPRINT-based BP-lowering strategy suggested by our individual participant data pooled analysis of observational studies because the absolute risk reduction would not be highly clinically significant. The BP-lowering effects of a SPRINT-based BP-lowering strategy suggested by our individual participant data pooled analysis of observational studies were small—with a mean effect of 0.1% per year reduction in absolute CID risk among all positive trial combinations (number needed to treat of 1,000) and a maximum effect of 0.26% per year reduction in CID (number needed to treat of ∼400/year). If the true relative and absolute risk reductions in CID from the SPRINT-based BP-lowering strategy are closer to those observed in SPRINT MIND, a trial of reasonable size and duration would have good power (sample size of ∼10,000 patients with a 5-year follow-up).

Simulated trials with a continuous outcome (global cognition) had greater statistical power than those with a binary outcome (incident CID). We were surprised that most simulated trials with the CID and global cognition outcomes had insufficient statistical power to detect a significant effect of BP-lowering when the individual participant data pooled cohort-estimated effects of BP-lowering on CID and global cognition from the BP-COG observational cardiovascular cohort studies are assumed to be causal and accurate. Even a 10-year trial with 50,000 participants with 100% follow-up is unlikely to identify these effects. Our simulation study reproduced the effects of the SPRINT-based BP-lowering strategy on the CVD outcome of stroke or MI found in SPRINT,^8,19 but did not reproduce the effects of the SPRINT-based BP-lowering strategy on CID found in SPRINT MIND.⁴

Comparing SPRINT MIND with MICROSIM and BP-COG observational cardiovascular cohort studies

One potential explanation for the differences in effect sizes between SPRINT MIND and the BP-COG observational cardiovascular cohort studies in MICROSIM is that the effect of BP-lowering on CID measured in SPRINT MIND or MICROSIM is incorrect. In MICROSIM, we observed that the SPRINT-based BP-lowering strategy was associated with an RR of CID of 0.98, which is an effect size at the upper limit of the 95% confidence interval for the effect of BP-lowering with antihypertensive agents versus control on CID from a recent meta-analysis of 12 randomized controlled trials including SPRINT MIND (odds ratio [OR], 0.93 [95% CI, 0.88–0.98]).²⁹ In contrast, the SPRINT-based BP-lowering strategy was associated with an RR of CID of 0.85 in SPRINT MIND,⁴ an effect size beyond the lower limit of the 95% confidence interval of the BP-lowering effect in the meta-analysis.²⁹ Assuming SPRINT MIND's baseline ASCVD and CID risks, sample size, and effect size of the SPRINT-based BP-lowering strategy on CID (RR, 0.85),⁴ a trial would have 67% power to identify a true effect of BP-lowering with the SPRINT-based BP-lowering strategy on CID, much less than the conventional power of 80%. Power would be 7% and only increase to 19% at a sample size of 50,000, assuming the smaller effect size (RR, 0.98) estimated by MICROSIM. The recent meta-analysis suggests that the effect size for BP lowering on CID ranges from a relative risk reduction of 2% to 12% ranges²⁹; whereas, a different meta-analysis of 5 randomized controlled trials suggested that the effect size for BP lowering on dementia alone ranges from a relative risk reduction of 1% to 25%.³⁰

Another distinction is that SPRINT MIND added more than the 1.1 BP medications per person that MICROSIM added, which used data from BP-COG observational cardiovascular cohort studies, likely reflecting a population with higher BP than the potential eligible population in the US. This difference in BP treatment intensity between SPRINT MIND and MICROSIM likely explains some of the effect size difference. The magnitude of the BP-lowering effect likely matters. For example, an intensive four-year BP-lowering intervention (target BP < 130/80) that reduced BP by an average of 22/9 mmHg compared to usual care (mean BP at 48 months, 128/73 mmHg versus 148/81 mmHg) was associated with a 15% lower risk of all-cause dementia in people with hypertension in rural China.³¹ This difference in BP medications added may also reflect subtle differences in population ages—SPRINT MIND had a minimum age of 50 for enrollment. In contrast, our simulations entered all adults into some trials. However, by stratifying trial populations based on ASCVD and dementia risks, the highest risk ASCVD and dementia groups disproportionately consisted of older patients with elevated BPs because ASCVD and dementia risks exponentially increase with age.¹¹ Given that the absolute effect size of BP-lowering was small even in the highest ASCVD and dementia risk groups, effect sizes would be similarly small if trials included older adults exclusively.

Another possible explanation is that SPRINT MIND and the BP-COG observational cardiovascular cohort studies in MICROSIM used different outcome measures. Both studies measured probable dementia by expert adjudication using diagnostic criteria, study-specific protocols, all available neuropsychological testing, informant and participant interviews, and medical record review. SPRINT MIND measured expert-adjudicated MCI, while our simulation study estimated cognitive impairment using an absolute cognitive score and regression approach^25,26 because some cohort studies did not measure MCI.^32,33 Some global cognition observations in the cohort studies were based on briefer cognitive screening assessments than those used in SPRINT MIND. Also, SPRINT MIND's expert adjudicators had functional assessment data, but the cohort studies’ expert adjudicators did not. Interestingly, SPRINT MIND did not detect a significant effect of intensive BP lowering on global cognitive function in a subset of participants,⁴ which contrasts with our BP-COG findings.⁷

Two additional explanations for the differences in the sizes of the effect of BP lowering on CID between SPRINT MIND and the BP-COG observational cardiovascular cohort studies in MICROSIM are selection/confounding and measurement error. In the observational studies, patients with hypertension who are treated more intensively may have higher true BPs and/or higher ASCVD risk than those who are not treated as intensively. Regarding measurement error, SPRINT MIND likely reduced BP measurement error. So, many SPRINT MIND participants with a “5-mmHg reduction in BP” probably had a true 5-mmHg reduction in BP. Whereas, in the BP-COG observational cardiovascular cohort studies, there is more regression to the mean among participants with hypertension. As a result, participants with a “5-mmHg reduction in BP” may only have had a true 3-mmHg reduction in BP.

Strengths

We used a simulation model that includes both dementia and ASCVD events and applies to the US population of adults. A prior simulation model that included dementia and ASCVD events focused on the UK population with diabetes.³⁴ While many simulation models include ASCVD events, our simulation model uniquely quantifies the potential effects of BP treatment strategies on CID, cognition, and ASCVD events in the US and trial populations. We used the best observational data from individual participant data pooled analyses of well-characterized US cardiovascular cohort studies with longitudinal cognitive outcomes to estimate effect sizes when trial evidence was unavailable. The modeling design allowed the absolute effect of BP treatment on CID to increase with age, as BP treatment was modeled with a constant RR reduction, and age was the dominant predictor of incident CID. We stratified trials by dementia and ASCVD risks.

Limitations

The assumptions and input data limit the results of our simulation. Trial evidence is limited in some areas (e.g., the effect of BP treatment on global cognition). We addressed this limitation by estimating effects using individual participant data pooled cohort analyses of US cardiovascular cohort studies. MICROSIM's model was based on robust estimates of age-adjusted dementia incidence in US adults. Although suboptimal precision of dementia incidence could bias results toward the null, incident dementia was estimated using individual participant data pooled analyses of expert-adjudicated incident dementia from US cohort studies. While the model might overestimate dementia incidence, this overestimation should not affect relative BP-dementia relationships. Our approach of using a mean BP correction to account for BP measurement differences between the SPRINT trial and NHANES/BP-COG is reasonable, but might not account for how BP measurements can widely vary within a person.²⁰ Many of the details of trial execution were not explicitly modeled in this simulation. So, for example, real-world power may be somewhat overestimated in these simulations as they do not account for potential recruitment delays, particularly for large theoretical trials. We modeled a constant RR for BP-lowering on CID across BP groups, but heterogeneity of treatment effect may exist, although it is difficult to evaluate with existing data. The simulation model does not account for anti-hypertensive treatment discontinuity or nonadherence. One cluster reported in our Voxel-Based Morphometry (VBM) analysis survived false discovery rate (FDR) correction but did not survive family-wise error (FWE) correction (cluster-level FWE p = 0.10), indicating a non-negligible chance of a false positive (Type I error) for that cluster. Using observational data to estimate causal effects has limitations, including residual confounding.

Implications

In this microsimulation study, we found that trials would need to be large with long follow-ups to identify the causal effects of BP-lowering on CID and cognition. However, assuming the 15% CID RRR in SPRINT MIND, trials with more feasible sizes and durations could detect those effects. Additional trials of identifying the causal effects of BP-lowering on CID are needed. Trials testing BP-lowering at younger ages to prevent later-life CID are of considerable interest and hold promise for population health.

Conclusion

In this microsimulation study, assuming the best observational estimates of BP-lowering on CID are causal, trials would be unlikely to identify these effects, and if they did, the effects are unlikely to be clinically significant. However, given that the observed effect in the SPRINT MIND trial is much larger than can be explained by our assumptions, experimental estimates of the effect of BP-lowering on CID might be greater than those from observational data.

Footnotes

Acknowledgements

The Atherosclerosis Risk in Communities (ARIC) Study is carried out as a collaborative study supported by National Heart, Lung, and Blood Institute contracts (75N92022D00001, 75N92022D00002, 75N92022D00003, 75N92022D00004, 75N92022D00005). The ARIC Neurocognitive Study is supported by U01HL096812, U01HL096814, U01HL096899, U01HL096902, and U01HL096917 from the NIH (NHLBI, NINDS, NIA, and NIDCD). The authors thank the staff and participants of the ARIC study for their important contributions.

The Coronary Artery Risk Development in Young Adults Study (CARDIA) is conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with the University of Alabama at Birmingham (75N92023D00002 & 75N92023D00005), Northwestern University (75N92023D00004), University of Minnesota (75N92023D00006), and Kaiser Foundation Research Institute (75N92023D00003).  This manuscript has been reviewed by CARDIA for scientific content.

The Cardiovascular Health Study (CHS) was supported by contracts HHSN268201200036C, HHSN268200800007C, HHSN268201800001C, N01HC55222, N01HC85079, N01HC85080, N01HC85081, N01HC85082, N01HC85083, N01HC85086, 75N92021D00006, and grants U01HL080295, U01HL130114 and R01HL172803 from the National Heart, Lung, and Blood Institute (NHLBI), with additional contribution from the National Institute of Neurological Disorders and Stroke (NINDS). Additional support was provided by R01AG023629 from the National Institute of Aging (NIA). A full list of principal CHS investigators and institutions can be found at CHS-NHLBI.org. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

The Framingham Heart Study (FHS) is a project of the National Heart, Lung, and Blood Institute (NHLBI) of the National Institutes of Health and Boston University School of Medicine. This project has been funded in whole or in part with Federal funds from the National Heart, Lung, and Blood Institute, National Institutes of Health (NHLBI), Department of Health and Human Services, under Contract No. 75N92019D00031.

The Northern Manhattan Study (NOMAS) has been funded at least in part with federal funds from the National Institutes of Health, National Institute of Neurological Disorders and Stroke by R01-NS29993.

ORCID iDs

Deborah A. Levine

Mitchell S. V. Elkind

Ethical considerations

The University of Michigan institutional review board approved the study. Participating institutions’ review boards approved the cohort studies.

Consent to participate

Participants provided written informed consent before inclusion in the cohort studies.

Consent for publication

Not applicable

Author contribution(s)

Deborah A. Levine: Conceptualization; Data curation; Funding acquisition; Investigation; Methodology; Project administration; Writing – original draft.

Rodney A. Hayward: Conceptualization; Investigation; Methodology; Supervision; Writing – review & editing.

Jeremy B. Sussman: Conceptualization; Investigation; Methodology; Writing – review & editing.

Rachael T. Whitney: Data curation; Investigation; Writing – review & editing.

Andrzej T. Gałecki: Methodology, Writing - review & editing.

Emily M. Briceño: Investigation, Writing - review & editing.

Alden L. Gross: Data curation; Investigation; Writing – review & editing.

Bruno J. Giordani: Investigation; Writing – review & editing.

Mitchell S. V. Elkind: Investigation; Writing – review & editing.

Darrell J. Gaskin: Investigation; Writing – review & editing.

Rebecca F. Gottesman: Investigation; Writing – review & editing.

Stephen Sidney: Investigation; Writing – review & editing.

Kristine Yaffe: Investigation; Writing – review & editing.

James F. Burke: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Supervision; Writing – review & editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research project was supported by National Institutes of Health (NIH) / National Institute of Neurological Disorders and Stroke (NINDS) R01 NS102715 (Levine, PI). The NINDS was not involved in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and the decision to submit the manuscript for publication of the funding agency reviewed the manuscript. The content is solely the authors’ responsibility and does not necessarily represent the official views of the NINDS or the NIH. Additional funding was provided by National Institute on Aging (NIA) grant RF1 AG068410 (Levine), NIA Claude Pepper Center grant P30 AG024824 (Galecki), NIA grant K01 AG050699 (Gross), NIA grant K23AG080035 (Briceño), a Methods Core supported by grant P30 DK020572 from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (Hayward), the NIA Michigan Alzheimer's Disease Research Center grant P30 AG072931 (Giordani), and the NINDS Intramural Research Program (Gottesman).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

Dr. James Burke had full access to all the study data and is responsible for its integrity and accuracy. The simulation model's code and input parameters are publicly available (). Data is available from the individual cohort studies based on their approval.

References

GBD 2019 Dementia Forecasting Collaborators. Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050: an analysis for the Global Burden of Disease Study 2019. Lancet Public Health 2022; 7: e105–e125.

Bai

Chen

Cai

, et al. Worldwide prevalence of mild cognitive impairment among community dwellers aged 50 years and older: a meta-analysis and systematic review of epidemiology studies. Age Ageing 2022; 51: afac173.

Livingston

Huntley

Sommerlad

, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. Lancet 2020; 396: 413–446.

Williamson

Pajewski

Auchus

, et al. Effect of intensive vs. standard blood pressure control on probable dementia: a randomized clinical trial. JAMA 2019; 321: 553–561.

Ghazi

Shen

Ying

, et al. Identifying patients for intensive blood pressure treatment based on cognitive benefit: a secondary analysis of the SPRINT randomized clinical trial. JAMA Netw Open 2023; 6: e2314443.

Burke

Copeland

Sussman

, et al. Development and validation of the Michigan ChROnic Disease SIMulation model (MICROSIM). PLoS One 2024; 19: e0300005.

Levine

Gross

Briceño

, et al. Association between blood pressure and later-life cognition among Black and White individuals. JAMA Neurol 2020; 77: 810–819.

Burke

Sussman

Yaffe

, et al. Effect of population-level blood pressure treatment strategies on cardiovascular and cognitive outcomes. Circ Cardiovasc Qual Outcomes 2024; 17: e010288.

Levine

Sussman

Hayward

, et al. The potential impact of optimal blood pressure treatment intensity to reduce disparities in dementia between Black and White individuals. J Alzheimers Dis 2025; 103: 506–518.

10.

Cheng

Kessler

Mackinnon

, et al. Reporting guidelines for health care simulation research: extensions to the CONSORT and STROBE statements. Simul Healthc 2016; 11: 238–248.

11.

Marma

Berry

Ning

, et al. Distribution of 10-year and lifetime predicted risks for cardiovascular disease in US adults: findings from the National Health and Nutrition Examination Survey 2003 to 2006. Circ Cardiovasc Qual Outcomes 2010; 3: 8–14.

12.

Arnold

Hogan

Colford Jr

, et al. Simulation methods to estimate design power: an overview for applied research. BMC Med Res Methodol 2011; 11: 94.

13.

Briceño

Gross

Giordani

, et al. Pre-Statistical Considerations for Harmonization of Cognitive Instruments: harmonization of ARIC, CARDIA, CHS, FHS, MESA, and NOMAS. J Alzheimers Dis 2021; 83: 1803–1813.

14.

The ARIC Investigators. The Atherosclerosis Risk in Communities (ARIC) Study: design and objectives. Am J Epidemiol 1989; 129: 687–702.

15.

Friedman

Cutter

Donahue

, et al. CARDIA: study design, recruitment, and some characteristics of the examined subjects. J Clin Epidemiol 1988; 41: 1105–1116.

16.

Fried

Borhani

Enright

, et al. The Cardiovascular Health Study: design and rationale. Ann Epidemiol 1991; 1: 263–276.

17.

Feinleib

Kannel

Garrison

, et al. The Framingham Offspring Study: design and preliminary data. Prev Med 1975; 4: 518–525.

18.

Sacco

Boden-Albala

Gan

, et al. Stroke incidence among White, Black, and Hispanic residents of an urban community: the Northern Manhattan Stroke Study. Am J Epidemiol 1998; 147: 259–268.

19.

Wright Jr

Williamson

Whelton

, et al. A randomized trial of intensive versus standard blood-pressure control. N Engl J Med 2015; 373: 2103–2016.

20.

Drawz

Agarwal

Dwyer

, et al. Concordance between blood pressure in the systolic blood pressure intervention trial and in routine clinical practice. JAMA Intern Med 2020; 180: 1655–1663.

21.

Turnbull

. Blood Pressure Lowering Treatment Trialists’ Collaboration. Effects of different blood-pressure-lowering regimens on major cardiovascular events: results of prospectively-designed overviews of randomised trials. Lancet 2003; 362: 1527–1535.

22.

Basu

Sussman

Hayward

. Detecting heterogeneous treatment effects to guide personalized blood pressure treatment: a modeling study of randomized clinical trials. Ann Intern Med 2017; 166: 354–360.

23.

Carey

Calhoun

Bakris

, et al. Resistant hypertension: detection, evaluation, and management: a scientific statement from the American Heart Association. Hypertension 2018; 72: e53–e90.

24.

Wolinsky

Unverzagt

Smith

, et al. The ACTIVE cognitive training trial and health-related quality of life: protection that lasts for 5 years. J Gerontol A Biol Sci Med Sci 2006; 61: 1324–1329.

25.

Unger

van Belle

Heyman

. Cross-sectional versus longitudinal estimates of cognitive change in nondemented older people: a CERAD study. Consortium to Establish a Registry for Alzheimer’s Disease. J Am Geriatr Soc 1999; 47: 559–563.

26.

Rossetti

Munro Cullum

Hynan

, et al. The CERAD Neuropsychologic Battery Total Score and the progression of Alzheimer disease. Alzheimer Dis Assoc Disord 2010; 24: 138–142.

27.

Seabold

Perktold

. Statsmodels: econometric and statistical modeling with python. SciPy 2010; 7: 92–96.

28.

Sprint Mind Investigators for the SPRINT Research Group Nasrallah

Pajewski

, et al. Association of intensive vs standard blood pressure control with cerebral white matter lesions. JAMA 2019; 322: 524–534.

29.

Hughes

Judge

Murphy

, et al. Association of blood pressure lowering with incident dementia or cognitive impairment: a systematic review and meta-analysis. JAMA 2020; 323: 1934–1944.

30.

Peters

Fitzgerald

, et al. Blood pressure lowering and prevention of dementia: an individual patient data meta-analysis. Eur Heart J 2022; 43: 4980–4990.

31.

Zhao

Zhong

, et al. Effectiveness of blood pressure-lowering intervention on risk of total dementia among patients with hypertension: a cluster-randomized effectiveness trial. In: American Heart Association Scientific Sessions, https://s3-amazonaws-com-443.web.bisu.edu.cn/cms.ipressroom.com/67/files/202310/202311100357/CRHCP+Dementia+abstract+23343.pdf (2023).

32.

Gillett

Thacker

Letter

, et al. Correlates of incident cognitive impairment in the REasons for Geographic And Racial Differences in Stroke (REGARDS) Study. Clin Neuropsychol 2015; 29: 466–486.

33.

Ivnik

Smith

Petersen

, et al. Diagnostic accuracy of four approaches to interpreting neuropsychological test data. Neuropsychology 2000; 14: 163–177.

34.

Breeze

Thomas

Thokala

, et al. The impact of including costs and outcomes of dementia in a health economic model to evaluate lifestyle interventions to prevent diabetes and cardiovascular disease. Med Decis Making 2020; 40: 912–923.