The association of intensive blood pressure treatment and non-fatal cardiovascular or serious adverse events in older adults with mortality: mediation analysis in SPRINT

Abstract

Aims

Randomized clinical trials of hypertension treatment intensity evaluate the effects on incident major adverse cardiovascular events (MACEs) and serious adverse events (SAEs). Occurrences after a non-fatal index event have not been rigorously evaluated. The aim of this study was to evaluate the association of intensive (<120 mmHg) to standard (<140 mmHg) blood pressure (BP) treatment with mortality mediated through a non-fatal MACE or non-fatal SAE in 9361 participants in the Systolic Blood Pressure Intervention Trial.

Methods and results

Logistic regression and causal mediation modelling to obtain direct and mediated effects of intensive BP treatment. Primary outcome was all-cause mortality (ACM). Secondary outcomes were cardiovascular (CVM) and non-CV mortality (non-CVM). The direct effect of intensive treatment was a lowering of ACM [odds ratio (OR) 0.75, 95% confidence interval (CI): 0.60–0.94]. The MACE-mediated effect substantially attenuated (OR 0.96, 95% CI: 0.92–0.99) ACM, while the SAE-mediated effect was associated with increased (OR 1.03, 95% CI: 1.01–1.05) ACM. Similar patterns were noted for intensive BP treatment on CVM and non-CVM. We also noted that SAE incidence was 3.9-fold higher than MACE incidence (13.7 vs. 3.5%), and there were a total of 365 (3.9%) ACM cases, with non-CVM being 2.6-fold higher than CVM [2.81% (263/9361) vs. 1.09% (102/9361)]. The SAE to MACE and non-CVM to CVM preponderance was found across all age groups, with the ≥80-year age group having the highest differences.

Conclusion

The current analytic techniques demonstrated that intensive BP treatment was associated with an attenuated mortality benefit when it was MACE-mediated and possibly harmful when it was SAE-mediated. Current cardiovascular trial reporting of treatment effects does not allow expansion of the lens to focus on important occurrences after the index event.

Graphical Abstract

Open in new tabDownload slide

Introduction

The primary analysis of most cardiovascular (CV) clinical trials utilizes a time-to-first event endpoint. All occurrences after the incident event are essentially ignored in the main data analysis. However, events may occur more than once, and an event, such as an index non-fatal major adverse cardiovascular event (MACE) or a non-fatal serious adverse event (SAE), may change the risk of mortality and the type of mortality. To address the issue of complete disease burden associated with an event, recurrent-event methodology is considered to be the best suited.¹ Understanding the relationship and mechanisms, after an index non-fatal event (MACE or SAE) to mortality, is best addressed by mediation analysis (how an effect occurs)²—the focus of this study.

The time-to-first event in the original Systolic Blood Pressure Intervention Trial (SPRINT) report, demonstrated that intensive blood pressure (BP) treatment reduced the incidence of non-fatal MACE, predominantly due to a lower incidence of heart failure. It also demonstrated an increased incidence of SAE driven by a higher incidence of hypotension, syncope, electrolyte abnormalities, and acute kidney injury.^3–5 The noteworthy goal of simultaneously decreasing both MACE and SAE could become the basis for the assessment of optimal healthy life expectancy without disease, disability, or injury.

There is a lack of clarity on whether the occurrence of an initial non-fatal event (MACE or SAE) can alter the relationship of intensive BP control to mortality. While the occurrence of a MACE has been associated with later disabilities of physical and cognitive function and death,^6,7 it has been suggested that an SAE in the context of intensive BP treatment is often reversible with dose de-escalation and not associated with lasting consequences.⁸ This, however, has not been fully evaluated.

We therefore conducted this post hoc analysis of SPRINT data to evaluate if the occurrence of an index MACE or an SAE mediates the association of intensive BP treatment on mortality. An enhanced appreciation of this complex relationship would improve the understanding of the totality of effects of intensive BP lowering beyond the effect on incident events.

Methods

SPRINT study design overview

The SPRINT design, eligibility, characteristics, and findings have been previously described.^3,5,9 In brief, SPRINT participants had to be ≥50 years with a systolic BP ≥130 mmHg and either have or be at high risk for CV disease (history of clinical or subclinical CV disease), chronic kidney disease, 10-year Framingham risk ≥15%, or age ≥75 years. Exclusion criteria were Type 2 diabetes mellitus, prior stroke, ejection fraction <35%, symptomatic heart failure within the prior 6 months, diagnosis or treatment of dementia, expected survival <3 years, unintentional weight loss >10% of body weight in the prior 6 months, nursing home residence, and orthostatic hypotension (systolic BP <110 mmHg after 1 min of standing). The institutional review board at each site approved the trial protocol.

Blood pressure measurements in the SPRINT were standardized as the average of three seated BP measurements during an office visit 5 min apart, after 5 min of quiet rest, using an automated measurement system (Model 907; Omron Healthcare).

The current study is a post hoc analysis using the publicly available SPRINT data set, SPRINT_POP research materials, obtained from the National Heart, Lung, and Blood Institute’s (NHLBI) Biologic Specimen and Data Repository Information Coordinating Center. To obtain the data set for the purposes of the current study, a Research Determination Official for Kaiser Permanente Northern California region reviewed the documents and determined that the project did not need a Kaiser Permanente Northern California Institutional Review Board review. Subsequently, an NHLBI Research Materials Distribution Agreement was signed by the principal investigator (A.K.), Kaiser Permanente’s Division of Research Compliance Manager, and an approved NHLBI authorized representative. The current study does not necessarily reflect the opinions or views of the primary SPRINT investigators or the NHLBI.

Exposure variable

The control arm comprised of participants randomized to a systolic BP target of <140 mmHg. The intensive arm included participants randomized to a systolic BP target of <120 mmHg.

Clinical outcomes

The primary outcome was all-cause mortality (ACM; available in the SPRINT data set). The secondary outcomes were CV mortality (CVM; available in the SPRINT data set) and non-CVM (defined as ACM minus CVM).

Mediating events

Clinical research goals are not only to estimate the association between treatment and outcome but also to understand mechanistic principles underlying the effects. Learning the pathways between treatment and outcome and clarifying possible causes of the outcome are the principal purviews of the investigation of mediation (Figure 1). We considered two types of events as potential mediators of the causal pathway between exposure and outcome. The first mediator variable was a composite of four non-fatal CV (MACE) events that included myocardial infarction, other acute coronary syndromes, stroke, or heart failure. The second mediator variable was the composite of six non-fatal adverse events (hypotension, syncope, arrhythmia, injurious falls, electrolyte abnormalities, or acute kidney insufficiency) which was sufficiently serious to be classified as an SAE or led to an emergency department visit. An SAE was defined as a clinically significant event that resulted either in hospital admission, hospital stay prolongation, surgical intervention, or persistent disability.

All of the individual SAE components were collected as part of the SPRINT (available at https://www.sprinttrial.org/public/Protocol_Current.pdf). Hypotension was defined as BP <110 mmHg and may be asymptomatic or accompanied by dizziness, lightheadedness, feeling faint, syncope, or other symptoms. Syncope was defined as the temporary loss of consciousness, also known as fainting or passing out. An arrhythmia was defined as an abnormality in the heart rhythm and can be slow (bradyarrhythmia) or fast (tachyarrhythmia). An injurious fall was defined as ‘a sudden, unintentional change in position in which the participant comes to rest on the ground, floor, or a lower level, not as the result of syncope or overwhelming external force’. An electrolyte abnormality was defined as serum sodium ≤132 or >150 mEq/L, serum potassium <3.0 or >5.5 mEq/L, increase in serum creatinine by at least 50% to a value ≥1.5 mg/dL since the last study lab usually 6 months apart. Acute kidney injury was coded if the diagnosis was listed in the hospital discharge summary and was believed by the SPRINT safety officer to be one of the top three reasons for admission or continued hospitalization.

Other covariates

The baseline variables available for the current study were age, sex, body mass index, systolic BP, diastolic BP, the presence of prior clinical CV disease, laboratory values (serum creatinine, estimated glomerular filtration rate), and medication use (statin, aspirin, number of antihypertensive medications). These covariates were available in the SPRINT data set and were felt to be appropriate to address the mediation–outcome relationship.

Statistical analysis

Clinical characteristics of participants and crude outcomes were compared by event categories (no MACE or SAE, MACE, SAE, or both MACE and SAE). We obtained non-adjusted incident rates (IRs)/1000 person-years. We used the paramed command in STATA Version 17 software (StataCorp, College Station, TX, USA) to perform causal mediation analyses using parametric regression models. We created one model for the mediator (MACE or SAE) conditional on treatment (randomized BP intensity) and a second model for the outcome(s) conditional on treatment, mediator, and covariates, allowing for the presence of treatment–mediator interactions (if present) in the outcome regression model using counterfactual definitions.^10,11 The specific assumptions made were the following: no unmeasured exposure–outcome confounding, no unmeasured exposure–mediator confounding, no unmeasured mediator–outcome confounding, temporality (treatment intensity occurring before mediating events, and mediating events occurring before mortality), no selection bias, and no measurement error.^2,11

We obtained the natural direct effect (NDE), mediated (natural indirect) effect, and marginal total effects (MTEs) for the association of exposure (intensive BP treatment compared with standard BP treatment) to the primary and secondary outcomes. The NDE expresses the change in outcome with intensive BP treatment assignment in the absence of the mediator (the effect if exposure–mediator pathway disabled; Figure 1, Pathway 3). The mediated effect expresses change in outcome with intensive BP treatment assignment in the presence of the mediator (Figure 1, Pathway 4). The MTE is defined as the extent of outcome change with intensive BP treatment.² We used logistic regression modelling for analysis of causal and non-causal pathways. We also report the controlled direct effect (CDE) that takes into account the effect across a population, if a mediator were fixed at a given level, which has been recommended for use in population-based health policy evaluation.²

Results

Overview

All 9361 SPRINT participants were used for the current analysis. Those who had either a MACE (70.6 ± 9.1 years) or an SAE (71.2 ± 9.8 years) were older than those who did not (67.1 ± 9.1 years) but younger than those who had both (74.1 ± 9.9 years). The highest proportion of women was seen in those who had an SAE (40.6%), while those who had only a MACE had the highest proportion of current smokers (16.9%). The highest systolic BP, lowest diastolic BP, the greatest proportion of clinical CV disease, highest serum creatinine, lowest glomerular filtration rate, and highest number of baseline medications were seen in those who had both events (MACE + SAE). There was slight heterogeneity among the groups with respect to laboratory values of total cholesterol, glucose, HDL-cholesterol, and triglycerides (Table 1).

Major adverse cardiovascular event, serious adverse event, and mortality types

Over an average trial duration of 3.2 years, 17.1% (1605/9361) experienced either a MACE or an SAE. The proportion of those who experienced an SAE was 13.7% (1280/9361) and was 3.9-fold higher than those who experienced a MACE [3.5% (325/9361)]. We found that <1% (61/9361) of participants had ≥2 MACE and 4.2% (393/9361) had ≥2 SAEs, while 2.0% (190/9361) experienced both a MACE and an SAE. The IR of MACE and SAE is shown in Figure 2A.

The total mortality was 3.9% (365/9361) with a 2.6-fold higher incidence of non-CVM [2.81% (263/9361)] compared with CVM [1.09% (102/9361)]. Among those who did not die (N = 8996), 81.9% did not have an event (MACE or SAE), 3.1% had a MACE, 13.4% had an SAE, and 1.6% had both a MACE and SAE. Among those who died (N = 365), 55.1% did not having a MACE or an SAE, 13.4% had a MACE, 19.4% had an SAE, and 12.0% had both a MACE and an SAE. The IR of mortality types are shown in Figure 2B.

In Table 2, we demonstrate the IR of mortality types as a function of MACE and SAE. All-cause mortality was highest in those with both MACE and SAE, driven predominantly by MACE events. Cardiovascular mortality was almost completely driven, as expected, by MACE. However, non-CV mortality was balanced between MACE and SAE but higher when both were present.

Major adverse cardiovascular event, serious adverse event, and mortality types stratified by chronological age

The IR of MACE, SAE, and mortality types as a function of age is shown in Figure 3. In Figure 3A, we show the rising incidence of events (MACE and SAE) with increasing age groups. Of note is the SAE preponderance compared with MACE across all age groups, with the highest difference between SAE and MACE in the ≥80-year-old group. Similarly, we show an increase in ACM with higher age groups with the ratio between non-CV mortality and CV mortality the highest in those ≥70 years old. Supplementary material online, Figures S1–S3 display the crude IR between intensive and standard BP therapy with MACE, SAE, or no MACE/SAE on each of the mortality types stratified by age.

Pathways 1, 2, and 3

We demonstrate that intensive BP treatment significantly lowered the incidence of MACE [odds ratio (OR) 0.80, 95% confidence interval (CI): 0.67–0.97], but significantly increased the incidence of SAE (OR 1.28, 95% CI: 1.14–1.43)—Pathway 1 of Figure 1. For Pathway 2, MACE significantly increased the odds of all mortality types (OR 4.98, 95% CI: 3.80–6.54) with the highest for CV mortality (OR 16.3, 95% CI: 10.6–25.1). An SAE occurrence significantly increased ACM and non-CV mortality (OR 2.14, 95% CI: 1.63–2.82) but not CV mortality (OR 1.19, 95% CI: 0.74–1.93). For Pathway 3, intensive BP treatment lowered ACM and CV mortality but did not lower non-CV mortality (OR 0.81, 95% CI: 0.63–1.04; Table 3).

Causal mediation modelling methods—Pathway 4

Using causal mediation modelling methods, we determined the effect size of intensive BP treatment (compared with standard therapy) on mortality types (all-cause, CV, and non-CV) in the presence of a mediator (MACE or SAE). This is displayed in Figure 1 as Pathway 4. We found that any mortality benefits noted by the natural direct were substantially attenuated or reversed when the mediated effect was taken into account (Table 4).

After a MACE, the mediated effect sizes of intensive BP treatment on mortality types were significantly attenuated (OR 0.96, 95% CI: 0.92–0.99) for ACM, significantly attenuated (OR 0.87, 95% CI: 0.78–0.97) for CV mortality, and was non-significant for non-CV mortality (OR 0.99, 95% CI: 0.97–1.002). After an SAE, the mediated effect size of intensive BP treatment on mortality types was significantly increased (OR 1.03, 95% CI: 1.01–1.05) for ACM, non-significant (OR 1.002, 95% CI: 0.98–1.03) for CV mortality, significantly increased (OR 1.03, 95% CI: 1.01–1.06) for non-CV mortality (Table 4).

The mediated effects were substantiatively similar even after stratifying at the age of 70 years old (Table 5). The CDEs, which aid in understanding the effect of the mediator across an estimated population, are shown in Table 6. It demonstrated that when mediated through a MACE or an SAE, intensive BP treatment did not decrease all-cause, CV, or non-CV mortality.

Discussion

This novel investigation, utilizing the publicly available SPRINT data set, sought to assess the effect of an incident non-fatal MACE or SAE on the relationship between BP treatment intensity and mortality outcomes. Our primary finding was that the occurrence of an index MACE or an index SAE altered the association of intensive BP treatment to all-cause, driven through, respectively, CV and non-CV mortality. In fact, the mediated pathway after a MACE revealed a significant effect size attenuation of intensive BP treatment on ACM. Moreover, when mediated through an SAE, there was a 3% increase in the odds of ACM with intensive BP treatment.

Our findings suggest a need for a more comprehensive approach in clinical trials on occurrences after the incident event. We believe this will more optimally balance appreciation of benefit (MACE reduction) and harm (SAE reduction)—important points in the care of older adults. We complemented the mediation analysis by further stratifying events (MACE and SAE) and mortality types by age groups. We demonstrated that the incidence of SAE was ∼3-fold greater than the incidence of MACE, and the incidence of non-CV mortality was 2.6-fold higher than CV mortality. Moreover, across all age groups, we found a higher IR of SAE compared with MACE with the highest difference between SAE and MACE in the ≥80-year-old group. Also, as expected, all types of mortality increased with age. However, the consistently increased ratio of non-CV mortality over CV mortality was in fact the highest in those ≥70 years old. Furthermore, our findings were conducted within a large randomized clinical trial and align now more closely with prior real-world population studies that have suggested the potential for harm with excessive BP reduction in older adults.^12–14 Based on these findings, future guidelines should consider a balanced approach to reductions of MACE and SAE in order to achieve optimal health not only without disease but also without disability and injury.

Our study goals and findings are distinct from previously published data. Prior studies of the effects of intensive BP control in SPRINT participants focused on the associations between intensive treatment and the first occurrence of SAE, MACE, and mortality, both overall and in various subcohorts—obesity,¹⁵ heart failure,¹⁶ chronic kidney disease,¹⁷ and age categories.¹⁸ While some have documented safety concerns, there has been no work to our knowledge studying mediation between BP treatment intensity and mortality mediated through an index MACE or SAE within the context of a clinical trial.

Our methodological approach is highly relevant for an older adult population. The findings demonstrate that commonly used analyses may overlook the scope of true risk in patients with multimorbidity, frailty, and other geriatric conditions that increase the complexity of care.^19,20 The mediation analyses attempted to explain pathways by which a treatment leads to the outcome and to provide mechanistic clarification by following a counterfactual or potential outcomes model.

We have shared our findings of the CDEs. The CDE is thought to aid in policy evaluation as it considers what the effect of the treatment would be across a population if a mediator was fixed at a given level. Our findings suggest that there could be a substantial attenuation in the effect of intensive BP treatment across a SPRINT-like population as the number of MACE and SAE increase—a particularly applicable finding in older adult patients with a high comorbidity burden and increased risk for MACE and treatment-related SAE.¹¹

Holistic perspectives on the effect of interventions in patients who have a shorter lifespan are a critical component of geriatric cardiology. Focusing solely on MACE reduction while ignoring or minimizing the impact of an SAE may not align with patient priorities²¹ or with optimal healthy life expectancy.²² Healthy life expectancy has been defined as the average number of years in a state of maximal health without disease, disability, or injury. It has been estimated that at the age of 65 years, the average additional life expectancy is ∼16.6 years. Of these remaining years, ∼44–47% is expected to be in a poor health state.²³ Our findings suggest that time-to-first-event analysis, such as used in the original SPRINT with relatively short-term follow-up, provides a non-comprehensive picture.

Our study has limitations. This was a post hoc analysis that was not pre-determined. In Pathways 1 and 3 (Figure 1), the exposure (BP treatment intensity) was randomized. However, intent-to-treat analysis based on randomization may lead to inappropriate conclusions if sufficient crossover occurs during follow-up (e.g. changes in treatment intensity may have occurred based on MACE or SAE occurrences). The mediator–mortality pathway (Pathway 2) was non-randomized and subject to residual confounding that may not have been fully accounted for by the available variables in the current data set. Other variables that better represent geriatric conditions such as frailty, disability, cognitive function, and multimorbidity indices, if available, may have better refined the model and our understanding of the complex relationship between BP treatment intensity and mortality types (all-cause, CV, and non-CV). Our methods of mediation were done using currently accepted causal inference modelling methods.¹⁰ Cardiovascular mortality was ascertained by study investigators with inherent potential for misclassification.

Conclusions

The analytic techniques used in the current study, not often employed in routine randomized clinical trial reporting, demonstrated that intensive BP treatment was potentially associated with an attenuated mortality benefit when mediated through MACE and possibly harmful when mediated through an SAE. We also discovered varying associations between MACE and SAE and the type of mortality (all-cause, CV, or non-CV) that addresses possible mechanisms. Moreover, we believe that the current reporting of randomized clinical trial treatment effects does not allow expansion of the lens to focus on important occurrences after the index event—a particularly poignant reminder when caring for complex older adults.

Supplementary material

Supplementary material is available at European Journal of Preventive Cardiology.

Author contributions

A.K. had full access to all study data and takes complete responsibility for data integrity and statistical analysis accuracy. Study concept and design: A.K. and M.C.O. Acquisition of study data: A.K. Statistical analysis: A.K. and M.C.O. Interpretation of data: A.K., M.W.R., M.J.K., P.G., D.E.F., A.A.D., M.S., J.S.R., D.M.K., and M.C.O. Initial draft of the manuscript: A.K. Final approval of the manuscript: A.K., M.W.R., M.J.K., P.G., D.E.F., A.A.D., M.S., J.S.R., D.M.K., and M.C.O. Study supervision: A.K., M.C.O.

Funding

M.W.R. is supported by NIA: R01 AG 060499, R01 AG 078153, NHLBI: R01 HL 147862, R01 HL 151431. P.G. is supported by American Heart Association grant 20CDA35310455 and National Institute on Aging grant K76AG064428. D.E.F. receives funding from NIA: R01 AG060499, R01 AG058883, P30AG024827, as well as VHA: HSR&D I01 HX003518, RR&D 1I21RX004409, and PCORI 1U19AG065188-01. A.A.D. receives research funding from the Claude D. Pepper Older Americans Independence Center funded by the National Institute on Aging on Aging P30-AG021334 and receives mentored patient-oriented research career development award from the National Heart, Lung, and Blood Institute K23-HL153771-01.

Data availability

The SPRINT data set is a public data set and can be obtained at https://biolincc.nhlbi.nih.gov/home/.

References

Author notes