Midlife Blood Pressure Variability and Risk of All-Cause Mortality and Cardiovascular Events During Extended Follow-up

Abstract

BACKGROUND

Studies demonstrate an association between visit-to-visit blood pressure variability (BPV) and cardiovascular events and death. We aimed to determine the long-term cardiovascular and mortality effects of BPV in midlife in participants with and without cardiovascular risk factors.

METHODS

This is a post-hoc analysis of the Atherosclerosis Risk in the Community study. Long-term BPV was derived utilizing mean systolic blood pressure at Visits 1–4 (Visit 1: 1987–1989, Visit 2: 1990–1992, Visit 3: 1993–1995, Visit 4: 1996–1998). The primary outcome was mortality from Visit 4 to 2016 and secondary outcome was cardiovascular events (fatal coronary heart disease, myocardial infarction, cardiac procedure, or stroke). We fit Cox proportional hazards models and also performed the analysis in a subgroup of cardiovascular disease-free patients without prior stroke, myocardial infarction, congestive heart failure, hypertension, or diabetes.

RESULTS

We included 9,578 participants. The mean age at the beginning of follow-up was 62.9 ± 5.7 years, and mean follow-up was 14.2 ± 4.5 years. During follow-up, 3,712 (38.8%) participants died and 1,721 (n = 8,771, 19.6%) had cardiovascular events. For every SD higher in systolic residual SD (range 0–60.5 mm Hg, SD = 5.6 mm Hg), the hazard ratio for death was 1.09 (95% confidence interval [CI] 1.05–1.12) and for cardiovascular events was 1.00 (95% CI 0.95–1.05). In cardiovascular disease-free participants (n = 4,452), the corresponding hazard ratio for death was 1.12 (95% CI 1.03–1.21) and for cardiovascular events was 1.01 (95% CI 0.89–1.14).

CONCLUSION

Long-term BPV during midlife is an independent predictor of later life mortality but not cardiovascular events.

Graphical Abstract

Graphical Abstract.

Higher long-term visit-to-visit blood pressure variability (BPV) has been shown to have negative effects on multiple organ systems, independent of presence, or severity of hypertension.^1–5 Several studies have shown a specific association between higher long-term visit-to-visit BPV and increased risk of cardiovascular disease or stroke.^6–8 Prior research has also demonstrated that patients with higher long-term visit-to-visit BPV are at elevated risk of all-cause mortality.^1,9–15 In a study of 3,394 initially younger patients who had long-term visit-to-visit BPV measured at 5 visits over 10 years, BPV was associated with all-cause mortality in midlife, up to 20 years later.⁹ In an unselected population of 240,662 adults, the long-term visit-to-visit BPV in the electronic medical record from routine clinical appointments over 3 years was associated with subsequent all-cause mortality.¹³

To further investigate these associations, we performed an analysis of the Atherosclerosis Risk in the Community study (ARIC).¹ We measured long-term visit-to-visit BPV over a 12-year period, followed by ascertainment of death and cardiovascular events during a follow-up period of up to 17 years. The longitudinal nature of ARIC and the reliability of its data provide an excellent opportunity to further study and validate the association between higher long-term visit-to-visit BPV and the outcomes of death and cardiovascular events in patients who were followed into later life and in a subgroup of patients without cardiovascular risk factors at baseline.

METHODS

This is a post hoc analysis of the ARIC study,¹⁶ a prospective, community-based, biracial cohort study of the risk factors for atherosclerosis and cardiovascular disease. With a local IRB waiver, we obtained the anonymized ARIC dataset from the NHLBI Biologic Specimen and Data Repository Information Coordinating Center. The ARIC codebooks and procedure manuals are publicly available.¹⁷ The study exposure was long-term BPV, derived from the visit-to-visit systolic blood pressures from 1987 to 1998 (Figure 1) (Visits 1–4; Visit 1: 1987–1989, Visit 2: 1990–1992, Visit 3: 1993–1995, Visit 4: 1996–1998). The mean (range) number of days between Visits 1 and 2 was 1,064 (920–1,948), between Visits 2 and 3 was 1,115 (420–2,001), and between Visits 3 and 4 was 1,083 (306–2,042). Blood pressure was measured manually via a random zero sphygmomanometer in Visits 1–4 by a certified trained technician in accordance with the ARIC manual of procedures.¹⁷ Participants were asked to rest in a sitting position for at least 5 minutes prior to blood pressure measurements at each visit. The final 2 resting blood pressure measurements at each visit were averaged for a visit mean systolic blood pressure.¹⁷ Thus, each participant had 4 visits with a visit mean systolic blood pressure, for a total of 4 blood pressures per patient. We excluded participants with missing blood pressure readings or demographic data.

Figure 1.

Diagram of the study design.

The primary outcome of our study is all-cause mortality. The secondary outcome is a composite of cardiovascular events, which includes myocardial infarction, cardiac procedure, fatal coronary heart disease, or ischemic stroke. Outcome ascertainment began after Visit 4, at which point participant death and the cardiovascular composite were recorded every year until 2016. For the individual models fit to the primary or secondary outcome, we excluded patients who had the outcome during the exposure period when BPV was being measured. Participants were censored after experiencing an outcome.

The exposure, long-term BPV, was measured using 5 indices of the systolic blood pressure: SD, residual SD (rSD), average real variability, successive variation, and coefficient of variation. The derivation of these variables is well described in the Appendix of an article by Rothwell et al.¹⁸ We represented the BPV indices both as continuous variables and in quintiles. We reported continuous systolic BPV values per SD increment. We focused on rSD because prior data have shown it may be a superior methodology for long-term visit-to-visit BPV measurement, mainly because it is less affected by underlying trends in blood pressure over time.¹⁹ rSD is calculated using the formula: $\mathrm{r}\mathrm{S}\mathrm{D}=\sqrt{\frac{\sum {(\mathrm{B}{\mathrm{P}}_{\mathrm{i}}-\mathrm{B}{\mathrm{P}}_{\mathrm{e}\mathrm{s}\mathrm{t}})}^2}{n-2}}$, where BP_i is the blood pressure average for an individual study visit and BP_est is the expected blood pressure from a linear regression of blood pressure from all the patient’s visits.

We selected potential covariates based off data points in ARIC that have previously been shown to be predictive of death.^20–22 We used least absolute shrinkage and selection operator regression analysis to select the final covariates for the adjusted model.^23,24 Baseline covariates were measured at the beginning of follow-up (Visit 4) and included patient age, sex, history of myocardial infarction, hypertension, diabetes, congestive heart failure, prior stroke, current anticoagulation, waist-to-hip ratio, exercise or sport activity (Visit 3 variable), current smoking, current drinking, and glucose level. We also included derived variables from the exposure period (Visits 1–4): the mean systolic blood pressure, the mean patient weight in kg, the difference in systolic blood pressure between Visits 4 and 1, and heart rate variability during the exposure period, defined as the SD of heart rate from Visit 1 to 4. We tested the interaction term between rSD and 3 stratification variables: mean systolic blood pressure during exposure ≥140 vs. <140 mm Hg, increase vs. decrease in systolic blood pressure from Visit 1 to 4, and if patients were taking an antihypertensive medication at the beginning of follow-up vs. not. If the interaction term was statistically significant (P < 0.05), we performed an analysis in both levels of the stratification.

We fit Cox proportional hazards models to the outcomes during follow-up. We also fit an unadjusted time-to-event model to generate Kaplan–Meier curves. We confirmed the proportional hazards assumption of the final Cox model by testing the Schoenfeld residuals and its fit by graphing the Cox–Snell residuals with the Nelson–Aalen cumulative hazard function (Supplementary Figure S2 online). We also fit our Cox models in a subgroup of cardiovascular disease-free patients, who were free of the following at the beginning of follow-up: prior stroke, myocardial infarction, congestive heart failure, hypertensive, or diabetes. P values <0.05 were considered statistically significant. All analysis was performed in Stata 16.0 (StataCorp, College Station, TX).

RESULTS

We included 9,578 patients in the primary cohort (Supplementary Figure S1 online). Baseline demographics and medical comorbidities are shown in Table 1. The mean ± SD age at the beginning of follow-up (Visit 4) was 62.9 ± 5.7 years and the mean ± SD per-patient duration of follow-up was 14.2 ± 4.5 years. Death during follow-up was seen in 3,712 (38.8%) patients at a mean of 10.2 (4.6) years. The rSD (mm Hg) during the exposure period was higher in patients who died vs. survived (9.4 ± 6.4 vs. 7.7 ± 5.1, P < 0.001), as were all other BPV indices (Table 2).

Table 1.

Baseline (Visit 4) demographics for the full cohort and stratified by participants who died during follow-up vs. those who survived

Variable	Full cohort (n = 9,578)	Died during follow-up (n = 3,712)	Survived or lost to follow-up (n = 5,871)	P value
Age (mean ± SD)	62.9 ± 5.6	65.4 ± 5.4	61.3 ± 5.2	<0.001
Male sex (n, %)	4,288, 44.8%	1,945, 52.4%	2,343, 39.9%	<0.001
White race (n, %)	7,852, 82.0%	2,958, 79.7%	4,894, 83.4%	<0.001
Education (n, %)
Less than high school	1,688, 17.6%	911, 24.6%	777, 13.3%	<0.001
High school or vocational	4,113, 43.0%	1,553, 41.8%	2,560, 43.6%
College or above	3,777, 39.4%	1,248, 33.6%	2,529, 43.1%
Exercise or sport activity (Visit 3) (n, %)	6,385, 66.7%	2,376, 64.0%	4,099, 68.3%	<0.001
Hypertension (n, %)	4,420, 46.2%	2,102, 56.6%	2,318, 39.5%	<0.001
Diabetes (n, %)	1,276, 13.3%	767, 20.7%	509, 8.7%	<0.001
Congestive heart failure (n, %)	214, 2.2%	177, 4.8%	37, 0.6%	<0.001
Prior stroke (n, %)	193, 2.1%	136, 3.7%	57, 1.0%	<0.001
Prior myocardial infarction (n, %)	676, 7.1%	455, 12.3%	221, 3.8%	<0.001
Coronary heart disease (n, %)	797, 8.3%	519, 14.0%	278, 4.8%	<0.001
Current alcohol use (n, %)	4,901, 51.2%	1,692, 45.6%	3,209, 54.7%	<0.001
Current smoking (n, %)	1,366, 14.3%	734, 19.8%	632, 10.8%	<0.001
Cholesterol medication use (n, %)	1,410, 14.7%	633, 17.1%	777, 13.3%	<0.001
Antihypertensive medication (n, %)	3,385, 35.4%	1,622, 43.7%	1,763, 30.1%	<0.001
Aspirin use (n, %)	5,437, 56.8%	2,240, 60.4%	3,197, 54.5%	<0.001
Anticoagulant use (n, %)	193, 2.0%	142, 3.8%	51, 0.9%	<0.001
LDL cholesterol (mg/dl, mean ± SD)	122.6 ± 33.3	121.6 ± 34.5	123.2 ± 32.5	0.022
HDL cholesterol (SI, mean ± SD)	1.29 ± 0.42	1.24 ± 0.43	1.32 ± 0.42	<0.001
Glucose level (SI, mean ± SD)	6.12 ± 2.04	6.49 ± 2.40	5.89 ± 1.74	<0.001
Waist-to-hip ratio (mean ± SD)	0.95 ± 0.07	0.96 ± 0.07	0.94 ± 0.07	<0.001
Body mass index, Visits 1–4 (mean ± SD)	27.9 ± 5.1	28.4 ± 5.4	27.6 ± 4.9	<0.001
Weight in kg, Visits 1–4 (mean ± SD)	79.5 ± 16.4	81.2 ± 17.1	78.4 ± 15.8	<0.001
Heart rate, Visits 1–4 (mean ± SD)	64.8 ± 8.4	65.7 ± 8.7	64.3 ± 8.1	<0.001
Heart rate SD, Visits 1–4	5.5 ± 3.3	5.9 ± 3.9	5.3 ± 2.9	<0.001

Abbreviations: HDL, high-density lipoprotein; LDL, low-density lipoprotein.

Table 2.

Systolic blood pressure values for the full cohort and stratified by participants who died during follow-up vs. those who survived

Variable	Full cohort (n = 9,578)	Died during follow-up (n = 3,712)	Survived or lost to follow-up (n = 5,866)	P value
Mean (mean ± SD)	122.5 ± 15.2	126.5 ± 15.8	119.9 ± 14.2	<0.001
SD (mean ± SD)	10.0 ± 5.7	11.2 ± 6.2	9.2 ± 5.1	<0.001
rSD (mean ± SD)	8.3 ± 5.6	9.4 ± 6.4	7.7 ± 5.1	<0.001
ARV (mean ± SD)	12.2 ± 7.8	13.7 ± 8.6	11.3 ± 7.0	<0.001
SV (mean ± SD)	13.9 ± 9.4	15.5 ± 9.4	12.8 ± 7.7	<0.001
CV (mean ± SD)	8.0 ± 4.1	8.7 ± 4.4	7.6 ± 3.9	<0.001

Abbreviations: ARV, average real variability; CV, coefficient of variation; rSD, residual SD; SV, successive variation.

The full multivariable Cox model fit to the primary outcome with hazard ratios for all covariates is shown in Supplementary Table S1 online. All BPV indices were associated with death during follow-up in the Cox models (Table 3). For every SD higher rSD (range 0–60.5 mm Hg, SD = 5.6 mm Hg), the hazard ratio for death was 1.09 (95% confidence interval [CI] 1.05–1.12). For every SD higher SD (range 0.5–50.7, SD = 5.7 mm Hg), the hazard ratio for death was 1.12 (95% CI 1.08–1.16). Similarly, for every SD higher successive variation (range 5.8–92.7, SD = 9.4 mm Hg), the hazard ratio for death was 1.12 (95% CI 1.08–1.16).

Table 3.

Cox proportional hazards model fit to the primary outcome of death and the secondary outcome of cardiovascular disease (CVD) events, showing hazard ratios for a SD higher in the measure of blood pressure variability (BPV)

BPV indices	Mean ± SD	Hazard ratio for deatha	95% CI	P value	Hazard ratio for CVD eventsb	95% CI	P value
SD (range 0.5–50.7)	10.0 ± 5.7	1.12	1.08–1.16	<0.001	1.02	0.97–1.08	0.383
rSD (range 0–60.5)	8.3 ± 5.6	1.09	1.05–1.12	<0.001	1.00	0.95–1.05	0.877
ARV (range 0.3–90.0)	12.2 ± 7.8	1.10	1.07–1.14	<0.001	1.05	1.00–1.11	0.049
SV (range 0.6–70.0)	13.9 ± 9.4	1.12	1.08–1.16	<0.001	1.05	1.00–1.11	0.070
CV (range 0.4–33.3)	8.0 ± 4.1	1.12	1.08–1.16	<0.001	1.02	0.97–1.08	0.359

Abbreviations: ARV, average real variability; CI, confidence interval; CV, coefficient of variation; CVD, cardiovascular events; rSD, residual SD; SV, successive variation.

^aAdjusted for baseline age, sex, race, history of myocardial infarction, hypertension, diabetes, congestive heart failure, prior stroke, current anticoagulation, waist-to-hip ratio, exercise or sport activity (Visit 3), level of education, current smoking, current drinking, glucose level, and mean systolic blood pressure from Visit 1 to 4, mean weight from Visit 1 to 3, heart rate variability, and difference in systolic blood pressure between Visits 4 and 1, showing hazard ratios for a SD higher in BPV (see Table 2 for SDs).

^b n = 8,771, due to loss of patients for CVD events during the exposure period.

After dividing rSD into quintiles, a Kaplan–Meier curve shows the risk of death increases across the quintiles (Figure 2), from a hazard ratio of 1 (quintile 1, reference) to 1.05 (quintile 2, 95% CI, 0.94–1.18) to 1.34 (quintile 3, 95% CI, 1.20–1.49) to 1.45 (quintile 4, 95% CI, 1.30–1.61) to 2.01 (quintile 5, 95% CI, 1.81–2.22) (log-rank P value <0.001). A similar increase across the quintiles was seen in an adjusted Cox model (Supplementary Table S2 online). In the Cox model fit to death, the interaction terms between rSD and mean systolic blood pressure during the exposure period of ≥140 vs. <140 mm Hg, increase vs. decrease in systolic blood pressure from Visit 1 to 4, and if patients were taking an antihypertensive medication at the beginning of follow-up vs. not, all lacked significance (P > 0.05), so further stratified analyses were not performed. The results of the Cox model after stratification by mean systolic blood pressure during the exposure period of ≥140 vs. <140 mm Hg is seen in Supplementary Table S3 online.

Figure 2.

Kaplan–Meier curve showing the survival estimates for the quintiles of systolic blood pressure residual SD.

We did not find that BPV was associated with our secondary outcome of cardiovascular events including fatal coronary heart disease, myocardial infarction, cardiac procedure, or ischemic stroke. This outcome occurred in 1,721 (19.6%) of 8,771 patients available for this analysis. For every SD higher rSD, the hazard ratio for cardiovascular events was 1.00 (95% CI 0.95–1.05). The remainder of the BPV indices can be seen in Table 3 and, apart from a modest association for average real variability (hazard ratio 1.05, 95% CI 1.00–1.11), were not associated with cardiovascular events.

In the subgroup of 4,452 patients who were cardiovascular disease-free at baseline (without prior stroke, myocardial infarction, congestive heart failure, hypertensive, or diabetes), 1,228 (27.6%) died and 574 (12.9%) had cardiovascular events during follow-up. For every SD higher rSD, the hazard ratio for death was 1.12 (1.03–1.21) and for cardiovascular events was 1.01 (0.89–1.14). The hazard ratios for the other indices of BPV are seen in Table 4.

Table 4.

Cox proportional hazards model fit to the primary outcome of death and the secondary outcome of cardiovascular disease (CVD) events in the cohort of cardiovascular disease-free patients (n = 4,452)

BPV indices	Mean ± SD	Hazard ratio for deatha	95% CI	P value	Hazard ratio for CVD eventsb	95% CI	P value
SD (range 0.5–25.4)	7.6 ± 3.7	1.21	1.11–1.32	<0.001	0.99	0.86–1.13	0.832
rSD (range 0–30.8)	6.5 ± 3.9	1.12	1.03–1.21	0.006	1.01	0.89–1.14	0.905
ARV (range 0.3–42.0)	9.3 ± 5.1	1.21	1.06–1.31	<0.001	0.97	0.85–1.11	0.683
SV (range 0.6–42.9)	10.6 ± 5.6	1.24	1.13–1.36	<0.001	0.99	0.86–1.15	0.917
CV (range 0.4–27.9)	6.8 ± 3.2	1.18	1.10–1.27	<0.001	0.98	0.87–1.10	0.692

Abbreviations: ARV, average real variability; BPV, blood pressure variability; CI, confidence interval; CV, coefficient of variation; rSD, residual SD; SV, successive variation.

^aAdjusted for baseline age, sex, race, current anticoagulation, waist-to-hip ratio, exercise or sport activity (Visit 3), level of education, current smoking, current drinking, glucose level, and mean systolic blood pressure from Visit 1 to 4, mean weight from Visit 1 to 3, heart rate variability, and difference in systolic blood pressure between Visits 4 and 1, showing hazard ratios for a SD higher in BPV (see Table 2 for SDs).

^b n = 4,328, due to loss of patients for CVD events during the exposure period.

Discussion

In this post hoc analysis of the ARIC cohort, we demonstrated an independent association between long-term BPV measures and all-cause mortality but not cardiovascular events. The association with mortality remained statistically significant in the subgroup of patients who were cardiovascular disease-free at the beginning of follow-up. Our findings regarding long-term BPV and risk of death are consistent with prior studies^1,9–15 and further add to the body of literature by expanding the studied population, specifically by including a community-based population with prolonged follow-up and a subgroup of patients without cardiovascular risk factors. Because we did not find that long-term BPV was associated with cardiovascular events, the association with mortality may be explained by noncardiovascular mechanisms.

Prior research looking at long-term visit-to-visit BPV and cardiovascular events has reported conflicting results. In a comprehensive analysis of the United Kingdom transient ischemic attack (UK TIA) and Anglo-Scandinavian Cardiac Outcomes Trial Blood Pressure Lowering Arm (ASCOT-BPLA) trials, Rothwell et al. demonstrated an association between visit-to-visit BPV, measured every 4 months, and stroke and coronary events among patients with previous history of transient ischemic attack and hypertensive patients.⁸ A subsequent analysis of the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT) trail found similar results with visit-to-visit BPV calculated from 7 measurements during the first 28 months.⁶ These findings were further supported by a systematic review and meta-analysis demonstrating an association between visit-to-visit BPV and cardiovascular events.²⁵ However, a recent article by Saladini et al. failed to show an association between long-term visit-to-visit BPV, measured at 7 visits over a year, and cardiovascular events,²⁶ but found a statistically significant association for short-term BPV measured with ambulatory blood pressures over 24 hours. Likewise, a large study of 28,790 patients in 2 clinical trials by Mancia et al. failed to show an association between long-term visit-to-visit BPV, measured at visits over 2 years, and myocardial infarction or stroke.²⁷

The mechanism underlying the association between BPV and clinical outcomes is not well understood. Clark et al. found a link between atheroma progression and BPV in patients with underlying coronary artery disease.²⁸ Nwabuo et al. found an association between higher BPV and biomarkers of cardiovascular disease, including higher left ventricular mass index, worse diastolic function, higher left ventricular filling pressures, and worse global longitudinal strain.²⁹ Furthermore, prior studies demonstrated an association between higher BPV and lower aortic distensibility, lower large and small artery elasticity and common carotid artery stiffness.^30,31 Each of these factors—atheroma progression, left ventricular dysfunction, and arterial stiffness—are established risk factors for cardiovascular events and mortality.^32,33

It is conceivable that ARIC participants with poorly controlled hypertension (e.g., mean systolic ≥140 mm Hg) had higher BPV, therefore, it can be argued that the association between higher BPV and mortality is a marker of underlying hypertension. In the present study, though, our analyses were adjusted for mean blood pressure and we did not find heterogeneity in the association between BPV and mortality in patients with mean systolic blood pressure ≥140 vs. ≤140 mm Hg. Other factors that may play a role in the association between BPV and mortality include poor medication adherence or change in antihypertensive treatments.³⁴ Although, analyses of randomized trials have suggested that the association is independent of changes in treatment or poor medication adherence.^2,8,35

The type of antihypertensive treatment could be important for the management of BPV. In a post hoc analysis of ASCOT-BPLA, BPV was lower in the amlodipine group compared with atenolol group.³⁶ In another analysis of ALLHAT, BPV was most pronounced in the lisinopril group and least in the chlorthalidone group.⁶ Finally, a systematic review and meta-analysis showed that BPV was reduced with calcium channel blockers and thiazides and increased with angiotensin-converting enzymes inhibitors.³⁷ A small pilot study recently provided additional evidence that BPV could be reduced with calcium channel blockers.³⁸ However, large-scale clinical trials with specific interventions aimed at reducing BPV are needed to identify the most promising therapeutic interventions for reducing BPV.

Despite the large sample size and the long follow-up period in this study, our study has limitations that need to be considered. We excluded 5,208 individuals who had discontinued the ARIC study or had the study outcome before the beginning of our follow-up period. It is possible they were different from the rest of the cohort with respect to BPV or mortality, and thus introduced selection bias. We may be underestimating the impact of BPV given the prolonged exposure and outcome periods. Detailed information on the dosage, adherence, and duration of therapy of antihypertensive treatment was not reliably available in this study, so we cannot address the effect of antihypertensive treatment on the association between BPV and mortality. We also did not account for the time-varying nature of the risk factors for our outcomes, which is not possible given limitations of the dataset. Finally, we cannot establish the causality of BPV in increasing the risk of death, which would require a prospective observational study or clinical trial of BPV reduction.

Long-term visit-to-visit BPV is an independent predictor of all-cause mortality in the general population with and without cardiovascular risk factors. These findings are important because of the attributes of the ARIC study, which allowed prolonged follow-up of participants from late middle age into older age.

FUNDING

Dr de Havenon is supported by NIH-NINDS K23NS105924.

DISCLOSURE

Dr de Havenon receives investigator initiated funding from AMAG and Regeneron Pharmaceuticals.