The Association between Antihypertensive Medication Nonadherence and Visit-to-Visit Variability of Blood Pressure: Findings from the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT)

Abstract

Low adherence to antihypertensive medication has been hypothesized to increase visit-to-visit variability (VVV) of blood pressure (BP). We assessed the association between antihypertensive medication adherence and VVV of BP in the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT). VVV of BP was calculated using standard deviation independent of the mean (SDIM), standard deviation, and average real variability across study visits conducted 6 to 28 months following randomization. Participants who reported taking <80% of their antihypertensive medication at one or more study visits were categorized as nonadherent. Participants were followed for cardiovascular events and mortality after the assessment of adherence and VVV of BP. SDIM of BP was higher for nonadherent (n=2,912) versus adherent (n=16,878) participants; 11.4±4.9 versus 10.5±4.5 for systolic BP (SBP); 6.8±2.8 versus 6.2±2.6 for diastolic BP (DBP) (each p<0.001). SDIM of BP remained higher among nonadherent versus adherent participants after multivariable adjustment [0.8 (95%CI 0.7–1.0) higher for SBP and 0.4 (95%CI 0.3–0.5) higher for DBP]. Standard deviation and average real variability of SBP and DBP were also higher among nonadherent versus adherent participants. Adjustment for nonadherence did not explain the association of VVV of BP with higher fatal coronary heart disease or non-fatal myocardial infarction, stroke, heart failure, or mortality risk. In conclusion, improving medication adherence may lower VVV of BP. However, VVV of BP is associated with cardiovascular outcomes independent of medication adherence.

Introduction

Visit-to-visit variability (VVV) of blood pressure (BP) is associated with increased risk for stroke, coronary heart disease (CHD), and mortality.^1–5 The mechanisms underlying the association between VVV of BP and these outcomes remain poorly understood.⁶ It is biologically plausible that missed antihypertensive medication doses could explain fluctuations in BP between visits among treated individuals. Nonadherence to antihypertensive medications has been associated with increased risk for cardiovascular events.^{7, 8} Thus, antihypertensive medication nonadherence could explain the adverse prognosis found in individuals with higher VVV of BP and may be a modifiable cause of VVV of BP.

Antihypertensive drug type may be relevant to the association between medication nonadherence and VVV of BP. In the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT), mean VVV of systolic BP (SBP) and diastolic BP (DBP) were lower among participants randomized to chlorthalidone and amlodipine compared with their counterparts randomized to lisinopril.⁹ Also, a meta-analysis of group level data from randomized clinical trials comparing antihypertensive regimens concluded that use of calcium-channel blockers and thiazide-type diuretics was associated with lower VVV of BP than use of angiotensin converting enzyme (ACE)-inhibitors, angiotensin receptor blockers, or beta-blockers.¹⁰ One potential explanation for this finding is that antihypertensive medications with longer half-lives and less pronounced BP rebound after missed doses may limit the influence of nonadherence on VVV of BP.¹¹ Accordingly, there may be an interaction between antihypertensive drug type and medication adherence on VVV of BP.

The goal of the current study was to determine the association between antihypertensive medication adherence and VVV of BP in ALLHAT. We evaluated the association, overall, and by ALLHAT randomization assignment to different antihypertensive drug classes. We also evaluated whether participants who had a change in adherence between two time periods had a change in VVV of BP. Finally, we conducted a mediation analysis to determine whether medication nonadherence explained the association between VVV of BP and cardiovascular and mortality outcomes. ALLHAT is well-suited to answering these questions as the study included a large sample of treated individuals with hypertension who had repeated measures of medication adherence and BP and prospective follow-up for outcomes.

Methods

Details of ALLHAT have been described previously.^{12, 13} Briefly, ALLHAT was a randomized double-blind, multicenter trial to determine whether treatment with calcium channel blockers, ACE-inhibitors, or alpha-adrenergic blockers, all newer antihypertensive classes at the time, was superior to treatment with a thiazide diuretic for lowering risk for fatal CHD or non-fatal myocardial infarction. Participants were randomized to chlorthalidone, amlodipine, lisinopril, or doxazosin, and treated to a goal systolic/diastolic BP of less than 140/90 mm Hg. All participants in ALLHAT provided written informed consent and all centers received institutional review board approval.

Eligibility for the Current Analyses

The doxazosin arm in ALLHAT was stopped early due to a 25% higher incidence of cardiovascular events, including a nearly two-fold higher risk of heart failure, accompanied by very low probability of reaching a statistically significant difference in the primary end point.¹⁴ Therefore, participants in this randomization arm were excluded from the current analyses. Also, the sample for the primary analyses was restricted to ALLHAT participants who attended, and had BP and medication adherence data, from at least 5 of the 7 study visits conducted 6 to 28 months following randomization. A minimum of 5 visits was chosen to ensure a robust assessment of VVV of BP.¹⁵ Visits in the first 6 months following randomization were excluded as large reductions in BP occurred before this time point while participants were being titrated to achieve their goal BP, and this could have inflated estimates of VVV of BP. This approach is consistent with prior analyses of VVV of BP in ALLHAT.⁹ The sample in analyses assessing the association between VVV of BP and outcomes was restricted to participants who had no outcome events (i.e., no fatal CHD or non-fatal myocardial infarction (MI), stroke, all-cause mortality, and heart failure) prior to the 28 month study visit. In secondary analyses, we additionally required participants to have BP and medication adherence data from at least 5 of the 7 study visits conducted 32 to 56 months following randomization.

Measures

Antihypertensive medication adherence

Adherence to medication was assessed at each visit by a study clinician using the Adherence Survival Kit developed for ALLHAT.¹⁶ Specifically, participants were asked whether they had taken at least 80% of their assigned study drug since the last follow-up visit. For our primary analyses, participants were categorized as nonadherent if they reported having taken less than 80% of their assigned antihypertensive medication at 1 or more visits during the 6 to 28 month time period after randomization. We also conducted secondary analyses in which participants were categorized as nonadherent if they reported having taken less than 80% of their assigned medication at 1 or more visits during the 32 to 56 months after randomization. In a sensitivity analysis, participants were categorized as nonadherent if they reported taking less than 80% of the prescribed antihypertensive medication at 2 or more visits during the 6 to 28 months post-randomization time period.

VVV of BP

VVV of BP was calculated during early (6 to 28 months following randomization) and, separately, late time periods (32 to 56 months following randomization) using three metrics based on each ALLHAT participant’s BP measurements: standard deviation independent of mean (SDIM), standard deviation, and average real variability (Supplemental Figure, S1).^{3, 17} The BP used for these calculations was the mean of two measurements taken during each follow-up study visit according to a standardized BP measurement protocol.¹² While we considered other VVV of BP metrics, the correlation between VVV of BP metrics is high, suggesting analyses with additional metrics would not provide new information.¹⁵ Our primary VVV of BP metric was SDIM.

Covariates

Covariates were selected a priori based on their potential to serve as confounders of the association between adherence and VVV of BP. Covariates included sociodemographic characteristics, practice site characteristics, medical comorbidities, randomization assignment, BP at baseline and during follow-up, and BP medication prescribing prior to and after randomization.

Cardiovascular and Mortality Outcomes

Outcomes included fatal CHD or non-fatal myocardial infarction (MI), stroke, all-cause mortality, and heart failure. Details of the event ascertainment process are provided elsewhere.^{12, 13} Participants were followed from the end of the VVV of BP assessment period to the date of each outcome, their date of death, or end of active ALLHAT follow-up (October 1, 2001 through March, 31, 2002).

Statistical Analysis

Participant characteristics and SDIM of SBP were calculated for nonadherent and adherent participants, separately. We used linear regression to calculate the adjusted mean difference in the SDIM of SBP in nonadherent versus adherent (reference group) participants. Three levels of adjustment were performed: 1) an initial model with adjustment for age, race/ethnicity, gender, geographic region of practice site and practice type; 2) a second model with additional adjustment for eGFR, diabetes, body mass index, smoking status, history of MI or stroke, history of coronary revascularization, history of other atherosclerotic cardiovascular disease, major ST depression, left ventricular hypertrophy, baseline systolic and diastolic blood pressure, use of BP medications prior to study randomization, and randomization assignment; and 3) a third model with further adjustment for covariates from follow-up visits including mean SBP during follow-up, mean number of antihypertensive medications taken during months 6 to 28, change in antihypertensive medication regimen during follow-up, antihypertensive medication classes being taken, and statin use. The above analyses were repeated using SDIM of DBP and alternate metrics of VVV of SBP (i.e., standard deviation and average real variability) as the measure of VVV, and after defining nonadherence as two or more visits with <80% adherence. As a further sensitivity analysis, the above analyses were repeated while restricting the analysis to participants who remained on a single drug class (i.e., only the drug to which they were randomized) from month 6 to 28 of follow-up. The above analyses were also repeated stratified by antihypertensive drug class randomization assignment. We tested for effect modification between adherence and randomization drug class on VVV of BP in linear regression models with the full population and multiplicative interaction terms between medication adherence and randomization assignment. In secondary analyses, we determined whether within-participant change in adherence between the early study period (6 to 28 months post-randomization) and the late study period (32 to 56 months post-randomization) was associated with their change in SDIM of SBP during this same interval. For this analysis, participants were grouped as being adherent in both the early and late study periods, nonadherent in both study periods, nonadherent in the early period but adherent in the late period, and adherent in the early time period but nonadherent in the late period. Mean change in SDIM of SBP between the early and late study periods was calculated for each of these groups.

In a final analysis, we assessed whether medication nonadherence explained the association between VVV of BP and major cardiovascular and mortality outcomes by considering two multivariable Cox proportional hazards regression models, one that assessed the hazard ratios (HR) for outcomes associated with VVV of BP without medication adherence and another with medication adherence included.¹⁸ For these analyses, the parameter of interest was the difference in the estimated association of SDIM of SBP with outcomes between the two models, and the 95% confidence interval for this estimate was calculated using a 1,000 iteration bootstrap. Data analyses were conducted in Stata Version 13 (Stata Inc. College Station, TX).

Results

Of the 33,357 participants randomized to chlorthalidone, amlodipine, or lisinopril in ALLHAT, 9,353 were excluded as they had fewer than 5 visits with BP measured and an additional 4,214 were excluded as they had fewer than 5 visits at which adherence was assessed. This left 19,970 participants meeting the eligibility criteria for our primary analyses. Fifteen percent of these participants (n=2,912) were nonadherent. Compared to adherent participants, nonadherent participants were slightly older and more likely to be Hispanic or Black (Table 1). Nonadherent participants were more likely to have evidence of end-organ damage as signified by major ST depression or T wave inversion or LVH on electrocardiogram, but were less likely to have a prior history of MI, stroke, or coronary revascularization. They were also less likely to have used BP medications prior to randomization and less likely to use statins during follow-up. Nonadherent participants were also more likely to have changes in BP medication classes during follow-up, were more likely to have uncontrolled BP between 6 and 28 months after randomization, and had higher mean SBP and DBP at those visits. The association between nonadherence status and higher BP remained statistically significant in adjusted analyses (Supplemental Table S2).

Table 1.

Characteristics of ALLHAT Participants Eligible for Primary Analyses by Antihypertensive Medication Adherence Status (N=19,790)^*

Participant Characteristics	Adherent† N=16,878; 85.3%	Nonadherent‡ N=2,912; 14.7%	P-Value
Variables from the baseline visit
Age in years, mean (SD)	66.5 (7.4)	66.9 (7.7)	0.008
Male sex, n (%)	9483 (56.2)	1600 (55.0)	0.21
Race/ethnicity, n (%)
Non-Hispanic White	9266 (54.9)	1182 (40.6)	<0.001
Non-Hispanic Black	4873 (28.9)	1046 (35.9)	<0.001
Hispanic White	1578 (9.4)	431 (14.8)	<0.001
Hispanic Black	335 (2.0)	73 (2.5)	0.08
Diabetes, n (%)	5934 (35.2)	1006 (34.6)	0.52
History of MI or stroke, n (%)	3861 (22.9)	618 (21.2)	0.05
History of coronary revascularization, n (%)	2344 (13.9)	315 (10.8)	<0.001
History of other ASCVD, n (%)	4094 (24.3)	712 (24.5)	0.82
Major ST depression or T wave inversion, n (%)	1675 (10.0)	341 (11.9)	0.003
LVH on electrocardiogram, n (%)	2591 (15.4)	575 (19.8)	<0.001
eGFR in ml/min/1.73 m2, mean (SD)	74.8 (17.4)	74.2 (17.6)	0.069
BMI in kg/m2, mean (SD)	29.8 (6.0)	29.6 (6.2)	0.19
Current smoker, n (%)	3541 (21.0)	659 (22.6)	0.046
Use of antihypertensive medication prior to randomization, n (%)	15395 (91.2)	2607 (89.5)	0.004
Baseline SBP in mm Hg, mean (SD)	145.4 (15.6)	147.1 (15.5)	<0.001
Baseline DBP in mm Hg, mean (SD)	83.5 (10.0)	84.9 (9.9)	<0.001
Randomization Assignment, n (%)
Chlorthalidone	8073 (47.8)	1328 (45.6)	0.03
Amlodipine	4684 (27.8)	815 (28.0)	0.79
Lisinopril	4121 (24.4)	769 (26.4)	0.02
Variables from the VVV of BP assessment period (6 to 28 months post-randomization)
Change in antihypertensive drug regimen, n (%)	7130 (42.2)	1370 (47.1)	<0.001
Number of antihypertensive drugs, mean (SD)	1.48 (0.64)	1.46 (0.63)	0.09
Use of statin medication, n (%)	6209 (36.8)	895 (30.7)	<0.001
SBP in mm Hg, mean (SD)	136.6 (10.9)	139.6 (12.3)	<0.001
DBP in mm Hg, mean (SD)	78.3 (7.0)	80.0 (7.5)	<0.001
Uncontrolled BP§, n %	5979 (35.4)	1295 (44.5)	<0.001

Abbreviations: ASCVD, atherosclerotic vascular disease; LVH, left ventricular hypertrophy; eGFR, estimated glomerular filtration rate; BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; SD, standard deviation.

^*Eligibility criteria for primary analyses in the current study included availability of blood pressure and medication adherence data from at least 5 of the 7 study visits conducted between 6 and 28 months post-randomization.

^†Adherent participants reported taking more than 80% of all antihypertensive medication at all study visits in early time period (study visits from month 6 to 28)

^‡Nonadherent participants reported taking less than 80% of antihypertensive medication at one or more study visits during same time period.

^§Uncontrolled BP defined by mean SBP ≥140 mm Hg and/or mean DBP ≥ 90 mm Hg between study visits from month 6 to 28.

SDIM of SBP was higher among nonadherent compared to adherent participants (11.4 ± 4.9 versus 10.5 ± 4.5; P<0.001; Table 2). After full adjustment, nonadherent participants had 0.8 (95% CI, 0.7, 1.0; P<0.001) higher SDIM of SBP than adherent participants. Also, compared to adherent participants, those who were nonadherent had higher standard deviation and average real variability of SBP. This same pattern was present when restricting the sample to 11,290 participants without antihypertensive medication changes (Supplemental Table S3). The association between adherence status and VVV of SBP was consistent across antihypertensive drug randomization assignment (P>0.8 for interaction term for all definitions of VVV of SBP; Supplemental Table S4). Nonadherent participants also had higher VVV of DBP (Supplement Table S5).

Table 2.

Mean and Adjusted Differences in Visit-to-Visit Variability of Systolic Blood Pressure by Antihypertensive Medication Adherence Classification (N=19,790)

Mean and adjusted differences in VVV of SBP	Adherent* N=16,878; 85.3%	Nonadherent† N=2,912; 14.7%	P-Value
Standard deviation independent of the mean
Mean (SD)	10.5 (4.5)	11.4 (4.9)	<0.001
Adjusted difference (95% CI)
Model 1‡	0 (ref)	0.8 (0.6, 1.0)	<0.001
Model 2§	0 (ref)	0.8 (0.6, 1.0)	<0.001
Model 3||	0 (ref)	0.8 (0.7, 1.0)	<0.001
Standard deviation
Mean (SD), mm Hg	10.4 (4.9)	11.8 (5.5)	<0.001
Adjusted difference (95% CI)
Model 1‡	0 (ref)	1.3 (1.1, 1.5)	<0.001
Model 2§	0 (ref)	1.2 (1.0, 1.4)	<0.001
Model 3||	0 (ref)	0.8 (0.6, 1.0)	<0.001
Average real variability
Mean (SD), mm Hg	11.4 (5.9)	13.0 (6.8)	<0.001
Adjusted difference (95% CI)
Model 1‡	0 (ref)	1.6 (1.3, 1.8)	<0.001
Model 2§	0 (ref)	1.4 (1.2, 1.7)	<0.001
Model 3||	0 (ref)	1.0 (0.8, 1.2)	<0.001

Abbreviations: VVV of SBP, visit-to-visit variability of systolic blood pressure; SD, standard deviation; CI, confidence interval

^*Adherent participants reported taking more than 80% of all antihypertensive medication at all study visits in early time period (study visits from month 6 to 28)

^†Nonadherent participants reported taking less than 80% of antihypertensive medication at one or more study visits during same time period.

^‡Model 1 includes adjustment for age, race/ethnicity, gender, geographic region of practice site, and practice type.

^§Model 2 includes variables in model 1 and body mass index, smoking status, estimated glomerular filtration rate, diabetes, history of MI or stroke, history of coronary revascularization, history of other atherosclerotic cardiovascular disease, major ST depression, left ventricular hypertrophy, baseline systolic and diastolic blood pressure, use of blood pressure medications prior to study randomization, and randomization assignment (chlorthalidone, amlodipine, or lisinopril).

^||Model 3 includes variables in model 2 and the following variables measured during study visits conducted between 6 and 28 months post-randomization: mean systolic blood pressure, antihypertensive medication classes taken beyond the primary randomized drug, mean number of hypertension medications prescribed, change in antihypertensive medication regimen, and statin use.

Overall, 4.6% of participants had two or more visits with less than 80% adherence. SDIM of SBP was higher among nonadherent participants versus adherent participants according to this more stringent categorization of nonadherence (11.0 ± 4.6 versus 10.6 ± 4.6; P=0.01). After full multivariable adjustment, SDIM of SBP was 0.5 (95% CI, 0.2, 0.9; P=0.001) higher among nonadherent versus adherent participants (Supplemental Table S6). The same pattern of results was found when using the other definitions of VVV of SBP and VVV of DBP (Supplemental Table S7).

Participants who were nonadherent in both the early and late study periods had higher SDIMs of SBP compared with those who were adherent in both study periods (Table 3). There was minimal change in the SDIM of SBP between the early and late study periods for participants who were adherent in both study periods and nonadherent in both study periods. A substantial number of participants, however, had a change in adherence between the early and late study period with 6.5% switching from adherent to nonadherent and 10.0% switching from nonadherent to adherent. Compared to participants who were adherent in both time periods, participants who changed from adherent to nonadherent had an increase in SDIM of SBP (0.9, 95% CI 0.5, 1.3; P<0.001) while participants who changed from nonadherent to adherent had a decrease in SDIM of SBP (−0.7, 95%CI −1.0, −0.3; P<0.001).

Table 3.

Mean and Adjusted Differences in Within-Person Change in Standard Deviation Independent of the Mean of Systolic Blood Pressure Between Month 6 to 28 (Early Time Period) and Month 32 to 56 (Late Time Period) According to Change in Adherence Category Between the These Time Periods (N=11,521)

Mean and adjusted differences in within-person change in SDIM of SBP between time periods	Adherent† during both periods (n=9235)	Nonadherent‡ during both periods (n=385)	Adherent during early period, nonadherent during late period (n=753)	Nonadherent during early period, adherent during late period (n=1,148)
SDIM of SBP
Early time period Mean (SD)	10.3 (4.3)	11.2 (4.8)	10.8 (4.8)	11.2 (4.6)
Late time period Mean (SD)	10.2 (4.5)	11.0 (5.3)	11.7 (5.3)	10.5 (4.9)
Change in SDIM of SBP*
Mean (SE)	−0.1 (0.1)	−0.2 (0.3)	0.9 (0.2)	−0.7 (0.2)
Adjusted change in SDIM of SBP* (95% CI); p-value
Model 1§	0 (ref)	−0.1 (−0.7, 0.5); p=0.78	1.0 (0.6, 1.4); p<0.001	−0.6 (−0.9, −0.2); p=0.002
Model 2||	0 (ref)	−0.1 (−0.7, 0.5); p=0.72	1.0 (0.6, 1.4); p<0.001	−0.6 (−1.0, −0.2); p=0.001
Model 3¶	0 (ref)	−0.3 (−0.9, 0.3); p=0.32	0.9 (0.5, 1.3); p<0.001	−0.7 (−1.0, −0.3); p<0.001

Abbreviations: SDIM of SBP, standard deviation independent of the mean of systolic blood pressure; SD, standard deviation; SE, standard error; CI, confidence interval

^*Change in SDIM of SBP was calculated as SDIM of SBP during late period minus SDIM of SBP during early period

^†Adherent participants reported more than 80% of all antihypertensive medication at all study visits over specified time period

^‡Nonadherent participants reported taking less than 80% of antihypertensive medication at one or more study visits over specified time period.

^§Model 1 includes adjustment for age, race/ethnicity, gender, geographic region of practice site, and practice type.

^||Model 2 includes variables in model 1 and body mass index, smoking status, estimated glomerular filtration rate, diabetes, history of MI or stroke, history of coronary revascularization, history of other atherosclerotic cardiovascular disease, major ST depression, left ventricular hypertrophy, baseline systolic and diastolic blood pressure, use of blood pressure medications prior to study randomization, and randomization assignment (chlorthalidone, amlodipine, and lisinopril).

^¶Model 3 includes variables in model 2 and the following variables measured during study visits conducted between 6 and 28 months post-randomization: mean systolic blood pressure, antihypertensive medication classes taken beyond the primary randomized drug, mean number of hypertension medications prescribed, change in antihypertensive medication regimen, and statin use.

Among participants in the primary analysis without a cardiovascular event before the 28 months visit (N=18,442), being in the highest versus lowest quintile of SDIM of SBP was associated with increased risk of fatal CHD or non-fatal MI, stroke, heart failure, and all-cause mortality after multivariable adjustment (Table 4). In a mediation analysis, further adjustment for adherence status did not explain the association between SDIM of SBP and any of our cardiovascular or mortality outcomes.

Table 4.

Hazard Ratios for Cardiovascular and Mortality Outcomes Associated with the Highest Versus Lowest Quintile of Visit-to-Visit Variability of Systolic Blood Pressure (Standard Deviation Independent of the Mean) Before and After Adjustment for Adherence (N=18,442)^*

Outcome	Model 1† Without adjustment for adherence Hazard ratio (95% CI); p-value	Model 2‡ With adjustment for adherence Hazard ratio (95% CI);p-value	% Mediation (95%CI)	P-value comparing model with versus without adherence
Fatal CHD or non-fatal MI	1.24 (1.01,1.53); p=0.04	1.24 (1.01,1.52); p=0.04	−1.5% (−16.3%, 14.2%)	0.45
Stroke	1.43 (1.05,1.95); p=0.02	1.43 (1.05,1.94); p=0.02	−0.1% (−7.8%, 7.2%)	0.96
Heart failure	1.41 (1.11,1.80); p=0.004	1.42 (1.12,1.81); p=0.004	1.8% (−2.0%, 8.4%)	0.27
All-cause mortality	1.56 (1.30,1.87); p<0.001	1.56 (1.30,1.87); p<0.001	−0.3% (−2.9%, 1.9%)	0.71

Abbreviations: CHD, coronary heart disease; MI, myocardial infarction; CI, confidence interval

^*Participants who had an outcome before the month 28 visit were excluded. Participants who reported taking more than 80% of all antihypertensive medication at all study visits in early time period (study visits from month 6 to 28) were categorized as adherent; participants who reported taking less than 80% of antihypertensive medication at one or more study visits during same time period were categorized as nonadherent.

^†Model 1 includes the following covariates from baseline: age, race/ethnicity, gender, geographic region of practice site, practice type, body mass index, estimated glomerular filtration rate, diabetes, smoking status, total cholesterol, history of myocardial infarction or stroke, history of coronary revascularization, history of other atherosclerotic cardiovascular disease, major ST depression, left ventricular hypertrophy, low high-density lipoprotein cholesterol, baseline systolic and diastolic blood pressure, use of blood pressure medications prior to study randomization, and statin use. Covariates obtained during study period months 6 to 28 include antihypertensive medication regimen being taken at each visit, randomization group, mean systolic blood pressure, mean number of antihypertensive drugs prescribed, and change in antihypertensive regimen

^‡Model 2 includes all the covariates in Model 1 plus medication adherence status.

Discussion

In this post-hoc analysis of ALLHAT data, self-reported medication nonadherence was associated with higher VVV of BP. These results were consistent for VVV of SBP, VVV of DBP, for three commonly used metrics of VVV of BP, remained present after adjustment for differences in participant characteristics, and were consistent for participants randomized to chlorthalidone, amlodipine, and lisinopril. Change in adherence was associated with change in VVV of BP, further supporting an association between antihypertensive medication nonadherence and higher VVV of BP. Although there was a statistically significant association between medication nonadherence and VVV of BP, nonadherence did not explain the association between VVV of BP and cardiovascular and mortality outcomes.

The finding that antihypertensive medication adherence was associated with VVV of BP is consistent with two prior studies. The first study involved a sample of 1,391 patients taking antihypertensive medication enrolled in a Medicare managed care program.¹⁹ In that study, SDIM of SBP was 1.2 units higher in nonadherent as compared with adherent patients. That study was restricted to BP measurements abstracted from patient charts; BP was not measured using a standardized protocol. The second study assessed the association between adherence and VVV of BP in the African American Study of Kidney Disease and Hypertension (AASK) Trial.²⁰ Among 988 participants with established chronic kidney disease in AASK, self-reported and pill count measures of nonadherence were associated with higher VVV of BP.

The current study extends these prior findings in several ways. ALLHAT includes a racially diverse population, and hence, increases the generalizability of the finding that nonadherence is associated with increased VVV of BP. The current study also shows that individuals who changed their adherence to BP medications had changes in their VVV of BP. This suggests that consistent use of BP medications may reduce VVV of BP. Third, the current study shows that the association between adherence and VVV of BP is independent of drug type, and accordingly, that differences in the effects of nonadherence to specific antihypertensive medications on BP may not be a substantial factor related to VVV of BP.

While the current study confirmed that VVV of BP is associated with cardiovascular outcomes and mortality, it also demonstrated that nonadherence was not a mediator of this association. Thus, improving adherence is unlikely to offset the increased risk associated with VVV of BP found in treated hypertensive patients. VVV of BP has been associated with adverse prognosis in both treated and untreated hypertensive patients.^{2, 5} Given the lack of mediation, other pathways are likely to underlie the adverse influence of VVV of BP on outcomes in both patient groups.²¹ Small experimental studies suggest that large fluctuations in BP are associated with endothelial injury and end-organ damage.²² There is a need for additional studies to elucidate the mechanisms underlying VVV of BP and its effect on cardiovascular outcomes.

The current findings should be interpreted in the context of several limitations. First, adherence was assessed using a self-report rather than an objective measure. Prior studies comparing self-reported adherence with electronic or biochemical (blood and urine assays for drugs and metabolites) measures of adherence have demonstrated that individuals over-report their adherence.^23–25 Accordingly, estimates of the prevalence and frequency of nonadherence may be conservative. However, nonadherence on this measure maintained strong associations with higher SBP, higher DBP, and uncontrolled BP. Further, self-reported adherence has been correlated with objective measures including pill count, pharmacy fill rates, and electronic monitoring.²⁶ These data suggest the adherence measure used in ALLHAT was valid. To our knowledge, there are no large studies with repeated measures of BP and cardiovascular prognosis that have data on electronically-measured adherence to antihypertensive medication. In this context, reliance on self-reported measures of adherence, while imperfect, is the best available option for studying whether nonadherence is a mediator of the association between VVV of BP and prognosis among individuals with medication-treated hypertension. We did not assess day-to-day changes in adherence to antihypertensive medications; adherence is a dynamic process in many patients and variability in adherence may be more robustly associated with VVV of BP than the summary classification of adherence used in this study. As with all randomized trials, participants may not be representative of the general population. ALLHAT, however, was a large simple trial that attempted to replicate real world practice, and the physiology of VVV of BP is unlikely to be different in ALLHAT compared to the general population. Participants’ BP was closely monitored and medications were carefully titrated to achieve and maintain SBP/DBP <140/90 mm Hg during follow-up. This may have reduced the extent of VVV of BP in the sample. Estimates of SDIM were derived from the distribution of BP in ALLHAT, and therefore caution should be applied in generalizing the current results to other populations. A substantial number of participants were excluded from these analyses as they had fewer than five visits with assessment of antihypertensive medication adherence. The association of VVV of BP with fatal CHD or non-fatal MI, stroke, and all-cause mortality was similar in the current study as previously reported in ALLHAT.²⁰ While VVV of BP was associated with a larger hazard ratio for heart failure after excluding people with fewer than five visits with assessment of adherence, an increased risk for heart failure was present in both the current and prior analyses.

Perspectives

Results from the current study suggest that antihypertensive medication nonadherence contributes to VVV of BP, and that efforts to reduce nonadherence have the potential to decrease VVV of BP. The difference in the VVV of BP attributable to nonadherence was similar to or greater than the difference observed between various types of antihypertensive drug classes.^{9, 10, 27, 28} However, nonadherence did not mediate the association between higher VVV of BP and increased risk for CVD outcomes. Further work is needed to examine both the mechanisms underlying the association between VVV of BP and CVD outcomes and whether decreasing VVV of BP can improve outcomes.

Novelty and Significance.

1) What Is New?

• Medication nonadherence was associated with increased visit-to-visit variability of blood pressure (VVV of BP).

• Medication nonadherence did not explain the association between VVV of BP and cardiovascular risk.

2) What Is Relevant

• Few studies have tested the association between medication adherence and VVV of BP.

• Prior studies have not assessed whether nonadherence mediates the association between VVV of BP and prognosis.

3) Summary

• Although medication nonadherence contributed to VVV of BP, nonadherence did not explain why individuals with higher VVV of BP were at increased cardiovascular risk. Additional research is needed to understand how VVV of BP influences prognosis.