Proton Pump Inhibitors Are Associated With Higher Risk of Cardiovascular Disease and Heart Failure

Abstract

Objective:

To examine associations of cumulative exposure to proton pump inhibitors (PPIs) with total cardiovascular disease (CVD; comprised of stroke, coronary heart disease, and heart failure [HF]) and HF alone in a cohort study of white and African American participants: the Atherosclerosis Risk in Communities Study.

Patients and Methods:

PPI use was assessed via pill bottle inspection at Visit 1 (January 1, 1987– 1989) and up to 10 additional times before baseline (Visit 5; 2011–13). We calculated cumulative exposure to PPIs as days of use from Visit 1 to baseline. Participants (n=4346 free of total CVD at Visit 5; mean age=75) were followed for incident total CVD and HF events through December 31, 2016. We used Cox regression to measure associations of PPIs with total CVD and HF.

Results:

After adjustment for potential confounding variables, participants with a cumulative exposure to PPIs of >5.1 years had a 2.02-fold higher risk of total CVD (95% confidence interval [CI], 1.50–2.72) and a 2.21-fold higher risk of HF (95% CI, 1.51–3.23) than nonusers.

Conclusions:

Long-term PPI use was associated with twice the risk of total CVD and HF compared to nonusers. Our findings are in concordance with other research and suggest another reason to be cautious of PPI overuse.

INTRODUCTION

Proton pump inhibitors (PPIs) are used for treatment of gastrointestinal acid-related disorders, such as heartburn, and are among the most commonly used drugs in the world. PPI use in the United States doubled from 3.9% of adults in 1999 to 7.8% in 2012.^{1, 2}

In addition to high prevalence of use, studies consistently indicate that PPIs are overused.³ In a study of hospital inpatients in Michigan, United States, only 10% of patients on PPIs had an acceptable indication.⁴ This problem extends outside the United States, where a large fraction of patients taking PPIs do not meet their country’s criteria for taking the drug (63%, 33%, and 67% in Australia,⁵ Ireland,⁶ and the UK,⁷ respectively).

Particularly important given its high prevalence and overuse, PPIs have been implicated as a risk factor for cardiovascular disease (CVD).⁸ In murine models, PPIs increased the level of asymmetrical dimethylarginine ⁹ and, subsequently, endothelial dysfunction.¹⁰ Under this mechanism, it would likely take months or even years for PPIs to affect vascular health because atherosclerosis, which is caused by endothelial dysfunction, is a slowly progressing disease. Thus, it is important to account for cumulative exposure to PPIs. Yet, many human studies, most of which reported a positive association between PPIs and CVD, as detailed in the discussion section, have not. Additionally, most studies of PPIs and CVD were in predominantly white cohorts and did not report heart failure (HF) separately. Examining HF separately may be important because PPIs and HF may be linked through a unique mechanism. Specifically, PPIs have been found to depress cardiac contractility in vitro.¹¹ Therefore, we examined associations of cumulative exposure to PPIs with incidence of total CVD (defined as a composite outcome of stroke, coronary heart disease [CHD], or HF) and HF only in a population-based cohort study with white and African American participants. Secondary outcomes were incident stroke and CHD. These were not included as primary outcomes due to the low number of events and, thus, low power to detect an association with PPIs.

METHODS

Study Population

The Atherosclerosis Risk in Communities (ARIC) Study is an ongoing, community-based cohort study designed to examine risk factors for CVD.¹² In 1987–1989 (Visit 1, as early as January 1, 1987), ARIC recruited and examined 15 792 participants aged 45 to 64 years living in 4 U.S. communities: Forsyth County, NC; Jackson, MS; suburban Minneapolis, MN; and Washington County, MD. Subsequently, ARIC contacted participants annually (twice annually since 2012) by telephone and periodically conducted examination visits. ARIC’s Visit 5 (2011–2013), attended by 6538 participants, was the first time point at which both PPI use was common and potential confounding variables were measured. Therefore, ARIC’s Visit 5 served as the baseline visit for the present study.

Of the 6538 participants at baseline, we excluded individuals who did not complete the medication questionnaire, which ascertained PPI use (n=228); had a history of CVD (CHD, HF, or stroke) at baseline (n=1,154); attended Visit 5 before the 2011 annual phone call (excluded due to complications of calculating cumulative exposure where this happened; n=22); were African American from Washington County or Minneapolis suburbs (excluded due to small numbers; n=18); were of a race other than African American or white (again, excluded due to small numbers; n=16); had missing data at baseline for variables in main analysis (n excluded = 681: 188 missing diabetes status, 469 missing smoking status, 12 missing systolic blood pressure, and 12 missing total cholesterol); or were missing outcome ascertainment (n=73). Thus, our final sample size for main analyses was 4346. The Institutional Review Boards of the collaborating institutions approved ARIC, and all participants gave informed consent.¹² Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research. All authors had access to the study data and reviewed and approved the final manuscript.

Measurement of Exposure

The use of PPIs was measured at all ARIC in-person visits (Visit 1 [1987–1989], Visit 2 [1990–1992], Visit 3 [1993–1995], Visit 4 [1996–1998], and Visit 5 [2011–2013]) through direct visual inspection of pill bottles, which could include over-the-counter and prescription medications used during the preceding two weeks. Participants also reported their medication names from prescription bottles as part of the annual telephone calls from 2006 through 2011, during which use of PPIs was identified. More specifically, participants were asked to assemble all medications they were currently taking and to “read the names of all the medications prescribed by a doctor…please do not include over-the-counter medications unless prescribed by a doctor.”

Measurement of Potential Confounding Variables

We describe the measurement of potential confounding variables in the Supplemental Material.

Ascertainment of Outcomes

Outcomes were ascertained from Visit 5 (2011–2013) through December 31, 2016. Incident HF was defined as the first occurrence of either a 1) hospitalization that included a primary or secondary diagnosis of International Classification of Diseases (ICD)-9th Revision discharge code of 428 (428.0 to 428.9) or 2) a death certificate with an ICD-9 code of 428 or an ICD-10 code of I50 among the listed underlying causes of death.^{13, 14}

Incident stroke was defined as definite or probable stroke, as described previously.¹⁵ In brief, for patients hospitalized for potential strokes (any type), ARIC abstractors recorded signs and symptoms and photocopied neuroimaging (computed tomography or magnetic resonance imaging) and other diagnostic reports. Using criteria adopted from the National Survey of Stroke,¹⁶ definite or probable strokes were classified by computer algorithm and separate review by a physician, with disagreements resolved by a second physician.

Incident CHD was defined as a validated hospitalization¹⁷ for definite or probable myocardial infarction or a definite CHD death. The criteria for definite or probable myocardial infarction were based on combinations of chest pain symptoms, electrocardiographic changes, and cardiac biomarker levels.¹⁸ The criteria for definite fatal CHD were based on chest pain symptoms, history of CHD, underlying cause of death from the death certificate, and any other associated hospital information or medical history, including that from an ARIC study clinic visit.¹⁸ Out-of-hospital deaths were investigated by means of death certificates and, in most cases, by an interview with ≥1 next of kin and a questionnaire completed by the patient’s physician. Coroner reports or autopsy reports, when available, were abstracted for use in validation.

An incident total CVD event was defined as the first occurrence of 1) HF; 2) a definite or probable stroke; or 3) CHD, defined as a definite or probable myocardial infarction or definite fatal CHD. Notably, sudden cardiac death is included in this definition of CVD, which includes a sizable portion of arrhythmic deaths.

For a supplemental analysis of electrocardiographic left ventricular hypertrophy, Cornell voltage (SV3 + RaVL) at Visit 5 was derived from 12-lead electrocardiograms and dichotomized as left ventricular hypertrophy (yes, no) using sex-specific criteria (>28 mm men; >22 mm women), as described in detail elsewhere.¹⁹

Statistical Analyses

We detail statistical analyses in the Supplemental Material. Briefly, we calculated the cumulative exposure to PPIs as days of use before baseline, which was ARIC’s Visit 5. Using these PPI categories, we examined baseline participant characteristics by PPI exposure. We calculated person-years of follow-up as time elapsed from baseline to whichever came first: any CVD event (ie, CHD, HF, or stroke), loss to follow-up, death, or the administrative end of follow-up (ie, 12/31/2016). Notably, when we analyzed subtypes of CVD, we censored at the time of the first event. We calculated incidence rates for each outcome by dividing the number of events by person-years. We used Cox regression to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for total CVD and HF by cumulative exposure to PPIs. We performed all analyses using SAS statistical software, version 9.4.

RESULTS

The mean age of ARIC participants at baseline (ARIC Visit 5) was 75 years, and approximately 40% were male. Compared to participants who never used PPIs, participants with a history of PPI use at baseline had less formal education and were more likely to be physically inactive, non-smokers, non-drinkers, taking antihypertensive medications, have diabetes, taking lipid medications, and taking aspirin (Table 1). We observed a marked increase over time in use of PPIs (Visit 1 [1987–1989]: 0%, 2011 annual phone call: 23%), as Visit 1 was coincident with market introduction, and PPIs became more accessible in 2003, when they became available over-the-counter (Figure 1).

TABLE 1.

Baseline Participant Characteristics by Exposure to Proton Pump Inhibitors, Atherosclerosis Risk in Communities Study, 2011–2013^a,b,c

	Exposure to proton pump inhibitors before baseline
Baseline characteristics	0 d (n=3211)	1 d-3.8 y (n=378)	3.9–5.1 y (n=381)	>5.1 y (n=376)
Age (y)	75±5	75±5	75±5	76±5
African American	20	28	21	11
Male	42	33	39	31
More than a hivschool education	56	49	49	49
Physical activityd (highest quartile)	19	13	13	14
Current smoker	7	5	4	3
Current drinker	53	44	49	47
Dietary pattern scones (higiest quartile)
Western	25	23	27	25
Prudent	24	27	28	29
Body mass index (kg/m2)	28±5	29 ±5	30±5	29 ±6
Systolic blood pressure (mm Hg)	130±18	130±18	130±18	132±18
Antihypertensive medication use	69	75	81	79
Diabetes	28	30	32	35
Total cholesterol (mg/dL)	186±42	184±44	179±40	182±40
High-density lipoprotein cholesterol (mg/dL)	53±14	52±13	51±13	52±13
Lipid medication use	51	57	64	61
Aspirin use	65	74	74	73
H2 blocker use	6	10	4	6

^aData are presented as mean ± standard deviation for continuous variables and percentage for categorical variables.

^bTo convert cholesterol values to mmol/L multiply by 0.0259.

^cThe number shown for each column is the maximum number. Numbers are lower for certain variables because of missing data.

^dMeasired at ARIC visit I (1987 to 1989). not baseline (ie. ARIC visit 5: 2011 to 2013).

Figure 1.

Prevalence of Proton Pump Inhibitors at Each Time Point, Atherosclerosis Risk in Communities Study

We characterized the last-observation-carried-forward method, which we used to deal with missing PPI measurements when creating the days of PPI use variable, in Supplemental Table 1. No one was using PPIs at Visit 1. At Visits 2–4, almost all those missing PPI measurements were assumed to be nonusers based on a previous visit, which is a safe assumption given the low prevalence of use at these time points. The biggest gap between PPI measurements was from Visit 4 (1996–1998) to the 2006 annual phone call, which is an approximately 9-year gap. In 2006, 87% of participants (3769 out of 4346) did not have PPI use ascertained. Thus, their most recent previous PPI ascertainment was assumed to still be true. By the 2007 annual phone call, this percentage was down to 23% (985 out of 4346). From this time point onward, most PPI measurements were not carried forward for more than 5 years.

During the 18,820 person-years of follow-up (median 4.6, maximum 5.6 years), there were 374 incident total CVD events and 215 HF events (Table 2). Crude incidence rates of total CVD per 1,000 person-years were similar among the lowest three exposure categories for PPIs. Participants with more than 5.1 years of exposure to PPIs had a total CVD incidence rate of 37 per 1000 person-years, which is about twice as high as the other categories. This trend in crude incidence rates was observed for HF as well.

TABLE 2.

HRs for Incident Total Cardiovascular Disease and Heart Failure by Cumulative Proton Pump Inhibitor Use, Atherosclerosis Risk in Communities Study, 2011–2016

	Exposure to proton pump inhibitors before baseline
Outcome	0 d (n=3211)	1 d-3.8 y (n=378)	3.9–5.1 y (n=381)	>5.1 y (n=376)
Total cardiovascular disease
No. of total cardiovascular disease eventsa	247	37	35	55
No. of person-yearsb	13,934	1664	1735	1487
Crude incidence rate/1000 person-years (95% Cl)	18 (16–20)	22 (16–31)	20 (14–28)	37 (28–48)
Model lc HR (95% Cl)	1 (reference)	1.25 (0.88–1.76)	1.13 (0.79–1.61)	2.11 (1.57–2.83)
Model 2d HR (95% Cl)	1 (reference)	1.22 (0.86–1.73)	1.10 (0.77–1.57)	2.21 (1.65–2.97)
Model 3e HR (95% Cl)	1 (reference)	1.19 (0.84–1.68)	1.04 (0.73–1.48)	2.02 (1.50–2.72)
Heart failure
No. of heart failure events	143	21	17	34
No. of person-yearsb	13.934	1664	1735	1487
Crude incidence rate/1000 person-years (95% Cl)	10 (9–12)	13 (8–19)	10 (6–16)	23 (16–32)
Model lc HR (95% Cl)	1 (reference)	1.22 (0.77–1.93)	0.95 (0.57–1.57)	2.25 (1.55–3.27)
Model 2d HR (95% Cl)	1 (reference)	1.18 (0.75–1.87)	0.92 (0.56–1.52)	2.41 (1.65–3.51)
Model 3e HR (95% Cl)	1 (reference)	1.18 (0.75–1.88)	0.88 (0.53–1.46)	2.21 (1.51–3.23)

^aTwenty-one participants had 2 different types of events on the same day. and I participant had 3 on the same day.

^bWe calculated person-years of follow-up as time elapsed from the baseline date to vyhichever came first: any cardovasciiar disease event (ie coronary heart disease heart failure or stroke), loss to follow-up. death, or administrative end of follow-up (ie. December 31. 2016).

^cUnadjusted.

^dAdjusted for demographic characteristics: age sex. and race.

^eAdjieted for Framingham cardiovascular disease risk factors: model 2 covariates. diabetes, smoking status, systolic blood pressure antihypertensive medication use. total cholesterol and high-density lipoprotein cholesterol

Long-term PPI use was associated with twice the risk of total CVD and HF compared to nonusers (Figure 2 and Table 2). Specifically, before adjustment for potential confounding variables, participants with a cumulative exposure to PPIs of >5.1 years had a 2.11-fold higher risk of total CVD (Model 1 95% CI, 1.57–2.83) and a 2.25-fold higher risk of HF (Model 1 95% CI, 1.55–3.27) than nonusers. Adjustment for CVD risk factors only slightly attenuated associations (Model 3 HR for total CVD, 2.02; 95% CI, 1.50–2.72 and Model 3 HR for HF, 2.21; 95% CI, 1.51–3.23). In a sensitivity analysis (data not shown), these associations persisted where we assumed that, if a participant reported taking a PPI at a visit or annual call, it was only taken for half the time until the next visit or phone call versus the whole time, as in main analyses. Associations persisted after additional adjustment for potential confounding variables: lipid medication use, physical activity, diet, education, body mass index, aspirin use, drinking status, and cumulative exposure to H₂-blockers (Supplemental Table 2).

Figure 2.

Adjusted† Hazard rations for incident total cardiovascular disease and heart failure by proton pump inhibitor use, Atherosclerosis Risk in Communities Study, 2011–2016.

†Model 3 adjusted for Framingham cardiovascular disease risk factors: age, sex, race, diabetes, smoking status, systolic blood pressure, antihypertensive medication use, total cholesterol, and high-density lipoprotein cholesterol.

In a supplemental analysis, we assessed whether use of PPIs of >5.1 years compared to no use was also associated with left ventricular hypertrophy at Visit 5. There was no association (Model 3 odds ratio for left ventricular hypertrophy: 0.92; 95% CI, 0.52,1.62; data not shown).

Cumulative exposure to PPIs of >5.1 years was also associated with an approximately 80% higher risk of CHD and stroke compared to nonusers (Model 3 HR for CHD, 1.77; 95% CI, 0.97–3.25 and Model 3 HR for stroke, 1.78; 95% CI, 0.92–3.44) (Supplemental Table 3). Notably, there is less precision to these effect estimates than the PPI-total CVD and PPI-HF effect estimates, and the lower confidence interval is close to null, while the upper confidence interval is a non-null value of high practical importance.

As a post-hoc supplementary analysis, we additionally modeled our exposure as any PPI use before Visit 5 baseline (Supplemental Table 4). PPI ever use was associated with an approximately 40% higher risk of total CVD and HF compared to never use (Model 3 HR for total CVD, 1.38; 95% CI, 1.11–1.71 and Model 3 HR for HF, 1.37; 95% CI, 1.03–1.82). These associations with PPI ever use persisted after additional adjustment for potential confounding variables: lipid medication use, physical activity, diet, education, body mass index, aspirin use, drinking status, and cumulative exposure to H₂-blockers (data not shown).

In another supplementary analysis, we excluded participants who self-reported angina at any point during the study (n=381 out of 4,346 [8.8%]). Comparing participants with a cumulative exposure to PPIs of >5.1 years to nonusers, participants without evidence of angina during the study had a CVD HR of 1.89 (95% CI, 1.33, 2.69) and HF HR of 2.02 (95% CI, 1.28, 3.20), only slightly lower than for the full study sample.

Finally, long-term (>5 years) use of H₂-blockers was associated with an approximately 10% higher risk of total CVD and HF. However, there is a level of uncertainty to these results due to the wide confidence interval: the lower confidence interval includes values that indicate no association and an inverse association, while the upper interval includes values that indicate a meaningful positive association (Model 4 HR for H₂-blockers and total CVD, 1.14; 95% CI, 0.83–1.57 and Model 4 HR for H₂-blockers and HF, 1.07; 95% CI, 0.69–1.65) (Supplemental Table 5).

DISCUSSION

Supporting our hypothesis, persons with long-term PPI use (>5 years) had twice the risk of total CVD and HF compared to nonusers. This association persisted after adjustment for potential confounding variables and was robust to a sensitivity analysis in which we changed assumptions about duration of exposure to PPIs.

The positive association of PPIs with total CVD is in line with other research findings. A meta-analysis of 17 randomized trials (7540 participants) found an association for any versus no PPI use (risk ratio, 1.70; 95% CI, 1.13–2.56).²⁰ Similarly, a meta-analysis of heterogeneous (I², 81%) observational studies also found a positive association for any versus no PPI use (443,284 participants; effect estimate, 1.25; 95% CI, 1.11–1.42).²¹ However, the studies included in the meta-analyses did not report HF separately, which might be important because PPIs and HF could be linked through a unique mechanism, and were in predominantly white cohorts. Also, the observational studies did not fully adjust for potential confounding variables and did not account for duration or cumulative exposure to PPIs. When we modeled our exposure as PPI ever use to match the observational meta-analysis, the PPI-total CVD HR of 1.38 more closely matched the meta-analysis’ effect estimate of 1.25.

Several potential biological mechanisms exist by which PPIs might cause CVD. However, we want to be clear that we are speculating, as the literature on this is still in its infancy, these potential mechanisms need to be confirmed through future study.

First, in murine models, PPIs increase the level of asymmetrical dimethylarginine⁹ and, subsequently, increase endothelial dysfunction.¹⁰ Endothelial dysfunction, in turn, promotes atherosclerosis over time.²² This conceptual framework suggests that cumulative, long-term PPI exposure could be important.

Secondly, there are mechanisms that may be unique to the relationship between PPIs and HF. For instance, limited evidence suggests that PPIs can depress cardiac contractility in vitro by reduction of Ca²⁺ signaling and myofilament activity.¹¹ It also might be speculated that PPIs could contribute to ventricular hypertrophy, but our supplemental analysis did not show an association between PPIs and electrographically measured left ventricular hypertrophy. If we had found an association with left ventricular hypertrophy, it would have been supportive of our hypothesis that PPIs and HF are related. However, the lack of association does not preclude an association with PPIs and HF because 1) electrocardiographic left ventricular hypertrophy is fairly specific for left ventricular hypertrophy, but not the same as HF and 2) most HF cases accumulated later in follow-up.

Moreover, there is evidence that PPIs may cause low serum magnesium through reduced intestinal magnesium absorption,^23–25 which could lead to CVD outcomes. A meta-analysis of observational studies, most of which used magnesium levels <1.7 mg/dL to define hypomagnesemia, reported that PPI users had a 43% greater risk of hypomagnesemia compared to nonusers (95% CI: 1.08, 1.88).²⁴ In absolute terms, approximately 15.6% of PPI users had hypomagnesemia in a large, cross-sectional study compared to 11% of nonusers.²⁶ Hypomagnesemia, in turn, may be a risk factor for HF^{27, 28} and CHD^29–32 in the general population.

Studies of the PPI-CVD association that have parsed out stroke, CHD, and/or HF have generally reported positive associations. A heterogeneous (I², 81%) meta-analysis of observational studies reported a 21% higher risk of myocardial infarction in PPI users versus nonusers (95% CI, 1.09–1.34).²¹ Two observational studies reported that PPI users had a 13 to 36% higher risk of stroke than nonusers,^{33, 34} but another study reported no association after adjustment for lifestyle factors and indication.³⁵ Furthermore, in an observational study of patients with CHD, PPI use was associated with an increased risk of HF and death (HR, 5.71; 95% CI, 1.63–20.04).³⁶

There are unique strengths to this study. The ARIC Study offers almost complete capture (ie, minimal loss to follow-up) of validated cardiovascular outcomes, including CHD, HF, and stroke. It also has research-quality measurement of numerous potential confounding variables. Further, ARIC includes both white and African American participants, whereas most other research on the PPI-CVD association was in predominantly white cohorts. Finally, this design captures cumulative exposure to PPIs and, because PPIs were not introduced into the market until 1989, we are not missing PPI use prior to the study commencing.

Limitations

Study limitations also warrant discussion. This is an observational study, which unlike a trial does not have the benefit of randomization to control confounding. Therefore, it is possible that residual confounding remains and could cause a spurious association. In particular, confounding by indication may be present because gastroesophageal reflux disease, the most common indication for PPIs, could be associated with CVD.³⁷ However, PPI use of more than one year was associated with CVD even among patients with gastroesophageal reflux disease. It is unknown if gastroesophageal reflux disease per se is associated with CVD or merely other factors that are associated with it. Although we could not control for gastroesophageal reflux disease because it was not captured in ARIC, we did control for many other potential confounding variables that are associated with gastroesophageal reflux disease, such as body mass index.

An alternative, non-causal explanation for the observed association between PPIs and CVD could be that PPI use is simply a surrogate for angina. In other words, some patients having angina might mistakenly believe they have gastroesophageal reflux disease and taken PPIs as treatment.³⁸ To address this, we excluded participants who self-reported “angina, angina pectoris, or chest pain due to heart disease” at any point during the study (n=381 out of 4,346 [8.8%]) as a supplemental analysis. Point estimates were only slightly attenuated: comparing participants with a cumulative exposure to PPIs of >5.1 years to nonusers, participants without evidence of angina during the study had a CVD HR of 1.89 (95% CI, 1.33, 2.69) and HF HR of 2.02 (95% CI, 1.28, 3.20) compared to CVD and HF HRs of 2.02 (95% CI, 1.50–2.72) and 2.21 (95% CI, 1.51–3.23), respectively, in the total study sample. However, because this analysis only excluded self-reported angina, there may be other patients with unreported or undiagnosed angina that was mistaken for gastroesophageal reflux disease and hence used PPIs, but because the angina was not reported, they were not excluded from this secondary analysis. Thus, the true attenuation of the association may be larger than was observed. Additionally, we examined the relationship of H₂-blockers, a medication commonly taken for gastroesophageal reflux disease, with total CVD and HF. We reasoned that no association of H₂-blockers with total CVD and HF would enhance the argument for a specific relation of PPIs with total CVD and HF being causal. Our findings on H₂-blockers do not support a strong association between H₂-blockers and CVD, based on a HR of 1.14. Although there was a range of uncertainty to this HR estimate for H₂-blockers, as reflected by a confidence interval, an association with CVD restricted to PPIs lends credence to our findings.

PPI use was self-reported, over-the-counter PPI use was not ascertained at annual telephone calls, and we were not able to ascertain dose or frequency of use. Also, there was an approximately 9-year gap between PPI ascertainment at Visit 4 to the 2006 annual follow-up, and we assumed that the measurement at Visit 4 held true until the next time PPI use was ascertained. When we instead assumed use for half of the gap between measurements, the associations of PPI use with total CVD and HF remained. Finally, some participants were missing PPI measurements at various time points and we carried the last measurement of PPI status forward. Because PPI use became exponentially more prevalent over time, our last-observation-carried-forward technique may have misclassified some PPI use. Assuming that PPI misclassification did not differ according to future CVD, it would tend to bias associations towards the null. However, if PPI misclassification differed by an important confounding variable, the association might be biased unpredictably.

Another limitation of this study is that analyses were limited to participants who returned for ARIC’s Visit 5 and did not have CVD at this time. On average, people who participated in ARIC past Visit 1 had a lower burden of CVD risk factors than all ARIC participants (data not shown). Additionally, among those who attended Visit 5/baseline, exposure to PPIs was positively associated with CVD before baseline (data not shown). Selective attrition could have created a downward bias of the HR, meaning the true association between PPI use and CVD could be stronger than we observed.

Finally, we unfortunately were not able to characterize incident HF events as to whether they were from reduced versus preserved ejection fraction.

CONCLUSION

In conclusion, we found that long-term PPI use is associated with twice the risk of total CVD and HF compared to nonuse in a cohort of white and African American participants in the United States. Although there are limitations to our measurement of PPI use, findings are in concordance with other research and suggest another reason to be cautious of PPI overuse.

Abbreviations and Acronyms:

ARIC

Atherosclerosis Risk in Communities

CHD

coronary heart disease

CVD

cardiovascular disease

HF

heart failure

ICD

International Classification of Diseases

PPIs

Proton pump inhibitors