With broad advertisement and over-the-counter availability, global proton pump inhibitor (PPI) use is common (10%–21%) for the treatment of gastroesophageal reflux disease (GERD).^1,2 Unlike histamine 2 receptor antagonists (H2RA), PPIs impair endothelial nitric oxide synthase and vascular function that lead to inflammation, GERD, and thrombosis, and they may impair cardiac function.³ With conflicting evidence, there is ongoing concern about whether PPIs increase the risk of cardiovascular disease (CVD), the leading cause of adult death and disability worldwide.^2-15

Our nested case–control analyses of PPI use, and its recency and duration with first and second myocardial infarction (MI) and stroke events in the World Trade Center (WTC) General Responders Cohort, a multicenter open cohort, clarifies whether the relationship is causal. Evidence regarding the timing of PPI use and its association with CVD is scant with mixed results.^5,7,10,13,15 Five studies assessing the timing of PPI use included patients using antiplatelet agents, which confounds the associations of PPI and incident of CVD because antiplatelet agents are used for secondary prevention of recurrent CVD and treatment of other related conditions that may precipitate incident myocardial infarction and stroke.^5,7,10,13 These studies are also flawed by the lack of an unexposed comparison group,⁵ comparison of cases to themselves before PPI use,^7,10 or very short–term assessment of medications use that cannot fairly assess causality of chronic conditions such as CVD.^5,7,15

Without access to medical records, this study assessed self-reported physician’s diagnosis or treatment for first and second MI as well as stroke events, conditions that account for most CVD mortality,¹² from the WTC Health Program standardized medical monitoring visits (scheduled every 12–18 months) data.¹⁶ Of 50,577 members providing written voluntary informed consent through December 31, 2021, 47,070 had essential demographic information. The analyses excluded 3.1% (n 5 1462) reporting antiplatelet use, 1 person with unknown birth date, 7 who were <18 years at incident MI or stroke, and 412 with unknown earliest MI or stroke dates. Responders not reporting MI or stroke were classified as controls. Medication use reported in free-text fields was reviewed through visual inspection aided by Stedman Plus medical/pharmaceutical spell checker 2020. The recent analyses classified PPI use by its most recent use regardless of longer PPI use duration. The duration analyses classified PPI use by its longest duration. Medication dose was inconsistently reported and not analyzed. Those not reporting medication use were classified as nonusers.

Up to 4 controls were randomly selected from noncases and frequency matched without replacement to the case’s 5 year age group, sex (male or female), first visit body mass index (BMI <25, 25–<30 and ≥30), and the cases’ first and last monitoring visit dates plus/minus 3 years. Controls were assigned their matched case event (“index”) dates. When possible, the BMI estimate used the average of the first 2 visits’ height measurements. To retain continuity, second events are subsets of matched first events.

Fixed-effects, conditional, matched-pairs, logistic regression analyses were adjusted for GERD and H2RA use, major CVD comorbidity conditions, and for potential confounding before the index date.¹⁷ Where possible, incomplete event dates were imputed (month’s midpoint for day, June for month). All missing binary covariate data were derived using multiple imputation by logit chained equations except for “other heart disease,” which was coded to the modal value (0).¹⁸ Categorical missing was coded to their modal referent values (white n 5 55, nonsmoker n 5 15 and protective services n 5 211).

Sensitivity analyses included the following: (1) replaced duration with the percent of preindex visits when PPI and H2RA were reported to estimate their cumulative use; (2) limited index dates to 2007–2021 when event dates may be recalled more accurately; and (3) distinguished program certified GERD, potentially a more objective measure, from solely self-reported GERD.

Descriptive and matching analyses were conducted using SPSS 28.0.1.1(15) (IBM Corp, Armonk, NY), and multiple imputation and regression analyses were conducted using Stata (StataCorp LLC. 2019. Stata Statistical Software: Release 16. College Station, TX).

The principles outlined in the Declaration of Helsinki 1975 as revised in 2013 and national and institutional committees’ ethical standards on human experimentation were followed. The program’s 5 clinical centers’ and the data center’s Institutional Review Board at the Icahn School of Medicine at Mount Sinai approved the research.

Frequency matching produced substantively similar study groups. Few responders took PPI within 30 days before the index dates (Table 1). Most cases with very recent PPI use had long-term PPI use. Only very recent (≤30 days) PPI use was associated with higher MI and stroke risk; less recent and longer PPI use was associated with no or diminished risks (Table 2). Less adjusted and the sensitivity results were very similar (data not shown).

The study reveals biologically implausible associations, where only very recent PPI use increased the risk of MI and stroke, which are chronic diseases that generally develop over years. The results also counter a dose–response relationship as nearly all responders with very recent PPI use had much longer use, which was associated with null or lower risks. Both MI and stroke involve broad physiologic processes, and impending events may produce symptoms that prompt PPI use, providing some explanation of our and other studies’ associations identified with recent use.^3,17 Self-reported MI and stroke validity is generally good.²⁰ The ample longitudinal data supported matching and statistical adjustment that strengthened internal validity and minimized selection and followup bias. Adjustment for the major known CVD comorbidities and confounding provide indirect adjustment for other medication use. Except for being mostly male, the sample characteristics including the prevalence of major CVD risk factors are similar to the US population.¹⁷

Analysis of the recency and duration of PPI use, excluding people who have used antiplatelet therapy, in studies with unexposed referent groups could clarify the nature of the associations of PPI use and CVD and resolve the ongoing debate.