Risk of circulatory diseases associated with proton-pump inhibitors: a retrospective cohort study using electronic medical records in Thailand

Introduction

Proton-pump inhibitors (PPIs) have become a drug of choice for the treatment of acid-related disorders since Omeprazole was introduced in 1989. Following that, PPIs outperformed the effectiveness of other gastric-acid suppressants (e.g., histamine 2-receptor antagonists (H2RAs)).

In 2010, however, the U.S. Food and Drug Administration (FDA) warned manufacturers to include the risk for bone fractures on the product labels when a high dose or long-term use of PPIs was prescribed (Strand, Kim & Peura, 2017). Since then, many studies reported more potential risks of PPIs use, including renal impairment, infections, and cardiovascular diseases (CVD) (Corsonello & Lattanzio, 2018). A meta-analysis of randomized controlled trials showed that monotherapy of PPIs was associated with 70% increase in cardiovascular diseases (relative risk (RR) = 1.70, 95% Confidence Intervals 95% CI = [1.13–2.56]) (Sun et al., 2017). Likewise, a meta-analysis of observational studies reported that the rates of cardiovascular diseases in PPIs users were significantly higher than in non-PPIs users (Odds ratio (OR) = 1.54, 95% CI = [1.11–2.13]) (Shiraev & Bullen, 2018).

Regarding in vitro studies, the potential mechanism underlying the association between PPIs and CVD was the inhibition of the nitrate-nitrite-nitric oxide (NO) pathway. It was also observed that omeprazole prevented the effects of nitrite and nitrate; the rise of pH in the stomach inhibited NO and other NO-related species synthesis. This consequently diminishes a cardiovascular protective effect (Amaral & Tanus-Santos, 2017; Pinheiro et al., 2016). In addition to a recent in vivo study, the long-term use of PPIs might inhibit dimethylarginine dimethylaminohydrolase (DDAH) activity, impair endothelial NO production, and reduce vascular function (Ghebremariam et al., 2015; Nolde et al., 2021). This mechanism was possibly linked to the impairment of circulatory system, including ischemic strokes (IS) and peripheral vascular diseases (PVD).

Nevertheless, the association between PPIs use and IS remained controversial among large observational studies. For instance, Sehested et al. (2018) conducted a study among all registry-Danish populations who underwent an elective upper gastrointestinal endoscopy during 1997 to 2012, had a 29% (95% CI = [5–59%]) greater absolute risk of ischemic stroke than those people who had not used PPIs. On the contrary, Nguyen et al. (2018) studied sex-different cohorts of health professionals and found non-significant associations (Adjusted Hazard Ratio (aHR) = 1.08; 95% CI = [0.91–1.27]).

Noticeably, the results remained inconclusive because of the different populations included in each study. A recent systematic review and meta-analysis also observed that the Asian population reported a substantially higher effect of PPIs use on the risk of cardiovascular events than studies from other regions. However, there were only two Asian studies, which was not sufficient to decide whether this was a coincidence or an actual effect modification (Nolde et al., 2022). In addition, non-user of PPIs was applied as a comparator in many studies, which led to indication bias on estimated outcomes.

Moreover, the potential mechanism underlying the association between PPIs and impaired vascular function has not yet been studied for PVD as a primary outcome. Although there were systematic review and meta-analysis studies on PPIs use associated with CVD, results remained inconsistent owing to medium to high risk biases among the pooled studies (Nolde et al., 2022; Sherwood et al., 2015; Shiraev & Bullen, 2018).

Therefore, this study aimed to assess the association between PPIs use and circulatory diseases (CVD, IS, and PVD) using electronic medical records. A reproducible observational study of circulatory disease-related to PPIs use in the Asian population is required to address the aforementioned issues, whereas, using the real-world clinical database is to gain more meaningful in clinical practice.

Materials and Methods

Study design

A retrospective cohort study was conducted to compare patients, initiated by PPIs or H2RAs (as an active comparator). We identified all outpatient or inpatient visits, who were firstly prescribed either PPIs or H2RAs from October 1, 2010, to September 30, 2017. Patients were excluded if they had been prescribed either drug for 6 months before the entry year, which was the fiscal year 2011 (October 1, 2010, to September 30, 2011) or 2012 (October 1, 2011, to September 30, 2012) (Fig. 1).

Study population

Patient information was retrieved from medical and administrative databases at Songklanagarind Hospital, Thailand. A description of the study hospital information was provided in the Supplemental materials. The datasets consisted of inpatient, outpatient, medication, laboratory, and administrative data. For inclusion criteria, all patients who were over 20 years old and administered either PPIs or H2RAs during the entry years were recruited (Fig. 2).

Results

In the fiscal year 2011, 157,522 participants aged over 20 years were included in the study cohort. However, some participants were not eligible because of neither use of PPIs or H2RAs, incomplete sex, and date of birth, whereas, some participants were excluded because study outcomes occurred during the entry period to the index date. After exclusion, there were 59,322 participants left to follow up after their individual index dates (Fig. 2).

The baseline and sociodemographic characteristics of eligible participants are presented in Table 1. The proportion of PPI and H2RAs users aged >60 years was 33 and 27 percent, respectively. In addition, the percentage of PPI and H2RAs female users were mostly observed (63 and 67 percent, respectively). Of all participants, there were only 0.11% were diagnosed with diseases of the esophagus, stomach, and duodenum using PPIs and H2RAs. PPIs users who were diagnosed with hypertension at the baseline had a higher percentage than those who used H2RAs. Less than 1% of all participants were diagnosed with smoking and alcohol abuse. The frequency of hospital visits was higher among H2RAs users than among PPIs users (median (IQR): 4 (2–9) vs 3 (1,7), respectively), whereas, the average blood pressure among all participants was less than 130 mmHg and 80 mmHg for systolic and diastolic blood pressure, respectively (Table 1).

The number of primary outcomes (CVD, IS, and PVD) and the crude incidence rates of each event are shown in Table 2. The median follow-up time of each outcome was 0.25 years before and after PSM. Compared to H2RAs, the crude incidence rates of all study outcomes were higher among PPI users (CVD: 5.41 vs 8.28, IS: 1.22 vs 3.51, and PVD: 0.09 vs 0.89, for H2RAs vs PPIs).

Discussion

Our study was to observe the potential association between PPIs and circulatory diseases by each composite outcome (CVD, IS, and PVD). Regarding a recent systematic review and meta-analysis of observational studies, pooled HR for the association between PPIs and myocardial infarction was 1.05 (95% CI = [0.83–1.32]), whereas, pooled HR for the association between PPIs and acute cardiovascular event was 1.05 (95% CI = [0.96–1.15]) (Nolde et al., 2022). In our study, PPIs use, as compared with H2RAs, was significantly associated with CVD for the PSM model, while, it was insignificant for the stratified Cox-PH model. In our study, CVD, as a primary outcome, included heart failure, cardiac arrhythmia, and ischemic heart disease (or myocardial infarction).

Because we applied stratified CPH to unmatched data, it was assumed that the treatment effect was the same for all strata, particularly among covariates that violated the CPH assumption. However, stratified CPH might not perform well in terms of bias and power if there was a treatment-by-stratum interaction, while, treatment effects might truly differ across strata. In addition, conversion into a categorical variable as small strata as possible made stratified CPH less efficient and then led to bias estimation (In & Lee, 2019; Mehrotra, Su & Li, 2012).

According to PSM model with CVD outcome, though, the result showed a balanced covariate between PPIs and H2RAs users, the strength of association might be underestimated due to multiple biases. (Table S5) As compared with the unmatched population, changed baseline characteristics after PSM (Table S3) might affect the strength of association. To the best of our knowledge, therefore, the positive association should be interpreted deliberately.

Regarding ischemic stroke outcome, our results showed a positive association in both the unmatched and matched populations. As compared to our study, a recent systematic review and meta-analysis (Nolde et al., 2022) from five observational studies showed that the pooled HR was 1.08 (95% CI = [0.97–1.20]). Of all observational studies, however, there were conflicting results between Nolde et al. (2021) (HR = 0.98; 95% CI = [0.89–1.08]) and Wang et al. (2017) (HR = 1.11; 95% CI = [1.02–1.21]) studies, which were both assessed with moderate for overall biases. Our study resulted a significant and consistent HRs, regardless of unmeasured confounding effects from smoking and alcohol consumption (Figs. S5, S6), as well as, multiple biases with E-values for both unmatched and matched populations (Table S5).

Alternatively, the association between PPIs and peripheral vascular disease (PVD) was explored because of the potential mechanisms proposed in in vitro and in vivo studies. Additionally, long-term use of PPIs was proposed to inhibit dimethylarginine dimethylaminohydrolase 1 (DDAH1), with the accumulation of asymmetric dimethyl arginine (ADMA), while the lowering of nitric oxide (NO) as a vasoprotective molecule led to an increase in vascular cell proliferation, platelet adhesion and aggregation, and inflammation (Ghebremariam et al., 2015; Tommasi et al., 2017). Former evidence also linked PPIs intake and cardiovascular risk by lowering NO production in endothelial cells (Nolde et al., 2021), whereas, oxidative stress causing peripheral vascular disease might be associated with endothelial dysfunction by reducing NO bioavailability (Ismaeel et al., 2020). Although this study showed a positive association between PPIs and PVD, the effect size was large due to the small number of PVD events.

In our study, over 90% of PPIs given to patients were omeprazole because it was the first drug of choice for ulcer healing and variceal bleeding, recommended by the Thai national drug list (Ministry of Public Health, 2022). As compared to esomeprazole, dexlansoprazole, and lansoprazole, 8 weeks or longer use of omeprazole was more associated with cardiovascular events (Sun et al., 2017). This was relevant to our result that participants who were more likely to persistently use of PPIs during a 1-year observation period (MPR > 50%), had more likelihood of being CVD, IS, and PVD.

The major strength of our study is that patient data from a university hospital were collected systematically, while new users (NU) and active comparators (AC) were performed. For the NU approach, the time-varying hazard could be assessed, whereas, the temporality of covariate assessment was preserved (Lund, Richardson & Sturmer, 2015). The objective of selecting AC was to mitigate confounding by indication and other unmeasured participant characteristics such as baseline health status, frailty, and assignment mechanism to treatment (Lund, Richardson & Sturmer, 2015; Yoshida, Solomon & Kim, 2015). Although H2RAs were not completely replaced by PPIs indications, this was the best active comparator we had, based on recommendations in the Thai National Drug List (Ministry of Public Health, 2022).

In addition, we recruited participants who were given either PPIs or H2RAs and not only underwent endoscopy but were also used for other reasons in real-world clinical practice, such as concomitant use of PPIs with NSAIDs to prevent potential GI irritation. Since we were concerned about adherence to medication among the participants, we applied the MPR that the more persistent the use of PPIs, the higher risk of developing circulatory diseases was observed.

However, this study had some limitations. The medians of the follow-up period were short because we applied an on-treatment scheme with the right censor due to the switch of treatments to reflect the real-world use of drugs in data analysis. In addition, we did not have information on the precise indication for PPIs’ prescription. CVD diagnosis included myocardial infarction (MI), which sometimes presented with gastrointestinal manifestations. Therefore, PPIs and H2RAs may be administered to patients who first present with MI. This would lead to protopathic bias where MI was not diagnosed, and then confirmed and recorded in electronic medical records after patients were prescribed either PPIs or H2RAs (Chui et al., 2022). According to baseline characteristics of participants, however, there were less than 1% of eligible participants who were diagnosed with gastrointestinal-related disease, while, the average baseline blood pressure among participants ranged in normal (below 130/80 mm.Hg). Having said that, there was a slight effect of the misclassification of outcomes between MI and gastrointestinal symptoms in this study.

Additionally, exposure to PPIs was identified using dispensed prescriptions, whereas, self-treatment of PPIs information was not recorded, which might lead to misclassification of exposure. On-treatment with the exclusion of the immortal period (time from the entry date to the index date) was prone to selection bias. Unmeasured confounders, particularly smoking, alcohol consumption, physical activity, and individual BMI, may affect the estimation of HR for circulatory diseases based on Array-based and E-values for multiple-bias sensitivity analysis.

Missing information on outcome measures could not be avoided because there was the loss of follow-up among patients in the study hospital while those patients were diagnosed later by other hospitals. The lack of generalizability of the study was still a limitation because an only hospital was studied. Although there is no reliability test for ICD-10, all ICD codes were directly assigned from the physician based on the patient’s conditions. Finally, we cannot exclude the power issue with H2RAs in this study, as most patients were treated with PPIs. For further research, we suggest exploring a potentially explainable mechanism of PPIs associated with circulatory diseases through mediators anconfirming the association between PPIs and PVD. In clinical practice, the risks and benefits of PPIs use, including combination with other drugs, must be weighed out, whereas, the necessity of PPIs use should be reassessed during the treatment plan of patients who are at risk of circulatory diseases.

Conclusion

In conclusion, the effects of PPIs associated with ischemic stroke and peripheral vascular disease were significant in a Cox proportional hazards and propensity score matching analysis in a Thai hospital-based cohort. This is the first study to report an association between PPIs use and PVD, meanwhile, we provided alternative results from the Asian population to compare with results from other regions.

Supplemental Information