Introduction

In recent decades, gastrointestinal diseases have been on the rise in older populations [1]. In this demographic, proton pump inhibitors (PPIs) are commonly used to treat heartburn, gastroesophageal reflux disease, peptic ulcer disease, and Helicobacter pylori infections [2, 3]. In an Irish study among adults aged 40 years or older with intellectual disabilities, the largest population of PPI users was aged 50–64 years (53.4%), and 30.2% were aged 65 years or older [4].

Concerns of potentially severe adverse effects of PPIs have increased, particularly in older adults. A recent review demonstrated that the use of PPI in older adults was associated with an increased risk of osteoporotic-related fractures, Clostridium difficile infection, community-acquired pneumonia, vitamin B12 deficiency, kidney disease, and dementia [5]. These potential side effects are of particular concern to the elderly because this vulnerable population is already more likely to suffer from an increased risk of these conditions and consequently, severe morbidity. As a result, the guidelines recommend avoiding the use of PPIs for longer than 8 weeks in the elderly except in high-risk patients and discontinuing or reducing PPIs in older adults with more than 8-week usage of PPIs for uncomplicated peptic ulcer disease or erosive peptic esophagitis [5,6,7].

Additionally, recent studies have reported that PPIs were associated with increased mortality in the general population and/or patients with cancer [8,9,10,11]. The association between PPIs and mortality in the elderly population is undefined, but it has been shown that they are potentially affected [10, 11]. Since previous studies have yielded inconsistent results for mortality in older adults who take PPIs and the excess risk of death associated with PPI use has not been systematically investigated, pooled estimates that combined relative risks for mortality from each study are needed. Thus, we performed a systematic review and meta-analysis of the association between PPIs and mortality in older adults.

Methods

Literature search and study selection

We conducted searches on MEDLINE, EMBASE, and the Cochrane Library for articles published up to August 24, 2022. To search for relevant articles, we used MeSH terms and text words related to outcome, such as “mortality,” “death,” and “fatality,” and intervention-related search terms, including the drug, brand, and chemical names of “proton pump inhibitor” (i.e., benatoprazole, dexlansoprazole, esomeprazole, lansoprazole, omeprazole, pantoprazole, rabeprazole, and tenatoprazole). The search strategy is presented in Supplement Table 1.

Table 1 Study characteristics of included studies
Full size table

All studies that met the following criteria were included: (1) investigated the association between PPIs and the risk of mortality in adults aged 50 years or older; (2) compared PPIs and control such as placebo or active comparator; (3) reported the quantified relative risk of death; (4) conducted randomized controlled trials (RCTs), non-randomized controlled studies, and observational studies; and (5) written in English. If the participants’ age was not clearly mentioned in the inclusion criteria for the study, we included studies with participants of both a mean or median ≥ 66 years of age and an interquartile range of ≥ 56 years. Two independent reviewers examined the study selection, and the third reviewer was consulted when there was a disagreement. We registered the study protocol in PROSPERO (CRD42020179631) and followed the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines [12].

Quality assessment

Two independent researchers conducted a quality assessment of the included studies. For RCTs, we used version 2 of the Cochrane Risk of Bias (ROB 2) tool, which is composed of five items (bias arising from the randomization process, bias due to deviations from intended interventions, bias due to missing outcome data, bias in measurement of the outcome, and bias in selection of the reported result), to assess the quality of studies with three levels (low, some concerns, and high risk of bias) [13]. The Risk of Bias in Non-randomized Studies of Interventions (ROBINS-I) was used to evaluate the quality of non-RCTs. ROBINS-I consists of five items from three domains: pre-intervention, at-intervention, and post-intervention domains [14]. Pre-intervention domain includes bias due to confounding and bias in selection of participants into the study. At-intervention domain includes biases in classification of intervention, and post-intervention domain includes bias due to deviations from intended interventions, bias due to missing data, bias in measurement of the outcome, and bias in selection of the reported result. Four levels were used to assess the risk of bias in included studies: low, moderate, serious, and critical.

Data extraction

We collected information on the study and demographic characteristics (country, study design, data source, study period, mean age, percentage of male participants), exposure (definition and number of PPI users and controls, follow-up period), and outcomes (number of death of PPIs and comparators, relative risk [RR] of mortality) from the included studies. The confounding variables were extracted in regression analysis when applicable.

The primary outcomes were the unadjusted and adjusted estimates of the risk of mortality associated with PPI use. We used the best-adjusted relative risks with a 95% confidence interval (CI) after controlling the confounding variables from each included study for the meta-analysis.

Statistical analysis

To estimate the pooled RRs with CIs, the inverse-variance random effect model was used. We calculated log RR and standard error (SE) using the 95% CIs or P-values reported in the studies. Each study reported a different type of relative risk, such as RR, hazard ratio (HR), or odds ratio (OR). In our meta-analysis, HRs were considered RRs [15, 16], and ORs were converted to RRs using the method described by Zhang and Yu, which uses OR and the incidence of mortality in the control group [17]. The studies that reported OR were included in the meta-analysis when we could calculate RRs from the data on OR and the incidence of mortality in the control group. In addition, we performed subgroup and sensitivity analyses according to follow-up period, population, country, study design, the median age of included studies, and quality of the studies. We conducted a duration analysis of the risk of death among PPI users by follow-up period: ≤ 6 months, > 6 months– ≤ 1 year, > 1– ≤ 2 years, > 2– ≤ 3 years, > 3– ≤ 4 years, and > 4 years. To further evaluate disease-specific mortality, we evaluated the association between PPIs and mortality among patients with cancers, cardiovascular disease, and kidney diseases as well as people who were institutionalized and hospitalized. To assess the impact of different health care systems, the subgroup analysis by countries was conducted: USA/Canada, Europe, Australia, and Asia. The association of subgroups was also examined by patient’s age to take into account differences in mortality by age groups (≤ 75, 76–85, and > 85 years old) as well as by quality assessment of included studies (low, moderate, and serious/critical risk of bias) to incorporate assessment into the analyses. We also analyzed the association in studies that included and did not include Charlson Comorbidity Index (CCI) as a covariate.

Heterogeneity was assessed using the I2 test and Q statistic, with the significance of the Q-statistic test being considered at p < 0.05. Heterogeneity was considered for I2 values of more than 50% [13]. The funnel plot was used to estimate possible publication bias owing to the tendency to publish studies with positive results. We used Review Manager 5.3 software (The Nordic Cochrane Centre, The Cochrane Collaboration, 2014).

Results

Literature search

An initial search was performed on 15,202 articles; we identified 13,544 relevant ones after excluding 1658 duplicate articles (Fig. 1). In the screening process, 12,992 articles were removed during the title/abstract review, and 511 articles were excluded during the full-text review due to one of the following reasons: no elderly patients, no PPI therapy, ineligible study design, no comparator group available, no outcomes of interest reported, and non-original studies. Finally, 49 cohorts among 41 studies (four RCTs and 37 observational studies) were included in our systematic review [11, 18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57] since five studies included more than one cohort in their studies [19, 20, 23, 26, 50]. In the meta-analysis, 4 cohorts among 4 RCTs and 44 cohorts among 36 observational studies were included. One study/cohort was not included in the meta-analysis as we could not calculate RR from OR due to a lack of data on the number of deaths [29].

Fig. 1
figure 1

PRISMA flow diagram of study selections

Full size image

General characteristics of the included studies

We included four RCTs, 34 cohort studies, and three case–control studies (Table 1). In RCTs, PPIs including omeprazole and pantoprazole were included as interventions with comparators of placebo. In observational studies, most studies included all kinds of PPIs as interventions and non-PPI users as controls including H2RA. The studies were conducted in the USA, Canada, Europe, Australia, and Asia. The study participants included the following: patients with cancer (n = 9), cardiovascular disease (n = 11), kidney disease (n = 4), institutionalization (n = 6), and hospitalization (n = 5). All studies included 2,515,079 participants with a median age ranging from 67 to 96 years and the percentage of males ranging from 18.4 to 99.5%. Study follow-up periods were from 21 days to 13.8 years. When estimating adjusted HRs or ORs, baseline demographics, disease-related clinical factors, comorbidities, and medications were included as confounding variables (Table 1).

Quality assessment

The included RCTs had a low risk of bias from missing outcome data, measurement of the outcome, and selection of the reported result (Supplement Fig. 1). For measurement of the outcome, we determined a low risk because mortality was not affected by any measurement methods. We evaluated some concerns for deviations from intended interventions for half of the included RCTs because PPIs can also be used as over-the-counter (OTC) drugs.

All cohort studies had a low risk of bias from deviations from intended interventions, missing data, measurement of the outcome, and selection of the reported result (Supplement Table 2). Among 37 cohort studies, 10 studies had moderate, serious, or critical risk of bias because appropriate confounding covariates for the adjusted estimates were not included, and 19 had a moderate risk for classification of interventions because OTC use of PPIs could not be captured. The overall risk of bias was assessed as low at 27%, moderate at 57%, and serious or critical risk at 16%.

Table 2 Subgroup analysis of the association between proton pump inhibitors and the adjusted risk of mortality of observational studies
Full size table

Proton pump inhibitors and mortality in RCTs

Among 4 RCTs, the death rates were 11.8% in PPI users (1286/10,918) and 11.4% in nonusers (1250/10,928). The PPI use and risk of mortality in unadjusted RR among individuals aged 50 years or older were not significantly associated (RR 1.03; 95% CI, 0.91–1.16), with moderate heterogeneity was moderate (I2 = 45%, p = 0.14 (Fig. 2A). When estimating adjusted risk ratio among 20,859 elderly people from 2 RCTs, the association between PPI users and death was not significantly associated compared to nonusers (aRR, 1.01; 95% CI, 0.94–1.08), with 0% of heterogeneity (I2 = 0%, p = 0.60) (Fig. 2C).

Fig. 2
figure 2

The association between proton pump inhibitors and mortality. A Unadjusted risk ratio in randomized controlled trials. B Unadjusted risk ratio in observational studies. C Adjusted risk ratio in randomized controlled trials. D Adjusted risk ratio in observational studies

Full size image

Proton pump inhibitors and mortality in observational studies

In 26 observational studies (29 cohorts), there were 75,675 deaths in 512,263 PPI users (14.8%) and 94,428 deaths in 1,731,521 nonusers (4.9%). We found no significant association between the use of PPI and mortality from unadjusted RR in the elderly (RR 1.14; 95% CI, 0.90 − 1.45) (Fig. 2B). There was significant heterogeneity (I2 = 100%, p < 0.001), and funnel plots showed no evidence of publication bias (Fig. 3A). For adjusted estimates, 41 cohorts from 33 studies of the association between PPIs and death included 2,429,961 individuals aged 50 years or older. The use of PPI was significantly associated with a 15% increased risk of mortality compared to non-use (aRR, 1.15; 95% CI, 1.10–1.20) (Fig. 2D). Significant heterogeneity was detected (I2 = 93%, p < 0.001), and there was no evidence of publication bias based on the funnel plot (Fig. 3B).

Fig. 3
figure 3

Funnel plot of observational studies. A Unadjusted risk ratio. B Adjusted risk ratio

Full size image

Additionally, we conducted subgroup and sensitivity analyses of adjusted risk ratios among observational studies. PPI use among the elderly was significantly associated with an increased risk of mortality (by 17–32%) compared to non-use in the follow-up period between more than 6 months and less than 4 years, while the association was not significant in a shorter follow-up of a less than 6 months (aRR, 1.02; 95% CI, 0.45–2.33) and ˃ 4-year follow-up (aRR, 1.06; 95% CI, 0.99–1.14) (Table 2). The mortality risk of PPI users among patients with cancer, cardiovascular diseases, kidney diseases, and institutionalization showed a similar trend. When investigating the risk of death by country, the association between PPIs and increased mortality was significant among studies conducted in Europe (aRR, 1.27; 95% CI, 1.18–1.38), while the association was towards significant among studies conducted in USA/Canada, Asia, and Australia. The significant association between PPI use and mortality appeared consistent among cohort studies (aRR, 1.14; 95% CI, 1.09–1.20) and case–control studies (aRR, 1.31; 95% CI, 1.23–1.40). According to the median age of participants in the included studies, the results of the association between PPI use and mortality remain similar to the base-case analysis results. The lower the risk of bias of the included cohort studies, the larger the RR (aRR, 1.13; 95% CI, 1.01–1.28 in low risk of bias vs aRR, 1.08; 95% CI, 1.02–1.15 in serious/critical risk of bias). The association of risk of death and PPI users among both studies included CCI as a covariate and studies that did not include CCI as a covariate remained significant.

Discussion

This systematic review found a 15% increase in the risk of death among elderly people taking PPI compared to those not taking PPI. In particular, the association between PPI use and mortality was significant in studies among senior adults with cancer, cardiovascular diseases, and kidney diseases as well as studies with longer follow-ups and cohort studies.

Our results are aligned with those of previous studies that included not only adults but also elderly [9, 58]. A previous study using administrative data from Veterans Affairs demonstrated an excess risk of death among adults taking PPIs with a mean age of 61 years [9]. A systematic review also showed that all-cause mortality increased in adults treated with PPI, especially older adults treated with long-term PPI, compared to patients not treated with PPI [58]. Although a clear mechanism is not known for the association between PPIs and the risk of death, a potential biological mechanism was suggested. Prolonged exposure to PPI impairs endothelial function, increases oxidative stress, slows lysosomal acidification and protein accumulation in endothelial cells, and accelerates human endothelial aging by shortening telomere length [59]. PPIs also upregulate the expression of protein levels and mRNA resulting in increased activity of the heme oxygenase-1 enzyme in gastric and endothelial cells, which reduces beneficial effects, including antiapoptotic, anti-inflammatory, antioxidant, antiproliferative, and immunomodulatory effects in vascular cells [9, 60, 61]. Elderly patients may be more vulnerable to side effects during long-term PPI use on account of their comorbidities, use of multiple drugs, and poor nutrition [36]. PPIs have also been significantly overused in the elderly, contrary to the recommendation to restrict the PPI treatment period to fewer than 12 weeks [58].

Our subgroup and sensitivity analyses demonstrated that PPI use significantly affected the increased death among older patients with cancer, cardiovascular disease, and kidney disease. These results are similar to those of previous studies that were not limited to the elderly, which correlated PPI usage with excess mortality from cardiovascular disease, chronic kidney disease, and upper gastrointestinal and lung cancers [10, 62, 63] and also suggested an association between PPI usage and increased risk of the aforementioned diseases [62, 64]. The authors suggested a mechanism leading to the excess cause-specific mortality related to PPI use is linked to the exacerbation of underlying disease or the development of a new disease that increases the risk of cause-specific death. This mechanism was also suggested as a unifying mechanism [59]. In addition, inappropriate prescription of high-dose PPI during or after hospitalization is frequent as antithrombotic agents are widely used in elderly patients [36].

Notably, in cohort studies and studies with long-term follow-up (more than 1 year to 3 years), 28–32% of increased risk of death was clearly observed. However, this was not observed when the follow-up period was shortened (less than 6 months). A previous review also showed that there was no immediate apparent increase in all-cause mortality in adults taking PPIs in one RCT, while increased mortality with PPI use was observed in the observational studies followed up to 1 year [58]. Although RCTs have a higher hierarchy of an evidence-based approach than observational studies because of randomization, the findings from observational studies can show the results with all confounders, including unidentified confounders in the real-world population [58]. Our systematic review showed that significant results were obtained in terms of real-world evidence rather than in RCT settings, suggesting that real-world evidence should be considered when applied to clinical settings for patient care, such as pharmacovigilance criteria.

Our systematic review has strengths that, to our best knowledge, this is the first meta-analysis study to explore the association between PPI use and mortality in the older adult population with a large sample size. The significance of the pooled estimates was higher in high-quality studies with a low overall risk of bias. Our diverse subgroup results categorized by the follow-up period, disease, study design, country, and risk of bias provide comprehensive and detailed information. Notably, our results clearly showed that PPI use was significantly associated with increased death in older patients with cancer, cardiovascular disease, and kidney disease. In addition, we provide real-world evidence based on results from cohort studies and long-term follow-up studies.

Our study has several limitations. Primarily, we were not able to investigate the impact of mortality by the duration or dosage of PPI use due to a lack of data. However, our best available evidence supports the association the PPIs and mortality in elderly people. Secondly, the level of heterogeneity was low in RCTs, and it may be caused by study design or the small number of included studies. However, the pooled estimates in observational studies have significant heterogeneity although we conducted several subgroup analyses to find the cause. It may be caused by various populations, different levels of controlling covariates, and unknown factors. Despite the diversity of included studies, results were consistent across subgroups. Thirdly, half of the cohort studies in the meta-analysis showed a moderate risk of bias in the classification of intervention due to self-administrated OTC without prescriptions in most countries. However, the significant association between PPI use and mortality remained consistent in studies with low, moderate, and serious/critical risk of bias. Furthermore, the relative estimates can be considered conservative as individuals who may take OTCs may be included as both PPI users and nonusers. Finally, we cannot identify the causal relationship between PPI use and the risk of mortality since most of the included studies investigated the association between PPI use and the risk of mortality. Also, there was no clear association between PPIs and death from the four RCTs in the meta-analysis.

Conclusions

Our meta-analysis showed that the association between PPIs and the risk of mortality of RCTs was inconsistent with the association from observational studies in the elderly. The meta-analysis of RCTs showed that PPI use was not powered to detect an increased risk of death compared to nonusers. However, the increased risk of death was identified in observational studies and studies with long-term follow-ups, and the association was consistent in the subgroup population, including among patients with cancer, cardiovascular disease, and kidney disease. Our findings highlight the need to increase awareness of increased mortality due to PPI use and to restrict PPI prescriptions to elderly people, wherein the benefits outweigh the potential risks.