Proton Pump Inhibitors and Risk of Dementia: A Hypothesis Generated but Not Adequately Tested

Abstract

We reviewed the evidence on proton pump inhibitors (PPIs) and dementia. PPIs are among the most widely utilized drugs in the world. Dementia affects roughly 5% of the population of the United States (US) and world aged 60 years and older. With respect to PPIs and dementia, basic research has suggested plausible mechanisms but descriptive and analytic epidemiological studies are not inconsistent. In addition, a single large-scale randomized trial showed no association. When the evidence is incomplete, it is appropriate for clinicians and researchers to remain uncertain. Regulatory or public health authorities sometimes need to make real-world decisions based on real-world data. When the evidence is complete, then the most rational judgments for individual patients the health of the general public are possible At present, the evidence on PPIs and dementia suggests more reassurance than alarm. Further large-scale randomized evidence is necessary to do so.

We reviewed the evidence on proton pump inhibitors (PPIs) and dementia. PPIs are among the most widely utilized drugs in the world. ¹ It is important to note that the World Health Organization has placed omeprazole, one of the first marketed PPIs, on their List of Essential Medications. ² PPIs sold over the counter (OTC) are far less expensive and generally used to treat frequent heartburn by decreasing gastric acid secretion, while prescription brands are used to treat gastroesophageal reflux disease, esophagitis, and peptic ulcer disease. ³

With respect to short-term use, PPIs are well tolerated and have relatively low rates of reported side effects, which include headache, nausea, diarrhea, abdominal pain, constipation, dizziness, fatigue, rash, and pruritus. ⁴ As regards their long-term use, it should be noted that chronic kidney disease (CKD) has been reported to be increased by 25-50% among users of PPI in observational data from the Atherosclerosis Risk in Communities (ARIC) study, a prospective cohort. ⁵ While observational data are useful to test hypotheses for moderate to large effects, they are hypothesis formulating for small to moderate effects.^6-8 This is because the amount of uncontrolled and uncontrollable confounding inherent in such designs may be as large as or larger than the effect sizes being sought. For pharmacologic therapies, confounding by indication is a particularly vexing issue. For example, nonsteroidal anti-inflammatory drugs are a major risk factor for CKD which may explain, at least in part, the observed finding. ⁵

Several years ago, descriptive data contributed to the formulation of the hypothesis that PPIs were associated with dementia. Dementia affects roughly 5% of the population of the United States (US) and world aged 60 years and older. Due, at least in part, to the aging of the populations of the US and the world, the number of people with dementia in the world is projected to increase to 82 million in 2030 and 152 million by 2050. ⁹

In this mini-review, we address the available evidence on PPIs and dementia and conclude that existing totality is incomplete. Basic research has suggested plausible mechanisms but the descriptive and analytic epidemiological studies are, not surprisingly, inconsistent. In addition, perhaps the most informative data were derived from a single large-scale randomized trial showing no association.

Basic research has suggested that the pathogenesis of Alzheimer’s disease (AD) involves the accumulation of amyloid-beta species in the brain. Amyloid-beta species is derived from the sequential cleavage of the amyloid precursor protein by beta and gamma secretases. ¹⁰ In one study contributing to the formulation of the hypothesis, PPIs, specifically lansoprazole, affected the metabolism of amyloid, both in vivo and in vitro. ¹⁰

Claims data are collected in very large sample sizes for administrative purposes and are, therefore, at best, hypothesis generating. Observational analytic studies are hypothesis testing for large effects but hypothesis generating for small to moderate effects.^6-8 Randomized trials are the most reliable design strategy to test hypotheses of small to moderate effects.^6-8

In a large pharmacoepidemiologic database of 73,679 patients aged 75 and over, free from dementia at baseline and currently on prescribed PPIs, were found to have small to moderate increased risks of dementia. Specifically, the hazard ratio (HR) was 1.44 (95% confidence interval (CI), 1.36–1.52. ¹¹

In another study, 3484 individuals age 65 years and older were screened for dementia every 2 years using time-varying PPI exposure and total standardized daily doses (TSDDs), dispensed to an individual in the prior 10 years. Over a mean follow-up of 7.5 years, 827 participants (23.7%) developed dementia. Subjects who were either self-selected for or were prescribed PPIs had no increased risk of dementia. The HR and 95% CIs for specific levels of cumulative exposure as measured by TSDDs when compared to no use were .87 (65-1.18) for 365 days or less (< 1 year), .99 (.75-1.30) for 366-1095 days (1-3 years), and 1.13 (.82-1.56) for 1096-1825 days (3-5 years). ¹²

In the Aging, Cognition and Dementia in Primary Care Patients (AgeCoDe) study of 3327 patients aged 75 years and older, 431 developed dementia and 260 developed Alzheimer’s disease. After controlling for the available data on confounders, compared to non-users, patients who self-selected for or were prescribed PPIs had small to moderate significantly increased risks of dementia (HR = 1.38, 95% CI: 1.04-1.83 and AD (1.44, 1.01-2.06). ¹³

In a subgroup of 13,864 women in the Nurses’ Health Study II who had completed a self-administered computerized neuropsychological test battery, the mean score difference for PPI users versus non-users was −.06; (−.11 to .00) a finding which achieved significance (P = .03). After control for the self-selection for H2 receptor antagonists (H2RA) use, the magnitude of this difference was attenuated and no longer significant. ¹⁴

In a study of research volunteers with normal or mild cognitive impairment (MCI) at baseline, 884 had self-selected for PPIs at every visit, 1925 took PPI’s intermittently, and 7677 did not ever take the drug. When compared with never users, those who self-selected for PPIs at every visit had a significantly lower decline in cognitive function (.78, .66-.93) and lower risk of conversion to MCI or AD (.82, = .69-.98). Further, when compared with never users, those who self-selected for intermittent use of PPIs had significantly lower declines in cognitive function (.84, = .76-.93) and conversion to MCI or AD (.82, .74-.91). This lower risk was also apparent among those persons with normal or MCI. ¹⁵

Meta-analyses of randomized trials are hypothesis testing when the component trials are designed a priori to test the same hypothesis but otherwise, hypothesis generating.^6-8 Meta-analyses of observational studies are always hypothesis generating because the individual studies are no longer randomized. In meta-analyses of observational studies, chance is always reduced but bias and confounding are also always introduced. A meta-analysis of 11 observational studies included 642,949 subjects (64% women; 158,954 who self-selected or were prescribed PPIs). Treatment durations ranged from 5 to 10 years. The pooled HR was 1.10 (.88-1.37) for dementia and 1.06 (.72-1.55) for AD. ¹⁶

In a population-based nested case-control analysis within a large population-based claims dataset in Taiwan, elderly patients prescribed PPIs had no increase in AD either overall or by dose or duration. ¹⁷

The one published randomized trial directly testing whether PPIs are associated with dementia is a 3 × 2 partial factorial double-blind design of 17,598 participants with stable cardiovascular disease and peripheral artery disease. The independent comparisons for the PPI hypothesis include 8791 assigned at random to pantoprazole 40 mg daily and 8809 given a placebo. Patients were treated and followed for a median of 3.01 years with 53,152 patient-years of follow-up. There were no significant differences between those assigned at random to pantoprazole or placebo with respect to dementia. Specifically, the odds ratio was 1.2 (.81-1.71) (P = .36). Data were collected every 6 months on a large number of safety variables. There was no statistically significant difference between pantoprazole and placebo in the proportion of participants who experienced prespecified non-cardiovascular in pneumonia, fracture, new diagnosis of diabetes mellitus, chronic kidney disease, dementia, chronic obstructive lung disease, or gastric atrophy but there was a possible but non-significant 33% increase in enteric infections (1.01-1.75). ¹⁸

Conclusion

The available evidence on PPIs and dementia is incomplete. These considerations emphasize to clinicians, policy makers, and regulatory authorities the crucial need to consider strengths and limitations of different types of evidence. When the evidence is incomplete, it is appropriate to remain uncertain. This position, however, while tenable for clinicians treating patients, may not be for regulatory or public health authorities who sometimes need to make real-world decisions based on real-world data in the absence of a complete totality of evidence.^6-8 This position, however, may not be tenable for regulatory or public health authorities who sometimes need to make real-world decisions based on real-world data.^6-8

The hypothesis has been formulated but not adequately tested. Perhaps the most reliable evidence derives from one published randomized trial directly testing this hypothesis and showing no statistically significant association.

In considering the evidence for any formulated hypothesis, we believe that clinicians as well as regulatory and public health authorities rely on the results of randomized trials because they provide the most reliable evidence upon which to detect the most plausible small to moderate benefits or harms. In contrast, there should not be undue reliance on observational data for small to moderate effects which can only be hypothesis generating. Descriptive data in general and claims data, in particular, regardless of the effect size, are at best, useful to formulate but not test hypotheses, and, at worst, grossly misleading.^6-8

When the evidence is complete, then clinicians can make the most rational judgments for individual patients and policy makers for the health of the general public. At present, the existing evidence on PPIs and dementia, while incomplete, suggests more reassurance than alarm. Further large-scale randomized evidence is necessary to distinguish between the hypothesized small to moderate hazard and no effect.