Abstract: Proton pump inhibitors (PPIs) are effective and suppressing acid, and therefore have efficacy against gastric acid-related disorders. The long-term safety of PPIs is less clear and there have been a number of studies raising concerns regarding risk of pneumonia, fracture, Clostridium difficile, chronic renal failure, and dementia. This latter concern is addressed by a study in this issue of AJG using health care registry data and found there was no association between PPI use and Alzheimer’s dementia. Furthermore, there was no increased risk of dementia with long-term use of PPIs or higher doses of PPIs. Discrepancies between studies probably relate to multiple testing and residual confounding and currently there is insufficient evidence to suggest that the association between PPIs and dementia is causal.
Proton pump inhibitors (PPIs) have been one of the major innovations in gastroenterology over the last 30 years. They are extremely effective in treating gastro-esophageal reflux disease (1) and indications have expanded to include diverse conditions such as prevention of peptic ulcer bleeding and functional dyspepsia (2). PPIs have also been extremely lucrative for the pharmaceutical industry earning over $25 billion each year globally (https://www.fiormarkets.com/report/global-proton-pump-inhibitor-drug-market-research-report-54096.html). Extessnsive randomized controlled trial data have shown these drugs are safe with adverse event rates very similar to placebo in the short term (1). Success stories in medicine often have a cycle of enthusiastic uptake followed by a period where caution is urged as reports regarding potential harm start to accrue. PPIs have been no exception to this phenomenon. The emergence of medical administrative databases such as the UK General Practice Research Database have allowed the evaluation of possible long term adverse effects of drugs using large numbers of patients with long term outcome data. Exploration of these databases suggested a link between PPIs and pneumonia (3) and this was soon followed by associations of PPI use and fracture (4), C. difficile risk (5), ischemic heart disease (6), chronic renal failure (7), and even all-cause mortality (8). The latest concern has been the association between PPI use and risk of dementia (9). A study (9) of 73,679 participants over the age of 75 yaers registered with a German statutory health insurer found regular PPI users had an increased risk of developing dementia with a hazard ratio of 1.44 (95% confidence interval (CI) 1.36–0.152).
It is important for others to replicate the association of PPIs with dementia as worldwide 47 million people have some form of dementia, a heterogeneous group of conditions which result in significant decline in cognitive domains such as memory, executive function and social cognition (10). As the population ages, it is predicted that the number of people with dementia will increase to close to 75 million in 2030 and 132 million in 2050 (11). As a result of these projected figures, dementia is recognised by the World Health Organization as a public health priority (12). Given the immense societal, financial and emotional cost of the increased number of persons living with dementia, it is important the we identify any drugs that might be contributing to these disorders, particularly Alzheimer’s disease (AD) as this accounts for 60-70% of cases diagnosed.
The association between PPIs use and AD was evaluated in a nested case–control study using a Finnish nationwide health care registry reported by Taipale et al. (13) in this issue of AJG. There was no statistically significant association between PPI use and AD (adjusted odds ratio for ≥3 years of PPI use=0.99; 95% CI=0.94 to 1.04). There was no increase in risk with longer duration of PPI use or higher dose of PPI therapy. To understand how two large database studies came to different conclusions, it is important to compare them so that we can obtain greater clarity as to whether there is an underlying causal effect between PPI therapy and dementia.
Both studies adjusted for a similar amount and type of confounding factors. The designs were different with one being a cohort study (9), whereas Taipale et al. (13) is a nested case control design. Usually, the cohort design would be considered the most robust approach if these were prospective research studies designed to answer this specific question. However, these both rely on retrospectively analysed registry data where information was collected without this specific hypothesis in mind. As such both designs have similar risks of bias and confounding. Taipale et al. (13) is by far the largest to date with 70,718 AD cases and 282,862 controls. Another strength of Taipale et al. (13) is that it evaluates the impact of a lag window between the use of PPI therapy and the onset of AD. The rationale for this is that no plausible biologic mechanism for PPIs causing dementia such as inducing B12 deficiency (14) or enhanced production of amyloid-B (15) that will act instantly and all will take years to develop the disease. Another reason for introducing lag period is that patients that are concerned about dementia may consult a clinician for other reasons initially as they do not want to address this sensitive topic. A patient that is embarrassed to talk about concerns around memory loss may initially bring up dyspepsia as a reason they want to see the doctor (after all 20% of the population have upper GI symptoms) and be prescribed a PPI well before the diagnosis of dementia is eventually made. The “hidden consultation” (16) is a well-known phenomenon and may be a confounding factor in the association with any drug and dementia if a lag period is not introduced. The authors analysed data without a lag period, with a 3-year and with a 5-year lag period. Whichever lag period was used, the association between AD and PPI use was not impressive, but there was an association between the highest doses of PPI use and AD and this could have led uncritical researchers to conclude there was a risk in this high dose group. Introducing a lag period of either 3 or 5 years completely nullified any impact of dose of PPI on risk of AD. Furthermore, the association between PPI therapy and AD was most prominent (although still very modest) in those taking PPI for <1 year and there was no difference in PPI use between those with and without AD in taking PPIs for 3 or more years. It is extremely unlikely that short term PPI use could cause dementia whereas long term PPI use would not so this suggest any association is due to bias or confounding factors.
Finally, Taipale et al. (13) specifically evaluated AD whereas the previous German study (9) assessed any cause of dementia and this difference in outcome measure may explain the divergent results. Taipale et al. (13) used the NINCDS—ADRDRA criteria (17). These were first published in 1984 and since then our knowledge of the clinical and neurobiological features of AD have advanced considerably. In addition, we now have a greater understanding of other causes of dementia such as Lewy body disease, vascular disease, and frontotemporal lobar degeneration and their distinguishing features, which can present with which can present with similar clinical features to AD (18). The diagnosis of AS relies upon a comprehensive clinical assessment and, due to the absence of a conclusive diagnostic test, definitive diagnosis still requires histopathological confirmation, which is typically only available at post mortem. As such, any clinical diagnosis of AD will be inaccurate and so evaluating specific types of dementia outside of post mortem studies will suffer from misclassification bias. This is unlikely to be the explanation for the AJG study being negative; however, a previous smaller cohort study (involving 3,327 patients) had found an association between PPI use and patients specifically diagnosed with AD, as well as any cause of dementia (19). This is not surprising, as AD accounts for up to 70% of all dementia diagnoses. Athough this study (19) reported a statistically significant association, the P-value was modest (P=0.04 for the association between PPI and AD) and the authors did not adjust for multiple testing. The cohort was set up to look for any risk of dementia and not PPIs specifically and the authors should therefore have adjusted their analysis for multiple testing. Had they done this, the results would not have been statistically significant but the authors can be forgiven for not making these adjustments. Studies that evaluate databases or cohort studies that evaluate a wide range of risk factors rarely adjust for multiple comparisons and those that do never take a conservative approach in this adjustment. If they did their results would almost always be negative and therefore less easy to publish and less likely to get cited.
The issue of multiple testing is a well-known problem. Ioannidis et al. (20) calculated that in genetic epidemiology it is likely to be that <1% of positive associations found between a candidate gene and a disease will turn out to be truly causal, because researchers do not sufficiently adjust for the play of chance when they are testing so many genes for so many diseases. This may explain, in part, why the promise of human genetics has so far failed to deliver in terms of dramatic improvements to human health. The same phenomenon is likely to occur in database studies where many drugs are evaluated for a huge number of adverse outcomes. Chance alone will dictate that some results will be positive and this will be amplified by confounding factors that will tend to drive an apparent association between PPIs and disease. Nearly all PPI studies have shown those that are prescribed PPIs are likely to be sicker than those not given these drugs (21, 22). If a person is ill from one (or a number of) disease(s), they are likely to succumb to others such as dementia. It is not that PPIs are the cause of dementia it is just that they are a marker for sicker patients that are more likely to get other diseases. Researchers adjust for this in the analyses, of course, but they cannot adjust for all possible confounders, particularly in administrative databases not set up to answer this specific question.
We live in an time where fame is highly valued and this may explain why current researchers may sensationalize their findings and are sometimes less willing to entertain the possibility that the association they report may be a chance finding or due to confounding or bias. There are many good studies evaluating the risks of PPI therapy but some papers do not sufficiently address the possibility that the association is spurious. The discussion section seems to be just a vehicle for some authors to point out a possible weakness and then congratulate themselves on why this is unlikely to impact on the results because their data are so robust. The reality is that, for weak associations, the data from administrative databases is rarely sufficiently robust to make strong claims regarding the harms of PPIs or any other drug and this is why these need addressing in randomized controlled trials. There is currently a large randomized controlled trial (23) underway that specifically addresses the possible harms of PPIs and the results of this are awaited with interest. Until then, the paper by Taipale et al. (13) is a timely reminder that we cannot be certain about many of the long-term harms of PPI therapy such as dementia.
PM conducts research evaluating the link between GI and psychiatric disease through the Inflammation, Microbiome and Alimentation: Gastro-intestinal and Neuropsychiatric Effects (IMAGINE) network (www.imaginespor.com). This is a Canadian Institutes of Health Research (CIHR) Strategy for Patient Oriented Research (SPOR) chronic disease network. The IMAGINE network is supported by CIHR (grant number SCA-145105), McMaster University, University of Calgary, University of Alberta, Alberta Innovates, Manitoba Research, Queen’s University, Dalhousie University, University of Montreal, Takeda Pharmaceutical Company, and Allergan Incorporated.