Prolonged Use of Proton Pump Inhibitors and Risk of Type 2 Diabetes: Results From a Large Population

It is still debated whether prolonged use of proton pump inhibitors (PPIs) might affect metabolic health.

To investigate the relationship between prolonged use of PPIs and the risk of developing diabetes.

We performed a case-control study nested into a cohort of 777 420 patients newly treated with PPIs between 2010 and 2015 in Lombardy, Italy. A total of 50 535 people diagnosed with diabetes until 2020 were matched with an equal number of controls that were randomly selected from the cohort members according to age, sex, and clinical status. Exposure to treatment with PPIs was assessed in case-control pairs based on time of therapy. A conditional logistic regression model was fitted to estimate the odds ratios and 95% CIs for the exposure-outcome association, after adjusting for several covariates. Sensitivity analyses were performed to evaluate the robustness of our findings.

Compared with patients who used PPIs for  2 years, respectively. The results were consistent when analyses were stratified according to age, sex, and clinical profile, with higher odds ratios being found in younger patients and those with worse clinical complexity. Sensitivity analyses revealed that the association was consistent and robust.

Regular and prolonged use of PPIs is associated with a higher risk of diabetes. Physicians should therefore avoid unnecessary prescription of this class of drugs, particularly for long-term use.

Proton pump inhibitors (PPIs) lead to long-lasting suppression of both basal and meal-stimulated acid secretion by irreversible inhibition of the H+/K+ ATPase (or proton pump) on gastric parietal cells (1). As a result of these pharmacological properties, they have become the first-choice therapy in patients with acid-related disorders such as gastroesophageal reflux disease, Barrett esophagus, and peptic ulcers and to prevent gastrointestinal bleeding while on nonsteroidal anti-inflammatory drugs (2). Because of their efficacy, the introduction of generic compounds and their use as over-the-counter medications in several states, the market for these drugs has progressively increased in the past 3 decades, placing them among the top 10 most commonly used medications worldwide (3).

In parallel, the expansion of the PPI market has seen increasing concerns regarding the misuse of these drugs in clinical practice (4), as well as the possible side effects (5). Several studies have identified various potential adverse reactions to their prolonged use including fractures, hypomagnesemia, gastric carcinoids, chronic kidney disease, dementia, and Clostridium difficile diarrhea (6-9). More recently, it became evident that PPIs can alter the normal bacterial milieu at the distal esophagus, stomach, small bowel, and colon (10). Importantly, changes in the gut microbiome have been postulated to play a role in the pathophysiology of metabolic diseases including obesity, insulin resistance (11), nonalcoholic fatty liver disease (12), and diabetes (13).

Nonetheless, clinical data on the association between PPI use and diabetes are limited. To our knowledge, no randomized clinical trial has been specifically designed to evaluate this possible link, but observational analyses performed in different ethnic groups showed conflicting results. Evidence on the topic has been recently summarized in a systematic review and meta-analysis including 3 cohort studies for a total of 244 439 participants. Although no significant association between PPI use and incident diabetes (pooled risk ratio, 1.10; 95% CI, 0.89-1.34; P = 0.385) was found, it should be stressed that a high degree of heterogeneity was identified (I2 = 93.5%), leading the authors to state that evidence was insufficient and inconsistent to make any definite conclusions (14).

The present study was therefore conceived to investigate the relationship between PPI use in terms of duration of and adherence to treatment and the risk of diabetes in the general population. To achieve these goals, we performed a large, nested case-control study in the real-world setting of the Italian Lombardy region.

The data used in the present study were retrieved from the Healthcare Utilization Databases of Lombardy, a region that accounts for almost 16% of Italy’s population (with > 10 million residents). In Italy, all citizens have equal access to health care provided by the National Health Service (NHS). An automated system of databases is used to manage health services in each Italian region. Healthcare Utilization Databases include a variety of information on residents, such as diagnosis at discharge from public or private hospitals, outpatient drug prescriptions, copayment exception for diagnosed chronic disease (including diabetes), and specialist visits and diagnostic examinations provided fully or partly free of charge by the NHS. These various types of data can be interconnected because a unique individual identification code is used by all databases for each NHS beneficiary. To preserve privacy, each identification code is automatically anonymized, and the inverse process is only allowed to the regional authority upon request of judicial authorities. Further details on Healthcare Utilization Databases in pharmacoepidemiological studies are available in previous studies (15, 16).

The target population included residents of Lombardy aged ≥ 40 years who were beneficiaries of the NHS. Of these, those who received at least 2 consecutive prescriptions of PPIs (ie, not more than 6 months apart) between 2010 and 2015 were identified, and the date of the second prescription recorded during this period was defined as the index date. The present analysis was deemed exempt by the review board at our institution, as the dataset used in the analysis was completely deidentified.

Exclusion criteria were as follows: patients who, within the 5 years before the index date (1) were not beneficiaries of the NHS; (2) received at least 1 PPI drug prescription; (3) received at least 1 prescription of histamine H2 receptor antagonist; and (5) had shown signs of presence of diabetes (ie, diagnosis of diabetes, exemption for diabetes, and/or at least 1 antidiabetic drug prescription).

The remaining patients were included into the cohort whose members accumulated person-years of follow-up from the index date until the earliest date among the onset of diabetes (see the following section), death, emigration, or August 31, 2020.

When the effect of time-dependent exposure on rare events is investigated by means of large health care databases, the case-control study design is a useful alternative to the cohort design, achieving similar results with superior computational efficiency (17).

A case-control study was nested into the cohort of PPI drug users. The outcome of interest was the diagnosis of diabetes, whose date of onset was defined as the date corresponding to the first event among (1) hospitalization with diagnosis of diabetes, (2) prescription of antidiabetic drug, or (3) activation of the copayment exception for diabetes. Cases were cohort members who experienced the event during the follow-up. For each patient, 1 control was randomly selected from among the cohort members to be matched for sex, age, and clinical status. Controls had to be at risk of the outcome when the matched case was taken in charge for diabetes.

Exposure to PPI treatment was assessed on case-control pairs in terms of time of therapy. For each patient, all PPI drugs prescribed during the follow-up were identified. The period covered by an individual prescription was calculated by means of the defined daily dose metrics. For overlapping prescriptions, the patient was assumed to have taken all drugs contained in the first prescription before starting the second one. The use of the drug has been classified in 4 categories: < 8 weeks, 8 weeks to 6 months, 6 months to 2 years, and > 2 years.

Baseline characteristics measured at index date included sex, age, clinical status, comorbidities (previous hospitalization for cardiovascular disease, cancer, depression, and respiratory and kidney diseases), and cotreatments for hypertension and dyslipidemia and use of anticoagulation and antiplatelet drugs, nonsteroidal anti-inflammatory drugs, digitalis, nitrates, antidepressants, and drugs for pulmonary diseases. In addition, the class of PPI with which each subject began the therapy was considered (ie, omeprazole, pantoprazole, lansoprazole, rabeprazole, and esomeprazole) and any combination of them.

For each cohort member, clinical status was assessed by the Multisource Comorbidity Score (MCS) (ie, a prognostic index based on 34 morbidities), which has been shown to predict mortality better than the Charlson, Elixhauser, and Chronic Disease Scores in the Italian population (18, 19). A weight proportional to its strength in predicting mortality was assigned to each condition, and the index was generated as the sum of the conditions’ weights suffered by the patient. The International Classification of Diseases, ninth revision, clinical modification, and anatomical therapeutic chemical (ATC) codes of diseases and conditions were included in the MCS, and corresponding weights were chosen as reported in a previous manuscript (18). The score was then categorized to identify the following groups of clinical status: good (MCS = 0), intermediate (1 ≤ MCS ≤ 4), poor (5 ≤ MCS ≤ 14), and very poor (MCS ≥ 15).

Standardized mean differences for binary covariates were used when appropriate to test between-group differences. Clinical equipoise was considered reached when the between-group comparison of covariates had a mean standardized difference of  2 years) than in those with shorter duration of drug therapy (< 8 weeks). The Statistical Analysis System Software (version 9.4; SAS Institute, Cary, NC, USA) was used for all analyses.

Of the 1 903 379 patients older than age 40 years receiving treatment with PPI drugs during 2010 through 2015, 777 420 met the inclusion criteria (Fig. 1). The cohort subjects accumulated 4 783 445 person-years of observation (mean, 6.2 years per patient) and generated 50 540 diagnoses of diabetes, with an incidence rate of 10.6 cases per 1000 person-years.

Flowchart of inclusion and exclusion criteria.

Among the 50 540 patients diagnosed with diabetes during the follow-up, 50 535 were matched with a control patient. Table 1 shows the characteristics of cases and controls. Approximately 50% of patients were men, and the average age was 66 years. The most prescribed PPI classes were pantoprazole and omeprazole. Cases and controls showed similar baseline characteristics, except for the use of antihypertensive and lipid-lowering drugs, which was greater among cases. During follow-up, cases spent more time with PPIs than controls.

Characteristics of the case patients and of the corresponding controls included into the study

Abbreviations: MV, matching variable; NSAID, nonsteroidal anti-inflammatory drug; PPI, proton pump inhibitor.

aFour categories were considered for the clinical profile according to the Multisource Comorbidity Score (MCS): good (MCS = 0), intermediate (1 ≤ MCS ≤ 4), poor (5 ≤ MCS ≤ 14), and very poor (MCS ≥ 15).

Characteristics of the case patients and of the corresponding controls included into the study

Abbreviations: MV, matching variable; NSAID, nonsteroidal anti-inflammatory drug; PPI, proton pump inhibitor.

aFour categories were considered for the clinical profile according to the Multisource Comorbidity Score (MCS): good (MCS = 0), intermediate (1 ≤ MCS ≤ 4), poor (5 ≤ MCS ≤ 14), and very poor (MCS ≥ 15).

Adjusted estimates of ORs for the risk of diabetes onset, related to time in therapy with PPI, are reported in Table 2. There was a trend of increased risk with increased PPI therapy duration. Compared with patients who used PPIs for  2 years, respectively.

OR and 95% CI for diabetes associated with the use of PPIs and other baseline characteristics

Google Scholar

WorldCat

Targownik LE, Metge C, Roos L, Leung S.

Google Scholar

WorldCat

Savarino V, Marabotto E, Zentilin P, et al. 

Google Scholar

WorldCat

Freedberg DE, Kim LS, Yang Y-X.

Google Scholar

WorldCat

Kwok CS, Yeong JK-Y, Loke YK.

Google Scholar

WorldCat

Gomm W, von Holt K, Thomé F, et al. 

Google Scholar

WorldCat

Janarthanan S, Ditah I, Adler DG, Ehrinpreis MN.

Google Scholar

WorldCat

Lazarus B, Chen Y, Wilson FP, et al. 

Google Scholar

WorldCat

Jackson MA, Goodrich JK, Maxan M-E, et al. 

Google Scholar

WorldCat

Vrieze A, Van Nood E, Holleman F, et al. 

Google Scholar

WorldCat

Sharpton SR, Ajmera V, Loomba R.

Google Scholar

WorldCat

Forslund K, Hildebrand F, Nielsen T, et al. 

Google Scholar

WorldCat

Peng CC-H, Tu Y-K, Lee GY, et al. 

Google Scholar

WorldCat

Rea F, Ciardullo S, Savaré L, Perseghin G, Corrao G.

Google Scholar

WorldCat

Corrao G, Compagnoni MM, Cantarutti A, et al. 

Google Scholar

WorldCat

Essebag V, Platt RW, Abrahamowicz M, Pilote L.

Google Scholar

WorldCat

Corrao G, Rea F, Martino M.D, et al. 

Google Scholar

WorldCat

Corrao G, Rea F, Carle F, et al. 

Google Scholar

WorldCat

Austin PC.

Google Scholar

WorldCat

Andrade SE, Kahler KH, Frech F, Chan KA.

Google Scholar

WorldCat

Arfè A, Corrao G.

Google Scholar

WorldCat

Schneeweiss S.

Google Scholar

WorldCat

Yuan J, He Q, Nguyen LH, et al. 

Google Scholar

WorldCat

Lin H-C, Hsiao Y-T, Lin H-L, et al. 

Google Scholar

WorldCat

Imhann F, Bonder MJ, Vila AV, et al. 

Google Scholar

WorldCat

Mehal WZ.

Google Scholar

WorldCat

Kolodziejczyk AA, Zheng D, Shibolet O, Elinav E.

Google Scholar

WorldCat

Ciardullo S, Monti T, Perseghin G.

Google Scholar

WorldCat

Czarniak P, Ahmadizar F, Hughes J, et al. 

Google Scholar

WorldCat

Strom B.

Google Scholar

Google Preview

WorldCat

These authors contributed equally to this work and served as co-first authors.

These authors also contributed equally to this work and served as co-lead authors.