Proton pump inhibitor induced hypomagnesemia and mortality: mediation analyses in the Diabetes Care System cohort

doi:10.21203/rs.3.rs-3745675/v1

Background: Chronic proton pump inhibitor (PPI) use and hypomagnesemia are common in people with type 2 diabetes (T2D), and both are associated with increased mortality and cardiovascular disease (CVD). Since PPI use can cause hypomagnesemia, we investigated if magnesium is a mediator of the association between PPI use and mortality and CVD in people with T2D.

Methods: This study was performed in 4,037 participants of the Dutch prospective Diabetes Care System cohort. Serum magnesium was measured between 2008-2014 and PPI use was determined the year preceding magnesium measurement. The primary outcome was all-cause mortality with follow-up until 2020, and the secondary outcome was fatal and non-fatal CVD. We performed causal mediation analyses.

Results: Mean serum magnesium was 0.80 mmol/l (SD 0.08) and 1,079 (26.7%) participants received PPI treatment. Median follow-up was 7.0 years [IQR 6.4-11.1], 711 (17.6%) participants died, and 608 (15.6%) had a cardiovascular event. PPI use was associated with a low serum magnesium level and an increased incidence of CVD and mortality. The association between PPI use and all-cause mortality was not mediated by magnesium (natural indirect effect (NIE): survival time ratio (STR) 0.99[95%CI 0.98;1.00] and magnesium was only a minor mediator in the association of PPI use and CVD (NIE: STR 0.98[95%CI 0.97;<1.00].

Conclusion: Serum magnesium was not a mediator of the association between PPI use and all-cause mortality. Magnesium explains only a minor part of the total effect of PPI use on CVD. Additional mediators of the association between PPI and adverse outcomes should be explored.

Health sciences/Cardiology

Health sciences/Endocrinology

Health sciences/Medical research

Health sciences/Health care/Disease prevention

Health sciences/Health care/Therapeutics/Drug therapy

Mediation analyses

mediator

indirect effect

magnesium

hypomagnesemia

proton pump inhibitor

mortality

cardiovascular disease

In this large cohort of people with type 2 diabetes in primary care, proton pump inhibitor use is associated with low serum magnesium and an increased incidence of cardiovascular disease and mortality. Low serum magnesium has an inverse association with the incidence of mortality and fatal and non-fatal cardiovascular disease.
The association between proton pump inhibitor use and all-cause mortality is not mediated by serum magnesium.
Magnesium is a small, but likely not clinically relevant mediator of the association between proton pump inhibitor use and fatal and non-fatal cardiovascular disease.
Additional pathophysiological mechanisms that explain the association between proton pump inhibitor use and adverse outcomes should be explored.

Proton pump inhibitors (PPI’s) are among the most frequently used pharmaceuticals, with almost 2.5 million users among the approximately 17.5 million Dutch residents in 2021 [1]. Due to the broad spectrum of indications, including prevention of medication-induced gastropathy, PPI use has been incorporated in many guidelines and PPI’s are frequently prescribed to people with type 2 diabetes (T2D) [2]. PPI’s are effective and generally well tolerated, yet, an increasing number of studies in various populations have demonstrated that PPI use is independently associated with adverse outcomes such as cardiovascular disease (CVD) and mortality [3–6]. However, the mechanisms behind these associations are not yet fully understood.

Magnesium (Mg²⁺) may play an important role in the association between PPI use and the increased risk of CVD and mortality. Multiple case reports described PPI induced severe hypomagnesemia because of inhibition of intestinal Mg²⁺ absorption, causing electrolyte disorders and cardiac arrhythmias [7–11]. Observational studies confirmed these associations, making PPI induced hypomagnesemia nowadays a well-known phenomenon[12–14]. Mg²⁺ is a cation that is essential for normal neuromuscular and cardiovascular (CV) function, with calcification inhibiting properties. Therefore, it has cardiovascular protective properties [15–18]. Multiple observational studies observed an inverse association between serum Mg²⁺ and the incidence of mortality, hypertension, vascular calcification, and CVD [19–26]. Also, in the Diabetes Care System (DCS) cohort the association between serum Mg²⁺ and heart failure (HF), atrial fibrillation (AF) and microvascular complications has been established [27].

Although the associations between PPI use, CVD and mortality, and hypomagnesemia are supported by evidence, a possible mediating role of Mg²⁺ in the association between PPI use and CVD, and mortality has only recently been suggested and not yet investigated [28–30]. Due to the higher prevalence of comorbidities that form an indication to prescribe PPIs, these drugs are more frequently used in people with T2D. People with T2D also have a higher prevalence of CVD and hypomagnesemia compared to the general population [31]. Therefore, investigating lower Mg²⁺ levels induced by PPI as a potential mediator is of particular interest in this population with a high CV risk.

Insight into the underlying mechanism of the negative effects of PPI use may contribute to the interdisciplinary debate on the benefits and safety of these frequently prescribed drugs and contribute to personalised medicine. Also, knowledge of potential mediators of the adverse effect of PPI’s can potentially contribute to clinical practices counteracting the adverse effects of chronic PPI use. Therefore, we investigated serum Mg²⁺ as a mediator of the association between PPI use and all-cause mortality and fatal and non-fatal CVD in a prospective cohort of people with T2D.

Study Design and Population

We used data from the DCS cohort. This is a prospective and dynamic cohort study of people with T2D and is described extensively elsewhere [32]. Clinical characteristics, anthropometric measurements, and fasting blood samples were collected during routine annual examinations. Inclusion criteria were primary diabetes treatment and confirmed T2D. We studied a subsample of participants with adequate biobank serum samples for Mg²⁺ measurement (n=4,400), with vital status and annual medication registration. We excluded participants with missing T2D diagnosis, those with a revised type of diabetes and those where the blood sampling was done at inclusion in DCS without preceding medication information, resulting in a sample of 4,037 participants (Figure 1). Due to the absence of follow-up visits for 42 cases and therefore missing information on CVD, the analysis of fatal and non-fatal CVD was performed in a sample of 3,995 participants.
Ethical clearance for the DCS cohort including biobanking was provided by the Medical Ethical Review Committee of the VU University Medical Center, Amsterdam and written informed consent was obtained from each participant. This study is reported according to the RECORD (REporting of studies Conducted using Observational Routinely-collected health Data) and AGReMA (A Guideline for Reporting Mediation Analyses) statements (supplementary A and B) [33, 34]

Exposure

During the annual visits, participants were requested to bring their current medication for registration. PPI use is defined as the use or non-use of any type of PPI (Anatomical Therapeutic Chemical classification (ATC)-codes A02BC01-A02BC05) at the visit preceding the year of biobanking (T0) and/or the visit during which biobanking was performed (T1) (Figure 2).

Mediator
Fasting serum samples were collected once between 2008-2014 (T1) and frozen at -80 °C. Mg²⁺ was measured in 2019 at the department of Laboratory Medicine of the Radboud UMC, the Netherlands, using calibrated standardised methods (coefficient of variation of 1.98%, Cobas C8000; Roche Diagnostics, Risch-Rotkreuz, Switzerland). The reference range of serum Mg²⁺ is 0.70-1.05 mmol/l and clinical hypomagnesaemia is defined as a Mg²⁺ value of <0.70 mmol/l.

Outcomes

The primary outcome was all-cause mortality and the secondary outcome was a composite of combined fatal and non-fatal CVD, with follow-up until January 2020 and April 2018, respectively. Vital status was verified every six months in the municipal registry and cause of death was determined using general practitioners’ records and electronic patient registration at regional hospitals. Cause of death and morbidity was coded using the International Classification of Diseases (ICD)-9 codes (390-459 and 798). The composite outcome of CVD included myocardial infarction, angina pectoris, HF, stroke, transient ischemic attack, and peripheral arterial disease and was registered based on self-reported events. Most of the self-reported CVD has been verified by medical records of the hospitals (sensitivity 86%, specificity 90%) [32].

Demographic characteristics and confounding variables

Information on demographic characteristics and confounding variables were collected during annual visits. We adjusted our analyses for potential confounders of the associations between 1) PPI-Mg²⁺, 2) Mg²⁺-clinical outcomes, and/or 3) PPI-clinical outcomes. These variables were: age, sex, smoking, body mass index (BMI), systolic blood pressure (SBP), Haemoglobin A1c (HbA1c), estimated glomerular filtration rate (eGFR) and low-density lipoprotein (LDL). HbA1c, fasting glucose, lipid levels and eGFR were determined based on fasting glucose blood samples drawn annually [32]. We also adjusted for the use of anticoagulant therapy, non-steroidal anti-inflammatory drugs (NSAIDs) or systemic corticosteroids, since these medications are often prescribed together with PPIs and may affect both Mg²⁺and clinical outcomes. For the analyses we included potentially confounding variables measured at the same time as PPI use (T0).

Statistical analyses

Baseline characteristics are presented for the total population and stratified by PPI use. Continuous variables are displayed as means and standard deviations (SD), or as medians and interquartile ranges (IQR) and for categorical variables as numbers and percentages. Due to the non-linear mediator-outcome association, Mg²⁺ was dichotomized by combining the middle and highest tertiles (Mg²⁺ >0.77 mmol/l, reference category) and comparing it to the lowest tertile (Mg²⁺ ≤0.77 mmol/l).

The following confounding variables were included a priori in the analyses: age, sex, prevalent CVD, BMI, SBP, smoking, HbA1c, eGFR and LDL (model 1), with addition of anticoagulant therapy, use of NSAIDs, systemic corticosteroids use, antihypertensive drugs and insulin dependence (model 2). We performed a stratified analysis for the presence of CVD at baseline for the primary endpoint. Effect modification by baseline CVD and sex was tested by adding interaction terms to the models.

Mediation analyses

We performed causal mediation analyses based on the potential outcome framework (Figure 3). [35]. We used logistic regression to estimate the association between PPI use (exposure) and serum Mg²⁺ (mediator) (i.e., the a-path). We used accelerated failure time (AFT) models to estimate the associations between serum Mg²⁺ and mortality and CVD (i.e., the b paths), and the associations between PPI use and mortality and CVD (i.e., c’ paths) [36, 37]. To assess possible exposure-mediator interaction, we added PPI-by-magnesium interaction terms to the AFT models. The most suitable error distributions for the AFT models were determined by comparing the Akaike Information Criterion (AIC) across multiple AFT models with different error distributions, where the smallest AIC indicated the best ‘fit’ [36, 38]. The no (unmeasured) confounding assumptions are addressed in Supplementary C. Based on the equations described in VanderWeele (2015) we estimated the natural indirect effect (NIE), natural direct effect (NDE), and total effect (TE) with corresponding 95% percentile bootstrap confidence intervals (CI) based on 1000 bootstrap resamples [35, 39]. The NDE, NIE and TE are presented as survival time ratios (STR). The proportion mediated was calculated as ln(NIE)/ln(TE).

Missing values

There were no missing data in the exposure or mediator variables and mortality registration was complete. Information on missing confounding values are presented in Supplementary D. Because of the low percentage of missing values (<5.0%, 194 cases) we performed a single set of imputations. All variables included in the analytic models were used as predictors for predictive mean matching (10 iterations).

Sensitivity analyses

Sensitivity analyses were performed to investigate the robustness of results. Complete case analysis and analyses using different cut-offs of the mediator were performed (hypo versus normomagnesemia and median Mg²⁺ value as the cut-off). Also, we excluded those participants who changed from PPI category during the preceding year (159/1079) and participants with baseline CVD. To explore the influence of competing risk within the analyses of CVD, we compared the effect estimates for the fully adjusted separate mediator-outcome and exposure-outcome associations, based on Cox-regression and competing risk regression, with non-cardiovascular mortality as a competing risk. Due to the not necessarily atherosclerotic origin of HF and AF (ICD-9 427.3-427.33), we also performed mediation analyses for HF and AF as separate outcome variables. Since it is suggested that diuretics can have profound Mg²⁺ lowering effects as well [40, 41], we adjusted for diuretics, apart from the pooled antihypertensive medication confounding variable.

All analyses were performed in Stata SE 14.1 (StataCorp., College Station, TX), after data cleaning and preparation in SPSS Statistics 24.0 (IBM Corp., Armonk, NY). All results are presented as effect estimates with 95%CI.

Patient characteristics

Mean age was 64.2 years (SD10.7), 2184 (56.1%) of the participants were men and mean serum magnesium was 0.80 mmol/l (SD0.08) (Table 1). The 1,079 PPI users (26.7%) were older, had higher BMI, lower eGFR and had a higher prevalence of CVD and co-medication use at baseline, compared to nonusers. Mean SBP and HbA1c were comparable in both groups, yet antihypertensive medication use and insulin therapy were more frequent in PPI users.

During a median follow-up of 7.0 years [IQR 6.4-11.1], 711 (17.6%) participants died, of which 134 (18.8.%) were classified as death due to CVD. During a median follow-up of 5.5 years (IQR4.3-9.6], 608 (15.2%) participants suffered one or more CV events.

Mediation analyses

PPI use was associated with a shorter survival time (TE: STR 0.86[95%CI 0.76;0.97]) in multivariable adjusted models (Table 2, Figure 4). With PPI use associated with lower Mg²⁺ (a-path: OR1.28[95%CI1.10;1.50] and lower Mg²⁺ associated with shorter survival time (b-path: STR 0.90[95%CI 0.80;<1.00], the fully adjusted indirect effect estimate of PPI use on all-cause mortality through Mg²⁺ was small (NIE:STR 0.99[95%CI 0.98;1.00], 4.1% mediation). The fully adjusted direct effect of PPI use on mortality (not through Mg) also demonstrated a shorter survival time 0.87[95%CI 0.77;0.97]. Analyses based on people with baseline CVD only did not considerably change the effect estimates (NIE:STR 0.98[95%CI 0.96;1.01]).
The mediation effect estimates for fatal and non-fatal CVD as the outcome were comparable with the effect of the primary outcome, with a TE of 0.86[95%CI 0.71;1.06], NDE of 0.87[95%CI 0.71;1.07] and a NIE of 0.98[95%CI 0.97;<1.00], 8.4% mediation. The fully adjusted estimates of the NIE for the separate outcomes of HF and AF were comparable to the estimates for fatal and non-fatal CVD (STR 0.98[95%CI 0.95;<1.00], 16.3% and STR 0.97[95%CI 0.94;0.99], 25.3%, respectively). No interactions (i.e., mediator-exposure interaction; interactions with sex, or baseline CVD) were present.

Sensitivity analyses

Complete-case analyses and the sensitivity analysis adjusting for diuretics as a separate confounding variable in addition to the combined antihypertensive variable, demonstrated comparable results to the main analyses. Also, different definitions of the mediator or the exposure and exclusion of participants with baseline CVD did not change the results considerably (SupplementaryE & F).

In this prospective T2D cohort, we demonstrated that the association between PPI and all-cause mortality is not mediated by serum Mg²⁺. However, serum Mg²⁺ is a partial mediator of the association between PPI use and fatal and non-fatal CVD. Since this mediation effect is very small, the clinical relevance can be debated and the need to explore other mediating mechanisms remains.

To our knowledge this is the first study investigating Mg²⁺ as a mediator of the association between PPI use and adverse outcomes. However, there is considerable previous evidence supporting our finding of associations among the variables in our mediation model. A meta-analysis of observational studies investigating PPI induced hypomagnesemia demonstrated a pooled adjusted OR of 1.71[95%CI 1.33;2.19]. Our results of an inverse association between serum Mg²⁺ and the incidence of mortality and CVD are in line with several meta-analyses in various populations and recently confirmed in this DCS cohort [20, 21, 27]. The association between PPI use and mortality has been described in several cohorts with reports of 45 excess deaths per 1000 PPI users in 10 years and an increased risk of 1-year mortality (HR 1.51[95%CI 1.03;2.77]) and myocardial infarction (HR 1.58[95%CI 1.11;2.25]) [3-6]. In our study we also found associations between PPI use and mortality as well as CVD (STR 0.86, 95%CI 0.75;0.97 and STR 0.86[95%CI 0.71;1.06]), however, less pronounced than in the aforementioned studies.

Differences in the observed effect estimates compared to previous studies may be explained by differences in study populations and confounder sets. The higher prevalence of hypomagnesemia (ranging between 13.5% up to 47.7%) reported in most diabetes cohorts compared to 9.0% hypomagnesemia in our study could be because other studies included mainly hospitalised patients with severe chronic kidney disease or longstanding diabetes [31, 42]. The DCS cohort consists of people with T2D in primary care, with low complication rate and close monitoring and therefore, participants of this cohort are less severely affected compared to populations that receive hospital treatment [32].

Our results indicate that serum Mg²⁺ is not a substantial mediator; hence, other pathophysiological mechanisms may explain the association between PPI use and mortality and CVD if causality exists. This may include (CYP2C19 related) drug interactions, tubulointerstitial inflammation, the effect of PPI on gastric and small intestine microbiota balance and pH related intestinal absorption of several nutrients and minerals [29, 30]. It is also reasoned that the combined effect of PPI use on gastric pH, expression of endothelial nitric oxide synthase and asymmetric dimethyl arginine activity, all reduce nitric oxide (NO) synthesis. Impaired NO synthesis in turn leads to endothelial dysfunction, with subsequent increased risk of CV events [29, 43].

Strengths & limitations
Since Mg²⁺measurement is not included in regular clinical examinations, this large cohort with Mg²⁺ measurements and the prospective design with a median follow-up of >5.5 years is unique. This cohort reflects the most prevalent course of T2D among people treated in primary care with a good external validity, contrary to most T2D cohorts, which typically include more severely affected people in secondary care. All Mg²⁺ samples have been measured at the same laboratory (single batch), and therefore variability because of calibration or technical differences in the Mg²⁺ measurement is low. Also, we performed causal mediation analyses, whereas most mediation analyses with a survival outcome are still based on traditional methods with inadequate reporting of the underlying assumptions [44]. Lastly, by performing several sensitivity analyses we were able to demonstrate robustness of our results.
The main limitation of our study is the influence of residual confounding and confounding by indication. Causal mediation analyses rely on assumptions of no unmeasured confounding of the underlying associations (Supplementary F), but residual confounding can never be excluded [45-47]. In this paper, we adjusted for a relatively large number of potential confounders to reduce the likelihood of residual confounding. Plausible confounders that we were not able to adjust for include diet, alcohol intake, physical activity, multimorbidity, the use of selective serotonin reuptake inhibitors and other co-medication (apart from those in model 2). Therefore, residual confounding could have resulted in overestimation of the mediation effects and may affect the causal interpretation of the mediation effect estimates.

Another limitation is the uncertainty in the classification of PPI users. Although the data were carefully collected and coded, the registration of PPI prescription does not guarantee compliance of intake. High incompliance of PPI intake could therefore have caused misclassification of PPI use [48, 49]. Misclassification could also be due to the over-the-counter availability of PPI’s (since 2010), which could have been missed with medication registration during the annual visits and could have led to an underestimation of the mediation effects. We were not able to take into account dose or duration of the PPI prescription. It has been demonstrated that the Mg²⁺ lowering effect of PPI’s are dose as well as duration dependent [12, 13]. Another limitation is that serum Mg²⁺ has a tightly regulated equilibrium and complex homeostasis, with only less than 1% of total body Mg²⁺ present in the serum [16]. Therefore, PPI use could affect Mg²⁺ body stores more severely than is reflected by the serum Mg²⁺ concentration. However, serum Mg²⁺ is easily obtained, affordable and generally accepted as indicator for Mg²⁺ status. Lastly, cause of death is difficult to determine and the relatively low incidence of CV death (18.9%, whereas T2D literature reports up to 50% of deaths due to CVD [50]) might indicate incorrect classification of some cases as non-CV deaths. This could have attenuated the results for our secondary endpoint.

Clinical impact and future research
Because of the very small mediation effect of Mg²⁺on the association of PPI use and fatal and non-fatal CVD, these analyses should be repeated in other cohorts. If results are consistent, it may be relevant to investigate if supplementing Mg²⁺ for chronic PPI users may prevent CVD. Since serum Mg²⁺ is no mediator in the association of PPI use and morality and only a small mediator in the association with CVD, future research should focus on the exploration of other pathways that can explain the remainder of this association, such as NO and intestinal microbiota balance, or identify individuals that do benefit from raising magnesium level. By creating more awareness of the harmful effects of PPI use in addition to their highly effective gastro-protective effect, we aim that this research will contribute to more careful evaluation and possibly restrictive use of PPI prescription and its over-the-counter availability in some countries.

In this large cohort of people with T2D in primary care, the association between PPI use and all-cause mortality is not mediated by serum Mg²⁺. However, serum Mg²⁺ is a partial, but very small mediator in the association of PPI use and fatal and non-fatal CVD. Therefore, future research should focus on other pathophysiological mechanisms that may explain the association between PPI use and adverse outcomes.

Data availability statement

Please find our data availability statement in Supplementary A ‘The RECORD statement checklist of items’ and the cohort profile of the Hoorn Diabetes Care System Cohort: The Hoorn Diabetes Care System (DCS) cohort. A prospective cohort of persons with type 2 diabetes treated in primary care in the Netherland (van der Heijden AA, et al., BMJ Open. 2017)

Additional Information

Funding: This research was funded by grants from the Netherlands Organization for Scientific Research (NWO Veni 016.186.012), the Dutch Diabetes Research Foundation (2017-81-014), and the NIGRAM2+ consortium. The NIGRAM2+ collaboration project is co-funded by the PPP Allowance made available by HealthHolland Top Sector Life Sciences & Health, to stimulate public-private partnerships (LSHM17034) and the Dutch Kidney Foundation (16TKI02). For additional measurements in (subsamples of) the DCS cohort, funding has been received from several instances including the Amsterdam UMC, Dutch Federation of University Medical Centres, health insurers, Dutch Science Organisation NWO, Dutch Organization for Health Research and Development ZonMw, Dutch Diabetes Foundation, European Foundation for the Study of Diabetes, International Diabetes Federation, European Innovative Medicine Initiative and European Unio. J.W.J.B. is supported by a ZonMW NWO-Vidi grant (91 71 8304).

Partners NIGRAM2+ consortium: Fresenius Medical Care, Calciscon and Nedmag Industries Mining en Manufacturing BV (NIMM).

Conflicts of interest: No potential conflicts of interest relevant to this article were reported.

Author contributions statement

EV, JR, TH, JB and MV all have made substantial contributions to the design, analysis and interpretation of this study. MB and PE have made substantial contributions to the design and acquisition of data of the cohort. LO and JdB have made substantial contributions to additional acquisition of data. EV, JR, TH, MV and JB have contributed to the manuscript. All authors have approved the submitted version and agreed to be accountable for their contributions.

SFK. Data en feiten 2022. Het jaar 2021 in cijfers. Data en feiten. https://www.sfk.nl/ Stichting Farmaceutische Kengetallen.
Westergaard N, Tarnow L, Vermehren C. Use of Clopidogrel and Proton Pump Inhibitors Alone or in Combinations in Persons with Diabetes in Denmark; Potential for CYP2C19 Genotype-Guided Drug Therapy. Metabolites. 2021;11(2).
Shah NH, LePendu P, Bauer-Mehren A, Ghebremariam YT, Iyer SV, Marcus J, et al. Proton Pump Inhibitor Usage and the Risk of Myocardial Infarction in the General Population. PLoS One. 2015;10(6):e0124653.
Xie Y, Bowe B, Yan Y, Xian H, Li TT, Al-Aly Z. Estimates of all cause mortality and cause specific mortality associated with proton pump inhibitors among US veterans: cohort study. Bmj-Brit Med J. 2019;365.
Shih CJ, Chen YT, Ou SM, Li SY, Chen TJ, Wang SJ. Proton pump inhibitor use represents an independent risk factor for myocardial infarction. Int J Cardiol. 2014;177(1):292-7.
Maggio M, Corsonello A, Ceda GP, Cattabiani C, Lauretani F, Butto V, et al. Proton pump inhibitors and risk of 1-year mortality and rehospitalization in older patients discharged from acute care hospitals. JAMA Intern Med. 2013;173(7):518-23.
Regolisti G, Cabassi A, Parenti E, Maggiore U, Fiaccadori E. Severe hypomagnesemia during long-term treatment with a proton pump inhibitor. Am J Kidney Dis. 2010;56(1):168-74.
Hoorn EJ, van der Hoek J, de Man RA, Kuipers EJ, Bolwerk C, Zietse R. A case series of proton pump inhibitor-induced hypomagnesemia. Am J Kidney Dis. 2010;56(1):112-6.
Epstein M, McGrath S, Law F. Proton-pump inhibitors and hypomagnesemic hypoparathyroidism. N Engl J Med. 2006;355(17):1834-6.
Lameris ALL, Hess MW, van Kruijsbergen I, Hoenderop JGJ, Bindels RJM. Omeprazole enhances the colonic expression of the Mg2+ transporter TRPM6. Pflug Arch Eur J Phy. 2013;465(11):1613-20.
Cundy T, Dissanayake A. Severe hypomagnesaemia in long-term users of proton-pump inhibitors. Clin Endocrinol (Oxf). 2008;69(2):338-41.
Srinutta T, Chewcharat A, Takkavatakarn K, Praditpornsilpa K, Eiam-Ong S, Jaber BL, et al. Proton pump inhibitors and hypomagnesemia: A meta-analysis of observational studies. Medicine (Baltimore). 2019;98(44):e17788.
Kieboom BCT, Kiefte-de Jong JC, Eijgelsheim M, Franco OH, Kuipers EJ, Hofman A, et al. Proton Pump Inhibitors and Hypomagnesemia in the General Population: A Population-Based Cohort Study. Am J Kidney Dis. 2015;66(5):775-82.
Nakashima A, Ohkido I, Yokoyama K, Mafune A, Urashima M, Yokoo T. Proton Pump Inhibitor Use and Magnesium Concentrations in Hemodialysis Patients: A Cross-Sectional Study. PLoS One. 2015;10(11):e0143656.
ter Braake AD, Tinneman PT, Shanahan CM, Hoenderop JGJ, de Baaij JHF. Magnesium prevents vascular calcification in vitro by inhibition of hydroxyapatite crystal formation. Sci Rep-Uk. 2018;8.
Jahnen-Dechent W, Ketteler M. Magnesium basics. Clin Kidney J. 2012;5(Suppl 1):i3-i14.
Maier JAM, Malpuech-Brugere C, Zimowska W, Rayssiguier Y, Mazur A. Low magnesium promotes endothelial cell dysfunction: implications for atherosclerosis, inflammation and thrombosis. Bba-Mol Basis Dis. 2004;1689(1):13-21.
Leenders NHJ, Vervloet MG. Magnesium: A Magic Bullet for Cardiovascular Disease in Chronic Kidney Disease? Nutrients. 2019;11(2).
Khan AM, Lubitz SA, Sullivan LM, Sun JX, Levy D, Vasan RS, et al. Low Serum Magnesium and the Development of Atrial Fibrillation in the Community The Framingham Heart Study. Circulation. 2013;127(1):33-8.
Del Gobbo LC, Imamura F, Wu JH, de Oliveira Otto MC, Chiuve SE, Mozaffarian D. Circulating and dietary magnesium and risk of cardiovascular disease: a systematic review and meta-analysis of prospective studies. Am J Clin Nutr. 2013;98(1):160-73.
Leenders NHJ, Vermeulen EA, van Ballegooijen AJ, Hoekstra T, de Vries R, Beulens JW, et al. The association between circulating magnesium and clinically relevant outcomes in patients with chronic kidney disease: A systematic review and meta-analysis. Clin Nutr. 2020.
Han H, Fang X, Wei X, Liu Y, Jin Z, Chen Q, et al. Dose-response relationship between dietary magnesium intake, serum magnesium concentration and risk of hypertension: a systematic review and meta-analysis of prospective cohort studies. Nutr J. 2017;16(1):26.
Hruby A, O'Donnell CJ, Jacques PF, Meigs JB, Hoffmann U, McKeown NM. Magnesium intake is inversely associated with coronary artery calcification: the Framingham Heart Study. JACC Cardiovasc Imaging. 2014;7(1):59-69.
Taveira TH, Ouellette D, Gulum A, Choudhary G, Eaton CB, Liu S, et al. Relation of Magnesium Intake With Cardiac Function and Heart Failure Hospitalizations in Black Adults: The Jackson Heart Study. Circ Heart Fail. 2016;9(4):e002698.
van Zuijdewijn CLMD, Grooteman MPC, Bots ML, Blankestijn PJ, Steppan S, Buechel J, et al. Serum Magnesium and Sudden Death in European Hemodialysis Patients. Plos One. 2015;10(11).
de Roij van Zuijdewijn CL, Grooteman MP, Bots ML, Blankestijn PJ, Steppan S, Buchel J, et al. Serum Magnesium and Sudden Death in European Hemodialysis Patients. PLoS One. 2015;10(11):e0143104.
Oost LJ, van der Heijden A, Vermeulen EA, Bos C, Elders PJM, Slieker RC, et al. Serum Magnesium Is Inversely Associated With Heart Failure, Atrial Fibrillation, and Microvascular Complications in Type 2 Diabetes. Diabetes Care. 2021;44(8):1757-65.
Ago R, Shindo T, Banshodani M, Shintaku S, Moriishi M, Masaki T, et al. Hypomagnesemia as a predictor of mortality in hemodialysis patients and the role of proton pump inhibitors: A cross-sectional, 1-year, retrospective cohort study. Hemodial Int. 2016;20(4):580-8.
Corsonello A, Lattanzio F, Bustacchini S, Garasto S, Cozza A, Schepisi R, et al. Adverse Events of Proton Pump Inhibitors: Potential Mechanisms. Curr Drug Metab. 2018;19(2):142-54.
Fossmark R, Martinsen TC, Waldum HL. Adverse Effects of Proton Pump Inhibitors-Evidence and Plausibility. Int J Mol Sci. 2019;20(20).
Pham PC, Pham PM, Pham SV, Miller JM, Pham PT. Hypomagnesemia in patients with type 2 diabetes. Clin J Am Soc Nephrol. 2007;2(2):366-73.
van der Heijden AA, Rauh SP, Dekker JM, Beulens JW, Elders P, t Hart LM, et al. The Hoorn Diabetes Care System (DCS) cohort. A prospective cohort of persons with type 2 diabetes treated in primary care in the Netherlands. BMJ Open. 2017;7(5):e015599.
Benchimol EI, Smeeth L, Guttmann A, Harron K, Moher D, Petersen I, et al. The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement. PLoS Med. 2015;12(10):e1001885.
Lee H, Cashin AG, Lamb SE, Hopewell S, Vansteelandt S, VanderWeele TJ, et al. A Guideline for Reporting Mediation Analyses of Randomized Trials and Observational Studies: The AGReMA Statement. JAMA. 2021;326(11):1045-56.
Valeri L, Vanderweele TJ. Mediation analysis allowing for exposure-mediator interactions and causal interpretation: theoretical assumptions and implementation with SAS and SPSS macros. Psychol Methods. 2013;18(2):137-50.
Ochoa LB, Rijnhart JJM, Penninx BW, Wardenaar KJ, Twisk JWR, Heymans MW. Performance of methods to conduct mediation analysis with time-to-event outcomes. Stat Neerl. 2019.
Gelfand LA, MacKinnon DP, DeRubeis RJ, Baraldi AN. Mediation Analysis with Survival Outcomes: Accelerated Failure Time vs. Proportional Hazards Models. Front Psychol. 2016;7:423.
Swindell WR. Accelerated failure time models provide a useful statistical framework for aging research. Exp Gerontol. 2009;44(3):190-200.
VanderWeele T. Explanation in Causal Inference, Methods for Mediation and Interaction: Oxford University Press Inc; 2015.
Whang R, Chrysant S, Dillard B, Smith W, Fryer A. Hypomagnesemia and Hypokalemia in 1,000 Treated Ambulatory Hypertensive Patients. J Am Coll Nutr. 1982;1(4):317-22.
Chrysant SG, Chrysant GS. Association of hypomagnesemia with cardiovascular diseases and hypertension. Int J Cardiol Hypertens. 2019;1:100005.
Kurstjens S, de Baaij JH, Bouras H, Bindels RJ, Tack CJ, Hoenderop JG. Determinants of hypomagnesemia in patients with type 2 diabetes mellitus. Eur J Endocrinol. 2017;176(1):11-9.
Sukhovershin RA, Cooke JP. How May Proton Pump Inhibitors Impair Cardiovascular Health? Am J Cardiovasc Drugs. 2016;16(3):153-61.
Lapointe-Shaw L, Bouck Z, Howell NA, Lange T, Orchanian-Cheff A, Austin PC, et al. Mediation analysis with a time-to-event outcome: a review of use and reporting in healthcare research. Bmc Med Res Methodol. 2018;18.
Lange T, Hansen JV. Direct and indirect effects in a survival context. Epidemiology. 2011;22(4):575-81.
Imai K, Keele L, Tingley D. A General Approach to Causal Mediation Analysis. Psychological Methods. 2010;15(4):309-34.
VanderWeele TJ, Vansteelandt S. Conceptual issues concerning mediation, interventions and composition. Stat Interface. 2009;2(4):457-68.
Duh MS, Gosselin A, Luo R, Lohoues H, Lewis BE, Crawley JA. Impact of compliance with proton pump inhibitors on NSAID treatment. Am J Manag Care. 2009;15(10):681-8.
Gosselin A, Luo R, Lohoues H, Toy E, Lewis B, Crawley J, et al. The impact of proton pump inhibitor compliance on health-care resource utilization and costs in patients with gastroesophageal reflux disease. Value Health. 2009;12(1):34-9.
Einarson TR, Acs A, Ludwig C, Panton UH. Prevalence of cardiovascular disease in type 2 diabetes: a systematic literature review of scientific evidence from across the world in 2007-2017. Cardiovasc Diabetol. 2018;17(1):83.

Table 1. Baseline characteristics of participants of the DCS cohort subsample.
	Total cohort (n = 4037)	PPI users* (n = 1079)	Nonuser (n= 2958)
Demographics & clinical characteristics
Age, years	64.2 (10.7)	66.0 (10.5)	63.6 (10.7)
Men, n (%)	2266 (56.1)	473 (51.4)	1711 (57.8)
Education level, n (%)
Low	1639 (40.6)	426 (47.0)	1151 (40.2)
Intermediate	1630 (40.4)	335 (39.2)	1223 (42.7)
High	612 (15.2)	117 (13.2)	475 (16.6)
Smoking status, n (%)
Current	722 (17.9)	160 (17.3)	535 (18.1)
Former	1983 (49.1)	486 (52.1)	1421 (48.0)
Never	1295 (32.1)	271 (30.2)	969 (32.8)
BMI, kg/m²	30.3 (5.4)	31.0 (5.7)	30.1 (5.3)
Systolic blood pressure, mmHg	141.7 (19.8)	142.4 (20.2)	141.5 (19.7
Prior cardiovascular disease, n (%)	932 (23.1)	329 (35.5)	459 (18.6)
Time since diagnose of T2DM (years)	6.3 (5.9)	6.2(5.8)	6.3 (6.0)
Laboratory values
Magnesium mmol/l	0.80 (0.08)	0.79 (0.9)	0.80 (0.07)
Magnesium dichotomised**, n (%) - Low Mg (range 0.44≤0.77 mmol/l) - High Mg (range 0.78-1.04 mmol/l)	1418 (35.1) 2619 (64.9)	431 (39.9) 648 (60.1)	987 (33.3) 1971 (66.6)
Hypomagnesemia, <0.70 mmol/l (%)	362 (9.0)	122 (12.9)	223 (7.5)
HbA1c, mmol/mol	50.0 (11.3)	49.5 (9.6)	50.2 (11.8)
Fasting glucose, mmol//l	7.9 (2.0)	7.8 (1.8)	8.0 (2.1)
Triglycerides, mmol/l	1.5 [1.1-2.1]	1.6 [1.2-2.3]	1.5 [1.1-2.1]
LDL cholesterol, mmol/l	2.5 (0.9)	2.5 (0.9)	2.5 (0.9)
HDL- cholesterol, mmol/l	1.2 (0.3)	1.2 (0.3)	1.3 (0.4)
eGFR, ml/min/1.73m²	79.2 (18.3)	75.9 (18.8)	80.4 (17.9)
Microalbuminuria	4.0 [2.9 -11.0]	4.0 [2.9 -11]	4.0 [2.9 -12]
Medication use
Insulin therapy, n (%)	774 (19.2)	209 (22.2)	535 (18.1)
Oral antidiabetic medication, n (%)	3046 (75.5)	691 (76.2)	2224 (75.2)
Anti-coagulant therapy, n (%)^	299 (32.2)	450 (48.1)	780 (26.4)
Antihypertensive medication, n (%)	2771 (69.0)	740 (79.1)	1918 (64.8)
Lipid lowering medication use, n (%)	2755 (68.2)	682 (77.3)	1957 (67.9)
Cardiac therapy, n (%)	330 (8.2)	157 (16.5)	153 (5.2)
NSAID, n (%)^	281 (7.0)	145 (16.5)	103 (3.5)
Systemic corticosteroids, n (%)^	70 (1.7)	40 (4.1)	26 (0.9)

Abbreviations: DCS, diabetes care system; BMI, Body Mass Index; T2DM, type 2 diabetes mellitus; HbA1c, Haemoglobin A1c; LDL, low-density lipoprotein; HDL, high-density lipoprotein; eGFR, estimated Glomerular Filtration Rate (CKD-EPI 2009); NSAID, non-steroid anti-inflammatory drugs.
* The most frequently prescribed PPI was omeprazole (45.9%), followed by pantoprazole (38.5%), esomeprazole (9.9%), rabeprazole (2.2%) and lansoprazole (1.3%).
**mean Mg concentration Low-Normal = 0.72 (SD 0.05) mmol/l, High-normal 0.84 (SD 0.05) mmol/l.
^ PPI prescription frequently indicated in combination with this medicine (indicated in clinical guidelines).
Note: Normally distributed variables are presented as mean ± standard deviation (SD), non-normally distributed data as median witch interquartile range, categorical data as number and percentage

Table 2. Results of the causal mediation analyses.
	a-path OR (95%CI)	b-path STR (95%CI)	Direct effect (NDE) STR (95%CI)	Indirect effect (NIE) STR (95%CI)	Total effect (TE) STR (95%CI)	% Mediation
PRIMARY OUTCOME All-cause mortality (n = 4037)
Crude	1.32 (1.15;1.53)	0.88 (0.78;0.98)	0.64 (0.57;0.72)	0.99 (0.98;<1.00*)	0.64 (0.56;0.72)		1.9%
Model 1	1.31 (1.12;1.52)	0.90 (0.80;<1.00)	0.83 (0.74;0.92)	0.99 (0.98;<1.00)	0.82 (0.73;0.92)		3.4%
Model 2	1.28 (1.10;1.50)	0.90 (0.80;<1.00)	0.87 (0.77;0.97)	0.99 (0.98;1.00)	0.86 (0.76;0.97)		4.1%
Stratified analyses for presence of baseline CVD**
Without CVD	1.16 (0.96;1.41)	0.95 (0.83;1.09)	0.81 (0.70;0.95)	1.00 (0.99;1.00)	0.81 (0.68;0.96)		0.8%
Baseline CVD	1.57 (1.17;2.10)	0.86 (0.72;1.01)	0.91 (0.77;1.08)	0.98 (0.96;1.01)	0.88 (0.75;1.06)		15.1%
SECONDARY OUTCOME Fatal & non-fatal CVD (n = 3995)
Crude	1.31 (1.13;1.51)	0.79 (0.65;0.95)	0.54 (0.44;0.66)	0.99 (0.97;<1.00)	0.53 (0.44;0.65)		2.3%
Model 1	1.29 (1.11;1.51)	0.78 (0.64;0.93)	0.80 (0.66;0.97)	0.99 (0.97;<1.00)	0.79 (0.66;0.96)		6.0%
Model 2	1.27 (1.08;1.48)	0.79 (0.65;0.95)	0.87 (0.71;1.07)	0.98 (0.97;<1.00)	0.86 (0.71;1.06)		8.4%
*CVD specific outcomes: fatal and non-fatal heart failure* (n = 3995) and atrial fibrillation (n = 3987)*******
HF	1.27 (1.08;1.49)	0.61 (0.43;0.87)	0.88 (0.60;1.29)	0.98 (0.95;<1.00)	0.86 (0.57;1.31)		16.3%
AF	1.26 (1.08;1.48)	0.47 (0.33;0.66)	0.90 (0.62;1.31)	0.97(0.94;0.99)	0.87 (0.63;1.26)		25.3%
Abbreviations: AF, atrial fibrillation; HF, heart failure; OR, Odds ratio; CI, confidence interval ; STR, survival time ratio; NDE, natural direct effect; NIE, natural indirect effect; TE, total effect; CVD, cardiovascular disease; BMI, body mass index; HbA1c, hemoglobin A1c, eGFR, estimated glomerular filtration rate; LDL, low density lipoprotein; NSAID, non-steroid anti-inflammatory drug; < Because of rounding to two decimals 1.00, but actually < 0.999 and therefore significant; * <1.00, number between 0.995 and 0.999 Only fully-adjusted model 2 analysis are presented: 3015 cases without baseline CVD, 932 cases with baseline CVD (stratified analysis and therefore CVD as confounder excluded from this model, interaction for baseline CVD P=0.10). * Only fully-adjusted model 2 analysis are presented , 131 and 202 events for HF and AF, respectively. a-path = association between proton pump inhibitor use and magnesium; (low/normal magnesium = 0, normal/high magnesium =1) b-path = association between magnesium and outcome corrected for determinant (PPI use) Model 1 = age and sex, baseline CVD, BMI, systolic blood pressure, smoking, HbA1c, eGFR and LDL Model 2 = model 1 + anticoagulant therapy, antihypertensive use, use of NSAID, systemic corticosteroids use and insulin dependence

No competing interests reported.

SupplementarydataProtonpumpinhibitorinducedhypomagnesemiaandmortalitymediationanalysesintheDiabetesCareSystemcohort.docx

Proton pump inhibitor induced hypomagnesemia and mortality: mediation analyses in the Diabetes Care System cohort

Status:

Version 1

Abstract

Figures

Key messages

Introduction

Methods

Statistical analyses

Results

Discussion

Conclusion

Declarations

References

Tables

Additional Declarations

Supplementary Files

Status:

Version 1