Nonprescription Magnesium Supplement Use and Risk of Heart Failure in Patients With Diabetes: A Target Trial Emulation

Abstract

Background

Both diabetes and low magnesium‐containing food intake may increase the risk of heart failure (HF). However, the effect of nonprescription magnesium supplements on the risk of HF or major adverse cardiac events in patients with diabetes is unknown.

Methods and Results

Using a target‐trial‐emulation approach, we assembled a national cohort of 94 239 veterans ≥40 years with diabetes, without prior HF or magnesium use, who received ambulatory care in the US veterans‐health care system documented by electronic clinic notes between January 1, 2006 and December 31, 2020. A natural language processing approach was used to detect self‐reported magnesium‐supplement use from clinic notes, n=17 619 were identified as users versus n=76 620 as nonusers. Using inverse probability treatment weighting, we constructed a cohort balanced in 88 baseline characteristics between users and nonusers. The primary outcome was incident HF. Secondary outcomes were major adverse cardiac events (myocardial infarction, stroke, HF hospitalization, or death). Hazard ratios (HRs) associated with magnesium‐supplement use and outcomes were estimated in the inverse probability treatment weighting weighted cohort using Cox regression. The inverse probability treatment weighting weighted cohort had a mean age of 67.4±10.3 years; 18.4% were Black, and 5.1% were women. The mean duration of magnesium‐supplement use was 3.5±3.1 (interquartile range, 1.1–5.1) years. Incident HF occurred in 8.0% of users and 9.7% of nonusers of magnesium supplements (HR, 0.94 [95% CI, 0.89–0.99]). Magnesium‐supplement use was also associated with a reduced risk of major adverse cardiac events (HR, 0.94 [95% CI, 0.90–0.97]).

Conclusions

Long‐term nonprescription magnesium supplement use was associated with a lower risk of incident HF and major adverse cardiac events in patients with diabetes. These findings should be replicated in randomized controlled trials.

Nonstandard Abbreviations and Acronyms

IPTW

inverse probability treatment weighting

MACE

major adverse cardiac events

VHA

Veterans Health Administration

Clinical Perspective.

What Is New?

• Magnesium supplements are associated with a small but significant risk reduction in heart failure and major adverse cardiovascular events over time in patients with diabetes, with benefits becoming apparent after 3 years of use.

What Are the Clinical Implications?

• The findings suggest that long‐term magnesium supplementation may offer cardiovascular benefits for individuals with diabetes, particularly those at risk of low serum magnesium levels, even if serum magnesium levels are unknown.

Over 50% of adult Americans use dietary supplements. ¹ , ² However, little is known about their efficacy and safety as they are not subject to approval by the US Food and Drug Administration. A strategic goal of the National Institute of Health Office of Dietary Supplements is to expand scientific knowledge about dietary supplements, and the 21st Century Cures Act calls for efficient use of real‐world data. ³ , ⁴ Ideally, every dietary supplement should be optimally evaluated by large, well‐designed, randomized controlled trials (RCT) to establish efficacy and safety. However, RCTs are cost prohibitive for nutritional supplements that require large sample sizes and long‐term follow‐up to determine potentially modest effect sizes. ⁵ , ⁶ , ⁷ When RCT evidence is not available, evidence from observational studies that mimic hypothetical RCTs may be the best alternative. ⁸ , ⁹ , ¹⁰ Recent advances in analytical technologies (eg, target trial emulation) have allowed investigation of clinical effectiveness of therapeutic modalities, which minimize the potential biases existing in observational data. ¹¹

Magnesium is an intracellular mineral that is an integral part of numerous major human enzymatic pathways. ¹² Available evidence indicates that dietary magnesium intake is inadequate in most adult Americans, and deficiency is common. ¹³ Hypomagnesemia is a risk factor for diabetes and is common in patients with diabetes. ¹⁴ Hypomagnesemia causes inflammation, endothelial dysfunction, insulin resistance, and atherosclerosis, ¹⁵ , ¹⁶ which can potentially be corrected by magnesium supplementation. ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ Magnesium supplementation may lower the risk of diabetes ¹⁴ , ²² , ²³ and improve insulin sensitivity in diabetes. ²⁴ , ²⁵ Laboratory studies have suggested that magnesium repletion can improve cardiac diastolic dysfunction in diabetes. ²⁶ We and others have demonstrated that diabetes is a risk factor for heart failure (HF). ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ Our prior work has shown a link between low dietary magnesium intake and a higher risk of HF, especially among those with diabetes. ³⁵ , ³⁶ However, less is known about the effect of magnesium supplements in reducing the risk of HF in patients with diabetes. There are yet no studies directly investigating magnesium supplementation and cardiovascular outcomes among patients with diabetes. In this study, we emulated a hypothetical target clinical trial to estimate the effects of magnesium supplement use on preventing heart failure in a large cohort of US military veterans with diabetes.

METHODS

Study Design

We conducted a retrospective cohort study using interconnected clinical databases from the Veterans Health Administration (VHA)'s national electronic health records data housed at the Corporate Data Warehouse. The study received VHA's Central Institutional Review Board approval (IRBNET#1710776‐1, 04/23/2021), which waived the requirement to obtain informed consent. Because of the sensitive nature of the data collected for this study, requests to access the data set from qualified researchers trained in human subject confidentiality protocols may be sent to Washington, DC Veterans Affairs Medical Center at helen.sheriff@va.gov. Given that our goal is to estimate potential treatment effects, if any, of magnesium supplement use and the risk of HF, we used a target trial design to emulate a controlled clinical trial ¹¹ to minimize potential biases from an observational study.

Study Population

The VHA's national electronic health records data were used to identify patients with diabetes who would be eligible to be enrolled in an emulated target trial during the period between January 1, 2006 and December 31, 2020 according to the selection criteria described later. diabetes was operationally defined based on the International Classification of Diseases, Ninth Revision (ICD‐9) and Tenth Revision (ICD‐10) codes (ICD‐9: 250.X, 357.2X, 362.0X, 366.41, 648.0X; ICD‐10: E11, O24.1X), antidiabetic medication use (Table 1), and hemoglobin A1C. As such, diabetes was confirmed if a patient had at least 1 of the following conditions: (1) ICD code for diabetes diagnosis in at least 2 clinical encounters, (2) ICD code for diabetes diagnosis once plus antidiabetes medication(s) prescribed, or (3) ICD code for diabetes diagnosis once plus a hemoglobin A1C ≥6.5%. The first date when any 1 of the 3 conditions was satisfied was used as the date of the first diagnosis of diabetes.

Table 1.

Baseline Characteristics of the Study Participants Before IPTW Weighting

Baseline characteristics	Overall (N=94 239)		Magnesium nonusers (N=76 620)		Magnesium users (N=17 619)		ASD before IPTW (%)
	Mean/N	SD/%	Mean/N	SD/%	Mean/N	SD/%
Demographics
Age, y	67.3	10.3	66.9	10.3	69.1	9.9	22
Male sex	89 613	95.1%	73 229	95.6%	16 384	93.0%	11
Race
White	66 868	71.0%	53 003	69.2%	13 865	78.7%	22
Black	17 461	18.5%	15 663	20.4%	1798	10.2%	29
Other*	9910	10.5%	7954	10.4%	1956	11.1%	2
Hispanic ethnicity	5511	5.8%	4912	6.4%	599	3.4%	14
Environmental Justice Index Social‐Environmental Percentile ranking	49.9	22.7	50.9	22.8	46.0	22.0	22
Diabetes information
Hemoglobin A1C, %	6.8	1.3	6.8	1.3	6.8	1.3	0
Duration between diabetes diagnosis and index date, y	5.9	4.2	5.8	4.2	6.3	4.4	12
Comorbid conditions
Smoking	34 852	37.0%	29 332	38.3%	5520	31.3%	15
Alcohol abuse	18 447	19.6%	15 674	20.5%	2773	15.7%	12
Opioid abuse	2499	2.7%	2138	2.8%	361	2.0%	5
Hypertension	85 277	90.5%	69 185	90.3%	16 092	91.3%	4
Myocardial infarction	8210	8.7%	6704	8.7%	1506	8.5%	1
Atrial fibrillation	8608	9.1%	6328	8.3%	2280	12.9%	15
Chronic kidney disease	14 693	15.6%	11 780	15.4%	2913	16.5%	3
Neurological disorders	34 800	36.9%	27 439	35.8%	7361	41.8%	12
Concurrent medications
Insulin	12 159	12.9%	10 001	13.1%	2158	12.2%	2
Metformin	38 292	40.6%	31 873	41.6%	6419	36.4%	11
Glucagon‐like peptide‐1	523	0.6%	396	0.5%	127	0.7%	3
Sodium‐glucose cotrasporter‐2 inhibitors	555	0.6%	417	0.5%	138	0.8%	3
Other diabetes medication	20 835	22.1%	17 531	22.9%	3304	18.8%	10
Multivitamin	21 182	22.5%	15 882	20.7%	5300	30.1%	22
Health care use
Hospitalization(s) in past year
0	86 501	91.8%	70 031	91.4%	16 470	93.5%	8
1	5912	6.3%	5030	6.6%	882	5.0%	7
2+	1826	1.9%	1559	2.0%	267	1.5%	4
Number of visits in past year	22.9	22.6	22.7	22.5	23.5	23.3	3
Vital signs and laboratory values
Serum magnesium, mg/dL	2.0	0.3	2.0	0.3	1.9	0.3	3
<1.7 mg/dL	1591	1.7%	704	0.9%	887	5.0%	24
1.7–2.5 mg/dL	14 909	15.8%	11 357	14.8%	3552	20.2%	14
>2.5 mg/dL	309	0.3%	236	0.3%	73	0.4%	2
Unknown	77 430	82.2%	64 323	84.0%	13 107	74.4%	24

ASD indicates absolute standardized difference; and IPTW, inverse probability of treatment weighting.

*Other included, Asian or Pacific Islander, American Native Hispanic, multiple racial groups or unknown.

Magnesium Supplement Use Identified Through Natural Language Processing

The main exposure was patients’ self‐reported current use of non‐prescribed magnesium supplements in clinic visit notes. We extracted patients’ self‐reported current use of magnesium supplements from clinic visit notes using natural language processing (NLP) methods. A detailed description of the NLP methods is provided in Data S1. Our model achieved an area under curve of 98.9%, a precision of 96.2%, and a recall of 87.7% for magnesium use identification.

Static Treatment Strategies—Remaining on Initiated Treatment of Magnesium Use Versus Nonuse in Follow‐Up

To ensure that the patient's magnesium supplement use did not change during follow‐up, the clinic visit notes of all study patients were checked every 12 months throughout the follow‐up period. If clinic visit notes did not report magnesium supplement use for >1 year, then it is considered a magnesium supplement discontinuation 1 year after the last documented magnesium supplement use. Similarly, if a magnesium nonuser had a clinic visit note documenting magnesium supplement use, it was considered a change of magnesium supplement status and censored from the follow‐up. Any study patient with no clinic visit notes for >1 year was considered lost to follow‐up and censored 1 year after the last clinic visit note.

Target Trial Emulation

We conducted emulated target trials by “enrolling” patients into the study cohort every 6 months from January 1, 2006 to December 31, 2020. The 6‐month periods of cohort enrollment were chosen to have a larger sample size of magnesium users for comparison against nonusers, given that most patients do not use magnesium supplements. A patient would be eligible for a trial if they satisfied the following conditions at an outpatient visit date: (1) already diagnosed with diabetes, (2) aged 40 years or older, (3) free of the primary outcome of interest—HF, (4) had a hemoglobin A1C test within the past 12 months (to ensure the patient was actively managed in the VHA for diabetes), (5) no medical prescription or supplement for magnesium in the medical record at any time before and (6) a clinic visit note within 30 days after they meet the trial eligibility criteria, given that the determination of magnesium supplement use via NLP was through the clinic visit note(s).

Because a patient can potentially be eligible for multiple emulated trials during the study period and a large majority of patients did not have magnesium supplementation in the 30‐day time window since the study entry date, we applied a single eligibility approach to enrolling patients. ³⁷ In a single eligibility approach, once a patient was enrolled in a trial, they would not be eligible to be reenrolled in subsequent trials. Therefore, we selected the first eligible date for the emulated trial enrollment for the magnesium supplement users to include as many magnesium supplement users as possible. Similarly, for the nonmagnesium supplement user (nonuser) group, we randomly selected 1% from the nonuser pool of every 6 months based on the first eligible date of their emulated trial enrollment. The index date (trial entry date) is defined as the first eligible date, as described. The process of a target trial emulation is displayed in Figure 1.*

Figure 1. An example of a target trial emulation using a study sample from January to June 2006.

*Enrollment procedure is applied repeatedly every 6 months. A patient can be enrolled in a trial once. HF indicates heart failure.

Study Covariates

Baseline covariates included the duration between the initial diabetes diagnosis and trial entry (index) date, enrollment year, demographics, comorbid conditions, multivitamin use, medication use, laboratory data, vital signs, and health care use. The list of covariates is included in Table 1 and Table S1. Comorbid conditions were identified using ICD codes. Multivitamin use was identified via NLP (Data S1) on the free text of clinic visit notes and prescription/fill records from the VHA electronic health records. Medication use was identified from prescription/fill records. Health care use was identified from outpatient and inpatient encounter records. Laboratory tests were identified using Logical Observation Identifier Names and Codes. Vital signs were identified from the vital sign records. Chronic conditions were captured from the electronic health records any time before the index or at the index date. Multivitamin use, medication use, laboratory values, vital signs, and health care use were captured from the time window of 1 year before the index date up to the index date, with the measures that were collected closest to the index date as the baseline data. The purpose of including so many baseline variables as covariates was to control for potential confounding bias as much as possible and to make the 2 treatment strategy groups comparable.

To account for the impact of socioenvironmental and economic status that could confound the study results, we added 2 variables: the median household income as well as the social‐environmental percentile ranking, based on the patients’ residential address zip code. The social‐environmental percentile ranking is a ranking of communities based on the composite data from the environmental justice index from the Centers for Disease Control and Prevention to measure the cumulative impacts of environmental burden on human health and the environment, where higher percentile ranking denotes higher cumulative burden of socioenvironmental impact on the residential community compared with lower ranking communities. ³⁸

Study Outcomes

The primary outcome was newly diagnosed (incident) HF. Incident HF was identified using ICD codes (ICD‐9: 428.X, 398.91, 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93; ICD‐10: I50.X, I11.0, I13.0, I13.2, I09.81, I97.13, I97.130, I97.131). An HF diagnosis was confirmed if the diagnosis was from an inpatient hospitalization or 2 outpatient visits, with the first date of diagnosis as the event date.

The secondary outcomes were major adverse cardiac events (MACE), defined as all‐cause death, hospitalization for HF, ischemic stroke, or acute myocardial infarction.

Patients were followed from the index date until censored due to magnesium supplement use status change (ie, stop use in users or start use in nonusers), prescribed magnesium was started, the outcome of interest occurred, or the end of 10th year since the index date, whichever occurred first.

Statistical Analysis

We applied the per‐protocol approach, which means no magnesium supplement discontinuation in users or magnesium supplement initiation in nonusers, to estimate the effect of magnesium supplement use and the risk of HF.

Propensity Score and Inverse Probability Weighting

In the primary analysis, we estimated the propensity score (or probability of being a magnesium supplement user) and inverse probability treatment weighting (IPTW) via a logistic regression model using all the baseline characteristics listed in Table S1 to predict magnesium supplement use. Standardized weights were calculated for each individual using the following formulas: ${\mathit{\text{standardized weights}}}_{\mathit{\text{exposed}}}=\frac{\mathit{\text{proportion exposed}}}{\mathit{\text{propensity score}}}$ for magnesium users and ${\mathit{\text{standardized weights}}}_{\mathit{\text{unexposed}}}=\frac{\mathit{\text{proportion unexposed}}}{1-\mathit{\text{propensity score}}}$ for nonusers. It is commonly accepted that a weight >10 is large, and any weight >10 should be trimmed down to 10. Therefore, we also followed this rule in our study. ³⁹ The weighting process is to generate a pseudo‐population that represents the characteristics of the overall cohort combined, including both magnesium users and nonusers. After weighting, magnesium users and nonusers were expected to share similar characteristics.

We calculated absolute standardized difference for each characteristic before and after weighting, with an absolute standardized difference >10% indicating an imbalance in characteristics between the 2 groups. ⁴⁰ We used the weighted cohort balanced on their baseline characteristics to estimate HF incidence rates by year between the 2 groups. We plotted Kaplan–Meier curves and used Cox regression modeling on the weighted cohort to evaluate the effect of magnesium supplement use and the risk of incident HF. We checked for proportional hazard assumption by adding an interaction term between magnesium supplement use and time in the Cox regression model to test for time effects, if any, on outcomes. We also checked for statistical interactions between magnesium supplement use and potential effect modifiers such as age (<65 versus ≥65 years), serum magnesium level, serum vitamin D level, vitamin D prescription, multivitamin use, and diuretic use by adding a product term to the regression models. When a product term was statistically significant, we conducted subgroup analyses. In subgroup analyses, we repeated the IPTW process to evaluate the effect of magnesium supplement use within each subgroup.

We also conducted a sensitivity analysis using the propensity matching method to evaluate the robustness of the treatment effect. As such, we matched nonusers to magnesium supplement users with a sample ratio of 1:1 based on their propensity score similarity.

Analyses were conducted using SAS (Version 9.4). A 2‐sided P‐value of <0.05 was considered significant.

RESULTS

Baseline Characteristics

A total of 94 239 patients (17 619 magnesium supplement users and 76 620 magnesium nonusers) were enrolled in the target trials. Patient characteristics before and after weighting are summarized in Tables 1 and Table 2, Tables S1 and S2. Before weighting, many key baseline characteristics had absolute standardized difference values >10%, suggesting a substantial between‐group differences. The magnesium users were older (69.1 versus 66.9 years), more likely to be female (7.0% versus 4.4%) but less likely to be Black (10.2% versus 20.4%) or Hispanic (3.4% versus 6.4%) than magnesium nonusers. Magnesium supplement users had a higher median household income, better social‐environmental percentile ranking (46.0 versus 50.9) and were less likely to have a history of alcohol abuse (15.7% versus 20.5%) or smoking (31.3% versus 38.3%), than nonusers. Magnesium supplement users were more likely to have a history of atrial fibrillation (12.9% versus 8.3%) but were similar in the prevalence of hypertension (91.3% versus 90.3%), myocardial infarction (8.5% versus 8.7%) and ischemic stroke (8.6% versus 8.5%) than nonmagnesium users. In addition, magnesium supplement users were also more likely to have a low serum magnesium level (5.0% versus 0.9%) and to take multivitamins (30.1% versus 20.7%); however, they were less likely to be treated with anti‐diabetes medications (ie, metformin) or antihypertensive medications (ie, angiotensin‐converting enzyme inhibitors, calcium channel blockers, or thiazides).

Table 2.

Baseline Characteristics of the Study Participants After IPTW Weighting

Baseline characteristics	IPTW‐weighted overall (N=93 887)		IPTW‐weighted magnesium nonusers (N=76 472)		IPTW‐weighted magnesium users (N=17 415)		ASD after IPTW (%)
	Mean/N	SD/%	Mean/N	SD/%	Mean/N	SD/%
Demographics
Age, y	67.4	10.3	67.4	10.3	67.4	10.0	0
Male sex	89 145	94.9%	72 649	95.0%	16 496	94.7%	1
Race
White	66 735	71.1%	54 293	71.0%	12 443	71.4%	1
Black	17 275	18.4%	14 143	18.5%	3132	18.0%	1
Other*	9878	10.5%	8038	10.5%	1840	10.6%	0
Hispanic ethnicity	5394	5.7%	4456	5.8%	938	5.4%	2
Environmental Justice Index Social‐Environmental Percentile ranking	49.8	22.7	49.9	22.8	49.6	22.5	1
Diabetes information
Hemoglobin A1C, %	6.8	1.3	6.8	1.3	6.8	1.3	0
Duration between diabetes diagnosis and index date, y	5.9	4.2	5.9	4.2	6.0	4.3	2
Comorbid conditions
Smoking	34 827	37.1%	28 288	37.0%	6539	37.5%	1
Alcohol abuse	18 558	19.8%	15 008	19.6%	3549	20.4%	2
Opioid abuse	2554	2.7%	2048	2.7%	506	2.9%	1
Hypertension	85 125	90.7%	69 249	90.6%	15 876	91.2%	2
Myocardial infarction	8251	8.8%	6681	8.7%	1570	9.0%	1
Atrial fibrillation	8696	9.3%	7040	9.2%	1656	9.5%	1
Chronic kidney disease	14 850	15.8%	12 005	15.7%	2844	16.3%	2
Neurological disorders	35 098	37.4%	28 395	37.1%	6703	38.5%	3
Concurrent medications
Insulin	12 341	13.1%	9961	13.0%	2379	13.7%	2
Metformin	38 288	40.8%	31 141	40.7%	7146	41.0%	1
Glucagon‐like peptide‐1	541	0.6%	435	0.6%	107	0.6%	0
Sodium‐glucose cotrasporter‐2 inhibitors	565	0.6%	463	0.6%	103	0.6%	0
Other diabetes medication	20 886	22.2%	16 954	22.2%	3932	22.6%	1
Multivitamin	21 350	22.7%	17 310	22.6%	4040	23.2%	1
Health care use
Hospitalization(s) in past year
0	86 012	91.6%	70 136	91.7%	15 876	91.2%	2
1	6030	6.4%	4844	6.3%	1186	6.8%	2
2+	1846	2.0%	1493	2.0%	353	2.0%	0
Number of visits in past year	23.2	23.2	23.0	23.2	24.1	22.9	5
Vital signs and laboratory values
Serum magnesium, mg/dL	2.0	0.3	2.0	0.3	2.0	0.2	0
<1.7 mg/dL	1596	1.7%	1300	1.7%	296	1.7%	0
1.7–2.5 mg/dL	15 054	16.0%	12 166	15.9%	2887	16.6%	2
>2.5 mg/dL	312	0.3%	255	0.3%	57	0.3%	0
Unknown	2.0	0.3	2.0	0.3	2.0	0.2	0

Abbreviation: ASD indicates absolute standardized difference; and IPTW, inverse probability of treatment weighting.

^*Other included, Asian or Pacific Islander, American Native Hispanic, multiple racial groups or unknown.

The mean±SD of IPTW weights were 1.00±0.36 (1.00±0.20 for nonmagnesium users and 0.99±0.71 for magnesium supplement users), with the minimum value of 0.20 (0.82 for nonmagnesium users and 0.20 for magnesium supplement users) and maximum value of 8.85 (6.75 for nonmagnesium users and 8.85 for magnesium supplement users).

After weighting, absolute standardized difference values for all measured baseline characteristics in the weighted cohort were <10% (Table 2; Table S2). The IPTW‐weighted pseudo cohort had a mean age of 67.4±10.3 years and comprised 18.4% Black, 5.7% Hispanic, and 5.1% female patients. There were 90.7% of patients with hypertension, 9.3% with history of atrial fibrillation, 8.8% with prior myocardial infarction and 8.6% with prior ischemic stroke. There were 19.8% of patients with history of alcohol abuse, 37.1% with history of smoking, and 24.1% of patients in the lowest quartile of median household income, with the average social‐environmental percentile ranking of 49.8±22.7. A total of 53.9% of patients were treated with antidiabetes medications and 76.7% with antihypertensive medications.

Nonprescription Magnesium Supplement Use and Heart Failure in the IPTW Weighted Cohort

The IPTW weighted cohort was followed up to 10 years, with a mean±SD) follow‐up duration of 4.7±3.3 years for magnesium supplement users and 5.4±3.1 years for nonusers. The mean duration of magnesium‐supplement use was 3.5±3.1 (interquartile range, 1.1–5.1) years for users. HF incidence was 8.0% (16.8 events per 1000 person‐years) in magnesium supplement users and 9.7% (18.0 events per 1000 person‐years) in nonusers, respectively in the weighted cohort.

The Kaplan–Meier curves of the weighted cohort indicated that magnesium users had a significantly improved HF‐free survival than nonusers (P<0.05 for the log‐rank test). Cox regression modeling showed that magnesium supplement use was associated with a significantly reduced risk of incident HF (hazard ratio [HR], 0.94 [95% CI, 0.89–0.99]; P=0.0268, Figure 2A). The interaction between magnesium supplement use and time was not statistically significant (P=0.3817), indicating that the proportional hazard assumption was not violated. The sensitivity analysis using propensity score matching analysis showed similar results (patient characteristics before and after matching were summarized in Table S3; HR for incident HF, 0.88 [95% CI, 0.82–0.95]; P=0.0009, Figure S1).

Figure 2. Kaplan–Meier curves of the outcomes in IPTW weighted cohort.

A, Kaplan–Meier curves of HF in IPTW weighted cohort; B, Kaplan–Meier curves of MACE in IPTW weighted cohort. HF indicates heart failure; HR, hazard ratio; IPTW, inverse probability treatment weighting; and MACE, major adverse cardiac events.

The association between magnesium supplement use and reduced incident HF risk was homogeneous across subgroups of the IPTW weighted cohort stratified by age (P=0.3508 for interaction), sex (P=0.2206), body mass index (P=0.3927), serum vitamin D levels (P=0.2395), prescribed vitamin D users (P=0.5455), multivitamin users (P=0.2203), and diuretic users (P=0.1379 for thiazides and P=0.3612 for loop diuretics). The association between magnesium supplement use and HF risk reduction appeared to be stronger in the White and Black subpopulations and less evident for the rest (P=0.0221 for interaction with race). Serum magnesium levels also significantly modified the effect of magnesium supplement use and incident HF, with a 20% risk reduction of HF in patients with low serum magnesium, compared with 8% in patients with normal magnesium and 6% in patients with unknown magnesium levels (P=0.0023 for interaction with serum magnesium) (Figure 3).

Figure 3. Subgroup analyses stratified by covariates that significantly interacted with magnesium supplement use.

As for outcome of HF, interaction between magnesium supplement use and serum magnesium: P=0.0023; interaction between magnesium supplement use and race: P=0.0221. As for outcome of MACE, interaction between magnesium supplement use and serum magnesium: P<0.0001; interaction between magnesium supplement use and serum vitamin D: P=0.0391; interaction between magnesium supplement use and multivitamin use: P=0.0004. HF indicates heart failure; HR, hazard ratio; and MACE, major adverse cardiac events.

Nonprescription Magnesium Supplement Use and MACE in the IPTW Weighted Cohort

MACE incidence was 18.2% (38.2 events per 1000 person‐years) in magnesium supplement users and 22.1% (40.8 events per 1000 person‐years) in nonusers, respectively, in the IPTW weighted cohort. Magnesium supplement use was associated with a significantly reduced risk of MACE (HR, 0.94 [95% CI, 0.90–0.97]; P=0.0009; Figure 2B). The interaction between magnesium supplement use and time was not statistically significant (P=0.3839), indicating the proportional hazard assumption was not violated. Among the subgroups of interest, significant heterogeneity in the association between magnesium supplement use and MACE was found only in subgroups stratified by serum magnesium levels (effect favoring lower levels), serum vitamin D levels (effect favoring higher levels), and multivitamin use (effect favoring users) (Figure 3).

DISCUSSION

In a national cohort of veteran patients ≥40 years with newly diagnosed diabetes and without prior HF, we showed that the use of magnesium supplements for a mean duration of 3.5 years was associated with a 6% relative reduction per year in the risk of incident HF and of MACE (death, HF hospitalization, ischemic stroke, or acute myocardial infarction), respectively.

Previous studies have demonstrated an association between intake of magnesium‐containing foods and reduced risk of HF in women ³⁶ and HF hospitalizations in the Black American population. ³⁵ However, the use of magnesium supplements and incident HF has not been demonstrated, especially in patients with diabetes. Moreover, the relationship between magnesium supplement use and MACE in patients with or without diabetes is uncertain. ⁴¹ , ⁴² Our findings provide suggestive evidence that for patients with diabetes, the use of magnesium supplements was associated with a small (~6% per year) but significant risk reduction of incident HF and MACE over time. Indeed, the survival analysis showed that the group's event rates did not start to separate until Year 3, which suggests that the potential benefits may only be realized with sustained use of magnesium supplements over a longer period.

Patients with diabetes may develop HF as a result of risk factors such as coronary artery disease and hypertension or may directly develop diabetic cardiomyopathy due to increased myocardial fibrosis, advanced glycation end‐product deposition, and increased cardiomyocyte resting tension. ⁴³ , ⁴⁴ Intracellular magnesium balance is important in maintaining peripheral glucose use, ⁴⁵ insulin signaling, ⁴⁶ oxidative stress, ⁴⁷ , ⁴⁸ and inflammation. ⁴⁹ A low magnesium state may lead to myocardial fibrosis, ⁵⁰ , ⁵¹ , ⁵² , ⁵³ left ventricular hypertrophy, and dysfunction, eventually leading to incident HF in patients with diabetes. ⁵⁴ , ⁵⁵ Therefore, it is mechanistically plausible that magnesium supplement use may delay HF onset in patients with diabetes by reducing left ventricular fibrosis and hypertrophy, thereby improving left ventricular systolic and diastolic function. In addition, given the magnesium effects on hypertension, endothelial function, and inflammation, ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ it is likely that these are also potential pathways involved in the association of magnesium supplement use and the reduced risk of myocardial infarction and stroke in our findings.

The implications of these findings are many because there are currently ~30 million Americans with diabetes ⁵⁶ , ⁵⁷ and 6 million Americans with HF. ⁵⁸ , ⁵⁹ Dietary supplements are inexpensive and generally safe. The findings are suggestive that long‐term magnesium supplement intake in patients with diabetes may provide cardiovascular benefits. The subgroup analyses that relate higher risk reductions with lower serum magnesium levels are supportive of the potential role of magnesium use among individuals with known, or who are at risk for, low serum magnesium levels. Of note, most patients in the study cohort did not have a serum magnesium level before their self‐reported magnesium supplement use despite showing risk reductions associated with magnesium supplement use. This finding is meaningful because it would be impractical to check serum magnesium levels for millions of potentially eligible individuals with diabetes who might be at risk for hypomagnesemia before taking an over‐the‐counter magnesium supplement.

Our study has some limitations. First, magnesium supplement use was identified from self‐reported data in the clinical notes via NLP. Although the NLP methodology has been validated with an area‐under‐curve score of 98.9%, misclassification of exposure can still occur, which could have diluted the strength of the association. Second, despite the use of state‐of‐the‐art analytic methods (target trial emulation design, IPTW weighted and propensity matching for sensitivity analysis) and an extensive list of 88 covariates to minimize bias and confounding, the possibility of residual confounding may still exist, given the observational nature of the study design. For example, there was a lack of information about the indications or reasons that individuals used magnesium supplements, which could result in indication bias. The magnesium supplement users who engaged in a range of health‐promoting behaviors were generally healthier than nonusers, which could incur healthy user bias. For example, the lower prevalence of antidiabetes and antihypertensive medication use for the magnesium supplement users versus nonusers despite similar hemoglobin A1c, serum creatinine, and systolic blood pressure values may suggest that the magnesium supplement user population likely had healthier health care behaviors at baseline versus nonusers since they needed less medications to achieve similar blood pressure and glycemic values. Thus, we used a new‐user design, target trial emulation, and IPTW that accounted for an extensive list of covariates, including socioeconomic and environmental covariates, to minimize selection bias.

Third, given the vast spectrum of magnesium supplements available in the market, we cannot ascertain the amount of elemental magnesium contained in the supplement being consumed by the patient. Besides, our study generalizability may be limited to older men with diabetes given the demographics of the veteran population.

Conversely, the strengths of the study include using a national electronic clinical database that contains a large, ethnically diverse veteran patient population with an extensive list of clinical and laboratory covariates that provided a comprehensive clinical picture of the study cohort, which allowed us to detect small but meaningful associations related to magnesium supplement use.

Conclusions

In conclusion, our study showed that long‐term magnesium supplementation was associated with a lower risk of HF and MACE in patients with diabetes. The potential benefits of magnesium supplement use should be further confirmed in an RCT.

Sources of Funding

Research reported in this article was supported by the National Health Lung and Blood Institute of the National Institutes of Health Office of the Director under award number 5R01HL156518 and with resources from the Office of Research and Development, Health Services Research and Development, and the use of facilities at the Washington, DC Veterans Affairs Medical Center. The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, the Department of Veterans Affairs, or the US Government.

Disclosures

None.

Supporting information

Data S1

Tables S1–S3

Figure S1