Risk of Ischemic Stroke Associated with Calcium Supplements and Interaction with Oral Bisphosphonates: A Nested Case-Control Study

Conflicting results about the association of calcium supplements (CS) with ischemic stroke (IS) have been reported. We tested this hypothesis by differentiating between CS alone (CaM) and CS with vitamin D (CaD) and between cardioembolic and non-cardioembolic IS. We examined the potential interaction with oral bisphosphonates (oBs). A nested case-control study was carried out. We identified incident IS cases aged 40–90 and randomly sampled five controls per case matched by age, sex, and index date. Current users were compared to non-users. An adjusted odds ratios (AOR) and 95% CI were computed through conditional logistic regression. Only new users were considered. We included 13,267 cases (4400 cardioembolic, 8867 non-cardioembolic) and 61,378 controls (20,147 and 41,231, respectively). CaM use was associated with an increased risk of cardioembolic IS (AOR = 1.88; 95% CI: 1.21–2.90) in a duration-dependent manner, while it showed no association with non-cardioembolic IS (AOR = 1.05; 95% CI: 0.74–1.50); its combination with oBs increased the risk of cardioembolic IS considerably (AOR = 2.54; 95% CI: 1.28–5.04), showing no effect on non-cardioembolic. CaD use was not associated with either cardioembolic (AOR = 1.08; 95% CI: 0.88–1.31) or non-cardioembolic IS (AOR = 0.98; 95% CI: 0.84–1.13) but showed a small association with cardioembolic IS when combined with oBs (AOR = 1.35; 95% CI: 1.03–1.76). The results support the hypothesis that CS increases the risk of cardioembolic IS, primarily when used concomitantly with oBs.


Introduction
Current guidelines for the prevention and treatment of osteoporosis [1][2][3] recommend a daily calcium intake of between 700 to 1200 mg and advise a daily dose of 800 IU of vitamin D (cholecalciferol) in postmenopausal women and older men (≥50 years old) to We performed initial computer research to identify all potential stroke cases using the ICPC-2 code K90 and the ICD-9-CM codes 434.x1, and 436, as well as keywords in diagnosis and comment fields. Stroke cases linked to drug abuse, vascular dissections, and aneurysms were excluded. Then, we grouped the potential cases identified into homogeneous subgroups according to the amount and type of information and extracted a random sample for each subgroup (totaling 1000 cases), and we conducted a manual revision of their complete clinical records. The case validation was carried out independently by two of the investigators (SRM and DBH), who were blinded to drug exposure, and discrepancies were solved by the whole research group (including a neurologist serving in a stroke unit (L.I.-E.) and a cardiologist (A.G.-L.). Further, we performed a second validation to identify the most probable pathophysiological subtype of IS (cardioembolic or non-cardioembolic), described elsewhere [17]. Briefly, an IS was considered cardioembolic if it met the following main criteria: (1) a note that it was cardioembolic accompanied the diagnosis of IS; or (2) a record of atrial fibrillation before or within 3 months after the index date; or (3) prescriptions of anticoagulants recorded before or within 3 months after the index date (74.1% of cardioembolic IS cases met at least two of them). A record of valvular prosthesis or mitral stenosis before the index date and prescriptions of class IC or class III antiarrhythmics recorded before or within three months after the index date, in addition to any of the criteria mentioned above, were considered as supportive information of the diagnosis of cardioembolic stroke. All IS cases not fulfilling these criteria were considered non-cardioembolic.
We considered the "index date" as the date of the first record of IS. Five controls per case individually matched with cases by exact age, sex, and index date were randomly selected from the underlying cohort following risk-set sampling. As controls arose from the underlying cohort, they fulfilled the same criteria as cases: aged 40 to 99 years, 1-year registry with their primary care physician (PCP), and no previous record of cancer or stroke. Such sampling of controls is incidence-density-based (the probability of control selection is proportional to the total person-time at risk) and allows for obtaining unbiased estimates of the underlying cohort's rate ratios (RRs) through the ORs computed in the case-control analysis [18].

New-User Design
We only selected initiators of calcium supplements among cases and controls. To this end, we excluded subjects with prescriptions of calcium supplements prior to the cohort start date.

Exposure Definition
New users of calcium supplements were categorized as "current users" when they had at least a recorded prescription that ended within 365 days before the index date and "past users" when the prescription ended beyond 365 days before the index date. Those patients without a recorded prescription of calcium supplements before the index date were considered "non-users". Among current users, we distinguished those who used them as monotherapy (CaM) or with vitamin D (CaD) (either in fixed-dose combinations or as separate medicinal products) and obtained the daily dose ("low" when it was less than 1000 mg of elemental calcium and "high" when it was equal to or over 1000 mg), and the treatment duration (sum of all consecutive prescriptions with a maximum accepted gap of 60 days between the end of the supply of one prescription and the start of the next one). The duration was categorized as equal to or less than 365 days and over 365 days.

Potential Confounding Factors
To compute the adjusted odds ratios (AORs), the following co-morbidities and other risk factors were considered as possible confounding factors using expert criteria (all of them recorded at the index date or before): Number of visits to primary care physician (PCP) during the year preceding the index date, body mass index (BMI), alcohol abuse, smoking, diabetes (recorded as such and/or by using glucose-lowering drugs as an indicator), hypertension, dyslipidemia (recorded as such and/or by using lipid-lowering drugs as an indicator), hyperuricemia (either asymptomatic or in the context of gout), peripheral artery disease (PAD), acute myocardial infarction (AMI), angina pectoris (recorded as such and/or use of nitrates), transient ischemic attack (TIA), thromboembolic disease, heart failure, chronic kidney failure, rheumatoid arthritis, and chronic obstructive pulmonary disease (COPD). All these comorbidities were recorded as such by the PCPs and further validated by the BIFAP team through the manual review of a small sample. For some comorbidities with specific drugs (diabetes, dyslipidemia, angina), we also used such drugs as disease indicators (as specified). Furthermore, current use of the following drugs was considered as a covariate: antiplatelet drugs, alpha-blockers, beta-blockers, calcium channel blockers, angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, diuretics, nonsteroidal anti-inflammatory drugs (both non-selective and COX-2 selective), paracetamol (acetaminophen), metamizole, opioids, systemic corticosteroids, proton pump inhibitors, H2-receptors antagonists, active forms of vitamin D (calcifediol, calcitriol, and alfacalcidol), hormonal replacement therapy, selective estrogen receptor modulators, strontium ranelate, oral bisphosphonates, and other agents for osteoporosis treatment (calcitonin, denosumab, and teriparatide). Atrial fibrillation and oral anticoagulant use were not considered potential confounding factors because they were part of the criteria used to define the cardioembolic IS (see Section 2.2).

Statistical Analysis
To estimate the association between calcium supplements and incident IS, we built a conditional logistic regression model and computed the odds ratio (OR) and its 95% confidence intervals (CIs), considering only the matching variables (exact age, sex, and index date). Further, we computed the adjusted odds ratios (AORs) by adding the confounding factors detailed above.
The potential interaction of calcium supplements with age (less than 70 and ≥70 years old), sex, background vascular risk (see below), CHA2DS2-VASc score (equal to 3 or less and greater), and current use of oral bisphosphonates was explored through stratified analyses. Background vascular risk was categorized into: (1) establish vascular disease, including patients with a record of PAD, AMI, or TIA; (2) vascular risk factors, including patients with a record of hypertension, diabetes mellitus, chronic renal failure, atrial fibrillation, dyslipidemia, current smoking, or BMI higher than 30 kg/m 2 without any of the vascular diseases aforementioned; and (3) no vascular risk factors, the remainder.
AORs obtained in different strata were compared using the interaction test described by Altman and Bland [19] (multiplicative interaction). The potential interaction of calcium supplements with oral bisphosphonates was also examined in the additive scale [20]. To that end, we built a variable with seven categories: (1) non-use of calcium supplements and non-use of oral bisphosphonates; (2) current use of CaM alone; (3) current use of CaD alone; (4) current use of oral bisphosphonates alone; (5) current use of CaM and oral bisphosphonates; (6) current use of CaD and oral bisphosphonates; and (7) the remainder (those patients who are: (a) past users of calcium supplements and past users of oral bisphosphonates; or (b) past users of calcium supplements and non-users of oral bisphosphonates; or (c) past users of oral bisphosphonates and non-users of calcium supplements).
We used unconditional logistic regression models for the stratified analysis by variables different from the matching ones, as conditional models produced unstable estimates.
We had missing values for BMI (30%) and smoking (46%) that were identified in specific categories within the variables. In a sensitivity analysis, however, we assessed their impact on the main results by applying multiple imputations by chained equation models (MICE) [21].
We only computed the AOR when there were more than five exposed cases per category. A p-value of less than 0.05 was considered statistically significant. No adjustment of multiple testing was carried out. All analyses were performed with STATA/SE 15 (StataCorp, College Station, TX, USA).

Ethical Aspects
The study was conducted according to the principles of the Helsinki Declaration (2013), as well as Spanish and European laws. The BIFAP Scientific Committee approved the study protocol of a bigger project on atherothrombotic events and the use of different drugs, which included the current one, on 26 May 2016 (project #04/2016). The present study was approved by the Research Ethics Committee of the University Hospital Príncipe de Asturias (Ref #CAL-BIS-CACO, #EOm 05/2022) on 29 March 2022. The data used were fully pseudonymized, and the Committee granted a waiver for the informed consent.

Results
A total of 13,267 incident IS cases (4400 classified as cardioembolic and 8867 as noncardioembolic), and 61,378 matched controls (20,147 with cardioembolic cases and 41,231 with non-cardioembolic cases) were included ( Figure 1). fully pseudonymized, and the Committee granted a waiver for the informed consent.

Results
A total of 13,267 incident IS cases (4400 classified as cardioembolic and 8867 as noncardioembolic), and 61,378 matched controls (20,147 with cardioembolic cases and 41,231 with non-cardioembolic cases) were included ( Figure 1). Baseline characteristics of cardioembolic and non-cardioembolic IS cases and their respective controls are shown in Table 1 (see Supplementary Material Table S2 for IS cases overall and their controls). As expected, cases had a higher prevalence of comorbidities and use of drugs for cardiovascular conditions than controls. Among controls, we explored the prevalence of comorbidities, cardiovascular risk factors, and co-medication of current users of CaM and CaD as compared to non-users. As shown in Figure S1 and Table  S1 (Supplementary Materials), CaM and CaD users presented a higher prevalence of comorbidities and co-medications than non-users, with hardly any difference between CaM and CaD users.  Baseline characteristics of cardioembolic and non-cardioembolic IS cases and their respective controls are shown in Table 1 (see Supplementary Material Table S2 for IS cases overall and their controls). As expected, cases had a higher prevalence of comorbidities and use of drugs for cardiovascular conditions than controls. Among controls, we explored the prevalence of comorbidities, cardiovascular risk factors, and co-medication of current users of CaM and CaD as compared to non-users. As shown in Figure S1 and Table S1 (Supplementary Materials), CaM and CaD users presented a higher prevalence of comorbidities and co-medications than non-users, with hardly any difference between CaM and CaD users. Cardioembolic IS cases were older (mean (SD) age: 76.5 ± 11.39 years old) than noncardioembolic IS cases (mean (SD) age: 73.2 ± 12.85) and also had a higher prevalence of comorbidities and co-medications.

CaM and Risk of Ischemic Stroke Overall and by Pathophysiological Subtype
Overall, IS cases presented a higher proportion of current users of CaM than controls (0.58% and 0.41%, respectively), yielding an AOR of 1.25 (95% CI: 0.95-1.63). The increased AOR was only significant with durations longer than 1 year (AOR >1 year = 1.78; 95% CI: 1.10-2.86 vs. AOR ≤1 year = 1.07; 95% CI: 077-149). No relationship with calcium daily dose was found (Table 2).  (Table 3). By contrast, for noncardioembolic IS, the AOR was 1.05 (95% CI: 0.74-1.50), and no change was observed with duration or daily dose (Table 4).  In the stratified analyses by subgroups of age, sex, background vascular risk, and CHA2DS2-VASc score, we observed that the current use of CaM was associated with an increased risk of cardioembolic IS in almost all the strata, with no evidence of statistical interaction with any of those variables. No increased risk associated with non-cardioembolic IS was observed among current users of CaM in any subgroup (Figure 2).
We did not find a significant increased risk for any IS subtype. For cardioembolic IS, the AOR was 1.08 (95% CI: 0.88-1.31), and no significant change was observed with duration or daily dose (Table 3). For non-cardioembolic IS, the AOR was 0.98 (95% CI: 0.84-1.13), and no change was observed with the daily dose of calcium; however, with a duration longer than 365 days, an AOR of 1.40 (95% CI: 1.11-1.77) was found (Table 4). There was no evidence of the interaction of CaD with age, sex, background vascular risk, and CHA2DS2-VASc score for both cardioembolic and non-cardioembolic IS (Figure 3).
We did not find a significant increased risk for any IS subtype. For cardioembolic IS, the AOR was 1.08 (95% CI: 0.88-1.31), and no significant change was observed with duration or daily dose (Table 3). For non-cardioembolic IS, the AOR was 0.98 (95% CI: 0.84-1.13), and no change was observed with the daily dose of calcium; however, with a duration longer than 365 days, an AOR of 1.40 (95% CI: 1.11-1.77) was found (Table 4). There was no evidence of the interaction of CaD with age, sex, background vascular risk, and CHA2DS2-VASc score for both cardioembolic and non-cardioembolic IS (Figure 3). Figure 3. Use of calcium supplements with vitamin D (CaD) and risk of the two main subtypes of IS (non-cardioembolic and cardioembolic) by sex, age, vascular risk, and CHA2DS2-VASc score. * VRF (vascular risk factors) included patients with at least one vascular risk factor: hypertension, renal failure, dyslipidemia, diabetes mellitus, atrial fibrillation, current smoking or body mass index higher than 30 kg/m 2 . ** EVD (establish vascular disease) included those with at least one record of peripheral artery disease, acute myocardial infarction, or transient ischemic attack.

Interaction of Calcium Supplements with Oral Bisphosphonates
The current use of CaM was associated with an increased risk of cardioembolic IS in the stratum of patients who concomitantly used oral bisphosphonates (AOR = 3.20; 95% CI: 1.21-8.47), while there was no significant association among non-users of oral bisphosphonates (AOR = 1.40; 95% CI: 0.78-2.53) (p for interaction = 0.1541). For non-cardioembolic IS, the current use of CaM was not associated with an increased risk regardless of the concomitant use of oral bisphosphonates. The use of CaD did not show a statistical interaction with oral bisphosphonates in any subtype of IS ( Figure 4).  (CaD) and risk of the two main subtypes of IS (non-cardioembolic and cardioembolic) by sex, age, vascular risk, and CHA2DS2-VASc score. * VRF (vascular risk factors) included patients with at least one vascular risk factor: hypertension, renal failure, dyslipidemia, diabetes mellitus, atrial fibrillation, current smoking or body mass index higher than 30 kg/m 2 . ** EVD (establish vascular disease) included those with at least one record of peripheral artery disease, acute myocardial infarction, or transient ischemic attack.

Interaction of Calcium Supplements with Oral Bisphosphonates
The current use of CaM was associated with an increased risk of cardioembolic IS in the stratum of patients who concomitantly used oral bisphosphonates (AOR = 3.20; 95% CI: 1.21-8.47), while there was no significant association among non-users of oral bisphosphonates (AOR = 1.40; 95% CI: 0.78-2.53) (p for interaction = 0.1541). For non-cardioembolic IS, the current use of CaM was not associated with an increased risk regardless of the concomitant use of oral bisphosphonates. The use of CaD did not show a statistical interaction with oral bisphosphonates in any subtype of IS ( Figure 4).
The potential interaction between CaM or CaD and oral bisphosphonates on each IS subtype was also examined in the additive scale. For cardioembolic IS, we found an increased risk associated with the concomitant use of CaM and oral bisphosphonates as compared to the non-use of any (AOR = 2.54; 95% CI: 1.28-5.04), while no significant increased risk was observed when either CaM or oral bisphosphonates were used singly (AOR = 1.47; 95% CI: 0.81-2.68; and AOR = 1.08; 95% CI: 0.76-1.56, respectively). Also, the concomitant use of CaD with oral bisphosphonates was associated with a significantly increased risk as compared to the non-use of any (AOR = 1.35; 95% CI: 1.03-1.76), but this was much smaller than with CaM. On the contrary, the combined use of CaM or CaD with oral bisphosphonates did not increase the risk of non-cardioembolic IS ( Figure 5).
compared to the non-use of any (AOR = 2.54; 95% CI: 1.28-5.04), while no significant increased risk was observed when either CaM or oral bisphosphonates were used singly (AOR = 1.47; 95% CI: 0.81-2.68; and AOR = 1.08; 95% CI: 0.76-1.56, respectively). Also, the concomitant use of CaD with oral bisphosphonates was associated with a significantly increased risk as compared to the non-use of any (AOR = 1.35; 95% CI: 1.03-1.76), but this was much smaller than with CaM. On the contrary, the combined use of CaM or CaD with oral bisphosphonates did not increase the risk of non-cardioembolic IS ( Figure 5).   compared to the non-use of any (AOR = 2.54; 95% CI: 1.28-5.04), while no significant in-creased risk was observed when either CaM or oral bisphosphonates were used singly (AOR = 1.47; 95% CI: 0.81-2.68; and AOR = 1.08; 95% CI: 0.76-1.56, respectively). Also, the concomitant use of CaD with oral bisphosphonates was associated with a significantly increased risk as compared to the non-use of any (AOR = 1.35; 95% CI: 1.03-1.76), but this was much smaller than with CaM. On the contrary, the combined use of CaM or CaD with oral bisphosphonates did not increase the risk of non-cardioembolic IS ( Figure 5).

Discussion
The results of the present study suggest that the use of calcium supplements without vitamin D (CaM) increases the risk of cardioembolic IS without affecting the risk of noncardioembolic IS. Such effect was observed with both low and high daily doses of calcium supplements and with any duration of use, though the highest risk appeared with a duration longer than 1 year. No interaction with sex, age, background vascular risk, or CHA2DS2-VASc score was detected. Notably, the results suggest the existence of an interaction with oral bisphosphonates, so the increased risk of cardioembolic IS associated with CaM was only observed when they were used combined with oral bisphosphonates, but not when both drugs were used singly.
The use of calcium supplements with vitamin D (CaD) was neither associated with IS overall nor with any of the main pathophysiological subtypes. However, a moderately increased risk of non-cardioembolic IS was observed with durations longer than 1 year. No interaction with sex, age, background vascular risk, or CHA2DS2-VASc score was found. A moderately increased risk of cardioembolic IS was detected when oral bisphosphonates and CaD were used concomitantly, but it was much smaller than the one found with CaM.
Several meta-analyses of randomized clinical trials (RCTs) that examined the association between calcium supplementation and the risk of cardiovascular events and stroke have been published, showing conflicting results. In 2010 and 2011, Bolland et al. [9,10] reported an association between the use of calcium supplements with an increased risk of myocardial infarction and stroke. In 2013, Mao P.-J. et al. [22] reported that there might be an increased risk of major cardiovascular events (myocardial infarction and coronary heart disease) associated with the use of calcium supplements with or without vitamin D. However, they did not find a statistically significant increased risk of stroke. In 2015, Lewis et al. [23] and in 2016, Chung et al. [11] reported that the use of calcium supplements was not associated with cardiovascular risk in generally healthy adults. In 2020, Yang et al. [24], in a meta-analysis including 26 prospective cohort studies and 16 RCTs, concluded that calcium intake from diet did not increase the risk of cardiovascular events or stroke, whereas calcium supplements could increase the risk of coronary heart disease, especially myocardial infarction, but not the risk of stroke. In 2021, Myung et al. [25] concluded that calcium supplements were associated with a significantly increased risk of cardiovascular events by about 15% and a non-significant increased risk of stroke by about 13% in healthy postmenopausal women. Finally, in 2023, Huo et al. [26] found with CaM a non-significant trend for both acute myocardial infarction (RR = 1.15; 95% CI: 0.88-1.51) and stroke (1.15; 95% CI: 0.90-1.46), while for CaD the trend was much smaller for AMI (1.09; 95% CI: 0.95-1.25) and almost null for stroke (RR = 1.02; 95% CI: 0.89-1.17).
All in all, the evidence from RCTs of the increased risk of using calcium supplements is weak either used alone or combined with vitamin D. Nevertheless, it is important to note that the duration was not analyzed. Also, as far as it concerns stroke, no consideration was given to the different pathophysiological types of strokes (hemorrhagic vs. ischemic, and among ischemic stroke, cardioembolic vs. non-cardioembolic). Also, no information is provided about the concomitant use of bisphosphonates. Our data suggest that all these points matter: duration of treatment, a precise definition of the type of ischemic stroke, and the concomitant use of bisphosphonates. Thus, in our view, the possible association of calcium supplements with ischemic stroke, particularly when given without vitamin D, is open.
Cardioembolic IS occurs as a consequence of endothelial injury, stasis, and hypercoagulability at the cardiac level [27]. Atrial fibrillation is believed to be the most prevalent cause of cardioembolic IS, although growing evidence suggests that atrial thrombogenesis can occur without a rhythm disorder. Kamel et al. [28] have proposed a new model in which an underlying atrial cardiopathy (with or without atrial fibrillation) would be the main biological substrate for thrombogenesis and posterior embolization. The mechanism by which CaM may increase the risk of cardioembolic IS is uncertain.
Calcium is a well-known coagulation factor, and it has been postulated that serum peak levels reached after the intake of calcium supplements may induce a transient hypercoagulation state that would promote thrombotic events [29]. However, this acute effect could not account for the null effect found with calcium supplements associated with vitamin D, nor would it explain the greater effect found with longer treatments and the fact that non-cardioembolic IS remained materially unaffected. The suggested interaction of calcium supplements with oral bisphosphonates is an interesting finding that may provide some clues. Our group recently reported a specific increased risk of cardioembolic IS with oral bisphosphonates [15], which was clearly duration-dependent. Based on the model proposed by Kamel et al. [28] for cardioembolism of atrial origin, we postulated that the long-term use of bisphosphonates might induce an atrial cardiopathy, associated or not with atrial fibrillation, which ultimately could be the primary biological substrate for thrombogenesis. Following this reasoning, the interaction with calcium supplements, especially when used without vitamin D, suggests that calcium and bisphosphonates may potentiate each other to promote long-run atrial cardiopathy, which could be the leading cause of the observed increased risk of cardioembolic IS. In this context, vitamin D may be acting as a preventive factor, which is consistent with abundant evidence suggesting a cardioprotective effect of vitamin D [30][31][32][33][34][35][36][37], including the recently published D-Health trial [38]. Nevertheless, two other major trials (VITAL [39] and VIDA [40]) did not show a preventive cardiovascular effect in a primary prevention setting. Interestingly, in the Finnish Vitamin D trial, Virtanen et al. [41] found that vitamin D 3 supplementation was associated with a reduced risk of atrial fibrillation.
The main strengths of our study are: (1) the PCPs collected the clinical information prospectively and filled the prescriptions using the computer system, making the misclassification of the exposure highly unlikely; (2) controls were randomly sampled from the source population, which prevents a control-selection bias; and (3) only new users were considered, avoiding a prevalent-user bias. Among the limitations are: (1) the present study is observational, and residual confounding due to unmeasured or unknown confounders is still possible; (2) treatment adherence, as in any clinical study, is not guaranteed; (3) despite the validation effort to determine the most probable pathophysiological subtype of IS (cardioembolic and non-cardioembolic), there may still be some misclassification.

Conclusions
The results of the present study suggest that the long-term use of calcium supplements without vitamin D increases the risk of cardioembolic IS, which seems particularly great when used in combination with oral bisphosphonates. Our results also suggest a moderately increased risk of cardioembolic IS when calcium supplements with vitamin D are used combined with oral bisphosphonates, but of much lesser magnitude. A moderately increased risk of non-cardioembolic ischemic stroke associated with the long-term use of calcium supplements with vitamin D is also suggested. These data urge us to reconsider the use of calcium supplements in preventing and treating osteoporosis [42,43], in particular as a complement to antiresorptive therapy with oral bisphosphonates.
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/jcm12165294/s1, Figure S1: Vascular and other risk factors (A), and comedication (B) among current users of calcium supplementation with or without vitamin D and non-users. All of them are from the control series of IS (overall); Table S1: Distribution of vascular and other risk factors and comedication in new users of CaM or CaD compared to non-users, among control series of the IS cases (overall); Table S2: Characteristics of ischemic stroke cases and controls; Table S3: Risk of ischemic stroke associated with the use of calcium supplements applying multiple imputations by chained equation models; Table S4: Risk of cardioembolic ischemic stroke associated with the use of calcium supplements, applying multiple imputations by chained equation models; Table S5: Non-cardioembolic ischemic stroke associated with use of calcium supplements, applying multiple imputations by chained equation models. References [16,17] are cited in the supplementary materials.