25-hydroxyvitamin D levels and risk of stroke: A prospective study and meta-analysis

Abstract

Background and Purpose

Despite evidence suggesting that vitamin D deficiency may lead to elevated cardiovascular disease risk, results regarding the association of 25(OH)D levels with stroke risk are inconclusive. We aimed to examine this association in a prospective study in women and to summarize all existing data in a meta-analysis.

Methods

We measured 25(OH)D levels among 464 women who developed ischemic stroke and an equal number of controls who were free of stroke through 2006 in the Nurses’ Health Study (NHS). We searched MEDLINE and EMBASE for articles published through March 2011 that prospectively evaluated 25(OH)D levels in relation to stroke.

Results

After multivariable adjustment for lifestyle and dietary covariates, lower 25(OH)D levels were associated with an elevated risk of ischemic stroke in the NHS: the odds ratio (95% CI) comparing women in the lowest vs. highest tertiles was 1.49 (1.01, 2.18; P_trend=0.04). We found 6 other prospective studies that examined 25(OH)D in relation to stroke outcomes. After pooling our results with these prospective studies that included 1,214 stroke cases in total, low 25(OH)D levels were associated with increased risk of developing stroke outcomes in comparison to high levels: the pooled relative risk (95% CI) was 1.52 (1.20, 1.85; I² = 0.0%, P_{heterogeneity}=0.63). In two studies that explicitly examined ischemic stroke, this association was 1.59 (1.07, 2.12; I² = 0.0%, P_{heterogeneity}=0.80).

Conclusions

These data provide evidence that low vitamin D levels are modestly associated with risk of stroke. Maintaining adequate vitamin D status may lower risk of stroke in women.

INTRODUCTION

Hypertension and diabetes are among the leading risk factors for stroke and targets for stroke prevention.¹ Interestingly, human observational studies strongly suggested beneficial effects of vitamin D, a hormone that primarily regulates calcium metabolism, on lowering the risk of both conditions.^2–4 In addition, mechanistic studies have provided mounting evidence supporting direct vascular effects of vitamin D that may lead to a lower risk of stroke.^{5, 6} Despite the basic science research evidence, epidemiologic studies examining vitamin D status, as measured by 25-hydroxyvitamin D [25(OH)D] levels in blood, in relation to risk of stroke morbidity or mortality have been inconsistent. Some,^7–9 but not all,^10–12 of these studies documented significant, inverse associations between 25(OH)D levels and stroke risk. The inconsistency of the results may be partially explained by insufficient statistical power, because most previous studies only accumulated a small number of stroke cases. Therefore, we conducted the largest prospective study, thus far, to examine the relationship between plasma 25(OH)D levels and risk of ischemic stroke among women who were initially free of stroke in the Nurses’ Health Study (NHS). We also summarized the existing evidence via a meta-analysis to shed light on this important association.

METHODS

Study populations

The NHS cohort consists of 121,700 female registered nurses aged 30 to 55 years who were residing in 11 U.S. states and enrolled in 1976 through responding to a mailed questionnaire on their medical history and lifestyle practices. In 1989–1990, blood samples were collected from 32,826 participants. Among these participants, we prospectively identified and confirmed 483 incident ischemic stroke cases from the date of blood draw through June 2006. We used a risk-set sampling scheme¹³ to randomly select one control for each case from the rest of the population who remained free of stroke when the case occurred; the probability of being selected as a controls is proportional to the length of follow-up. We further matched cases and controls for age at blood draw (±1 year), date of blood draw (±3 months), menopausal status (yes, no), use of postmenopausal hormone (current or non-user), race (white or other races), and smoking status (past, current, or non-smoker). After excluding 19 case-control pairs with missing 25(OH)D data, 464 pairs were available for analysis.

The study protocol was approved by the institutional review board of the Brigham and Women’s Hospital and the Human Subjects Committee Review Board of Harvard School of Public Health.

Ascertainment of ischemic stroke

In the NHS cohort, we requested permission of access to medical records of participants who reported having stroke diagnosis in follow-up questionnaires. Study physicians, who were blinded as to the exposure status of these participants, reviewed medical records including results of computerized tomography or magnetic resonance imaging. An ischemic non-fatal stroke diagnosis was confirmed if the medical records demonstrated a neurological deficit with sudden or rapid onset that persistedfor >24 hours or until death in accordance with the criteria of the National Survey of Stroke,¹⁴ with an underlying etiology of thrombotic or embolic mechanism. Fatal stroke cases were identified from reports of next of kin or postal authorities, or by searching the National Death Index if stroke was indicated as the cause of death in autopsy reports, hospital records, or death certificates. Fatal ischemic stroke cases were confirmed by reviewing medical or autopsy records, following the same diagnostic criteria as above.

Assessment of exposure and covariates

Plasma samples of each case-control pair were handled identically and analyzed in the same run by the same technicians in a random sequence under identical conditions. Plasma 25(OH)D levels were measured using the LIAISON 25 OH Vitamin D TOTAL Assay (DiaSorin, Stillwater, MN), which is a direct competitive chemiluminescence immunoassay performed on the DiaSorin LIAISON automated analyzer. Quality control samples were dispersed throughout each analytical run. Based on the measurements of these control samples, the average intra-assay coefficient of variation was 13.9% for 25(OH)D. Assessments of covariates were described in Supplementary Material.

Statistical methods

We categorized the cases and controls into tertiles according to the distribution of the 25(OH)D levels among the controls. Conditional logistic regressions, in which matching design is taken into account, were used to examine the association between 25(OH)D and stroke risk. In the multivariable analysis, we controlled for covariates that may confound the association of interest, including body mass index (BMI), physical activity, and dietary factors. P values for linear trend were calculated by entering an ordinal score based on the median value in each tertile of 25(OH)D into the models. In a secondary analysis, we calculated the season-adjusted residuals of 25(OH)D levels by regressing this biomarker on the month at blood draw in a linear regression model. We used the residuals that are statistically independent of season as exposures of interest and repeated the analysis.

All P values were 2-sided, and 95% confidence intervals (CI) were calculated for ORs or RRs. Data were analyzed with the Statistical Analysis Systems software package, version 9.2 (SAS Institute, Inc., Cary, NC).

Meta-analysis

We identified and included 6 studies^7–12 that explicitly evaluated 25(OH)D levels in relation to incident stroke or stroke mortality (Figure 1). Study selection and data extraction were described in Supplementary Material. Meta-analyses were performed by using Stata 11.0 (Stata-Corp, College Station TX). We fitted a fixed-effects model to derive summary estimates that were based on the logarithms of RRs and corresponding standard errors in each individual study. Heterogeneity among the results of these studies was evaluated using the Cochran Q test and I² statistic. To detect any non-linear dose-response relationship, we first used a restricted cubic spline regression model (Stata RC_SPLINE command) with three knots to create spline variables of 25(OH)D levels. We then fitted two log-linear dose-response regression models (Stata GLST command); in the first model we only included a continuous 25(OH)D variable, and in the second model we included both linear and spline terms of 25(OH)D levels. Lastly, we used the likelihood ratio test to examine the significance of any non-linearity by comparing the model with the linear term only to the model with both the linear and the cubic spline terms. For this dose-response analysis, we requested data on levels of 25(OH)D and person-time (if Cox regression was used) for each category for six studies.^8–12 Five studies^{7, 9–12} responded to our requests and provided these data. For one study that did not respond to our request,⁸ we multiplied number of subjects by average follow-up time to estimate the person-years. Begg funnel plots and Egger tests were used to assess potential publication bias.^{15, 16}

Figure 1.

Literature search and study selection.

RESULTS

The Nurses’ Health Study

Table 1 shows the baseline characteristics of cases and controls. The cases and controls had similar baseline levels of 25(OH)D. Among controls, 41.6% would be considered vitamin D deficient, with a 25(OH)D levels <50 nmol/L,¹⁷ compared with 45.3% of cases. We also examined the association of 25(OH)D with age, BMI, physical activity, systolic blood pressure, dietary vitamin D intake, race, and season of blood draw among controls. As expected, 25(OH)D levels were significantly (P<0.05) correlated with BMI [Partial Pearson correlation coefficient (r)=−0.23], physical activity (r=0.20), dietary vitamin D intake (r=0.19). In addition, white participants had higher 25(OH)D levels than African Americans [mean 25(OH)D levels were 56.9 vs. 44.7 nmol/L, respectively], and 25(OH)D levels were higher for women living in the Southern states (mean: 59.7 nmol/L) than those who lived in the Northern states (mean: 55.5 nmol/L) or when blood samples were collected during Summer months (mean: 59.8 nmol/L) in comparison to other seasons (mean: 55.8 nmol/L). Plasma 25(OH)D levels were not correlated with age (r=0.01) or systolic blood pressure (r=0.02).

TABLE 1.

Baseline characteristics of ischemic stroke cases and controls, the Nurses’ Health Study.

Characteristics*	Cases (n = 464)	Controls (n = 464)	P value†
Demography and lifestyle
Age (year)‡	60.8±5.9	60.8±6.0	0.93
European ancestry (%)‡	97.6	97.6	>0.99
Body mass index (kg/m2)	26.0±5.1	25.4±4.8	0.09
Physical activity (MET-hr/week)	15.1±19.6	16.2±18.4	0.41
Smoking status (%)‡,§			0.08
Never smoked	41.9	41.6
Past smoker	40.4	42.0
Current smoker	17.7	16.4
Menopausal status (%)‡			>0.99
Pre-menopause	6.3	6.3
Post-menopause, current hormone users	42.2	42.0
Post-menopause, not current hormone users	46.8	47.2
Post-menopause, hormone use status unknown	4.7	4.5
Aspirin use (%)§			0.02
<1 tablet/week	39.6	30.2
1–2 tablets/week	25.3	30.4
3–6 tablets/week	11.6	16.6
7–14 tablets/week	17.6	16.8
15+ tablets/week	6.0	6.1
Use of multivitamin (%)	37.9	34.3	0.25
History of hypertension (%)	47.8	34.3	<0.001
History of high cholesterol (%)	47.6	46.3	0.69
History of heart disease (%)	15.3	14.0	0.58
History of diabetes (%)	12.9	6.3	0.0005
Diet
Vitamin D (IU/day)	353.8±208.7	347.1±211.0	0.63
Calcium (mg/day)	1021.0±414.1	1042.3±378.0	0.41
Glycemic load	101.3±16.5	100.9±16.4	0.73
Polyunsaturated to saturated fat ratio	0.57±0.15	0.58±0.17	0.31
Trans fat (g/day)	1.7±0.4	1.7±0.5	0.77
Alcohol (g/day)	6.7±11.0	6.0±9.7	0.32
25(OH)D (nmol/L)	55.0±25.5	56.8±22.7	0.24

Abbreviations: MET-hr, Metabolic Equivalent Task-hour; 25(OH)D, 25-hydroxyvitamin D.

^*For continuous variables, values were expressed as mean±SD; for categorical variables, % was used.

^†P values were based on Student’s t test for continuous variables or Pearson χ² test for categorical variables.

^‡Matching factors.

^§Based on non-missing data.

In conditional logistic regression accounting for matching factors, women in the lowest 25(OH)D tertile had a non-significantly higher risk of ischemic stroke (Table 2). After adjustment for BMI and physical activity, the association was attenuated to 1.24 (0.89, 1.74; P_trend=0.20) comparing extreme tertiles. However, when we adjusted for other lifestyle risk factors and history of chronic conditions, the association was strengthened to some extent: the OR for stroke (95% CI) was 1.42 (0.99, 2.04; P_trend=0.06) for the women in the lowest tertile compared with those in the highest. Finally, after additional adjustment for dietary risk factors, including use of multivitamin, intakes of calcium and trans fat, polyunsaturated fat to saturated fat ratio, and glycemic load, the association was further strengthened, with an OR (95% CI) of 1.49 (1.01, 2.18; P_trend=0.04), comparing extreme tertiles. Further adjustment of estimated glomerular filtration rate (eGFR) or C-reactive protein (CRP) did not materially change the association; the ORs (95% CI) were 1.49 (1.02, 2.19; P_trend=0.04) and 1.46 (1.00, 2.14; P_trend=0.05), respectively. When we used alternate cut-points that were largely consistent with the only prior study of ischemic stroke,⁸ we observe similar associations. For example, in comparison to participants whose 25(OH)D levels >47 nmol/L (62.3% of total participants), women who had 25(OH)D levels ≤31 nmol/L (13.6% of total participants) had a multivariable OR (95% CI) of 1.53 (0.98, 2.39; P_trend=0.09) for stroke. When we used season-adjusted residuals of 25(OH)D levels as the exposure, we found a similar, inverse association, suggesting that our observations were not subject to seasonal variations of 25(OH)D levels: the multivariable-adjusted ORs (95% CI) were 1.41 (0.99, 2.02) and 1.48 (1.02, 2.17; P_trend=0.04) for the 2^nd and 1^st tertiles of 25(OH)D residuals, respectively.

TABLE 2.

Odds ratio (95% CI) of ischemic stroke for 25(OH)D levels at baseline, the Nurses’ Health Study.

	Plasma levels of 25(OH)D (nmol/L) at baseline
	Tertile 3 (highest)	Tertile 2	Tertile 1 (lowest)	Ptrend
Median (Range)	77.6 (65.5, 264.3)	55.0 (45.8, 65.4)	35.0 (9.2, 45.7)	-
Case/control n.	133/156	160/154	171/154	-
Model 1*	1.0	1.20 (0.88, 1.64)	1.31 (0.95, 1.82)	0.10
Model 2†	1.0	1.16 (0.84, 1.59)	1.24 (0.89, 1.74)	0.20
Model 3‡	1.0	1.25 (0.89, 1.76)	1.42 (0.99, 2.04)	0.06
Model 4¶	1.0	1.26 (0.89, 1.79)	1.49 (1.01, 2.18)	0.04

Abbreviations: 25(OH)D, 25-hydroxyvitamin D.

^*Model 1 was conditioned on all matching factors, i.e., age at blood draw (year), menopausal status (yes, no), use of postmenopausal hormone (current or non-user), White (yes, no), and smoking status (past, current, or non-smoker).

^†Based on model 1, model 2 was further adjusted for body mass index (kg/m²) and physical activity (MET-hr/week, in tertiles).

^‡Based on model 2, model 3 was further adjusted for use of aspirin (<1 tablet/week, 1–2 tablets/week, 3–6 tablets/week, 7–14 tablets/week, and 15+tablets/week), hypertension (yes, no), high cholesterol (yes, no), history of heart disease or diabetes (yes, no), and alcohol consumption (gram/day, in tertiles).

^¶Based on model 3, model 4 was further adjusted for use of multivitamin (yes, no) and intakes of calcium and trans fat, polyunsaturated fat to saturated fat ratio, and glycemic load (all in tertiles).

We further conducted several secondary analyses to explore the association of 25(OH)D in certain subgroups or with certain stroke subtypes. First, we examined interaction between plasma 25(OH)D levels and risk factors of stroke or factors that influence 25(OH)D levels, including age, BMI, physical activity, use of hormone therapy, season of blood draw, latitude of participants’ residence, eGFR, and CRP (Supplementary Table 1). We found a marginally significant interaction by current use of hormone therapy: comparing low vs. high tertiles of 25(OH)D, the OR (95% CI) was 1.06 (0.62, 1.84; P_trend=0.80) for nonusers, whereas this OR (95% CI) was 2.16 (1.17, 4.00; P_trend=0.01) among users, the majority (>75%) of whom used oral hormone therapy. Second, we examined the association of 25(OH)D with several subtypes of ischemic stroke: large artery infarction (n=142), lacunar infarction (n=173), or other types of ischemic infarction (n=149). The ORs (95% CI) comparing extreme tertiles of 25(OH)D levels were 1.78 (0.77, 4.10; P_trend = 0.20) for large artery infarction, 2.47 (1.15, 5.32; P_trend=0.02) for lacunar infarction, and 1.21 (0.59, 2.51; P_trend=0.59) for all other types of ischemic stroke. Lastly, we used cutpoints that were used to define vitamin D deficiency/adequacy¹⁸ to categorize the study participants and repeated the analysis. Although similar trend was observed, the association was somewhat attenuated: in comparison to women whose 25(OH)D levels ≥75 nmol/L, the ORs (95% CI) were 0.96 (0.65, 1.43) for levels in 50–74 nmol/L, 1.11 (0.72, 1.70) for levels in 30–49 nmol/L, and 1.39 (0.80, 2.42) for levels <30 nmol/L.

Meta-analysis

The characteristics of the 6 other prospective cohort studies that evaluated 25(OH)D levels in relation to various stroke outcomes are shown in Supplementary Table 2. Most studies included both men and women, except Bolland et al’s study¹¹ that included women only. Serum 25(OH)D levels were measured in all studies, whereas study outcomes varied among these studies: two studies examined fatal strokes only and four other studies evaluated total strokes. Only Kilkkinen et al’s study⁸ examined ischemic stroke and hemorrhagic stroke separately. Cox proportional hazard regression was used to model the association of interest in all but Pilz et al’s study,⁷ in which logistic regression was used to model the association. The number of stroke cases ranged from 42 to 293 in these studies. In total, including our study, there were 1,214 stroke cases included in the current meta-analysis.

When all data were pooled together, the combined RR (95% CI) of stroke comparing low vs. high vitamin D status was 1.52 (1.20, 1.85) (Figure 2). We did not observe evidence for heterogeneity of the results; the I² statistics was 0.0% and the P_{heterogeneity} was 0.63. When we pooled the results in the NHS cohort with Kilkkinen et al’s results specifically for ischemic stroke, the pooled RR (95% CI) was 1.59 (1.07, 2.12). Similarly, no evidence of heterogeneity was found between the results of these two studies. Of note, in this analysis for ischemic stroke we used RR comparing ≤ 31 nmol/L vs. >47 nmol/L in the NHS study to be consistent with the cutpoints used in the Kilkkinen et al’s study.⁸ The Begg funnel plot and Egger test did not suggest evidence of publication bias (Supplementary Figure 1).

Figure 2.

Pooled fixed-effects relative risk (95% CI) of stroke comparing high 25(OH)D levels with low 25(OH)D levels. A, total stroke; B, ischemic stroke. Bars indicate 95% CIs and P values are P for heterogeneity.

In the cubic spline regression model, we detected a potentially non-linear dose-response relationship; the P for non-linearity was 0.02. This dose-response relationship is shown in Figure 3. There was a clear threshold effect in that the risk of stroke did not start decreasing until the 25(OH)D levels exceeded certain levels around 40 nmol/L.

Figure 3.

Dose-response relationship between blood levels of 25(OH)D and risk of stroke. The solid line represents point estimates of the association between 25(OH)D levels and stroke risk, and the dotted lines are 95% CIs. The horizontal line is the reference line.

We further conducted several sensitivity analyses to examine the robustness of our observations. First, when we used a random-effects model, we observed essentially the same results. Second, when we excluded Drechsler et al and Pilz et al’s studies that included dialysis patients or patients referred to coronary angiography, respectively, the results did not change substantially: the pooled RR (95% CI) was 1.51 (1.18, 1.83). Lastly, when we examined the dose-response relationship among prospective cohort studies that used Cox regression to model the association,^8–12 a marginally significant P value for non-linearity of 0.06 was documented, probably because of low power for this stratified analysis.

DISCUSSION

In this prospective study among women without prior history of stroke, with multivariable adjustment for lifestyle and dietary factors, we documented a modest association between low plasma 25(OH)D levels and an increased risk of incident ischemic stroke. Associations appeared most robust for lacunar infarction. This observation was confirmed by a meta-analysis pooling our data with six other prospective studies that overall demonstrated the same inverse associations between blood 25(OH)D levels and various stroke outcomes. A potentially non-linear dose-response relationship between 25(OH)D levels and stroke risk was further suggested by the current investigation.

Independent of the effects of vitamin D on calcium metabolism, emerging evidence suggests several mechanisms through which vitamin D modulates cardiovascular health. For example, vitamin D receptor-null mice, have significantly increased expression of renin and plasma angiotensin, whereas injection of 1,25-dihydroxyvitamin D to wild type mice suppressed renin production, showing significant effects of vitamin D on this pathway.⁵ Independent of the tightly–controlled circulating 1,25-dihydroxyvitamin D levels, intracellularly-produced hormone may directly regulate the function and behaviors of vascular smooth muscles cells, endothelial cells, macrophages, and adipocytes.^19–22 In addition, local production of 1,25-dihydroxyvitamin D appears to be regulated in part by levels of 25(OH)D,²⁰ suggesting that the autocrine or paracrine effects of vitamin D may depend on the availability of 25(OH)D as well. Consistently, animal and human experiments provide data supporting the effects of vitamin D treatment/supplementation on increasing endothelium-dependent vascular relaxation,^{23, 24} inhibiting vascular smooth muscle cell growth,²⁵ improving insulin resistance and beta-cell dysfunction,^{26, 27} inhibiting production of inflammatory cytokines,^27–29 and regulating the reaction of monocytes to environmental stressors.³⁰ These mechanisms may also likely underlie the stronger association observed among current users of hormone replacement therapy; the effects low 25(OH)D status may be further exaggerated by the adverse effects of oral hormone therapy on inflammation, thromboembolism, and coagulation.³¹

Despite the basic science evidence, only limited prospective data existed regarding the relationship between vitamin D status and risk of stroke. In addition, most of these studies did not document significant findings probably because most of these earlier studies employed small sample sizes resulting in limited statistical power for detecting a significant association. However, when we pooled our findings with all previous prospective data, we found a significant association suggesting that low vitamin D status may lead to an increased risk of stroke. Moreover, we did not observe significant heterogeneity of individual study associations; six of seven studies suggested an inverse association. With respect to the associations for subtypes of stroke, only Kilkkinen et al examined this association for ischemic stroke and hemorrhagic stroke separately.⁸ They found a significant inverse association between baseline 25(OH)D levels and risk of fatal ischemic stroke (n = 175), although the inverse association for fatal hemorrhagic stroke (n = 43) was not significant.⁸ When we categorized our participants according to cutpoints of 25(OH)D levels that corresponded to those in this Finnish study, we observed similar associations for combined fatal and nonfatal ischemic stroke. Thus far, the current study was the only investigation that further examined 25(OH)D levels in relation to subtypes of ischemic stroke, and we found a stronger association for lacunar infarction. The effects of vitamin D on hypertension and diabetes may explain this stronger association because these diseases were believed to be major risk factors for lacunar stroke,³² although a recent meta-analysis suggested that these two conditions were equally associated with lacunar and nonlacunar strokes.³³ Apparently, more studies are warranted to derive a solid estimate on the effects of vitamin D on subtypes of stroke.

Our investigation is subject to a few limitations. First, our study population consisted of registered nurses with mostly European ancestry. It is unknown whether our current findings can be generalized to other races that, in general, have lower 25(OH)D levels than Caucasians.¹³ This restricted generalizability pertains to the meta-analysis results as well because the vast majority of participants in the contributing studies were also Caucasian. In addition, 25(OH)D deficiency (<50 nmol/L) was relatively common (>40%) and dietary calcium was high (>1000 mg/day) in our study population. These characteristics further restrict the generalizability of the results. Second, we only measured baseline 25(OH)D levels, and a single measure may not represent long-term levels. However, in a validation study among 71 NHS participants who provided three blood samples within 2–3 years, 25(OH)D levels at each time point were highly correlated, with an intraclass correlation (ICC) of 0.72.³⁴ In addition, in the current study among 102 controls who provided a second blood sample approximately 10 years after baseline blood draw, an ICC of 0.48 was observed. These data suggest that a single measure of blood 25(OH)D is a reasonable proxy for vitamin D exposure, with correlations over time that are similar to those for cholesterol and blood pressure.³⁵ Further, measurement error in the 25(OH)D assay, although relatively modest (CV=13.9%), may introduce additional random variation that likely attenuates the true association. Third, although the ability to adjust carefully for a wide array of lifestyle and dietary covariates in this study is a strength, residual confounding, in particular due to adiposity, cannot be fully excluded in this observational study. Uncontrolled and residual confounding in studies included in the meta-analysis cannot be excluded, either. The results can be biased by confounding toward either direction. Fourth, although we did not find statistical heterogeneity of associations in our meta-analysis, these prospective studies are heterogeneous in several aspects such as various stroke outcomes (stroke mortality and total, ischemic, and hemorrhagic strokes), diverse methods for identifying and confirming stroke outcomes, different assays for 25(OH)D measurements, varying baseline health status of the study populations (general population vs. patients with specific diseases or conditions), various covariates adjusted in each individual study, and dramatically different levels of 25(OH)D and comparison categories across studies. Large individual-level data are still needed to minimize this between-study heterogeneity and to model any does-response relationship precisely. Likewise, detection of interactions between vitamin D and other stroke risk factors also requires such data. Fifth, since vitamin D deficiency is quite common among prevalent cerebrovascular patients and a prior history of stroke is a strong risk factor of recurrent stroke,¹ previous investigations on total stroke or stroke mortality^{7, 10–12} that did not exclude baseline stroke patients may be subject to confounding bias, which can lead to erroneously strong associations. Lastly, some key factors involved in the biology of 25(OH)D, such as 1,25-dihydroxvvitamin D, parathyroid hormone, phosphate, and fibroblast growth factor 23, were not measured in the current analysis. Therefore, we were unable to examine the role of these molecules in the association between 25(OH)D and risk of ischemic stroke. In terms of strengths, the current study used incident ischemic stroke confirmed by medical records, and such homogeneity of outcome may help detect biologically meaningful associations. Other strengths of the current study include large sample size, long follow-up period, and rich data allowing comprehensive analysis.

CONCLUSIONS

In conclusion, we found a modest association between low 25(OH)D levels and risk of incident ischemic stroke among U.S. women without a prior history of stroke, and this observation was consistent with pooled results based on all prior prospective studies. Further studies are warranted to explore this association among other races and for stroke subtypes. In particular, large-scale clinical trials are needed to shed light on the effects of achieving optimal vitamin D status in the primary prevention of stroke. Given that U.S. populations have a high prevalence of both vitamin D insufficiency and stroke, solid evidence is especially needed to guide public health recommendations to lower the burden of both conditions.