Bone Mineral Density and Incidence of Stroke
European Prospective Investigation Into Cancer-Norfolk Population–Based Study, Systematic Review, and Meta-Analysis
Background and Purpose—The prospective link between osteoporosis and future risk of stroke requires evidence from large-scale population-based long-term studies.
Methods—Calcaneum broadband ultrasound attenuation was measured in the Norfolk cohort of the European Prospective Investigation into Cancer-Norfolk between 1997 and 2000. Incident strokes were ascertained by hospital record linkage and death certificates in March 2009 and December 2011, respectively. A search of MEDLINE and EMBASE was performed to evaluate the relationship between bone mineral density and incident stroke. After data extraction of relevant studies, pooled risk of stroke was estimated using meta-analysis.
Results—In 14 290 participants (mean follow-up of 9.3 years; total person-years 132 574), there were 599 incident strokes. Participants in the lowest 10% of the calcaneum broadband ultrasound attenuation distribution had an increased stroke risk (hazard ratio 1.41; 95% confidence intervals, 1.02–1.94) compared with those in the top 30% of the distribution after adjustments. A decrease of ~1 standard deviation in broadband ultrasound attenuation (20 db/MHz) was associated with a 17% increase in relative risk of stroke (95% confidence intervals, 5%–30%).
Meta-analysis of 4 studies (25 760 participants, 1237 cases of stroke) found that for every decrease in 1 standard deviation in bone mineral density, there was an increased risk of incident stroke among women (pooled relative risk 1.22; 95% confidence intervals, 1.09–1.37; I2=0%, 3 studies) but not in men (pooled relative risk 1.05; 95% confidence intervals, 0.94–1.17; I2=0%, 2 studies).
Conclusions—Bone mineral density predicts total stroke risk. The evidence is stronger in women with regard to the continuous relationship.
Osteoporosis and increased fracture risk are recognized complications after a stroke.1 However, the relationship between bone health and stroke is not straightforward, although there seems to be a complex mutual connection between stroke and bone health.2 The current evidence suggests that these 2 conditions are risk factors for each other albeit more clear support for the effect of stroke on bone health. To date, few population level evidence exists with regard to the association of low bone mineral density (BMD) with subsequent stroke risk.
BMD can be assessed using different methods. Dual-energy x-ray absorptiometry is considered as a standard noninvasive method to assess BMD, but it is costly and therefore not always available to use in daily clinical practice. Quantitative ultrasound at the peripheral sites is relatively cheaper, easy to perform, and hence has potential to be useful in daily practice. The validation studies against dual-energy x-ray absorptiometry suggest the usefulness of quantitative ultrasound at heel (calcaneus) in diagnosing osteoporosis and future fracture risk.3
We have previously reported that quantitative ultrasound of the calcaneus predicted total and hip fracture risk in men and women in the European Prospective Investigation into Cancer (EPIC)-Norfolk prospective population study.4 In this article, we examine the relationship between BMD assessed using broadband ultrasound attenuation (BUA) and velocity of sound (VOS) and subsequent stroke risk and performed a systematic review and meta-analysis to quantify the existing evidence of predictive value of BMD on subsequent stroke risk.
The Norfolk cohort of the European Prospective Investigation into Cancer (EPIC-Norfolk) comprises 25 000 men and women 40 to 79 years old at baseline who were resident in Norfolk, UK, at the time of recruitment. Data collection and follow-up in EPIC-Norfolk were expanded to enable assessment of chronic disease determinants; recruitment and study methods have been detailed elsewhere.5,6 Briefly, between 1993 and 1997, participants completed health questionnaires and attended a first clinic visit, at which detailed health and lifestyle measurements were taken.
Between 1997 and 2000, surviving participants were invited to attend a second clinic visit; ≈15 000 responded, a response rate of 58% of those mailed, after excluding those who had moved from the area or died. At this visit, ultrasound measurements of the calcaneum were obtained.4 Trained nurses examined participants. Height and weight were measured in light clothing without shoes. BUA (db/MHz) and velocity of sound (VOS; m/s) were measured at least twice on each calcaneum with a CUBA sonometer (McCue Ultrasonics, Winchester, UK). We used the mean of left and right ultrasound measures for analysis. The coefficient of variation was 3.5%. We recorded ambient temperature. The 5 CUBA machines used were calibrated daily with a physical phantom. Machines were also compared on 1 calcaneus.
Weight was measured with participants wearing light clothing without shoes. Height was measured up to the nearest 0.1 cm using a stadiometer with shoes removed. Body mass index (BMI) was calculated as weight (kg) divided by height squared (m2). Blood pressure (BP) was measured with an Accutorr monitor (Datascope, Huntingdon, UK) after the participant had been seated for 5 minutes. We used the mean of 2 BP measurements for analysis. From nonfasting venous blood samples, we measured serum total cholesterol with the RA 1000 (Bayer Diagnostics, Basingstoke, UK).
At the baseline survey (1993–1997), participants completed a detailed health and lifestyle questionnaire. Participant’s educational status, occupational social class, and physical activity were obtained from the baseline health and lifestyle questionnaire. Educational status was recorded as no qualification, O-level, A-level, degree, or higher qualification. Social class was classified according to the Registrar General’s occupation-based classification scheme.7,8 A 4-level physical activity index was derived from the validated EPIC short physical activity questionnaire designed to assess combined work and leisure activity.9
We ascertained medical history with a question on the health questionnaire repeated in second health survey (1997–2000)—“Has a doctor ever told you that you have any of the following?”—followed by a list of conditions, including cancer, stroke, myocardial infarction, and diabetes mellitus. Smoking history was obtained from the questions “Have you ever smoked as much as one cigarette a day for as long as a year?” and “Do you smoke cigarettes now?” Participants were also asked to report the medications they were taking (name, dose frequency, etc).
For comparisons of categorical variables with 1 standard deviation (SD) decrease in BUA and VOS, we grouped these into dichotomous variables: physical activity as physically inactive (physical activity categories 1 and 2) and physically active (categories 3 and 4), occupational social class as lower (class III-manual, IV, and V) and higher (class III-nonmanual, II, and I), educational attainment as lower (no or less than A-level qualification) and higher educational attainment levels (at least A level).
All participants were flagged for death by cause at the Office of National Statistics. Participants were also linked to National Health Service hospital information system, so that admission anywhere in the UK was notified to EPIC-Norfolk. They were also linked to ENCORE (East Norfolk COmmission Record) for admission episodes. Incident strokes were identified from the death certificates (Office of National Statistics) or hospital discharge code for International Classification of Diseases (ICD) 10–I60-I69. The follow-up time started at baseline for this study (date of second health check) and ended at the censor date defined as date of the event (date of death with stroke as cause of death in the death certificate or incident stroke) or last follow-up (December 2011). ENCORE data linkage was available up to end of March 2009, and the Office of National Statistics (ONS) linkage was available up to the end of December 2011. These ascertainment methods of EPIC-Norfolk for stroke incidence have been previously validated.10
No absolute diagnostic cut points exist for osteoporosis with quantitative ultrasound.11–13 Criteria in studies include a specified number of SDs below either the sex-specific and age-specific mean, or the sex-specific mean for a young adult. World Health Organisation recommended a cut point of >2.5 SDs below the mean for a young adult on the basis of bone densitometry. The paucity of data from men makes definition of cut points even more difficult.
For these analyses, we used an arbitrary lowest 10% of the sex-combined distribution for the ultrasound measures BUA and VOS to define a high-risk group. This absolute value was ~2 SDs below the mean at age 45 years for men and 1.5 SDs below the mean at age 45 years for women. We used sex-combined percentiles to enable direct comparison of men and women with the same absolute values. We categorized the remainder of the cohort by 40th and 70th percentile cut points to obtain 4 groups: the lower 10%, and then 3 equal groups of 30% of the population. We assessed stroke rates and relative risks (RRs) by these categories. We used the Cox multivariate regression model to examine the independent predictive value of calcaneum measures for stroke incidence.
Statistical analyses were performed using STATA version 11.2/SE (College Station, TX). Participants with missing data were excluded from analyses. After excluding participants with prevalent cancer and stroke at second health visit, multivariate adjustments were made to examine how far the association between BMD at the calcaneus and the risk of stroke might be explained by other known demographic, lifestyle, socioeconomic, and cardiovascular risk factors. Calcaneum ultrasound measures were used as continuous variables. Stepwise adjustments were constructed. We adjusted for age and sex in model A; age, sex, systolic BP, and cholesterol in model B; and we additionally adjusted lifestyle factors, such as smoking, physical activity, and BMI, in model C. Model D additionally adjusted for occupational social class and educational level, and model E was constructed as of model D but additionally adjusted for prevalent myocardial infarction (MI) and diabetes mellitus. The final model (model F) adjusted for variables as in model E along with lipid-lowering medication and antihypertensive medication use.
We also assessed the relative predictive value of 1 SD decrease in BUA and VOS using Cox multivariate model compared with other risk factors: every 1-year increase in age, being male, every 5-mm Hg increase in systolic BP, every 1-mmol/L increase in total cholesterol, being a current smoker, being physically active, having BMI >30 kg/m2, having no or lower than A-level educational attainment, being in the lower occupational social class, and having prevalent diabetes mellitus or MI. We further examined the association of calcaneum measures with stroke stratified by sex (male versus female), age group (<65 versus ≥65 years), and cigarette smoking status.
The sponsors of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The EPIC-Norfolk study was approved by the Norwich Research Ethics Committee.
Systematic Review and Meta-Analysis
Studies that reported the association between BMD and incident stroke were included in the systematic review. We searched PubMed and EMBASE from inception until July 2013 using the terms described in online-only Data Supplement I, with no language limitations, and we checked bibliographies of included articles. One reviewer (C.S. Kwok) screened abstracts and titles, and then 2 reviewers (C.S. Kwok and Y.K. Loke) independently reviewed all potentially relevant studies to confirm eligibility. Data extraction of included studies was performed by C.S. Kwok and J.K.-Y. Yeong, and checked by Y.K. Loke. Study validity was evaluated based on methods used in ascertainment of BMD and incident stroke events as well as steps taken to reduce confounding in the primary studies. We pooled data using the inverse variance method and random effects model in RevMan 5.2 software (Nordic Cochrane Center, Copenhagen, Denmark). For these comparisons, we used the multivariable adjusted measures of association (hazard ratios [HRs], RRs or odds ratios [OR]) for a 1 SD reduction in BMD and incident stroke. We performed analysis considering sex- and race-specific results. Heterogeneity was estimated using I2, and we considered a value >50% to demonstrate substantial heterogeneity.14 We planned to evaluate publication bias through asymmetry testing if there were >10 studies in the data set, and no evidence of significant heterogeneity.15
Of 15 786 participants who attended the second health check in 1997 to 2000 with available heel bone ultrasound measures, 382 participants were excluded from the analysis because they had a previous stroke, and 1136 participants were excluded because of a previous diagnosis of cancer. A total of 14 290 men and women 42 to 82 years old were included. During mean follow-up of 9.3 years (SD, 1.7; total person-years, 132 574), there were 599 incident strokes. Figure I in the online-only Data Supplement demonstrates the reasons for exclusions and numbers of participants included in the different models. A total of 12 795 participants were included in the final models (models D, E, and F). There were no significant differences of participants’ characteristics between the model A and final models with regard to age and sex.
Mean values and ranges of the percentile categories are 48.75 (14.95–56.12); 66.28 (56.13–74.59); 81.96 (74.61–89.84); and 103.10 (89.85–145.54) for BUA and 1563.75 (1456.25–1581.50); 1605 (1581.75–1623.25); 1638.92 (1623.33–1655.25); and 1682.75 (1655.50–1802.50) for VOS, respectively. Table 1 shows the baseline sample characteristics of participants included in the EPIC-Norfolk study who had calcaneum ultrasound performed during their second health check by BUA categories. People in the lowest 10% of BUA distribution were older and were more likely to be women. They had higher systolic BP and cholesterol levels. They were more likely to be current smoker, less likely to be physically active, more likely to have BMI <30 kg/m2. Despite overall statistical significant difference between the groups with regard to social class, the differences were small except for the lower proportion of professionals (5%) in the bottom 10% compared with 3 other categories (8%). The proportions of people with prevalent diabetes mellitus were lower in this group. Although there seemed to be no difference in percentage of people on lipid-lowering treatment, a higher proportion of people in the bottom group were on antihypertensive medication compared with other categories.
Table 2 shows consistent significant trends with significantly higher risk of stroke in those in the lowest 10% of BUA distribution compared with the highest 30%. Additional adjustments attenuated the estimates but remained statistically significant. Notably, adjusting for lifestyle behaviors attenuated the association. Further adjustments did not attenuate the associations further. Similar, but statistically nonsignificant, trends were observed for VOS analysis.
Table 3 shows the independent HRs for stroke risk during follow-up for every 1 SD decrease in BUA or VOS values in comparison with every 1-year increase in age, being male, every increase in systolic BP of 5 mm Hg, every increase in 1 mmol/L of total cholesterol, being a current smoker, being physically active, having BMI >30 kg/m2, having lower level of education and being in lower social class, having a prevent condition (diabetes mellitus and MI). For both measures, every decrease in 1 SD was associated with an increase hazard of stroke: HR 1.17 (1.05–1.30) and HR 1.12 (1.02–1.22) for BUA and VOS, respectively, independently of the covariates included. Every 1 SD decrease in BUA was equivalent in risk to 20 mm Hg increase in systolic BP or ≈1.5 years increase in age after taking into consideration the other factors included in the model.
Table 4 shows that the independent RRs of stroke associated with BUA and VOS was consistent in subgroups in stratified analyses.
Systematic Review and Meta-Analysis
We screened 1595 titles and abstracts and identified 6 relevant studies for the systematic review,16–21 and of them 3 studies16–18 were possible to be included in the meta-analysis (Figure II in the online-only Data Supplement). Two studies were performed in the United States,16,17 and the other study was performed in Sweden.18 Two of the studies examined the relationship for both sexes,17,18 but the remaining study was limited to women only.16 One of the 2 studies that examined both sexes reported results for men and women separately,17 whereas the other study reported a single sex-combined risk estimate.18 Therefore, including EPIC-Norfolk study, a total of 4 prospective cohort studies with the maximum follow-up of 18.7 years were included in the meta-analysis. A total of 25 760 participants and 1237 cases of incident stroke were included in the meta-analysis (Table 5).
Different methods were used for evaluating and ascertaining BMD and incident stroke across the studies (Table I in the online-only Data Supplement). Although absorptiometry was used in all the studies, only one of them evaluated bone density at the femoral neck,18 whereas the other 2 evaluated the bone density in the left hand17 and at 3 different sites (distal radius, proximal radius, and calcaneus).18 Two studies used ICD-9 codes to ascertain stroke cases,16,17 and 1 study collected stroke cases from a validated register.18 Three studies adjusted for potential confounders, and the adjustments were limited in 2 studies making these studies liable to residual confounding.16,18 One of these studies only adjusted for age,16 whereas the other adjusted for age, sex, and body mass index.18 The most highly adjusted study in the published literature accounted for age, smoking status, alcohol consumption, history of diabetes mellitus, history of heart disease, education, body mass index, recreational physical activity, and BP medications.17
Overall, a 1 SD reduction in BMD was associated with increased risk of incident stroke (pooled RR 1.12; 95% CI [1.04–1.22], I2=23%, 4 studies; Figure). The significant difference was primarily driven by the results from studies of women (pooled RR 1.22; 95% CI [1.09–1.37], I2=0%, 3 studies). The results for men were not significant (pooled RR 1.05; 95% CI [0.94–1.17], I2=0%, 2 studies).
Three studies were eligible to be included in the systematic review, but they were not included in the meta-analysis. One of these studies was a population-based case–control study, which evaluated 4175 participants in Taiwan.19 They found that osteoporotic vertebral fracture was significantly increased among patients with incident stroke (adjusted HR 2.71, 95% CI [1.90–3.86]). Another study that was not included in meta-analysis was a prospective observational study of 744 participants.20 This study found that the HR for BMD and incident stroke were not significant but no numeric results were reported. The third study was an observational study of cases and controls that included 251 patients and 63 cases of stroke and found that the highest quartile of BMD had a significantly higher risk of incident stroke compared with the lowest quartile for women but not for men.21 This study reported an increased OR for stroke associated with a drop in 1 SD of BMD, but the 95% CIs and statistical significance of this finding were not reported and, as such, we were unable to include it in the meta-analysis.
Publication bias was not formally assessed because there were <10 studies included in the analysis. However, we found clear evidence of selective reporting in 1 study20 where the authors reported in the text that there was no significant association between BMD and stroke, but the ORs and 95% CIs were not given. This null finding could not therefore be included in our meta-analysis, which means that the pooled data set could be reporting overinflated risk estimates.
We found a prospective relationship between BMD assessed by heel (calcaneum) ultrasound at the baseline and subsequent stroke risk in middle and older age apparently healthy general population. Men and women in the bottom 10% of BUA distribution had increased future risk of stroke compared with the top 30% of BUA (70%–100%; HR 1.41 [95% CI, 1.02–1.94]). There also seems to be a linear relationship with every 1 SD decrease in BUA/VOS being associated with a higher risk of stroke (17% increase in relative risk of stroke [95% CI, 5% to 30%]).
Our study has several strengths which include large sample size performed in apparently healthy community dwelling general population without previous known stroke or cancer, with complete follow-up (100% follow-up) using validated follow-up methods10 and our ability to control for a range of potential confounders including demographics, anthropometry, lifestyle, and social and medical risk factors. Overall, our results from both the observational study and systematic review yield supportive and consistent evidence that decreased BMD is associated with risk of incident stroke in women.
In an observational study, confounding and reverse causality issues require attention. We addressed these issues in this study in several ways. First, we adjusted for possible confounders that potentially relate to both BMD and known stroke risk factors. Second, we excluded people with stroke and cancer at the baseline and, third, we specifically adjusted for people with prevalent MI and diabetes mellitus and medications for hypertension and hypercholesterolemia in later models. The more consistent association with BUA suggests it is a more specific and sensitive measure than VOS.3
There is strong and consistent evidence to suggest osteoporosis, and subsequent fracture risk is substantial in people who sustained stroke.1 Immobility with generalized bone loss compounded by region-specific bone loss at hemiplegic side was considered as the major contributing factor for the development of osteoporosis after stroke. Several pathophysiological mechanisms have been proposed. Sato et al22 showed evidence of bone resorption as early as within 7-days poststroke. Remodeling imbalance at bone multicellular unit has also been suggested, evidenced by decline in serum biomarkers of bone formation.22 Factors such as duration of hemiplegia,23 degree of functional recovery,24 reduced vitamin D status,25,26 and the use of anticoagulants27 may influence bone loss after stroke.
On the contrary, the link between osteoporosis and stroke with the former being a risk factor for the latter is less well researched. It has been recognized that patients with stroke are highly likely to have preexisting osteoporosis; 40% of patients with stroke admitted to a rehabilitation unit in Japan with a mean time from onset of ≈40 days had established osteoporosis.28 Classical cardiovascular risk factors do not fully account for stroke mortality and burden at the population level especially in low-income countries.29 Therefore, understanding of the possible contribution of novel risk factors for stroke risk and stroke mortality is important.
Most recently, Qu et al30 systematically reviewed the literature, and they reported that low BMD was not associated with the risk of stroke mortality (HR 1.08; 95% CI [0.89–1.28]); however, there is dearth of data on the link between BMD and stroke incidence in the literature. One possible explanation of why the associations were null is because the studies were underpowered, and this is supported by the researchers reporting 95% confidence intervals that are broad. In addition, our systematic review builds on the findings of their review as we have only considered stroke incidence, and we reported results (including nonsignificant findings) from studies that were excluded from their meta-analysis. Our study provides further evidence by extending this link in men and women with wider age range (range 42–82 years) using more robust adjustment for various confounders that were not considered in previous studies. The point estimates for stroke risk at follow-up for every 1 SD decrease in men and women in our study were 1.08 and 1.25, respectively, and for those ≥65 years (including men) was 1.22 (1.08–1.38). Differences in population mix, sample sizes, and follow-up duration, and risk factors adjusted for may account for slight differences in estimates. We also considered publication bias or selective outcome reporting to be a problem and important limitation in our meta-analysis.
Whether osteoporosis is a marker rather than risk factor for stroke remains uncertain. Nevertheless, there are potential biological mechanisms underpinning the possible causal link. It has been proposed that excess calcium resorbed from the bone occurring in osteoporotic process may become incorporated into the vascular lining, in people with osteoporosis especially at atheromatous plaques making them more brittle and unstable as well as making blood vessels more rigid. Frost et al31 showed BMD was negatively correlated with arterial stiffness over carotid and femoral arteries. Fifty-nine percentage of osteoporotic women (BMD T-score < −2.5) had calcified plaque at ≥1 sites compared with 42% and 20% for women with osteopenia (T-score < −1) and normal BMD, respectively (P for trend=0.04).31
BMD is a measure of lifetime exposure to estrogen, and the observed age and sex differences in the association between BMD and stroke risk may stem from differences in levels of endogenous estrogen that could affect cardiovascular risk. Although it is tempting to speculate that high endogenous estrogen (and therefore higher BMD) might be associated with lowering of cardiovascular risk, the association between estrogen and stroke risk is still debated. A recent meta-analysis of randomized controlled trials on effect of hormone replacement therapy on cardiovascular outcomes demonstrated a significantly increased risk of stroke with exogenous estrogen.32 Moreover, a recent observational study found that higher endogenous estradiol levels (particularly in those with central adiposity) were associated with increased stroke risk.33
In another interesting study researchers assessed the potential role of estrogen signaling on this association through putative target genes (osteoprotegerin and interleukin-6); carriers of the osteoprotegerin-1181C/C genotype had a significantly increased risk of intracerebral hemorrhage (P=0.005).34 This implies that alterations in osteoprotegerin-mediated signaling in the vasculature may be involved in the pathophysiology of hemorrhagic stroke. There have also been a number of studies on the effect of statins on bone mass. In 2 recent meta-analyses on the effects of statins on BMD and fracture risk, statin use was associated with modest but significant increase in total hip and femoral neck BMD35 and significant reduction of fracture risk in case–control (OR, 0.62, 0.45–0.85) and cohort (OR, 0.77, 0.59–1.00) studies, but not in post hoc analyses of randomized trials (OR, 1.03, 0.91–1.16).36
Another possible link between bone health and stroke is through the action of vitamin D. Vitamin D is well known for its effect on bone; it has also been shown to be associated with cardiovascular risk factors, such as hypertension37 and diabetes mellitus38 as well as with markers of subclinical atherosclerosis such as intima-media thickness and coronary calcification.39 Hence, it was thought that vitamin D deficiency contributes to the development of cardiovascular events such as stroke through its association with risk factors. However, direct effects of vitamin D on the cardiovascular system may also be involved.40 Recent findings have shown that vitamin D receptors are expressed in cardiomyocytes, vascular smooth muscle cells, and endothelial cells, and that vitamin D affects inflammation and cell proliferation and differentiation in these tissues.36 A recent meta-analysis on the limited available data suggests that vitamin D supplementation at moderate-to-high doses may reduce cardiovascular risk.41 Further prospective placebo-controlled randomized trials however are less promising.42
Naturally, our study had limitations. Because of the requirement of participants to provide detailed health and lifestyle information and to be able to undergo health checks, the initial response rate was modest (40%). This may have introduced a healthy responder bias. Nevertheless, the baseline characteristics of the study population were similar to those of other UK population samples, except for a slightly lower prevalence of smokers.5 Furthermore, as previously reported, the sample characteristics of EPIC-Norfolk respondents who attended to second health check were not materially different with regard to age (0.5 years younger as a group) and other characteristics such as BMI (again 0.5 U difference).4 Moreover, the truncation of distribution because of healthy responders may attenuate the associations, but this should not have produced a spurious relation between quantitative calcaneum ultrasound measure (BUA) and stroke observed within the study participants; if anything, truncation of the distribution is likely to reduce the power of any associations. Subgroup analysis reduced the sample size and event rates in strata that are less prevalent in this cohort (eg, smoking). This may contribute to nonsignificant association observed in current smokers. There also remains a possibility of residual confounding as well as there may be unknown potential confounders that we were not able to adjust for such as depression.
We used death certification and a hospital record linkage system using ICD coding to identify stroke cases. Although follow-up with the use of these methods is virtually complete, this approach may underestimate incident nonfatal stroke cases that are not admitted to the hospital. The use of self-reported stroke to exclude prevalent cases may have missed some prevalent strokes. We were not able to separately examine stroke subtypes. Nevertheless, the primary focus of the study was to assess the risk prediction of clinical stroke event severe enough to lead to hospitalization or death regardless of stroke subtype. In any case, the misclassification of strokes was likely to only attenuate any associations. Only single measurements of BUA and VOS and other covariates, such as cholesterol and BP, were made at baseline at the second health check. These measures as well as lifestyle behaviors, which may affect BMD, may have changed during the follow-up period. Moreover, the blood sample taken was a nonfasting sample and was therefore less standardized for some of the variables (eg, cholesterol concentration) than was a fasting blood sample. Nevertheless, random measurement error was likely only to attenuate any relations observed between calcaneum quantitative ultrasound and stroke.
The meta-analysis had both limitations and strengths. The main limitations of the meta-analysis were that only 3 studies in addition to the current study were found, and 2 of these studies were subject to residual confounding because of minimal adjustments for potential confounders.16,18 Furthermore, there was methodological variation in the way BMD was assessed across the included studies, which may account for differences in results. However, the included studies found on the search were large (average of ≈3800 subjects) prospective cohort studies with long duration of follow-up (up to 18.7 years), and we were able to analyze the results for men and women separately.
In summary, there seems to be a complex connection between bone health and stroke. Both may share common biological or lifestyle antecedents and the association warrants further exploration. In the interim, people with low bone mass may not only be at high risk for fractures, but also increased risk for stroke and therefore candidates for targeted intervention not just for fracture prevention but also potentially cardiovascular risk assessment and control.
We gratefully acknowledge the participants and collaborating general practices of EPIC-Norfolk. We thank our funders and the staff of EPIC-Norfolk. The funders had no role in design, analysis, and interpretation of the data.
K.-T. Khaw and N.J. Wareham are principal investigators of EPIC-Norfolk. R.N. Luben is responsible for data management and record linkage. P.K. Myint designed the study and developed the analysis plan. A.B. Clark analyzed the EPIC-Norfolk data. Meta-analysis protocol was developed by P.K. Myint, C.S. Kwok, and Y.K. Loke. C.S. Kwok screened abstracts and titles. C.S. Kwok and Y.K. Loke independently reviewed all potentially relevant studies to confirm eligibility. Data extraction of included studies was performed by C.S. Kwok and J.K.-Y. Yeong, and checked by Y.K. Loke. P.K. Myint drafted the article with critical input from C.S. Kwok and Y.K. Loke. All coauthors contributed in writing of the article. P.K. Myint is the guarantor.
Sources of Funding
European Prospective Investigation into Cancer (EPIC)-Norfolk is funded by the Medical Research Council (UK) and Cancer Research UK.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.113.002999/-/DC1.
- Received August 8, 2013.
- Accepted November 19, 2013.
- © 2014 American Heart Association, Inc.
- Myint PK,
- Poole KE,
- Warburton EA
- Taal MW,
- Cassidy MJ,
- Pearson D,
- Green D,
- Masud T
- Khaw KT,
- Bingham S,
- Welch A,
- Luben R,
- Wareham N,
- Oakes S,
- et al
- Elias P,
- Halstead K,
- Prandy K
- Shohaimi S,
- Luben R,
- Wareham N,
- Day N,
- Bingham S,
- Welch A,
- et al
- Wareham NJ,
- Jakes RW,
- Rennie KL,
- Schuit J,
- Mitchell J,
- Hennings S,
- et al
- Miller PD,
- Njeh CF,
- Jankowski LG,
- Lenchik L
- 12.↵WHO. Assessment of Fracture Risk and Application to Screening for Postmenopausal Osteoporosis. Geneva: World Health Organization;1994.
- Higgins JPT,
- Green S
- Ioannidis JP,
- Trikalinos TA
- Browner WS,
- Pressman AR,
- Nevitt MC,
- Cauley JA,
- Cummings SR
- Mussolino ME,
- Madans JH,
- Gillum RF
- Jørgensen L,
- Engstad T,
- Jacobsen BK
- Sato Y,
- Maruoka H,
- Oizumi K,
- Kikuyama M
- Sato Y,
- Kuno H,
- Asoh T,
- Honda Y,
- Oizumi K
- Sato Y,
- Honda Y,
- Kunoh H,
- Oizumi K
- Frost ML,
- Grella R,
- Millasseau SC,
- Jiang BY,
- Hampson G,
- Fogelman I,
- et al