Among a cohort of elderly women, abdominal aortic calcification scored on baseline lateral spine densitometric images intended for vertebral fracture assessment was associated with subsequent myocardial infarction or stroke over a median 4-yr period, independent of clinical cardiovascular disease risk factors.
Introduction: Cardiovascular disease (CVD) risk among older women is not adequately captured by traditional CVD risk factors. Lateral spine images obtained on bone densitometers for vertebral fracture assessment (VFA) can detect abdominal aortic calcification (AAC), an important marker of subclinical CVD. Our objective was to estimate the association between AAC scored on VFA images and subsequent myocardial infarction (MI) or stroke in elderly women.
Materials and Methods: Among participants in a randomized controlled trial (women; age >75 yr) of clodronate versus placebo, those who sustained an MI or stroke during the median 4-yr follow-up study period were selected as cases (n = 408), and 408 controls were randomly selected from the remainder of the parent study population. Baseline VFA images were scored for AAC with a previously validated 24-point scale and a newer, simpler 8-point scale.
Results: The OR of incident MI or stroke for those in the middle and top tertiles, respectively, compared with the bottom tertile of AAC score were 1.14 (95% CI, 0.79–1.66) and 1.74 (95% CI, 1.19–2.56) for the 24-point scale and 1.42 (95% CI, 0.98–2.05) and 1.77 (95% CI, 1.22–2.55) for the 8-point scale, adjusted for age, high-density lipoprotein and low-density lipoprotein cholesterol, triglycerides, blood pressure, smoking, renal function, health status, and baseline diagnoses of diabetes mellitus, hypertension, angina, and prior stroke.
Conclusions: AAC scored on VFA images is independently associated with incident MI or stroke. Because bone densitometry is indicated for all women ≥65 yr of age, VFA imaging offers an opportunity to capture this CVD risk factor in postmenopausal women undergoing bone densitometry at very little additional cost.
Coronary heart disease and stroke are the first and third leading causes of mortality among elderly women,(1) and >60% of women who die of coronary disease have no prior symptoms of the disease.(2) Guidelines have been developed to identify women at higher than average risk of coronary heart disease based on traditional clinical risk factors of elevated blood pressure, dyslipidemias, cigarette smoking, obesity, and diabetes mellitus.(3,4) However, a substantial proportion of those who suffer morbid or fatal cardiovascular disease events are not at high risk when judged by these risk factors.(5–7)
For these reasons, a variety of diagnostic tests for subclinical cardiovascular disease (CVD) have been proposed for use with traditional clinical CVD risk factors to identify a larger proportion of those at high risk of incident CVD events. These include serum C-reactive protein,(8,9) carotid ultrasound to assess carotid intimal thickness (TMT),(10,11) measurement of ankle-brachial index,(12,13) and electron-beam CT to detect coronary calcium.(14–17)
Abdominal aortic calcification (AAC) detected on lateral lumbar spine radiographs is also predictive of incident coronary heart disease(13,18) and stroke,(19) independent of traditional CVD risk factors. Lateral spine images obtained on a bone densitometer are increasingly used to detect prevalent vertebral fractures, important BMD-independent predictors of future fracture.(20,21) Bone densitometry is now widely recommended for all women ≥65 yr of age,(22–25) and vertebral fracture assessment (VFA) imaging done at the time of densitometry carries little additional cost (2007 average Medicare reimbursement, $33)(26) and minimal additional radiation exposure (1/100th the effective radiation dose of a lateral lumbar/thoracic radiographs).(27) VFA imaging has also been shown to detect radiographic AAC with reasonable accuracy(28,29) and offers the potential of identifying subclinical cardiovascular disease in the postmenopausal female population at large.
However, no study to date has investigated whether AAC detected on VFA images is associated with any form of incident CVD. Our primary objective was to estimate ad hoc the association between AAC scored on baseline VFA images with a previously validated 24-point scale (AAC-24) and subsequent myocardial infarction (MI) or stroke in women ≥75 yr of age participating in a randomized trial of clodronate versus placebo, after adjustment for traditional risk factors.(30) Our secondary objective was to estimate the association with AAC scored with a newer, simplified 8-point scale (AAC-8) and incident MI and stroke.
MATERIALS AND METHODS
This study was approved by the Park Nicollet Institute Institutional Review Board.
The parent study population consisted of 5596 white British women ≥75 yr of age (mean age, 80 yr) recruited from general practice registers of South Yorkshire and North Derbyshire in the United Kingdom to participate in a randomized trial of a bisphosphonate drug (clodronate) versus placebo for the prevention of hip and other clinical fractures.(30) Participants were not required to have low BMD or any specific fracture risk factors. Exclusion criteria were prior bilateral hip arthroplasties, concurrent use of any pharmacologic fracture prevention medication, concurrent malignancy, or any other medical condition that would impede informed consent or compliance with study procedures. Participants were followed for a median of 4 yr.
For this study, a nested case-control study design was used. Because those with acute MI or stroke are routinely hospitalized in the United Kingdom, before commencement of the study, we chose a combined outcome of fatal and nonfatal incident MI or stroke as our primary measure of incident CVD, to assure as complete ascertainment of the outcome as possible. All 408 study participants who had a documented incident MI or stroke during the study follow-up period were selected as cases, and a sample of 408 control subjects were randomly selected from the rest of the study population (Fig. 1). This study design was chosen such that the primary hypotheses could be addressed with adequate power at modest cost, with controls selected to be representative of the entire subset of the parent study population that did not have an MI or stroke.
Ascertainment of incident MI and stroke
Incident CVD events were recorded as adverse events in the parent randomized control trial (RCT) at each study follow-up visit done every 6 mo. Discharge summaries of all hospitalizations that occurred during the prior 6 mo were reviewed at each study visit for serious adverse events, and the cause of all deaths during the study follow-up were ascertained using death reports from the NHS central registry.
Ascertainment of AAC
All study participants at baseline had a single-energy lateral spine image obtained on a Hologic QDR 4500A densitometer in the supine position for prevalent VFA. For this study, baseline VFA images could be located for 807 (98.9%) of the selected cohort (Fig. 1). Of these, 732 (90.7%), including 369 cases and 363 controls, had adequate space anterior to the lumbar spine to include the entire abdominal aorta and thus allow evaluation for AAC. One reader (JTS) evaluated the available soft-copy images for the cases and controls, with a previously validated AAC-24 and a newer simplified AAC-8 (Fig. 2). The reader was blinded to case status and all participant characteristics except age.
Details of both the AAC-24 and AAC-8 scales have been published elsewhere. Briefly, in the AAC-24 system, the anterior and posterior aortic walls were divided into four segments, corresponding to the areas in front of the lumbar vertebrae L1-L4. Aortic calcification scored as 0 if there was no calcification, as 1 if one-third or less of the aortic wall in that segment was calcified, as 2 if more than one-third but two-thirds or less of the aortic wall was calcified, or as 3 if more than two-thirds of the aortic wall was calcified. Scores could therefore range from 0–6 for each vertebral level, and the total score range was from 0–24.(31)
The AAC-8 scale has the same definition of aortic calcification as the AAC-24 scale but uses a simplified scoring system. In AAC-8, the score is the sum of the total length of calcification for each of the anterior and posterior aortic walls in front of vertebrae L1 to L4. A score of 0 is given if no calcification is seen, 1 if the aggregate length of calcification is ≤1 vertebral height, 2 if that length is >1 but ≤2 vertebral heights, 3 if that length is >2 but ≤3 vertebral heights, and 4 if the aggregate length of calcification is >3 vertebral heights.(28) The total score range is 0–8. In two previous cohorts, AAC-8 and AAC-24 scale scores have been shown to be highly correlated, but in our experience, AAC-8 is faster and easier to score than AAC-24 and may be easier to implement in clinical practice.
Ascertainment of CVD clinical risk factors
All participants in the parent study had systolic and diastolic blood pressure (BP), body mass index, and health status measured by the EuroQol visual analog scale.(32,33) Past medical events that may increase the risk of fractures, such as prior stroke, and other current self-reported comorbidities (including diabetes mellitus, hypertension, angina, and heart failure) were documented at the baseline study visit. Total hip BMD was assessed at baseline on a Hologic 4500A densitometer. After one quarter of the parent study population had been recruited, smoking status was assessed at baseline for the remainder of the study population. Lipid levels were not assessed during the parent study, but sera stored at −40°C were available to measure triglycerides and total, high-density lipoprotein (HDL), and low-density lipoprotein (LDL) cholesterol levels for 704 of those selected as cases or controls for this study. For the remainder, all of the stored sera had already been used in other studies.
Baseline characteristics of the cases and controls were compared with χ2 statistics for categorical variables and with Cochrane t-test statistics for continuous variables. Logistic regression models using SAS 9.1.3 software were used to assess the age-adjusted and multivariable-adjusted associations of AAC-24 and AAC-8 points and incident MI or stroke. Covariates included in the multivariable model were age, systolic BP, LDL and HDL cholesterol, triglycerides, smoking, renal function, treatment assignment (clodronate or placebo), self-reported diagnoses of diabetes mellitus, hypertension, angina, prior stroke, and health status. The multivariable models were done for that subset with complete covariate data, but also with multiple imputation using the PROC MI procedure in SAS 9.1.3 to impute missing covariate data.(34)
This study was designed to have 90% power to detect an OR of 1.75 for the primary outcome incident MI or stroke among those in the top tertile of AAC versus the bottom tertile, using either AAC scale. Although this study was not powered to analyze the association of AAC with incident MI and with incident stroke as separate outcomes, secondary analyses were done to estimate the multivariable-adjusted associations of AAC-8 and AAC-24 with each of these outcomes separately. An exploratory analysis was also performed examining the association of incident MI or stroke with AAC within tertiles of Framingham Point Score (FPS), an indicator of absolute 10-yr risk of coronary heart disease based on age, systolic BP, total and HDL cholesterol, smoking, and treatment of hypertension.(35,36) This was done to qualitatively gauge how predictive AAC may be of incident MI or stroke compared with FPS and to assess whether AAC might identify a subset of elderly women at high risk of incident MI or stroke not identified by FPS.
Among all women in the parent study who did not have an incident MI or stroke, the baseline characteristics of the 408 control patients selected for this study were nearly identical to the 4780 women that were not selected (data not shown), indicating that a reasonable random subset of the parent population without incident MI or stroke had been selected. Those with an unevaluable VFA image (n = 75) had higher body mass index, slightly lower total and LDL cholesterol levels, and better renal function, but were similar with respect to case status as those (n = 732) with an evaluable VFA image (p = 0.40; Table 1). Among those with an evaluable VFA image for AAC, cholesterol and triglyceride levels were available for 627 (85.7%) and smoking status for 512 (69.9%). Complete covariate data were available for 420 (57.4%).
Table Table 1.. Characteristics of Those With VFA Images Evaluable for AAC Different From Those With VFA Images Unevaluable for AAC
Among the cases, 191 women (51.8%) had an incident MI and 178 women (48.2%) had an incident stroke. Those who had an incident MI or stroke during the follow-up period had, at baseline, slightly higher triglycerides and systolic BP and lower HDL cholesterol, renal function, and health status. Cases were more likely to have angina, hypertension, and prior stroke at baseline and were less likely to have complete covariate data (Table 2). No association was noted between case status and either total hip BMD or clodronate treatment.
Table Table 2.. Baseline Characteristics of Cases and Controls
Among the 732 women with VFA images evaluable for AAC, there was a statistically significant linear trend for increasing risk of incident MI or stroke with increasing tertile of AAC-24 (p = 0.004) and with increasing tertile of AAC-8 (p = 0.002) using multiple imputation for missing covariate data. Those in the third tertiles of AAC-24 (score > 6) and AAC-8 (score > 3), respectively, had multivariable-adjusted excess odds of 74% and 77% for incident MI or stroke compared with those in the first tertiles of AAC score (Table 3). No statistically significant differences in OR were observed comparing the second tertile to the first or third tertiles. Results were very similar when analyses were limited to those with complete covariate data.
Table Table 3.. Association of AAC With Incident MI or Stroke (Primary Analyses)
When the associations of AAC with incident MI and incident stroke were analyzed separately, a statistically significant linear trend with increasing tertile of both AAC-24 (p = 0.003) and AAC-8 (p < 0.001) was noted for incident MI but not for incident stroke. Those in the third tertiles of AAC-24 and AAC-8, respectively, had excess odds of 106% and 114% for incident MI compared with the first tertiles. Those in the third tertiles of AAC-24 and AAC-8, respectively, had apparent excess odds of 49% and 51% for incident stroke compared with the first tertiles (Table 4). Those in the second tertile of AAC- 8 also had 105% excess odds of incident MI compared with the first tertile. Otherwise, no differences in the odds of either incident MI or incident stroke were noted between the third and second or between the second and first tertiles of AAC using either scale. Again, no associations were found between incident stroke and MI analyzed as separate outcomes and baseline total hip BMD or clodronate treatment.
Table Table 4.. Multivariable-Adjusted (With Multiple Imputation) Association of AAC With Incident MI and Stroke as Separate Outcomes
Those in the third tertile of the FPS (FPS > 23) had an OR of 1.74 (95% CI, 1.18–2.55) and those in the second FPS tertile (FPS = 21 or 22) had an OR of 1.06 (95% CI, 0.68–1.66) for incident MI or stroke compared with the first tertile (FPS < 16–20). When stratified by FPS tertiles, AAC tertiles remain statistically significant predictors of increased risk of MI or stroke (p value for trend = 0.001 for both AAC-24 and AAC-8). Among those in the second tertile of FPS, those in the third tertile of AAC-24 or AAC-8 score, respectively, had ORs of 2.95 (95% CI, 1.45–6.03) and 2.53 (95% CI, 1.29–4.98) for incident MI or stroke compared with those in the lowest tertiles of both AAC and FPS (Fig. 3).
Radiographic AAC has been shown to be a strong predictor of incident coronary heart disease,(13,18) stroke,(19) heart failure,(37) and intermittent claudication,(38) independent of traditional clinical CVD risk factors, and assessments of AAC on VFA images and standard radiographs have been shown to be highly correlated. In this study, a high level of AAC (score of ≥6 on the 24-point scale or score of ≥3 on the 8-point scale) on VFA images with either a previously validated 24-point scale or a simpler 8-point scale was predictive of incident MI or stroke among elderly women ≥75 yr of age, after adjustment for traditional clinical CVD risk factors. These results are consistent with previous studies documenting the association between CVD and AAC, most of which scored AAC with the same 24-point scale also used in this study. Supine lateral and lateral decubitus imaging yield spine images of similar quality, and on both, the aorta can be visualized sufficiently on VFA for AAC to be scored in 90% of postmenopausal women.(29)
The association between a high level of AAC and incident MI or stroke seemed to be as strong as the association between a high FPS (corresponding to an absolute 10-yr risk of incident coronary heart disease of 20% or more) and incident MI or stroke. Moreover, VFA may identify a subset of elderly women at high risk for incident MI or stroke that do not seem to be at high risk based on traditional clinical risk factors. Many women within the intermediate risk of coronary heart disease based on the FPS (10-yr coronary disease risk of 11–20%) may benefit the most from the detection of AAC on VFA images, in that if a VFA shows an AAC-8 score of 3 or higher or an AAC-24 score of 6 or higher, the 10-yr probability of incident coronary disease for them may exceed 20%. Current National Cholesterol Education Program guidelines recommend more aggressive treatment goals for those with a 10-yr risk >20% compared with those at lower risk.(36)
Low BMD itself has been shown to be associated both with vascular calcification(39,40) and with incident cardiovascular disease.(41) Unlike these prior reports, we found no association between total hip or femoral neck BMD and incident MI or stroke. However, our study population was much older (mean age, 80 yr) than the populations of those prior studies (mean ages ranging from 54 to 64 yr of age), and was not selected to have low BMD (only 21% of cases and 23% of controls had a femoral neck T-score < −2.5). If the association between BMD and incident cardiovascular disease weakens with increasing age (as has been shown between BMD and incident hip fracture with increasing age),(42) or if the association between BMD and incident CVD is stronger at lower ranges compared with higher ranges of BMD, this may at least in part explain this discrepancy. Further research is required to address these issues.
One prior study has found that etidronate reduced carotid intimal medial thickness, a risk factor for stroke,(43) whereas another small study found no effect of alendronate on progression of coronary artery calcification.(44) Neither our analysis nor that of the parent study found any association between clodronate treatment and incident cardiovascular disease. Our study was not designed to specifically test the hypothesis that clodronate may reduce incident stroke or MI. Moreover, whereas animal studies suggest that oral bisphosphonates may prevent vascular calcification,(45) our study population consisted of very elderly women of whom a substantial proportion had established vascular calcification. Finally, unstable atherosclerotic plaques, thought to be important in the pathogenesis of acute vascular events such as stroke and MI, may actually contain less calcium than stable plaques,(46) and hence the association between vascular calcification and cardiovascular disease events may be because of the fact that vascular calcification is strongly associated with the simultaneous presence of unstable plaques.(47) Therefore, reduction of MI or stroke with bone active agents may require mechanisms of action other than simply preventing vascular calcification.
Bone densitometry is now recommended for all women ≥65 yr of age to assess incident fracture risk,(23,24,48,49) and VFA is indicated for the detection of vertebral fracture in such women.(50) This may allow identification of a significant CVD risk factor among the majority of the female population ≥65 yr of age at minimal additional cost or radiation exposure. Based on identification of prevalent vertebral fracture alone, this procedure is cost effective for that subset of postmenopausal women who, in the absence of prevalent vertebral fracture, would not be candidates for fracture prevention therapy.(51) Simultaneous assessment for AAC on VFA images in this population may increase the use of the procedure even further. Although AAC is not as predictive of coronary events as coronary calcium,(52) VFA imaging has the advantage over other imaging modalities of lower cost and greater convenience, and in comparison with CT imaging, much lower radiation exposure.
However, as is true for other candidate tests to detect those at high risk of CVD that are not identified by traditional clinical CVD risk factors (such as C-reactive protein, homocystinemia, carotid IMT, and coronary calcium score), the precise role that imaging for AAC may play in strategies to prevent clinical cardiovascular disease remains undefined.(3) Future studies of VFA imaging with larger populations including younger postmenopausal women and men, with more complete covariate data collection, and with the specific pre hoc intention to assess the association between AAC and multiple cardiovascular disease endpoints (including mortality) are needed. These will greatly help define the role that VFA imaging may play in strategies to prevent incident clinical cardiovascular disease within the broader population.
Our study has several important strengths. This is the first study to show that AAC scored on VFA images is predictive of a form of incident cardiovascular disease and the first to show the predictive validity of the AAC-8 scale, which is practical for use in clinical practice. Assessments of AAC were done completely blinded to case status and covariate data. Although the parent study population was recruited to participate in a RCT, the parent study had very few exclusion criteria, such that our results are generalizable to elderly female white populations. Records of all hospitalizations and deaths were obtained during the study follow-up period for all participants, such that ascertainment of fatal and nonfatal MI and stroke sufficient to cause death or hospitalization is highly likely to be accurate. To the extent that misclassification of cardiovascular disease outcomes occurred, our results would be biased toward the null.
This study, however, has important limitations. This is an ad hoc analysis of a randomized clinical trial done for purposes unrelated to cardiovascular disease. Significant proportions of the study cohort had missing cholesterol and triglyceride levels and missing data regarding smoking status. This is mitigated, however, by the similarity of the results whether the multivariable analyses were restricted to those with complete covariate data or done with multiple imputation. This study did not have adequate power to assess the association of AAC with incident MI and stroke separately or to assess the association between AAC and cardiovascular mortality or other CVD endpoints. These results are strictly applicable only to white women ≥75 yr of age, and additional studies may be needed to confirm the applicability to men, younger postmenopausal women, and those of other ethnic backgrounds. However, because radiographically detected AAC has been confirmed as a CVD risk factor in men and younger women(13,18) and good agreement has been shown between radiographic and VFA detected AAC,(28,29) results using VFA in these populations to detect AAC may be similar to radiographically detected AAC.
In conclusion, a high level of AAC detected on VFA images is predictive of incident MI or stroke among elderly white women, independent of other clinical CVD risk factors. VFA imaging performed at the time of bone densitometry offers an opportunity to simultaneously assess for prevalent vertebral fracture and AAC at minimal additional cost. Our data strongly suggest that AAC incidentally detected on images obtained for the purpose of VFA should not be ignored, and the provider responsible for care of the patient should be notified of its presence and significance. Further studies in larger and more diverse populations will be needed to better define how VFA can be best incorporated into strategies to prevent incident cardiovascular disease.
Drs Taylor and McCloskey had full access to all of the data for this study.