Prevalent Vertebral Fracture (PVFx) and Abdominal Aortic Calcification (AAC) Independently Predict Incident Atherosclerotic Cardiovascular Disease (ASCVD) Events in Older Men

Abstract

Purpose:

To estimate the associations of prevalent vertebral fracture (PVFx) and abdominal aortic calcification (AAC) with incident ASCVD (myocardial infarction, fatal or non-fatal cerebrovascular accident, or coronary heart disease death).

Methods:

2799 older men (mean [SD] age 76.3 [5.5] years) enrolled in the MrOS sleep ancillary study had PVFx (SQ grade 2 or 3) assessed by human reader and AAC by automated convolutional neural networks on baseline lateral spine radiographs. Auto-AAC was categorized as low, moderate, or high (24-point scale score <2, 2 to <6, or ≥6). Men were contacted every 4 months for ascertainment of possible ASCVD events over a mean (SD) follow-up of 8.4 (3.1) years. Associations of PVFx and auto-AAC category with incident ASCVD were estimated with modified proportional hazards models accounting for non-CVD mortality.

Results:

PVFx was present in 7.3% of the cohort; 34.5% had moderate auto-AAC, 28.8% had high auto-AAC, and 396 (14.1%) had an incident ASCVD event. Compared to men with low auto-AAC, those with moderate (HR 1.34, 95% CI 1.01, 1.77) and high (HR 1.55, 95% CI 1.16, 2.06) auto-AAC had increased risk of ASCVD events adjusted for PVFx and other risk factors. Compared to men with no PVFx, those with PVFx had a higher risk of ASCVD events (HR 1.51, 95% CI 1.07, 2.14) adjusted for auto-AAC level and other risk factors.

Conclusion:

AAC assessed by automated methods and PVFx are independently associated with incident ASCVD events and may aid in ASCVD risk stratification in older men, but confirmatory studies are needed.

1. Introduction

Prevalent vertebral fracture (PVFx) and abdominal aortic calcification (AAC) can be simultaneously ascertained on lateral spine images, and are associated with incident hip, vertebral, and major osteoporotic fractures in older men[1] and women,[2, 3] adjusted for each other and other fracture risk factors. In addition, AAC is a robust predictor of incident cardiovascular disease events in middle-aged and older women and men.[4–7] Low bone mass,[8] high fracture risk,[9] and PVFx[8, 10, 11] are also associated with incident cardiovascular disease events in older women. PVFx is also associated with vascular calcification in patients with hemodialysis patients,[12] and lower limb arterial insufficiency in patients with CKD stages 3 to 5.[13]

However, it remains unclear if PVFx predicts incident cardiovascular disease events in older men, and if that association remains after accounting for clinical cardiovascular disease risk factors and the competing risk of non-cardiovascular mortality. It is also unknown whether both PVFx and AAC predict cardiovascular disease events independent of each other. Our primary aim was to estimate, in a large cohort of older community-dwelling men, the association of PVFx and AAC with incident atherosclerotic cardiovascular events (a composite outcome of myocardial infarction, fatal and non-fatal stroke, and coronary heart disease death), adjusted for each other and clinical cardiovascular disease risk factors. Our secondary aims were to estimate the associations of PVFx and AAC with incident ischemic congestive heart failure, peripheral vascular disease events, and any incident cardiovascular disease event, adjusted for each other and clinical cardiovascular disease risk factors.

2. Materials and Methods

The Osteoporotic Fractures in Men (MrOS) study enrolled 5994 men at six U.S. study sites (Birmingham AL, Minneapolis MN, Palo Alto CA, Pittsburgh PA, Portland OR, and San Diego CA) between the years 2000 and 2002, as described in prior publications, after signing informed consent documents approved by the Institutional Review Board review at each of the six sites. Eligible men were 65 years of age or older, were community dwelling and able to walk without assistance, and had not had bilateral hip arthroplasties. All men had lateral lumbar and thoracic spine x-rays taken at their baseline visit. Among them, a mean (SD) 3.4 (0.5) years later, 3135 men were enrolled in the MrOS sleep substudy (Figure 1). Those who did not enroll in the Sleep Substudy were excluded, because they were not rigorously followed up prospectively for incident cardiovascular disease events. Characteristics of men who enrolled in the sleep study at baseline were showed they were slightly younger and healthier than men who did not (Supplemental Table 1), although the differences in these characteristics were slight. Among those in the Sleep Substudy, 2799 men with baseline visit lateral thoracic and lumbar spine radiographs of sufficient quality to allow adequate adjudication for prevalent vertebral fracture status and scoring of abdominal aortic calcification comprise the analytical cohort for this study (Figure 1).

Figure 1 – Flow Diagram for Analytic Cohort.

2.1. Ascertainment of Prevalent Vertebral Fracture

Lateral spine thoracic and lumbar spine x-rays were adjudicated for PVFx using a two-stage process.[14] First, technicians triaged the x-rays for the presence of any vertebrae with an abnormal appearance. Participants for whom all vertebrae were judged to be clearly normal were classified as having no prevalent vertebral fracture. The remaining x-rays were read by one expert reader (author JTS) for prevalent vertebral fracture. The Genant semi-quantitative method was used with one modification, that grade 1 vertebral deformities had to exhibit broad endplate depression in order to be considered fractured. Tests of intra-rater reliability were done at three timepoints and showed kappa scores of 0.79 to 0.93. For the primary analysis, those with ≥ 1 SQ grade 2 or 3 vertebral fractures were considered to have PVFx, and the others to have no prevalent vertebral fracture. A secondary analysis was done defining PVFx as having any SQ grade (1 through 3) vertebral fracture.

2.2. Ascertainment of Abdominal Aortic Calcification (AAC)

Baseline lateral spine x-rays were semi-quantitatively scored using the Kauppila method developed in the Framingham study.[15] The anterior and posterior aortic walls are scored with each segment of the aorta anterior to each of lumbar vertebrae 1 through four are scored on a scale of 0 to 3 for proportion of the length of the aorta within that segment. The scores for all 8 segments are summed, yielding a possible score range of 0 to 24. One reader (PS) scored all x-rays for AAC, and for a subset of 40 individuals inter-rater agreement with a clinician expert (JTS) was excellent (ICC 0.94 [95% C.I 0.88 to 0.97]).[16] For the purposes of the primary analyses, AAC scores were categorized as low (score 0 or 1), moderate (score 2 to 5), or high (score ≥ 6).

We have used convolutional neural networks to develop an automated algorithm to score AAC on lateral spine radiographs in the MrOS, with very good to excellent agreement with AAC scores generated by a human reader (spearman correlation 0.92, kappa agreement between expert and auto-AAC categories 0.74).[17] Readings of AAC are time-consuming and require trained experts, and it is possible that widespread scoring of AAC on lateral spine imaging will require automated methods. For this reason, we chose to use the CNN-generated auto-AAC (also referred to as ML-AAC) score categories (low [<2], moderate [2 to <6], and high [≥ 6]) as the AAC predictor variable in our primary analyses. Secondary analyses were also done using AAC score categories generated by human reader as the AAC predictor.

2.3. Ascertainment of Covariate Predictors

All covariate predictors were measured at the first Sleep Substudy visit, unless otherwise noted below. Height and weight were measured on a wall-mounted Harpenden stadiometer and on a balance beam scale, respectively, and body mass index calculated as weight divided by height squared. Waist circumference was measured with a flexible tape measure three times and averaged. Systolic blood pressure was measured in the sitting posture in both arms and averaged. Smoking status and past physician diagnosis of coronary heart disease, peripheral vascular disease, or cerebrovascular disease was assessed by self-report. Self-reported physical activity was assessed with the Physical Activity Scale for the Elderly (PASE). Diabetes mellitus was defined at the first MrOS Sleep Substudy visit as fasting (>8 hours) glucose>126 mg/dl OR self-reported diagnosis of diabetes OR using hypoglycemic medications or insulin use. Oxidized LDL cholesterol was measured at the first Sleep Substudy visit with a Beckman Coulter Biomek NXp (Beckman Coulter, Inc, Fullerton, CA) using direct sandwich enzyme immunoassay and Oxidized LDL ELISA (manufacturer Mercodia AB, Uppsala, Sweden). HDL and LDL cholesterol and triglycerides were not assessed at the first Sleep Substudy visit, and hence values from the main baseline MrOS visit years 2000–2002 were used. These were measured using a Roche COBAS Integra 800 automated analyzer (Roche Diagnostics Corp., Indianapolis, IN). Total cholesterol values were calculated from HDL, LDL, and triglyceride levels. Participants were instructed to bring to their first Sleep Study visit bottles for all currently used prescription and over the counter medications. Those using a statin or anti-hypertensive medication (calcium channel blocker, beta blocker, angiotensin receptor blocker, angiotensin converting enzyme inhibitor, thiazide diuretic, hydralazine, aliskeren, or prazosin) were identified.

2.4. Ascertainment of Incident Cardiovascular Disease Events

MrOS Sleep Substudy enrollees were contacted by postcard (and if necessary, by phone) every four months to ask if they had any adverse heart or blood vessel health events since they were last contacted. Physician and hospital records were obtained for those that indicated they may have had a cardiovascular health event. For in-hospital fatal events, the death certificate was also obtained, and for out of hospital fatal events an interview with next of kin was conducted and records from the most recent hospital stay obtained. All records were reviewed and adjudicated by a physician according to a pre-specified protocol using criteria that have been used for cardiovascular intervention trials.[18] An expert cardiologist was available for additional review of equivocal cases.

The primary outcome was incident atherosclerotic cardiovascular disease (ASCVD) events, defined by the American Heart Association as myocardial infarction, stroke (fatal or non-fatal), or coronary heart disease death. Secondary outcomes were incident ischemic congestive heart failure, incident peripheral vascular disease (peripheral arterial surgery, acute arterial dissection, occlusion, or rupture), and any cardiovascular event. Any cardiovascular event was defined as ASCVD, incident CHF, PVD or hospitalization for unstable angina, coronary artery bypass surgery, mechanical coronary revascularization, transient ischemic attack, deep vein thrombosis, pulmonary embolism, or sudden death (due to coronary heart disease or not otherwise specified).

2.5. Statistical Analysis

All analyses were conducted using Stata version 18.0 (College Station, TX). Baseline characteristics were described stratified by auto-AAC level, and statistical tests across auto-AAC level were done using Kruskal-Wallis tests for continuous variables and chi-square statistics for categorical variables.

Proportional subhazards models, modified by the method of Lambert[19] to account for the competing risk of non-CVD mortality, were used to assess associations of predictor variables with the primary outcome (ASCVD events). Model 1 estimated the association of PVFx (Genant SQ grade 2 or 3) with incident ASCVD adjusted for study site, age, and self-reported past cardiovascular disease. Model 2 added auto-AAC category as an additional covariate.

Other candidate covariates (Supplemental Table 2), were tested individually in separate models for their association with incident ASCVD, adjusted for PVFx, auto-AAC, age, past CVD, and study site, accounting for competing risk of non-CVD mortality. The initial full multivariable model included the additional covariates associated with incident ASCVD at a p-value <0.20 along with Model 2 covariates. For the final multivariable model (Model 3), those covariates with a fully adjusted p-value >0.2 were removed. Model 4 was the same as Model 3 but excluded those with a previous history of CVD. We also tested the significance of interactions terms between PVFx status and past CVD, as well as between auto AAC category and past CVD, by adding these interaction terms separately to Model 3. The proportional hazards assumption was tested and was not violated for any model.

Cumulative incidence curves for ASCVD over 5-years were generated from Model 3 for subsets defined by PVFx status and auto-AAC level. The relative variable importance for each covariate in Model 3 was calculated as the increase of log-likelihood chi-square value when that covariate was singly removed from Model 3.

Secondary analyses with incident ischemic congestive heart failure, peripheral vascular disease, and any cardiovascular disease as the outcome were performed including all of the covariates of the Model 4. Secondary analyses were also performed defining prevalent vertebral fracture as any SQ grade (1, 2, or 3), and substituting human reader AAC score categories for those of auto-AAC.

3. Results

Baseline characteristics of the cohort stratified by auto-AAC level are shown in Table 1. Those with higher levels of auto-AAC were older, more likely to have a self-reported physician diagnosis of cardiovascular disease, to have a current or past history of smoking, to have diabetes mellitus, and to be baseline users of statin and/or anti-hypertensive medication. Those with higher levels of auto-AAC also had higher systolic blood pressure and waist circumference, and modestly lower HDL, oxidized LDL, and total cholesterol levels.

Table 1:

Baseline Characteristics Stratified by Auto-AAC Category

	Auto-AAC24 Score Category
Characteristic	Low (<2)	Moderate (2 to <6)	High ≥ 6	p-value
Number (%)	1,027 (36.7%)	965 (34.5%)	807 (28.8%)
Prevalent Vertebral Fracture (SQ 2 or 3), N (%)	58 (5.6%)	75 (7.8%)	71 (8.8%)	0.028*
Age, years (SD)	74.5 (4.8)	76.5 (5.4)	78.4 (5.8)	<0.001†
Past Diagnosis of CVD‡, N (%)	252 (24.5%)	386 (40.0%)	437 (54.2%)	<0.001*
Systolic Blood Pressure, mmHg (SD)	124.8 (15.6)	127.7 (15.8)	129.6 (17.4)	<0.001†
Body Mass Index, kg/m2 (SD)	27.0 (3.7)	27.1 (3.8)	27.1 (3.9)	0.737†
Waist Circumference, cm (SD)	98.9 (10.8)	99.8 (11.1)	100.1 (11.0)	0.051†
Smoking Status, N (%)
Never	495 (48.2%)	373 (38.7%)	240 (29.7%)
Past	513 (50.0%)	577 (59.9%)	541 (67.0%)	<0.001*
Current	18 (1.8%)	14 (1.5%)	26 (3.2%)
HDL Cholesterol, mg/dl (SD)	49.8 (13.9)	48.3 (14.0)	49.1 (14.9)	0.011†
Oxidized LDL Cholesterol Units/L (SD)	44.6 (11.9)	43.9 (11.8)	42.8 (12.6)	<0.001†
Total Cholesterol, mg/dl (SD)	195.1 (34.5)	194.2 (33.1)	191.1 (34.6)	0.018†
Diabetes Mellitus, N (%)	139 (13.5%)	146 (15.1%)	154 (19.1%)	0.005*
PASE Score (SD)	156.8 (73.7)	144.7 (72.6)	135.3 (69.8)	<0.001†
Baseline Statin Use, N (%)	212 (20.6%)	282 (29.2%)	274 (34.0%)	<0.001*
Baseline Anti-Hypertensive Medication Use, N (%)	412 (40.1%)	475 (49.2%)	493 (61.1%)	<0.001*

^*Likelihood Ratio Chi²

^†Kruskal-WaNis test

^‡Self-reported physician diagnosis of coronary heart disease, cerebrovascular disease, or peripheral vascular disease

Among candidate covariates, systolic blood pressure, HDL cholesterol, oxidized LDL cholesterol, body mass index, baseline statin use, anti-hypertensive medication use, and aspirin use met criteria for inclusion in the initial multivariable model (Supplemental Table 2). However, oxidized LDL cholesterol (p = 0.57), BMI (p=0.31), and aspirin use (p=0.37) had p-values above 0.20 for the association with incident ASCVD in the initial full model, and hence were removed from the final multivariable model.

3.1. Primary Analyses

Adjusted only for age, previous history of cardiovascular disease, and study enrollment site, men with PVFx (SQ grade 2 or 3) compared to those with SQ grade 0 or 1 had a subhazard ratio of 1.62 (95% C.I. 1.18, 2.22) for incident ASCVD (Table 2). This subhazard ratio did not meaningfully change with the addition of auto-AAC category (Model 2, Table 2). Compared with men with low or no AAC, those with moderate (SHR 1.38, 95% C.I. 1.08, 1.61) and high levels of AAC (SHR 1.62, 95% C.I. 1.23, 2.13), had increased risks of incident ASCVD, accounting for age, PVFx, and past CVD. These associations were only slightly attenuated, and remained significant, with further adjustment for systolic blood pressure, HDL cholesterol, baseline statin use, and anti-hypertensive medication use (Model 3, Table 2). Interaction terms of self-reported past CVD with PVFx and with autoAAC category were insignificant (p-values 0.97, 0.30, and 0.50 for PVFx, moderate auto-AAC, and high auto-AAC, respectively).

Table 2 -.

Associations Prevalent Vertebral Fracture & Auto-AAC with Incident ASCVD

	Model 1	Model 2	Model 3	Model 4*
Number of observations (Number of outcome events)	N=2740 (381)	N=2740 (381)	N=2528 (353)	N=1561 (171)
	HR (95% C.I.)	HR (95% C.I.)	HR (95% C.I.)	HR (95% C.I.)
Prevalent Vertebral Fracture (SQ 2 or 3)	1.62 [1.18 2.22]	1.60 [1.17 2.20]	1.51 [1.07 2.14]	1.52 [0.90 2.57]
Auto-AAC Category
Low (<2)		Reference	Reference	Reference
Moderate (2 to <6)		1.38 [1.06 1.81]	1.34 [1.01 1.77]	1.44 [1.00 2.08]
High (≥6)		1.62 [1.23 2.13]	1.55 [1.16 2.06]	1.47 [0.98 2.21]
Past Cardiovascular Diagnosis	1.61 [1.32 1.98]	1.49 [1.21 1.84]	1.55 [1.23 1.94]
Age (per 5 year increase)	1.34 [1.23 1.47]	1.29 [1.18 1.41]	1.28 [1.16 1.41]	1.46 [1.27 1.68]
Systolic Blood Pressure (per SD increase)			1.22 [1.11 1.35]	1.27 [1.09 1.47]
HDL Cholesterol (per SD increase)			0.86 [0.77 0.97]	0.87 [0.73 1.02]
Baseline Statin Use			0.83 [0.65 1.06]	0.82 [0.54 1.25]
Baseline Anti-Hypertensive Medication Use			1.18 [0.94 1.47]	1.32 [0.97 1.80]

^*Excluding those with past physician diagnosis of cardiovascular diagnosis (coronary heart disease, cerebrovascular disease, and/or peripheral vascular disease)

Hazard Ratios with p-value of association <0.05 are in Bold

Predicted ASCVD-free survival curves generated from Model 3 showed that the 5-year ASCVD-free survival varied from 0.93 in men without grade 2–3 PVFx and low auto-AAC to 0.84 in those with both PVFx and a high level of auto-AAC (Figure 2). Men with PVFx had lower ASCVD-free survival compared to those without grade 2–3 PVFx. Within each category of PVFx status, men with moderate auto-AAC vs those with low auto-AAC had a lower ASCVD-free survival, and men with high auto-AAC appeared to have a lower ASCVD-free survival than men with moderate auto-AAC. ASCVD-free survival appeared very similar in men without grade 2–3PVFx but with high auto-AAC and men with grade 2–3 PVFx but low auto-AAC. Increasing age was the most important ASCVD predictor, followed by systolic blood pressure and past history of CVD (Figure 3). Auto-AAC and PVFx were the fourth and sixth most important predictors, of similar magnitude to HDL cholesterol.

Figure 2 – Cumulative ASCVD-Free Survival in Subsets of Men Defined by auto-AAC level and Prevalent Vertebral Fracture Status.

Figure 3 – Relative Variable Importance for Prediction of Incident ASCVD Events.

*SQ grade 2 or 3 prevalent vertebral fracture

^†calcium channel blocker, beta blocker, angiotensin receptor blocker, angiotensin converting enzyme inhibitor, thiazide diuretic, hydralazine, aliskeren, or prazosin

When men with self-reported previous CVD diagnoses were excluded from the analysis, the sample size was substantially from 2528 to 1561. The point estimates of association between PVFx and auto-AAC category were similar those in to model 3, but 95% confidence intervals around the point estimates of associations all crossed 1.00 (Model 4, Table 2).

3.1. Secondary Analyses

The associations of PVFx and AAC with incident ASCVD events did not meaningfully change when human reader AAC score categories were used in place of auto-AAC score categories (Supplemental Table 3). However, the associations of PVFx with incident events were much weaker in magnitude when PVFx expressed as any SQ grade (1, 2, or 3, Supplemental Table 4) rather than SQ grade 2 or 3.

The associations of PVFx (SQ grade 2 or 3) with incident ischemic congestive heart failure events and any cardiovascular disease event were smaller in magnitude than its association with incident ASCVD and not significant (Supplemental Table 5).Men with PVFx grade 2 or 3 versus PVFx grade 0 or 1 appeared to have a higher risk incident peripheral vascular disease events, but the 95% confidence interval crossed 1.00 (HR 1.54, 95% CI 0.88 to 2.68). On the other hand, men with a high level of auto-AAC compared with men with a low level of auto-ACC had increased risks of incident ischemic congestive heart failure, peripheral vascular disease events, and any cardiovascular disease events (Supplemental Table 5).

4. Discussion

In this study of older men, prevalent moderate or severe vertebral fracture (PVFx) and higher auto-AAC were associated with increased risks of incident ASCVD events, adjusted for each other and other cardiovascular risk factors. Our findings suggest that both PVFx and auto-AAC may contribute to ASCVD risk stratification among older men, and indicate their potential for refining estimates of ASCVD risk and guiding clinical cardiovascular risk management decisions.

Men who have already been diagnosed with CVD are already known to be at high risk of incident ASCVD events. Hence, identifying prevalent vertebral fracture and/or auto-AAC has the potential to influence cardiovascular disease risk management decisions primarily in the subset of older men who do not have known clinical cardiovascular disease. When our analyses were restricted to men without a prior clinical CVD diagnosis, the point estimates of associations of PVFx and auto-AAC with incident ASCVD events were similar in magnitude to those in the entire analytic sample, but the confidence intervals around the point estimates of association crossed 1.00. Hence, additional studies with larger numbers of older men who do not have any prior clinical cardiovascular disease events are warranted to examine the associations of PVFx and auto-AAC with risk of ASCVD events in men without a clinical history of CVD.

Moreover, it is noteworthy that we did not find an association between PVFx and incident congestive heart failure, incident peripheral vascular disease, or any cardiovascular disease event. These results may in part be due to the low prevalence (approximately 7%) of moderate to severe vertebral fracture in our cohort of community dwelling men unselected for fracture risk factors. In a very large cohort using national insurance claims data in Taiwan, Lee and colleagues reported that those with clinical vertebral fracture had a subsequent higher incidence of hospitalization for CHF.[20] Larger studies of older men selected for higher fracture risk, with a higher prevalence of vertebral fracture, are needed to investigate if PVFx is associated with incident congestive heart failure or incident peripheral vascular disease in older men.

Our study findings nonetheless indicate that prior findings of an association between prevalent vertebral fracture and incident major adverse cardiovascular disease events in older women[10, 11, 21, 22] are also true for older men. Our results are consistent with prior analyses of older men enrolled in the STRAMBO study that reported an association of prevalent vertebral fracture with greater AAC severity.[23] Prevalent vertebral fracture in other studies has also been found to be associated with iliac[24] and coronary artery calcification.[25] Other studies in women have found associations of other indicators of bone fragility (such as high FRAX fracture risk estimates[9] or very low bone mineral density[8, 26]) with incident cardiovascular disease events, and with coronary artery calcification.[27] Our findings are broadly consistent with these other studies. With respect to AAC, our findings are also consistent with a large number of studies in middle-aged men showing that severe AAC predicts a variety of incident cardiovascular disease events after accounting for other cardiovascular disease risk factors.[4, 6, 7]

Osteoporosis and associated fractures and atherosclerotic cardiovascular disease share many clinical risk factors (such as smoking and alcohol abuse), pathogenic mechanisms such as activation of inflammatory pathways and oxidative stress, and expression of matrix proteins in both vascular walls and bone such as osteopontin, osteoprotegerin, and bone morphogenetic proteins.[28] Altered microvascular circulation related to atherosclerosis and increased arterial stiffness may also directly impair bone perfusion, bone mineral density, and strength.[29, 30] However, the complex causal web that links these conditions is not fully elucidated or understood.[28, 31]

We recognize limitations to our study. The MrOS population is 90% Caucasian, community dwelling, and slightly healthier than the age-matched general US older male population. Hence, our findings may not be generalizable to those of non-white ethnicity, long-term care residents, or older men with a high level of multimorbidity. Our finding of an association between PVFx and incident ASCVD events does not establish that PVFx is a causal factor for ASCVD events. There may be residual confounding from measured and unmeasured variables accounting at least in part for our findings. The Kaupila scale to score AAC is semiquantitative not as precise as fully quantitative ascertainment.[32] However, imprecision in our AAC measure would tend to bias the associations of AAC with incident ASCVD toward the null. Finally, prior cardiovascular disease was assessed by self-report, which has high specificity but moderate sensitivity compared to claims data for ascertainment of prior clinical cardiovascular disease events.[33]

However, there are important strengths to our study. In particular, both PVFx and AAC were ascertained with high intra- and inter-rater reliability. Incident ASCVD events were ascertained by expert review of medical records, consistent with adjudication of events in randomized controlled trial of cardiovascular disease interventions.

In conclusion, prevalent vertebral fracture and automated assessments of AAC are associated with incident atherosclerotic cardiovascular disease (ASCVD) events in older men, after accounting for each other and other cardiovascular disease risk factors. Further studies are needed in older men without a history of clinical cardiovascular disease and in study populations of older men with a higher prevalence of radiographic vertebral fracture.

Highlights.

• Prevalent vertebral fractures (PVFx) predict incident ASCVD, adjusted for AAC.

• AAC predict incident ASCVD, adjusted for PVFx and other clinical ASCVD risk factors

• Confirmatory studies are needed in older men with no prior cardiovascular disease