Bone mineral density and risk of cardiovascular disease in men and women: the HUNT study

The HUNT Study is the largest Norwegian population-based health study [25]. All adults residing in the northern part of Trøndelag county (n = 94 194 in 1995–1997, n = 93,860 in 2006–2008) were invited to undergo clinical examinations, blood sampling, interviews, and questionnaires in four surveys (HUNT1 1984–86, HUNT2 1995–97,HUNT3 2006–08 and HUNT4 2017–2019) [26]. Bone densitometry was not performed in HUNT1, thus, in the present study, we utilized data from HUNT2 and HUNT3 where a total of 65 215 (69.2% of those invited) and 50 796 (54.1%) individuals participated, respectively. Forearm bone densitometry was performed in 17 749 participants in HUNT2 (21,734 invited; 81.7% participated) and 14 774 participants in HUNT3 (22,490 invited; 65.7% participated) resulting in a total sample of 26 056 adults (Fig. 1). Of the total 26 056 sample, 6 467 individuals (24.8%) participated in both HUNT2 and HUNT3. Participants were selected for bone mineral density measurements in two different ways: 1) a random sample was selected (a 5% random sample from HUNT2 and 10% random sample from HUNT3 among all participants or as part of a 30% random sample from the female birth cohorts or random sample from young-HUNT1 participants) and 2) a symptom sample was selected for the HUNT Lung Study [26], where participants who were asked to participate in spirometry measurements were also asked to participate in bone mineral density measurements. Relevant symptoms included wheezing or breathlessness during the last 12 months at baseline, a history of asthma or ever use of asthma medication. The detailed participant selection process is presented in Figure S1.

Among participants with asthma or COPD we have excluded 1 191 individuals (4.6%) who were current users of inhaled corticosteroids, which could have influenced their BMD measurements [27]. To examine incidence, we excluded subjects with a prior history of cardiovascular diseases including AF, AMI, HF and stroke (n = 2 008, 7.7%), resulting in a total sample of 22,857. Flow chart of participant selection process is illustrated in Fig. 1.

In HUNT2, BMD was measured using single-energy X-ray absorptiometry (SXA) (DTX 100, Osteometer Meditech A/S, Copenhagen, Denmark). Daily calibration of the densitometers was performed with equipment-specific phantoms [28]. Measurements were taken in the non-dominant distal forearm, while the dominant arm was used in the case of previous fractures in the non-dominant arm (2.5% of cases). The distal region was 24 mm proximal to the point at which radius and ulna are 8 mm apart [29]. In HUNT3, BMD was measured using dual-energy X-ray absorptiometry (DTX200, Osteometer Meditech A/S, Copenhagen, Denmark) (n = 9 147) and DTX100 (n = 5 627).

BMD was standardized as T-scores. In both sexes separately, we calculated the T-score as observed BMD minus mean BMD from a reference population divided by standard deviation (SD) of reference population. The reference population was a healthy female population (excluded individuals with self-reported osteoporosis, arthritis, hip, or wrist fractures, hyper- or hypothyroidism and use of corticosteroids) aged 20–39 years from the HUNT Study [22]. Further, BMD T-score was categorized according to the WHO criteria as normal (T-score ≥ – 1.0), osteopenia (– 1.0 to – 2.5), and osteoporosis (≤ – 2.5) [30].

Incident cases were ascertained by linking the HUNT data with full hospital records in Nord-Trøndelag County from 1995 to 2015. The diagnoses were based on International Statistical Classification of Diseases and Related Health Problems (ICD).

Medical records were manually reviewed by fellow cardiologist and AF was diagnosed based on ICD-10 code I48. The patient was considered as having AF if the electrocardiogram (ECG) could be classified as AF or atrial flutter according to the standard criteria based on the American College of Cardiology consensus guideline [31]. If an ECG scan was not in the digital medical record, the written records were further reviewed for ECG interpretation and, in doubtful cases, the information was evaluated separately by specialist in cardiology and internal medicine, and then discussed in a consensus meeting [32]. In the cases where an ECG was not taken at all, but patients had described irregular heartbeats or periods of fast, irregular pulse, it was not considered AF in our study.

AMI was defined and diagnosed by the caregiving cardiologists and physicians according to the European Society of Cardiology/American College of Cardiology consensus guidelines and consisted of ICD-9 code 410 and ICD-10 codes I21 and I22 [31]. Criteria for AMI included: specific clinical symptoms according to case history information, changes in blood levels of cardiac enzymes, and electrocardiogram changes as defined in the American and European consensus guidelines. If the cardiologists or physicians judged the event to not be a valid AMI, the event was deleted from the registry. A small part of the AMI diagnoses (2%) from medical records have been manually validated [33] and an ongoing validation study (unpublished) found that 92% of the cases (n = 1194) was type 1 AMI.

Ischemic stroke consisted of ICD-9 codes 433 and 434 and ICD-10 code I63 (all positions), while hemorrhagic stroke consisted of ICD-9 codes 430, 431 and 432 and ICD-10 codes I60, I61 and I62. Electronic medical records and diagnostic imaging of hospital admissions for stroke in Norway has been shown to have high sensitivity and positive predictive values in validation studies [34, 35]. HF diagnosis was based on ICD-10 code I50. In addition, we aggregated AMI, any stroke and HF cases into any CVD.

A self-administrated questionnaire was used to assess participants’ smoking status (never, former and current), physical activity (inactive, low, medium and high), alcohol use (abstainers, light, moderate and heavy drinkers), education (< 10, 10–12, > 12 years) and medical history of common chronic diseases. A detailed description of the covariates can be found elsewhere [36]. Body mass index (BMI) was calculated by dividing body weight (kg) by height (m) squared (kg/m²). Estrogen users were defined as women reporting current or previous use of systemic estrogen pills or patches, while non-users as women reporting “never use” of estrogen. Postmenopausal women were defined as those who self-reported use of estrogen or answered negative to “Do you still menstruate?” question.

Baseline characteristics were presented using means (SDs) for continuous variables and numbers (percentages) for categorical variables. For the individuals participating at both HUNT2 and HUNT3, HUNT2 was regarded as baseline.

To investigate the prospective association between BMD T-score and outcomes we used Cox proportional hazard models to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). Risk time was calculated from baseline until the first event of interest, death, emigration or end of follow-up (30th November 2015), whichever came first. We used follow-up time as the time scale in our analysis. We tested the proportionality of hazards using log–log curves and Schoenfield's test. We fitted cause-specific models, thus participants with competing events (deaths) were censored at the time of the event. Missing data on covariates were imputed using multiple imputation with chained equations, M = 20.

The non-linearity in the relationship between BMD T-score and outcomes was assessed using the restricted cubic splines with three knots. The number of knots were determined using Akaike (AIC) and Bayesian information criterion (BIC). We observed no deviation from linearity by comparing the Cox proportional hazard models with and without cubic spline terms using Wald test and likelihood ratio tests.

We reviewed the literature and performed a Directed Acyclic Graph analysis to select covariates that could cause both BMD and CVD (Figure S2). A minimally adjusted model included age and age-squared (Model 1). The age-squared term was used to account for the possible non-linearity of age influencing the exposure of interest. Further to this, we controlled for BMI, physical activity, smoking status, alcohol use, and education level (Model 2). We additionally adjusted for estrogen and postmenopause in the analysis among women (Model 3). The E-value was calculated to quantify the strength of residual confounding that would require to explain away the estimates [37]. All analyses were prior stratified by sex due to the sex differences in BMD and bone turnover [38].

We performed the data analyses using Stata 13.1 for Windows 10 (StataCorp). The study received ethics approval from the Regional Committee for Medical Research Ethics (REK 2015/1462). All study participants gave informed written consent.

Among women free of any CVD events at baseline, there were 2 093 incident cases of CVD (13.5%), 670 AF (4.3%), 772 AMI (5.0%), 1 042 ischemic stroke (6.7%), 192 hemorrhagic stroke (1.2%), and 349 HF (2.3%). Among men free of any CVD events at baseline, there were 835 incident cases of CVD (11.3%), 248 AF (4.8%), 400 AMI (5.4%), 347 ischemic stroke (4.7%), 72 hemorrhagic stroke (1.0%), and 115 HF (1.6%).

We found no evidence for association between 1 unit decrease in distal forearm BMD T-score and CVD, AF, AMI, ischemic stroke, hemorrhagic stroke, and HF in men and women (Table 2). Among women the HRs (95% CI) were 1.01 (0.98–1.04) for CVD, 0.99 (0.94–1.05) for AF, 0.99 (0.94–1.04) for AMI, 1.03 (0.98–1.07) for ischemic stroke, 1.05 (0.95–1.16) for hemorrhagic stroke, and 1.02 (0.94–1.10) for HF (Model 2, Table 2). Similar results were observed in Model 3 (Table 2). Among men the HRs were 0.99 (0.94–1.03) for CVD, 0.95 (0.88–1.02) for AF, 0.96 (0.90–1.03) for AMI, 1.03 (0.96–1.11) for ischemic stroke, 1.12 (0.96–1.32) for hemorrhagic stroke, and 0.97 (0.84–1.11) for HF (Model 2, Table 2).