Study population

The present study is part of the on-going Tromsø Study (http://www.tromsostudy.com). Details of study design have been described elsewhere [24]. Briefly, the study is a longitudinal, population-based investigation in Tromsø (Norway), a city of nearly 70000 inhabitants. The study comprises six repeated surveys from its start in 1974 (Tromsø I) through 2007/08 (Tromsø VI) [24]. Only men were invited to the first survey, but from Tromsø II 1979/80 both women and men have been included. The participation rate have ranged from 65% to 77% [24]. Each survey has been conducted in two phases, with the most basic examination in phase 1 (height, weight, blood pressure, blood samples, and questionnaires) and more extensive examinations for a random sub-sample of the cohort in phase 2, depending on available financial resources. The Tromsø Study, including this study, has been approved by the Norwegian Data Inspectorate and the Regional Committees for Medical and Health Research Ethics (Rec North). All participants signed a written informed consent.

The present study is based on data from Tromsø VI, details described elsewhere [23]. In short, 9625 men and 10137 women were invited, and 6054 men (62.9%) and 6930 women (68.4%) attended phase 1 of the survey. Among those, a total of 11484 subjects were invited for phase 2, and 3141 men (61.5%) and 4166 women (65.3%) attended. Persons with valid BMD measuremenst from Tromsø V in 2001/2002 were invited for BMD measurements of the hip, i.e. a dual femur scan, and altogether 3854 persons attended. Among these, a VFA, also called lateral vertebral assessment (LVA), was performed in a randomly selected group of 2894 persons. Seven blurred VFA scans had to be excluded, leaving 2887 persons, 1206 men and 1681 women, with clearly measurable VFA scans and total hip measurements. Among the 2887 persons with VFA scans, we obtained the following numbers according to each vertebral level: T4 = 2350, T5 = 2743, T6 = 2845, T7 = 2863, T8 = 2875, T9 = 2878, T10 = 2885, T11-L3 = 2887, L4 = 2848.

Ascertainment of vertebral fracture

Vertebral fracture was ascertained by the VFA of the GE Lunar Prodigy, Lunar Corporation, Madison, WI, USA, version 12.20. Vertebral morphometry is a quantitative method developed for identification of osteoporotic vertebral fractures based on the measurement of vertebral heights, identifying the anterior, middle, and posterior heights of each vertebra. Depending on their relative relations according to a given reference, the software identifies three types of fractures: wedge, biconcave, and compression, according to three degrees of severity, ranging from mild through moderate to severe [25]. Although some authors suggest that spinal radiograph is the gold standard for the diagnosis of vertebral fractures [13, 26], the morphometric method is recognized as being easy, precise and using low radiation exposure [25, 27, 28], with high precision in measuring moderate and severe deformities [13]. In our dataset, only 1% of the deformities were identified as being mild, the majority were either moderate or severe [23]. All our scans were taken according to the standard set by GE Lunar Prodigy, and specially trained technicians did the scanning according to the standardized protocol. One of the technicians (the first author) performed the quality assessment of the total material afterwards. For precision analysis of the VFA, a random sample of 50 participants was re-analyzed by the same technician. The mean intra-class correlation coefficient was 0.84 for average height of the vertebrae [23].

Bone mineral density

BMD expressed as g/cm² was measured at the total hip and femoral neck by DXA, using the same densitometer as for the VFA (GE Lunar Prodigy). Daily phantom measurements were performed throughout the survey. Three technicians did the scanning according to a standardized protocol, and one of them performed the quality assessment of the total material afterwards. The short term in vivo precision error was 1.2 and 1.7% for total hip and femoral neck measurements, respectively [29]. For the main analyses of this study, we have included BMD measurement of the total hip, where 2738 valid measurements were available (either left or right hip). Based on the Lunar reference, 102 persons (3.5%) had a T-score ≤ −2.5 (osteoporosis) and 789 persons (27%) had a T score between −2.5 and −1.0 (osteopenia).

Questionnaire

Data on lifestyle variables were collected through questionnaires in both phases of the study. The questionnaires obtained information, among other, on smoking habits, physical activity, self-perceived health, and education. Smoking status was classified into three categories, namely: present, former, and never. These were further grouped into two categories: former and never smokers were categorized as “not smoking” and smokers as “smoking”. The question on leisure time physical activity level during the last year had four alternatives from sedentary (mostly tranquil activities), moderate (lightly active at least four hours a week), active (vigorously active at least four hours a week) and highly active (exercising regularly several times a week). The sedentary and moderate active were categorized into one “low active group”, and the active and highly active into one “high active” group. Five levels of self-perceived health (very good, good, neither good nor bad, bad, and very bad) were categorized into two, good (very good and good) and poor. Five levels of education were categorized into three levels: 1) primary school only (i.e. seven years), 2) up to four years more than primary school, and 3) more than four years after primary school.

Data analysis

Each individual was classified as having a vertebral fracture if there was a presence of at least one fracture as described in the “Ascertainment of vertebral fracture” section. The association between sex and vertebral fracture risk was assessed by Chi-square statistics. Univariate analyses (Chi-square statistics or Independent sample T-tests) were used to examine the association between baseline characteristics and vertebral fracture risk in women and men. Logistic regression was used to assess the association between each risk factors and vertebral fracture risk in women and men separately, adjusting for age. Odds ratio (OR) of fracture (and 95% confidence interval (CI)) was estimated per standard deviation (SD) of continuous risk factors (e.g., BMD, body mass index (BMI)). Multivariable logistic regression analysis was used to assess the association between each significant or combined risk factors and vertebral fracture risk. Testing for interaction was done by including the product of sex and BMD and sex and age in the model. The final and most optimal model was found using backward selection procedures. In order to assess the impact of risk factors on vertebral fracture, we estimated the population attributable risk fraction (PAR) for combined risk factors in the “final” multivariable model, for women and men separately. In this study context, PAR represents the proportion of vertebral fractures that can be attributed to a risk factor, if there is a causal relationship between the risk factor and fracture [30]. However, because vertebral fracture is associated with multiple risk factors, stratified models were used to estimate the fracture risk in four categories of exposure and the heuristic approach described by Hanley, J.A. [31] was used to calculate the PAR. The formula used to calculate the attributable risk (AR) (actually called attributable fraction) was the following:$AR=\frac{\left(R{R}_i-1\right)\times Pc{t}_i}{1+{\displaystyle {\sum }_i^n\left(R{R}_i-1\right)\times Pc{t}_i}}$(where (n) is the total number of the categories, and (i) takes value of each category risk (RR) and percentage (Pct)). All statistical analyses were performed with the SPSS statistical package (versions 19).