Participants
This study utilised data from men and women participating in the Geelong Osteoporosis Study (GOS), a large, ongoing, population-based study conducted in the Barwon Statistical Division in south-eastern Australia [21]. Initially, 1494 women (aged 20-94 years, participation 77.1%) and 1,540 men (aged 20-93 years, participation 67%) were randomly selected from the Commonwealth of Australia electoral rolls (where voting is compulsory), between 1994 and 1997, and 2001 and 2006, respectively. Participants have been invited for two to five yearly assessments. An additional sample of 246 women aged 20-29 years was recruited (participation 70.9%) between 2004 and 2008, allowing for continuing investigation of the full adult age range.
For this cross-sectional analysis, data collected at the 10-year follow-up for women and 5-year follow-up for men was utilised. Of the1127 women who participated in the 10-year follow-up, participants for whom bone data were not available were excluded, resulting in a final sample of 1070, aged 20-94 years. Of the 978 men who participated in their 5-year follow-up, similarly, participants for whom bone data were not available were excluded, resulting in a final sample of 926, aged 24-98 years.
Measurement of the outcome variables
BMD (g/cm2) was measured at the spine (PA projection, L2-4) and total hip for men (Lunar Prodigy, GE, Madison, WI, USA) and women (DPX-L, GE, Madison, WI) [21]. Trained technicians carried out all examinations and performed daily calibrations of the densitometers with an equipment-specific phantom.
At the same time as BMD was measured, bone quality was determined by calcaneus QUS (Achilles Insight Ultrasonometer, GE Lunar, Madison, WI, USA) of the left heel, yielding the following parameters: broadband ultrasound attenuation [BUA (dB/MHz)], reflecting microarchitecture and bone density, speed of sound [SOS (m/s)], reflecting elasticity and bone density and stiffness index [SI (%)], a calculated clinical index [22]. SOS and BUA measure the speed and frequency-dependent ultrasound attenuation of ultrasound signals passing through soft tissue and trabecular bone [23]. SI is a combined parameter, calculated from these primary measures [24].
Measurement of exposure variables
The following data were collected concurrently with the bone health assessments:
Medication use
Current medication use was determined via self-report. Participants were requested to bring a medication list or containers to their appointment to ensure accurate reporting. Exposure to anticonvulsants, and other medications known to affect bone, such as oral glucocorticoids, bisphosphonates, and thyroid medication were coded based on the Australian index of medications guidelines.
Questionnaire data
Information on daily alcohol use (g/day) and calcium intake (mg/day) was determined using a validated food frequency questionnaire [25]. Participants were classified as smokers if they reported current use at the time of assessment and habitual physical activity level was classified as active if vigorous or light exercise was performed regularly; otherwise participants were classified as sedentary.
Other markers of bone health
Osteoporosis status was identified as T-score < -2.5 at either the spine or total hip [26, 27]. Information regarding previous adult fracture was ascertained via radiological reports from medical imaging centres servicing the region. This method of fracture ascertainment has been previously validated [28].
Anthropometric measures
Body mass index (BMI) was calculated (kg/m2) from height, measured to the nearest 0.1 cm, and body weight measured to the nearest 0.1 kg.
Socioeconomic status
Area-based socio-economic status (SES) was established by matching the x-y coordinates of participants’ residential addresses to the Australian Bureau of Statistics’ Index of Relative Socioeconomic Advantage and Disadvantage (IRSAD) data to determine a score for each participant. The IRSAD consists of information regarding income and skill level. IRSAD scores for this study were determined according to cut points of the Barwon Statistical Division and categorised into quintiles whereby SES quintile 1 was considered most disadvantaged and SES quintile 5 was most advantaged.
Statistical analyses
Minitab (Version 18; Minitab, State College Pa) was used to perform statistical analyses. Differences between anticonvulsant users and non-users were detected using t-tests for continuous normally distributed variables, Kruskal-Wallis for non-parametric continuous variables and chi-square or Fisher’s exact test for discrete variables. Multiple regression was used to explore associations between anticonvulsant use and a) BMD and b) QUS measures (BUA, SOS and SI). Data for men and women were analysed using separate statistical models. Confounders including age, BMI, smoking status, SES, physical activity and medications known to affect bone were tested sequentially and retained in the final model when significant (p<0.05). Interactions between exposure variables were checked for effect modification in the final models.