We recruited participants from a community-based research database that was established through random sampling from the Victorian Electoral Roll between April 2002 and August 2003. Detailed information about recruitment of women from this database to a large cross-sectional study of androgens in women has been previously reported . Participants in this study of androgens who agreed to be contacted regarding further research were invited to participate in the current study. No incentives were offered to the subjects. Of the 754 women who were re-contacted, 542 (71.9%) agreed to be involved.
Demographic and prevalence data
Data were obtained for age, weight and height, smoking, alcohol consumption, and menopause, relationship and employment status. Body mass index (BMI; kg/m2) was calculated from height and weight measurements provided by the participants. Women were classified as 'partnered' if they were married, in a de-facto relationship or single with a partner. Employment was defined as work outside the home and included full-time, part-time, casual and shift work. Women were categorized into 3 groups, premenopausal, perimenopausal and postmenopausal, using a hierarchy of information including history of a bilateral oophorectomy, age (≥ 60 years), current use of hormonal contraception or systemic hormone therapy, history of a hysterectomy, bleeding pattern and the presence of vasomotor symptoms . Outliers that were identified in the demographic data were cross-checked with data from the previous cross-sectional study of androgens to ensure they represented meaningful values.
Low back pain prevalence
To obtain data on low back pain prevalence, participants were asked if they had experienced low back pain at the following time points/periods; now, past 6 months, past 12 months or ever. A body chart was provided, with the region from the border of the rib cage to the gluteal folds shaded, to define the region of pain.
Pain intensity and disability data
To obtain data on low back pain intensity and disability the Chronic Pain Grade Questionnaire (CPG) was administered. The CPG has been shown to be a reliable and valid instrument for use in population surveys of chronic pain, including low back pain [11, 12]. The CPG includes 7 questions from which a pain intensity score (0–100) and disability score (0–6) are calculated.
Based on these scores, participants are classified into one of the CPG grades:
Grade 0 – no pain or disability
Grade 1 – low pain intensity (and low disability)
Grade 2 – high pain intensity (and low disability)
Grade 3 – high disability, moderately limiting (regardless of pain intensity)
Grade 4 – high disability, severely limiting (regardless of pain intensity)
To investigate the relationship between pain intensity and beliefs about back pain, participants were classified into 3 groups based on their pain intensity score; no pain (= 0), low pain intensity (<50) and high pain intensity (≥ 50). Subjects were also categorized into 3 groups based on their disability score; no disability (= 0), low disability (<3) and high disability (≥ 3) and these groups were used to examine the relationship between disability and beliefs.
Beliefs about back pain
The Back Beliefs Questionnaire (BBQ) was used to examine individual's beliefs about back pain and its consequences, regardless of whether back pain had been previously experienced. The questionnaire has been reported to have good internal consistency (Cronbach: 0.7) and test-retest reliability (ICC: 0.87) . The questionnaire consists of 14 statements to which the respondent indicates their level of agreement on a 5 point scale. A score of 1 indicates complete disagreement and a score of 5 complete agreement. As 5 of the 14 statements are distractors, the scores of the 9 remaining statements are reversed and then summed to provide a total score ranging from 9 to 45. A lower score indicates the respondent has more negative beliefs about back pain. Two population-based studies undertaken in Australia  and Canada  have previously used the BBQ, and prior to the implementation of a public health campaign, reported mean back belief scores of 26.5 (95% CI: 26.1, 26.8) and 26.4 (SD: 6.4) respectively.
A study package including the participant information sheet, consent form and study questionnaire was mailed to participants in July 2006. Participants that did not return their questionnaire were sent a letter to ask whether they have received the questionnaire and if they needed assistance with its completion. If participants did not respond to this letter, phone contacted was attempted on three occasions. All procedures were conducted in accordance with the Declaration of Helsinki and approved by the Monash University Human Research Ethics Committee. Written informed consent was obtained from all participants.
Data relating to the characteristics of participants, their level of pain and disability, and their beliefs about back pain were tabulated. Univariable and multivariable analyses for pain and disability were performed using ordinal regression , as the outcome variables pain and disability were categorised into ascending levels based on the CPG methodology. However, the assumption of proportional odds on which this method was based was violated for the pain outcome when two independent variables, menopause status and back beliefs, were included in the multivariable model. Thus, we applied the generalized partial proportional odds model described by Williams (2006) to fit non-parallel regression lines for the three pain categories. A parallel regression line was used for other variables and hence they share the same regression coefficients. Given the proportional odds assumption was held for the disability outcome, only the parallel regression line was fitted. Multivariable analyses involved adjustment for the following potential confounders; age, BMI, smoking, alcohol consumption, and menopause, relationship and employment status. The univariable analysis was performed using the ordinary ordinal regression method . The statistical analyses were conducted using Stata software version 9.2 . A statistical test was considered to be significant if the associated p-value was less than 0.05.