After data entry, data was manually and statistically checked as a part of the data-cleaning process. Initially, descriptive statistics were calculated for all variables. To identify patterns of health behaviours in the study population, the Prudence Score was stratified by demographic characteristics of gender, age, education, marital and employment status. A bivariate analysis using the Chi2 test was performed to assess differences between meeting health behaviour recomm and the demographic variables of gender, education, work status, marital status, and area of residence. A two-sample t-test was used to assess the differences in the mean Prudence Score for gender, age, education, employment status, marital status and SEIFA.
The Prudence Score was further divided into 3 categories: low (0 to 5), medium (6 & 7) and high (8 to 10) as the primary outcome variable in the multivariate analysis. The multivariate logistic model was used to further assess the impact of each demographic factor on the combined score as well as determine characteristics of individuals at highest risk of having unhealthy lifestyles. Results of this model are reported as relative risk ratios. Records for missing values for any of the prudent variables were deleted from the total Prudence Score calculations. The statistical package STATA version 11 (Stata Corporation, 2008) was used for the analysis.
Of the 30 GPs approached, 25 showed initial interest. Four GPs subsequently withdrew because of leave planned during the recruitment period or inadequate computer systems to identify potentially eligible patients. Practice staff identified a total of 8,281 (32% men) potentially eligible patients for the remaining 21 GPs, of whom 4,678 completed and returned the questionnaire. As such, the participation rate at baseline was 59.9%, after notified deaths and returns to sender were omitted (5.2%; n = 412). Figure presents a study flow chart.
Baseline characteristics of the sample
Baseline socio-demographic characteristics for intervention and control groups are presented in Table . The average age of participating patients was 47 years and the majority were female (68.7%), married or living as married (68.8%), and with a diploma or university degree (56.6%).
Socio-demographic and health behaviour characteristics at baseline [mean(± SE) and number(percentages)]
Distribution of health behaviours
Table summarizes the distribution of the dietary and lifestyle behaviours for all the participants. Most (86%) were non-smokers whereas only 12.4% consumed 7 or more serves of vegetables and fruits daily. These ten behaviours are further stratified by gender and age (Tables not included). Women more soften reported adhering to several behaviours than men: eating 4 or less serves of meat per week (74.6% versus 57.2%, p < 0.05), drinking low fat milk (72.7% versus 61.8%, p < 0.05) and drinking alcohol within recommended limits (74.0% versus 56.6%, p < 0.05). For 46.4% of women BMI was within the recommended range, as compared to 29.6% of men. Older patients reported significant higher adherence to recommended behaviours, except for BMI where older patients had lower adherence. No age specific differences were noted for physical activity and salt intake. A significantly lower percentage of married patients had BMI within normal range when compared to patients who were single, widowed or divorced (40.9% versus 50.5%, p < 0.05).
Combinations of health behaviours
The data were analyzed to assess the proportion of participants that adhered to certain combinations of multiple health behaviours. Only 30% of patients reported they were non-smokers and adhered to physical activity and alcohol guidelines (see Table ). This proportion dropped to 5.1% when adding a fourth behaviour, adherence to recommended daily F&V intake. Only 2.8% of the study population adhered to all four behaviours and also had normal body weight (18.49 to 24.99 kg/m2).
Proportion of the general practice population adhering to recommendations for combined health behaviours
The mean Prudence Score for the study sample was 5.80 (95% CI 5.75-5.85), with scores approximating a 'normal' distribution. Less than 1% reported a Prudence score of '0', similarly only 2.8% adhered to five important recommendations: BMI within normal range, sufficient physical activity, within limits alcohol intake, being a non-smoker and sufficient F&V intake. Women had a significantly higher age-standardized mean Prudence Score than men (5.98 versus 5.41, t = 10.57; df = 4065; p < 0.001). Table shows the mean Prudence Score was lower in men than in women for all age-groups, but tended to increase with age in both genders. Patients with tertiary education had higher scores than those with high school education (5.98 versus 5.55; t = 8.02, df = 4050, p < 0.001). Regardless of educational background, marital status or employment status, women reported following a healthier lifestyle than men (Table ).
Distribution of mean Prudence Score by socio-demographic variables and effect of gender
Table displays the results of a multinomial regression model examining associations between socio-demographic factors and 3 categories of the Prudence Score. After simultaneous adjustment for socioeconomic status and all other factors in the model there was an increased risk of unhealthy diet and lifestyle for males (Relative Risk Ratio (RRR) = 3.03; 95% CI = [2.42-3.79]), younger age (RRR = 3.76; [2.83-5.01]) and lower educational attainment (RRR = 2.82; [2.11-3.76]). Of these factors, age between 18 to 39 years and educational attainment below high school were the strongest predictors of unhealthier lifestyle behaviours. As the interaction terms were significant for age and gender (χ2 = 180.4, df = 10, p < 0.05) as well as for education and employment status (χ2 = 83.0, df = 14, p < 0.05), the main effects model was adjusted for these potential confounders. Following the re-analysis marital status and employment status were no longer significantly associated with Prudence Score.
Multivariate associations between Prudence Score and socio-demographic variables