The Health Improvement and Prevention Study (HIPS) was a stratified cluster randomized controlled trial conducted in general practices in New South Wales, Australia (Australian New Zealand Clinical Trials Registry, ACTRN12607000423415). The methodology and outcomes of the trial have been described previously
The trial involved 30 practices who used electronic medical records and had expressed interest in the trial. Sixteen practices were randomly allocated to the study intervention group and 14 to the study control group. Patients were eligible to participate in the study if they had attended the practice in the preceding 12 months and were either aged 40–55 years with a diagnosis of hypertension and/or hyperlipidaemia or were aged 56–64 years with or without recorded risk factors. Exclusion criteria included diabetes, cardiovascular disease, current severe illness, inability to speak adequate English or understand the consent form. The majority of general practice patients aged over 64 years have already developed chronic disease so were not included. Each practice randomly selected up to 160 patients from practice records and invited them by mail to have fasting blood sugar and lipids levels; and attend their practice for a health check at baseline and 12 months.
The patients in the intervention practices attended for a structured health check visit. During the visit practice staff who had received prior training, assessed their vascular risk factors (blood pressure, lipids, fasting blood glucose, body mass index, waist circumference, smoking, nutrition, alcohol intake, and physical activity) and provided brief lifestyle advice and motivational interviewing. The brief intervention was modeled on the 5As framework (ask, assess, advise, assist and arrange)
. Patients were referred to the lifestyle modification program if they were found to be at high risk
. Sixty-three percent of the 301 eligible patients were referred to the program
. The lifestyle program was provided in the local area of each participating practice and coordinated by a program manager. It consisted of one individual visit with a dietitian or exercise physiologist for assessment and individual goal setting, followed by four, 1.5 hour, group sessions over 3 months; and a further two follow-up sessions at 6 and 9 months. The group sessions were adapted from the “Counterweight Program- CHANGE”
 and included education, physical activity and self management strategies (goal setting, self monitoring, developing practical skills and problem solving) aimed at promoting positive dietary and physical activity changes, reduction in alcohol intake, smoking cessation and weight loss.
Patients attending the control practices were also invited to attend the practice for a routine health check. At this they received usual general practice care for their risk factors by general practitioners who had not received the prior training.
Patients, who were blinded to the practice allocation, were mailed a questionnaire at baseline and 12 months and asked to complete it in private before attending their practice for the health check. The questionnaire, based on previous research, collected self-reported behaviours
 (baseline and 12 months), demographic characteristics
 (baseline) and the Kessler Psychological Distress Scale
 (baseline and 12 months).
Self-reported behaviours included:
Current smoking status
Serves of fruit and vegetables per day (diet risk, <7 serves per day)
Alcohol consumption (number of standard drinks on a typical day)
Physical activity level (included duration of vigorous and moderate physical activity) (score range 0–8, Inactivity, <4)
Demographic information included gender, age and markers of socioeconomic status (home ownership and employment status)
. Self-reported employment categories were full time employment, full time education, unemployed, unable to work, looking after family, retired or other. Self-reported home ownership categories were living in own home, living in rented accommodation, other living arrangements.
The Kessler Psychological Distress Scale (K10)
 is a ten item questionnaire measuring negative emotional states in the preceding four weeks. Responses are rated on a five point scale and summed to produce a score from 10 to 50. High scores (30–50) are strongly associated with a diagnosis of a psychiatric disorder. The instrument is used in clinical practice and is sensitive to changes resulting from interventions
Body Mass Index (BMI) was calculated using height and weight from audits of medical records at baseline and 12 months. A participant was considered “at risk” if their BMI ≥ 25 kg/m2.
A statistician who was not involved in the data collection used computer generated random numbers to randomly allocate practices to intervention and control groups, stratified by location. Data collection officers were blinded to the allocation of practices.
Sample size calculation
A priori sample size calculation for the secondary analysis on the K10 score confirmed that 350 patients in each group would have 80% power and 5% significance level to detect an effect size of 0.28 between intervention and control groups at 12 months assuming 15% lost to follow-up after adjustment for clustering (ICC = 0.025)
The effect size was based on expert opinion. Larger effect sizes have been demonstrated in groups with a diagnosis of psychiatric disorder undergoing specific therapy but this population did not represent the participants in our study
We conducted bivariate analyses using ANCOVA or Chi-square (for categorical data) to test for differences between intervention and control groups at 12 months, using SPSS, version 15 (SPSS Inc)
. An intention to treat analysis was conducted including those lost to follow up (dropouts) if data were available and patients had not requested withdrawal of their data. Missing data was not included and no data was imputed. Multilevel analysis allows for incomplete outcome data as long as a missing at random process can be assumed
. The characteristics of the dropouts were compared with other participants at baseline. Multilevel multivariable analysis using MLwiN (statistical software for multilevel models)
 was conducted for the sub group of participants who had complete K10 data at 12 months using list-wise deletion of missing values. Patients (level 1) were clustered within general practices (level 2). Initially, we fitted a baseline variance component model (no independent variables) for K10 at 12 months followed by the main model. The main multilevel model added covariates and included patients’ gender, age, home ownership status and employment.
Change in diet, BMI, physical activity and alcohol were examined as potential mediators of the lifestyle intervention effect on a change in psychological distress after adjustment for age, home ownership status and employment. Change variables were computed using the formula: 12 month score – baseline score. Smoking status was categorical and a change score was not computed. Therefore it was not included in the subsequent analysis.
To assess mediating effects, a product-of-coefficient test, appropriate for cluster randomized controlled trials, was used
. This test consists of (1) estimating the effect of the intervention on changes in the behavioural mediator (α coefficient) by regressing changes in the mediator onto the intervention; (2) estimating the independent effect of changes in the potential mediator on changes in psychological distress (β coefficient) by regressing changes in psychological distress onto the intervention and changes in the mediator; (3) computing the product of the two coefficients αβ, representing the mediated effect; (4) dividing αβ by its standard error. The estimates were obtained using multilevel linear regression models (ICC for the multilevel mediator model was 0.016), accounting for age, home ownership, employment and within-practice cluster effects. The ratio of the total mediated effect to total intervention effect was also estimated. The standard error of the mediated effect was computed using the multivariate delta method
. All mediation analyses were conducted using MLwiN and Microsoft Excel.
Ethics approval for the study was obtained from the University of New South Wales (UNSW) Human Research and Ethics Committee. All participants gave fully informed written consent.