A cluster-randomized trial of a telephone counselling intervention for physical activity and diet was conducted among adults from a low socio-economic community in Australia [17]. Data from 434 adult participants with type 2 diabetes or hypertension (mean age 58.2 [SD 11.8]; 61% female; mean BMI 31.1 [SD 6.8]) from a disadvantaged community in Queensland were used for cost-effectiveness modelling. Participants were recruited from ten primary care practices via electronic medical records searches for condition-eligible participants. Data were collected between February 2005 and November 2007 and analysed between January and October 2008.

Participants were randomised to Telephone Counselling (n = 228) or Usual Care (n = 206). Participants in the Telephone Counselling group received a 12-month intervention involving 18 telephone calls from trained counsellors. The intervention schedule was adapted from the work of King and colleagues to facilitate initiation and then maintenance of behaviour change [18], [19]. During the initiation phase, calls were delivered weekly for the first three weeks and then fortnightly up to four months. During the maintenance-enhancement phase, calls were made monthly between 4 and 12 months. Participants received an intervention workbook along with a pedometer, a self-monitoring form, and an exercise band. The workbook was adapted from the work of Demark-Wahnefried and colleagues [20]. To obtain approval from an ethics committee and to reduce attrition, the Usual Care group received more than would typically have been made available under existing practice conditions. Prior to the intervention, the Usual Care group were being managed by their general practitioner. They were consented into the trial and asked about their lifestyle behaviours over the course of the trial for the purpose of data collection. This involved three telephone interviews of approximately 45–60 minutes duration each. They were provided feedback on their behaviour after each assessment and sent off-the-shelf print materials about good health behaviour and a quarterly newsletter. The primary outcomes [17] show statistically-significant intervention effects (Telephone Counselling vs. Usual Care) from baseline to 12-months in: percentage calories from total fat (1.17% reduction); percentage energy from saturated fat (0.97% reduction); vegetable intake (increase 0.71 servings per week); fruit intake (increase 0.30 servings per week); and fiber (increase 2.23 grams per week). Increases in mean weekly physical activity at 12-months were reported by the Telephone Counselling (71 minutes) and Usual Care groups (84 minutes). Although to a lesser extent than the Telephone Counselling group, the Usual Care group experienced clinically minor but statistically significant improvements from baseline in intake of total fat, saturated fat and fruit.

All costs are reported in 2008 Australian dollars. The cost of delivering the Telephone Counselling intervention and the Usual Care alternative were measured during the primary trial. Single-use items such as workbooks and mail-outs were attributed to participants. The jointly used resources of telephone counsellor time, fixed overheads and the staff costs of managing the program were allocated to participants on the basis of the number and duration of calls. All costs incurred were allocated across successful contacts. The dollar costs of input factors were obtained by a review of local market prices. The participant use of health care services between baseline and 12 months were self-reported and collected by validated questionnaire [21]. These included: consultations with a general practitioner, psychiatrist/psychologist; visits to an emergency department; visits by a nurse, home health nurse or occupational therapists; the costs of hospital admissions based on the number of bed days used; and, visits to an outpatient department. Dollar valuations of these resources were obtained from the Commonwealth Government schedule of re-imbursements [22]. The SF-36 health survey was administered to all participants at baseline, 4 months and 12 months. The data were mapped onto the SF-6D using a validated algorithm [23]. The SF-6D provides a preference-based value of health outcomes derived from standard gamble questions. The SF-6D is appropriate for estimating quality adjusted life years (QALY) [24].

Diet and physical activity outcomes of the trial were assessed using telephone interviews at baseline, 4-months and 12-months. Validated instruments used in Australian population health surveys were employed [25]–[27]. Diet and physical activity were assessed relative to Australian guidelines: 150 minutes a week of accumulated moderate physical activity on five or more days per week [28]; at least five servings per day of vegetables; at least two daily servings of fruit; less than 30% of energy intake from total fat; less than 10% of energy intake from saturated fat and 30 grams or more of fibre per day [29].

Risk of death was based on the estimated future mortality risks of the subjects in the trial. Self-reported data were collected on age, sex, body mass index, cigarette smoking status, history of diabetes mellitus or cardiovascular events, hypertension and hypercholesterolemia. The latter two were converted to systolic blood pressure and total serum cholesterol by imputing the age and sex-specific average values [30]. Data from the 2003 Australian Burden of Disease and Injury study [31] were used to describe the population in terms of disease-specific and overall mortality rates and these were adjusted according to each individual's risk factors. The adjusted mortality rates were used to estimate life expectancies of participants in a Sullivan life table [32].

Cost-effectiveness was assessed using a state-transition Markov model, which predicted the progress of participants through five health states for ten years after recruitment. The structure of the model is shown in Figure 1. Individuals in a state of sub-optimal behaviour (SO) move to improved physical activity (PA) if they meet current Australian Physical Activity guidelines (more than 150 min/wk×5 days); or, move to improved diet (DIET) if they met at least three of the five current Australian recommended nutrition guidelines [29]. If both criteria are met then individuals move to improved physical activity and diet (PA&DIET). Participant movements through the health states are governed by the trial data on health behaviour and individual mortality risks. Risk of death increases with each cycle to reflect ageing and the modelling assumes – conservatively – that spending a cycle in a good health behaviour state does not reduce mortality risk. The model updates, and so individuals move, after one 4 month cycle and then at every 12 month time-point. Movements between all states are possible with the exception of the death state, which is absorbing [33]. The baseline to 4-month data informs the first transitions; the 4-month to 12-month data informs all subsequent transitions, until year 10. This introduces uncertainty, but allows costs and benefits arising in future time periods that were not observed within the timeframe of the trial, to be estimated. The observed cost and utility scores (SF-6D) of trial participants that occupied relevant health states for each model cycle were estimated from the trial data. They were then used to calculate the total cost and QALY outcomes for Telephone Counselling and Usual Care. Baseline values for costs and utilities from the whole sample were used to define an existing practice (Real Control) group. This group was assumed to make no changes to their health behaviours, costs or QALY outcomes over time. Costs and QALY outcomes accumulate as the model updates. To illustrate model output, we include the costs and utilities for the Telephone Counselling group at 12 months in Figure 1. The model is complete when all the costs and QALY outcomes for the intervention and comparators are simulated for a period of 10 years. All outcomes arising in future time periods are discounted at 3% [34]. A detailed description of the modelling method has been published [35].

The first comparison, which is based on the cluster randomised trial, is Telephone Counselling vs. Usual Care. Comparison groups in randomised trials are often inappropriate comparators for cost-effectiveness analyses and decision making [36]. Usual Care did not equate to an existing practice alternative that is relevant to decision makers. To compensate for this a Real Control group was defined who represent a decision to remain with existing practice and can be thought of as ‘never contacted’ about their diet and exercise behaviour. The second comparison was Telephone Counselling vs. an existing practice alternative (Real Control). As the Usual Care group in the intervention received a ‘brief intervention’, the third comparison was Usual Care vs. existing practice alternative (Real Control). To simplify all comparisons, 100 participants were simulated in each group. The simulation was informed by all of the data reported in the primary trial (n = 431); three participants were excluded due to death.

The model shown in Figure 1 was fitted using the WBDiff interface of the WinBUGS software package (MRC Biostatistics Unit, Cambridge University and Imperial College, School of Medicine at St Mary's, London, 1989). Beta distributions were fitted for the parameters that described transition probabilities and health -related utilities, and gamma distributions were fitted for parameters that described costs [37]. The change in total cost (ΔC) and QALY outcomes (ΔE) were estimated for all three comparisons. These were combined with the decision maker's willingness to pay for QALYs (γ) to calculate changes to monetary net benefits:

The probability an intervention was cost-effective vs. the relevant comparator was estimated by taking 2,000 random re-samples from the probability distributions for parameters; and then counting the number of times the monetary net benefit statistic was positive over the total number of re-samples [38]. The results of this process are plotted in the form of a cost-effectiveness acceptability curve that shows the probability a decision is cost-effective, given uncertainties in model parameters, for a range of the decision maker's willingness to pay for QALYs (γ). Fenwick et al. provides information about the use and interpretation of cost-effectiveness acceptability curves [38], [39].

This modelling study has strengths and weaknesses. Predicting the health behaviours of participants for ten years after recruitment has an advantage over trial-based evaluations that only capture events that happen during the period of data collection [50]. Information about the longer-term outcomes that follow a decision to implement the Telephone Counselling program is useful for decision makers and health planners. There are few studies with long-term follow up for health behaviour programs [51], and modelling future outcomes based on observed data is one solution [50]. This advantage is also the source of the major caveat in relation to our findings. Data that describe health behaviour for physical activity and diet for the time between 4 and 12 months were used to predict health-behaviour states for a further nine years, and this may not be accurate. There might be a decline in effectiveness in later years as compared to the changes observed between 4 and 12 months, as the novelty of the intervention wears off. If this is true, then the results are biased in favour of Telephone Counselling. There might, however, be a slower rate of attrition if those who adhered to the relevant health behaviour up to month 12 are more likely to adhere in future time periods. Under this case, the findings on the benefits of Telephone Counselling are conservative. Undertaking longer-term follow-up is the only way to remove this uncertainty; however, this would be an expensive proposition. The modelling assumed the program continued for ten years, and that participants continued to receive scheduled contacts from counsellors and feedback. Thus, the costs of implementing the maintenance part of the intervention for the Telephone Counselling group (month-4 to month-12) were included alongside the costs of Usual Care for 10 years. Because a ten-year trial is unlikely to be funded, a modelling approach to predicting future outcomes is parsimonious.

Risk of death was not reduced in the model when participants adhered to the good health behaviour. There is evidence that improving physical activity and diet reduce risk of premature death [2], [4]. The effect of using an evidence-based reduction in death risk, for those in a state of good health behaviour was tested (results not shown); not surprisingly, the cost-effectiveness of Telephone Counselling increased. The assumption of equal mortality risks for Telephone Counselling and Real Control is therefore conservative, and may underestimate the value of the Telephone Counselling program. Another conservative assumption is that Real Control participants did not deteriorate and cause higher health-care costs, and have worse health outcomes than were observed at baseline. It is likely those with hypertension and diabetes, at the ages observed among the sample, who fail to improve their diet and physical activity, will endure substantial morbidity and health-care costs over a ten-year period. That the sample was recruited from one socio-economically disadvantaged community, and 65% had three or more co-morbidities, suggests that the generalisation of results should proceed with caution. Finally, the model structure was based on health behaviour rather than disease states, because the study participants had a range of co-morbid chronic conditions. This flexible framework might be suitable for other interventions that aim to change health behaviour among a group of participants with heterogeneous health conditions. Choosing Telephone Counselling over existing practice (Real Control) is supported by evidence for efficacy and for cost-effectiveness; with the latter conditional on affordability and opportunity cost [52], [53]. Assumptions about the positive effects being achieved and maintained in broad-reach public health programs can be supported.