Study population

It is expected that MobileMums would be offered to all women with children under 1 year old in Queensland, Australia, regardless of their current level of physical activity. With 61 020 women giving birth in Queensland in 2010 and with 413 fetal deaths,14 the number of women eligible for the intervention in 2011 was 60 607. We expect around 60% of women who were offered the intervention would participate. This is based on the randomised control trial conducted in 2011,15 ,16 where of the 511 women assessed for eligibility 306 started the baseline assessment. This gives 36 364 women in Queensland who would be eligible and willing to participate in the MobileMums intervention in 2011, and this is the baseline cohort size considered for this study. This participation estimate of 60% is likely conservative, as the programme would not include the time-consuming assessments that were undertaken purely for research purposes. Given the uncertainty around this estimate, we consider the effects of reducing this cohort size by 50% to 18 182 women, and increasing it by 50% to 54 546 women.

Modelling health outcomes and costs

A state-based Markov model provides the framework for this analysis and is used to estimate the costs and consequences associated with MobileMums and usual care. The development of the model has been informed by the effectiveness results from a 9-month two-arm community-based randomised controlled trial undertaken in 2011.15 A total of 263 women from around Caboolture, Queensland, received usual care (n=130) or the MobileMums intervention (n=133).16 Data were collected prior to the intervention being received (time 1—T1: 0 months), after the 12-week MobileMums programme was completed (T2: 3 months) and again after a further 6 month no-contact maintenance period (T3: 9 months). Owing to an administrative error the trial was registered retrospectively with the Australian Clinical Trials Registry (ACTRN12611000481976) and 26 of the trial participants were already receiving MobileMums or usual care by the time of registration. However, none of these participants had passed T2 when the trial was registered.

The main efficacy findings from the trial have been reported in detail by Fjeldsoe et al.16 Briefly, while the intervention had a large and statistically significant beneficial effect on activity levels between T1 and T2, there was no statistically significant effect at T3, although the estimated increase in activity remained positive. These results suggest that MobileMums can only be expected to have an effect on activity levels in the short-term. Under the assumption that only long-term changes in activity levels affect the risk of an individual developing future chronic health conditions, the time horizon of the model used is 2 years.

There are just two states in the model with participants either ‘physically inactive’ or ‘physically active’, and an individual is required to be undertaking 30 min of moderate-intensity to vigorous-intensity physical activity on at least 5 days a week to be classified as active. An effective physical inactivity intervention increases the likelihood that inactive individuals become active (tpImprove) and/or reduces the likelihood that active individuals become inactive (tpRegress). Individuals move between states using monthly cycles, and spending a month as active or inactive has a cost and health outcome associated with it (described below). An outline of the model is shown in figure 1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Outline of the Markov model used to estimate the costs and effects of MobileMums and usual care.

Health effects

To estimate the value for money of MobileMums, health effects are expressed in terms of quality-adjusted life-years (QALYs). Given the design of the model used, MobileMums can only affect health-related quality-of life, with no mortality effects. The health-related quality-of-life associated with being physically active or inactive was estimated from participants’ responses to SF-12 questionnaires at T1, T2 and T3. Mean imputation was used for missing questionnaire data at each time period (1% of participants at T1, 13% at T2, and 32% at T3). Two errors were made in the printing of the SF-12 questionnaires. First, at T1 one question from the SF-12 was omitted in error, and so scores were randomly generated for this dimension. Second, one of the questions offered one too many potential responses at all time periods, and so those who selected this superfluous response were evenly split and moved into either the next best or next worst choice.

Questionnaire responses were transformed into the EQ-5D, a standardised measure of health outcomes, using an algorithm provided by Gray et al17 which provides utility scores close to group means, especially for individuals not in poor health. This approach generates health-related quality-of-life scores associated with spending a year as physically active or inactive which could range between 0 (equivalent to death) and 1 (equivalent to perfect health). Monthly scores were simply one-twelfth of this. QALYs and costs in the second year were discounted at 5% following the relevant guidelines.18 ,19

Costing perspective

This study is intended to inform decision-making regarding resource allocation across the health system in Queensland. Consequently, a health system perspective is taken, with only the costs borne by the health system included.20 While costs falling outside of the health system, such as the cost to participants of purchasing goods or services related to undertaking exercise, may be of interest, they are not are not of direct relevance given the perspective taken here and so have been excluded. All costs reported have been inflated to 2014 Australian dollars (AUD) and any costs accruing in the second year of the model have been discounted at 5% in line with guidelines for submission to the Medical Services Advisory Committee18 and the Pharmaceutical Benefits Advisory Committee19 in Australia.

The estimated cost of providing MobileMums across Queensland is based on the costs of delivering the intervention in the randomised controlled trial.16 To extrapolate these costs, assumptions have been required concerning number of behavioural counsellors and programme coordinators required for widespread dissemination. It is assumed that counsellors could be assigned to 30 participants per week, while coordinators could cover five counsellors and their participants per week. Counsellors and coordinators are assumed to be health practitioners with, on average, 2 years in their current role and, in terms of Queensland Health's salary scale,21 paid at a HP3 (6092 AUD per month) and a HP4 (8150 AUD per month) level, respectively. The costs of developing the computer programme to send text-messages, sending the text-messages and providing other programme materials are assumed to be the same as in the trial.

In addition to the costs of delivering the intervention, the costs relating to participants healthcare use have also been incorporated. If the intervention reduces future healthcare use then the cost saving associated will counterbalance the cost of providing MobileMums. Participants’ reported their use of healthcare services at T1, T2 and T3 and the average use of those who were physically active and inactive were estimated. As with the SF-12, mean imputation was used for missing data (0% of participants at T1, 12% at T2, and 31% at T3). The associated costs were estimated using the Medicare Benefits Schedule for July 201122 and Australian hospital statistics.23

Expected effects

The purpose of this evaluation is to estimate the expected value for money of the MobileMums intervention, which is indicated by the incremental cost-effectiveness ratio (ICER) for MobileMums. This ratio is given by the expected (mean) change in costs associated with the intervention divided by the expected change in QALYs.24 This ratio can then be compared against a cost-effectiveness threshold. The threshold used is 64 000 AUD which is based on the estimate by Shiroiwa et al25 of the willingness-to-pay for an additional QALY in Australia. If the cost-effectiveness ratio for MobileMums falls below 64 000 AUD then the intervention can be expected to be ‘cost-effective’.

Uncertainty

Parameter uncertainty was quantified using Monte Carlo simulations, with the model evaluated 10 000 times, with each simulation involving random draws from each parameter distribution. These distributions are based on the trial data, with transition probabilities and QALYs given beta distributions, while healthcare utilisation and its associated costs assigned gamma distributions and uniform distributions, respectively. This produces 10 000 pairs of incremental costs and effects, and these are presented on a cost-effectiveness plane along with the expected costs and effects and the cost-effectiveness threshold. The probability that MobileMums is cost-effective is given by the proportion of pairs of incremental costs and benefits at which the intervention would be considered cost-effective. The percentage of pairs where the change in QALYs is positive and the change in costs is negative is equal to the probability that MobileMums is cost-saving. It is also possible to estimate credible intervals around the expected change in costs and QALYs by taking percentiles of the costs and QALYs produced in the Monte Carlo analysis.26

Uncertainty also exists surrounding the modelling assumptions. In particular, three areas stand out for particularly onerous assumptions: transition probabilities after 9 months (T3), the number of programme counsellors and coordinators required, and the number of women who would be eligible and willing to participate in the trial. The assumptions used for these areas are the subject of scenario analyses. First, the model is reassessed under the assumption that after T3 all programme activity effects are mitigated entirely, and then again under the assumption that the estimated treatment effect observed at T3 is maintained for a further 15 months, at which point the treatment effect is entirely mitigated. Second, the number of counsellors and coordinators required is increased by 50% and reduced by 50%. And lastly, increasing the cohort size by 50% and reducing it by 50% is considered.