All simulations were based on a cohort the size of the US population aged 25 to 64 years of age in 2004. At the start of the simulation, we assumed that all cohort members were healthy, which we defined as the absence of 5 modeled diseases: type 2 diabetes, coronary heart disease, ischemic stroke, breast cancer, and colorectal cancer. This approach is conservative, because it considers only the preventive benefits of activity. In practice, community-level interventions reach adults with chronic diseases such as diabetes and heart disease. There are well-documented therapeutic effects of PA in many chronic diseases,¹³ and ignoring such benefits reduces the estimated cost-effectiveness estimates. The cohort, the size of the adult US population in 2004, was stratified by age, sex, and level of physical activity, using US Census Bureau and 2003 BRFSS data.^14,15 At the start of the simulation, all cohort members were assumed healthy.

The model then simulated how physical activity levels changed over the course of 20 years, and simulated the effects of these changes on incidence of the 5 diseases, quality of life, and mortality. The MOVE model incorporated data on how interventions moved cohort members across physical activity levels, and also incorporated data that specified how PA levels changed over time, in the absence of intervention (the “natural history model”).

In the first scenario, all cohort members, aged 25 to 64 years, were simulated to receive the exemplar interventions for 1 year, immediately upon entry into the model. Each cohort member had an intervention-specific probability of improving their level of PA. The effect of no intervention was also simulated. In the second scenario, the exemplar interventions were only applied to those adults 50 to 64 years of age, while the change in PA levels at 1 year for adults aged 25 to 49 years was determined using the natural history model.

The CDC MOVE model in this study retained parameter estimates used in the original study.¹² After the first year (ie, for the next 19 years), in each scenario, cohort members had an annual probability of either remaining at the same physical activity level or moving to a lower activity level, as it was conservatively assumed that the impact of an intervention would decline after the intervention had ended. For all of the interventions, with the exception of the enhanced access intervention, a 50% decline in physical activity in year 2 was modeled. For the enhanced access study, a 33% decline was modeled in this second year, because the environmental enhancements persisted long after the intervention had ended.

Following this assumed substantial decline (33% to 50%) in physical activity postintervention in year 2, cohort members were transitioned into a natural-history model, which modeled the general decline in physical activity that occurs with age. Thus each year participants faced sex- and age-specific probabilities of moving to a lower level or remaining at the same physical activity level. Among the healthy, the risk of developing one of the 5 diseases depended on activity level, age, and sex. The risk of death depended on age, sex, and disease status.

We characterized the 4 PA levels (inactive, irregularly active, sufficiently active, and highly active) in terms of a range of MET-minutes per week using BRFSS data (MET-minutes estimate the energy expended on PA, based upon the intensity, duration, and frequency of PA, and are used as a measure of total PA per week).¹⁴ The effect size of each intervention was converted into an increase in MET-minutes per week. To estimate the probability of moving to a higher PA level after intervention, we added intervention-specific MET-minutes per week to the current level, and noted the proportion of the cohort that moved from one level of PA to another as a result of an intervention. This proportion was then used as our transition probability. For example, if adding a certain number of MET-minutes per week to all persons in the inactive group caused 25% of them to move up to the irregularly active group, we estimated the probability of moving from inactive to irregularly active as 0.25 in the first year following the intervention. For the analysis focusing on the middle-aged cohort, we applied the increase in MET-minutes and intervention costs only to cohort members 50 to 64 years of age. The intervention effect size was assumed constant across age groups, but sex-specific.

We modeled a substantial decline (33% to 50%) in intervention effect following the 1-year intervention, based on limited data available on long-term maintenance of PA resulting from interventions.^44–46 Following this decline, further changes in activity levels over time were determined by the natural history model.^47–49 That is, for each year except the first, individuals were assigned sex- and age-specific probabilities of either moving to a lower PA level or remaining at the same level.

In both groups of simulations, we estimated the cost-effectiveness of each intervention as cost divided by quality-adjusted life year (cost/QALY), using methods consistent with the guidelines established by the Panel on Cost-Effectiveness in Health and Medicine.⁴² Over a 20-year time horizon, the MOVE model was used to project costs, gains in life-years (survival), and gains in QALYs associated with each intervention, as well as with no intervention (natural history). Consistent with the panel’s recommendations, the societal perspective was adopted, and future costs and benefits were discounted to the present at an annual rate of 3%.⁴² Under each scenario, the performance of each intervention compared with no intervention was assessed using a ratio of the additional expected cost of each program divided by additional expected QALYs gained relative to the no intervention alternative. In addition, the number of cases of disease prevented was also estimated.

To determine the robustness of the final results, we conducted probabilistic sensitivity analyses, with particular emphasis placed on intervention effect size and cost estimates.

To assess the impact of uncertain intervention cost and effect size parameter estimates on uncertainty in cost-effectiveness, we performed a probabilistic sensitivity analysis. When running such an analysis, we obtained not just a single cost-effectiveness ratio for each intervention, but a distribution of ratios reflecting joint parameter uncertainty of its cost and effectiveness.

Using the distributions from this analysis, we assessed the probability that the cost-effectiveness of each intervention was below various thresholds that are commonly used to determine whether interventions provide good value for money.

Tables 1 and 2 summarize the average costs, effectiveness, and cost-effectiveness ratios associated with a one-time application of each PA promotion intervention, relative to no intervention. Results are cumulative over a 20-year time horizon, but average per person values are reported in both tables.

In the first scenario (Table 1), interventions were applied to the entire cohort of adults aged 25 to 64 years, and compared with no intervention. With no intervention, the average discounted quality-adjusted life expectancy (total QALY) was predicted as 11.18 years, and 20-year cumulative costs were estimated at about $61,100. Intervention participation increased average total QALY by 0.008 to 0.063 QALYs (Table 1), or equivalently, intervention participation improved average healthy life expectancy by 0.42 to 3.28 weeks. Cost-effectiveness ratios ranged between ~$42,000 and ~$146,000 per QALY gained. The Lombard social support intervention⁹ was estimated as the most effective, with the largest gain in QALYs (0.063), compared with no intervention. The Reger community-wide campaign¹⁰ was estimated as the most cost-effective (about $42,500/QALY).

In the second scenario (Table 2), interventions were applied only to persons aged 50 to 64 years, and compared with no intervention. Intervention participation improved average QALYs by 0.003 to 0.024, which is equivalent to 0.16 to 1.25 weeks. Cost/QALY in middle-aged adults (age 50 to 64 years) were 38% to 47% lower depending upon the intervention, and ranged from ~$34,000 and ~$127,000 per QALY gained. That is, for each intervention, the cost-effectiveness was better when it was applied only to adults age 50 to 64.

All interventions reduced disease incidence in the simulations. The reductions ranged from 3 to 20 cases per 100,000 for colon cancer and from 55 to 420 cases per 100,000 for CHD.

Results from the probabilistic sensitivity analysis for the middle-aged cohort are shown in the acceptability curves in Figure 2. For example, there is about a 40% chance that the cost/QALY of the young intervention¹¹ is less than $50,000 per QALY (a traditionally used benchmark of cost-effectiveness), and virtually a 100% chance that it is below the more contemporary threshold of $200,000/QALY.⁵⁰ A review of several studies found that the median estimate of willingness to pay for gains in quality of life was $265,345 per QALY.⁵⁰ That is, despite model parameter and resultant cost/QALY ratio uncertainties, it is almost certain that this intervention is an acceptable use of societal resources.