Hester van Eeren

Framework for modelling the cost-effectiveness of systemic interventions | 2 21 | In the developed model two treatment alternatives were compared. To provide an example of a cost-effectiveness analysis of systemic interventions, a group receiving FFT therapy and a comparison group receiving TAU were evaluated. TAU refers to a comparable treatment, which delinquent youth would have received if they had not received FFT. As institutions offer diverse types of alternative therapies to FFT, TAU may differ between the different institutions. In one institution TAU may be MST, while another institution may offer Cognitive Behavioral Therapy (CBT) as an alternative to FFT. In our illustration subjects could not switch between FFT and TAU. For an extensive comparison between two systemic interventions, the model should include several types of cost categories. Table 1 depicts the common cost categories in health economic evaluations; direct and indirect costs inside and outside the health care system adapted to the field of crime. The included types of costs are derived from a combination of the costs commonly included in health economic evaluations and literature on cost of crime (Cohen, 2005). These costs not only pertain to costs incurred by the delinquent juvenile, e.g. costs due to criminal activities or treatment, but also to costs falling on family, caregivers and the society as a whole. For reasons of comparability with other interventions in health care, the model included all relevant societal costs in accordance with the Dutch manual for costing in economic evaluations (Hakkaart, Tan, & Bouwmans, 2010). Discount rates for future costs and effects were set consistent with guidelines for economic evaluations in the Netherlands (The Health Care Insurance Board, 2006). (Note that differential discounting is required in the Netherlands to account for the growth in the value of health over time. See for example Brouwer and colleagues (2005) for the rationale behind this. Therefore, by using these rates it was implicitly assumed here, that the value of a criminal activity free year (CAFY) will also increase over time, comparable to the rate of a QALY.) Data analytic procedures: Cost-effectiveness and scenario analyses In effect studies, uncertainty is generally represented as a confidence interval, i.e. the magnitude of uncertainty is expressed in standard deviations of the measurement error. This assumes that all relevant uncertainty is measurable in a single outcome measure, and that the distribution of the measurement error is reasonably normal. As both assumptions do not apply in typical health economic evaluations, normal t-tests and other parametric statistics are not particularly useful in health economic modelling. Instead, probabilistic analysis was conducted to take the uncertainty of the model parameters into account. In this analysis uncertainty was simulated by running the Markov model several times using a large cohort of subjects, each time with slightly different parameter values. These values were obtained by randomly sampling from each of the parameter distributions, i.e. gamma distributions for costs, and Dirichlet distributions for transition parameters (Briggs et al., 2006). One thousand Monte Carlo simulations were performed. In each simulation a random draw from the parameter distributions was taken, which creates a unique set of cost and effect parameters. The expected costs and effects were then calculated and could be plotted on a cost-effectiveness plane. Four additional scenarios