Hester van Eeren

Value of information in crime prevention: An illustration | 3 37 | Moffitt (Moffitt, 1993; Schawo et al., 2012). Dying because of committing crimes was not reflected in the CAFY. Adolescents were assumed to face a risk of death equivalent to the age specific mortality rates in the general population (Statistics Netherlands, 2015b). The second group consisted of costs of health-care use, productivity losses, and other societal costs such as costs of the criminal justice system. Both costs outside health care, and health care costs were included, such as the costs of visiting a psychiatrist or psychologist. As the family system is involved in the interventions provided, we included both the costs of the adolescent and those of one of the parents. The model state costs were fixed over time until the adolescent was 23 years. It was assumed that from that age onwards not all cost categories (such as a family guardian or foster care) would remain relevant. The third group comprised the intervention costs. The costs of one completed FFT treatment was calculated to be approximately €10,900 per adolescent, whereas the Course House was about €37,800 (retrieved from Slot et al. (Slot et al., 1992)). Both costs were extrapolated to 2013 Euro’s accounting for inflation based on the consumer price index (Statistics Netherlands, 2015a). The cost and effects in the model were discounted (i.e. Brouwer, van Hout, & Rutten, 2000; van Hout, 1998), according to the guidelines for economic evaluations in the Netherlands (The Health Care Insurance Board, 2006). A - criminal B - not criminal C - dead tpA2B(1-nmr) tpB2A(1-nmr) nmr nmr tpB2B(1-nmr) tpA2A(1-nmr) Figure 1. Markov model nmr = natural mortality rate tpA2A = transition probability of staying in state A tpA2B = transition probability of moving from state A to state B tpB2A = transition probability of moving from state B to state A tpB2B = transition probability of staying in state B To represent the uncertainty of each model parameter, we assigned parameter distributions (Table I in Supplemental Material). In a probabilistic analysis, uncertainty was simulated by running the model 10,000 times using a cohort of subjects and each time taking different parameter estimates from the parameter distributions (Briggs et al., 2006; Claxton, 2008). These 10,000 unique sets of parameter values were used to estimate the mean expected cost-effectiveness. For further details on the cost- effectiveness model, we refer to Schawo et al. (Schawo et al., 2012).