Hester van Eeren

Summary | 155 | the expected value of further information depends on this WTP value. The illustrative WTP value was based on and averaged over different values society wants to pay to prevent crimes like robbery and vandalism. The illustrative value of information analysis revealed that at a societal willingness-to-pay of €71,700 per criminal activity free year, further research to eliminate parameter uncertainty was valued at €176 million. This means that society should be willing to spend a maximum of €176 million in reducing decision uncertainty in the cost-effectiveness of the two interventions. In particular, most of the uncertainty was found in the effects of the interventions, which was translated to transition probabilities of adolescents moving from one state to another in the Markov model, and, to a lesser extent, in the intervention costs of the Course House and the direct non health-care costs in both model states. This illustrative analysis showed that the results were meaningful and can be interpreted according to health care evaluation studies. Moreover, it showed that this analysis can be helpful in justifying additional research funds to further inform the reimbursement decision with regard to youth care interventions. Finally, this study made important recommendations for applying this method more systematically in youth care, like defining a WTP value to the defined outcome and, as in the cost-effectiveness analysis itself, determining the range of effects and costs that should be taken into account. In Chapters 4 and 5 , the aim was to investigate and illustrate the use of available, non-randomized data when evaluating treatments in clinical practice. The propensity score (PS) method was used to control for initial differences due to the non-random assignment of adolescents to the treatments evaluated. In Chapter 4 , the feasibility of the univariate and multivariate PS method was demonstrated in subgroup analyses of outcomes research. The performance of using the univariate PS was tested using Monte Carlo simulations with additional adjustment on the subgroups. . The multivariate PS was estimated by combining the treatment groups and subgroup categories. The treatment effect and subgroup effects were estimated in a linear regression model adjusting for either of the two PS estimations. The bias and mean squared error showed minor differences between both PS methods, with marginally lower values of the bias and mean squared error when using the multivariate PS. Clinical practice data from a large effectiveness study on psychotherapy in personality disorders were used to compare the two methods. Using these data, the differences between short-term and long-term treatment were compared using the severity of patients’ problems as the subgroup of interest. Both the univariate and multivariate PS estimations yielded similar results. The results of this study support the use of the multivariate PS with slightly less biased estimated treatment effects. The choice of the subgroup of interest, however, should be clinically relevant and influences the choice for analyses and interpretations as well. In Chapter 5 , two youth care interventions, FFT and Multisystemic Therapy (MST) were compared on their effectiveness using non-randomized, clinical practice data from adolescents assigned to either one of these interventions (422 MST; 275 FFT) at the Viersprong, institute for personality disorders and behavioral problems in the Netherlands. Data were gathered within Routine Outcome Monitoring and the effectiveness of the two interventions was estimated using the PS method to control for initial measured differences between the treatment groups. The primary outcome