Hester van Eeren

General introduction | 1 9 | Not being randomized A Randomized Controlled Trial (RCT) is considered to be the golden standard for comparing the effectiveness of interventions; its results are considered as the highest scientific evidence. In an RCT the internal validity is high, because participants are randomized over an experimental and control condition and the aim is to obtain unbiased treatment effect estimates: the treatment selection bias is assumed to be zero (APA, 2006; Imai, Kind, & Stuart, 2008; Stuart, Cole, Bradshaw, & Leaf, 2011). Proven effectiveness in an RCT is also the highest level of evidence for interventions to be marked as effective in the DEI. However, conducting an RCT is not always feasible due to ethical or practical constraints (Black, 1996). Therefore, alternative options should be considered to obtain evidence for clinical practice. For example, a quasi-experimental study design using observational data could be used in evaluating the effectiveness of interventions in clinical practice. Using such a study design, regular clinical practice can be followed and participants are not randomly allocated to interventions. The external validity of the results found in such studies could be higher than in RCT’s, since participants represent the actual target population and the sample selection bias is closer to zero (Imai et al., 2008; Stuart et al., 2011). However, to increase the internal validity of non-randomized samples, and decrease the treatment selection bias, one should minimize the effect of allocation bias in evaluating the effectiveness of interventions. Without controlling for possible baseline differences due to allocation bias in non- randomized studies, the effect of the interventions could be confounded by these initial differences. To control for allocation bias in observational data, the propensity score (PS) method is a valid and frequently used method. The PS is defined as the conditional probability of assignment to an intervention given a set of observed pre-treatment differences (Rosenbaum & Rubin, 1983). When applying the PS, the strongly ignorability assumption should be considered carefully, meaning that the treatment assignment should be independent of the potential outcome, the treatment assignment should be independent given a set of measured covariates, and each adolescent should have a chance of being in either treatment arm (Shadish, 2013). Given the assumption of strongly ignorability, applying the PS to achieve balance in the treatment arms enables the achievement of results equivalent to randomized studies (Austin, 2011; Shadish, 2013; West, Cham, Thoemmes, Renneberg, Schulze, & Weiler, 2014). This thesis explores the possibilities of using the PS in youth care evaluation studies, because using the PS enables comparative effectiveness studies in youth care and increases the external validity of the results. External validity is especially important in health policy and, thus, in cost-effectiveness research, since the results should be translated to spending public money to reimburse cost-effective interventions.