In the present study, it could not be ruled out that the probability of response would systematically and considerably differ between patients with and without endpoint-assessments. However, since the primary analysis of effectiveness concerned the total treatment-offer, regardless of possible deviations from the protocol, statistical methods to correct for bias in the findings caused by missing endpoint-assessments, like multiple imputation or propensity score estimation, were only of limited applicability. It was therefore considered crucial to minimize the occurrence of missing endpoint-assessments as much as possible, by conducting intensive field work and by providing additional financial compensation for participating in the endpoint-assessments. Nevertheless it could not be excluded that some missing endpoint-assessments would occur in the study population. In case of such missing endpoint-assessments, and because of lack of satisfactory alternatives, the "last observation carried forward" (LOCF) method was used in the primary analysis.

Regarding LOCF within the framework of the study, it is important to emphasize that this technique does not provide an estimate of the patient's status at the time of the month 12 assessment, but rather of his status at the time of his last available assessment. Assuming a progressively effective treatment intervention as a patient participates longer in the treatment program, LOCF will underestimate the effectiveness of the treatment of patients with a missing endpoint-assessment, who still participated in the treatment after their last available assessment. In addition, it is not unlikely that participation in a study is more frequently discontinued at times when a patient is performing poorly, which may once again result in an underestimation of the treatment-effect of that particular participant when using LOCF. The influence of these factors on the outcome of the study depends upon the extent to which these factors occur differentially in the two treatment conditions, and result in a diverging likelihood of response among participants with and without an endpoint-assessment between the two treatment conditions. These issues will be explored in secondary analyses, testing the robustness of the findings.