Statistical analysis
All main analyses will be carried out at the end of the last follow-up assessments (that is, there will be no interim analyses) and will be based on the intention-to-treat principle, with due consideration being given to potential biases arising from loss to follow-up. Random effects regression models will be fitted to the repeated measures to estimate treatment effects for outcomes, controlling for treatment center, in-patient status and the corresponding baseline assessment for the outcome under investigation. We will allow for the presence of missing outcome data under the assumption that the data are Missing At Random (MAR), using the terminology of Little and Rubin [54], with the possible addition of inverse probability weighting to adjust for the possible role of non-adherence to allocated treatment and other intermediate outcomes as predictors of future loss to follow-up [55]. Stata will be used for these main analyses. Secondary analyses to investigate putative meditational mechanisms, but also the effect of receipt of an adequate dose of treatment (CACE estimation), will be carried out; these will use methods similar to those of Baron and Kenny [56] but also the newer approach of instrumental variables analysis to allow for the omitted variables problem that is, hidden confounding [55-58]. MPlus and Stata will be used for these analyses.