Chapter 5 116 Linear mixed models (LMM) were implemented to investigate between-group differences (IMR vs. CAU) over an 18-month period, comprising 12 months of treatment and six months of follow-up (59). To assess treatment effects, we adopted the following analysis protocol for all outcomes (6). We first determined the best-fitting model using a stepwise modeling procedure. Subsequently, time and conditions were entered into the model; a time× condition interaction term was included only if the inclusion of this term improved the model based on the Akaike Information Criterion. A statistically significant time × condition interaction was considered to represent a treatment effect. For all outcomes, the estimated marginal means and 95% confidence intervals (CI) were calculated for the three measurement moments. To assess between-group effect sizes, we calculated Cohen’s ds (0.2–0.3, small; 0.5, medium; >0.8, large) (60). As LMMs use the full data set, this methodology retains people with missing values; thus, imputation is not necessarily beneficial in this context (61). For all outcomes, we followed the same procedure to examine the robustness of our findings within a sensitivity analysis among IMR completers. A “completer” was defined as a patient who had attended ≥ 50% of all scheduled sessions. To explore the possibility of selective IMR non-completion, the patients’ attendance was analyzed via logistic regression using 12 variables representing baseline characteristics. Hospitalization was analyzed in two ways. First, to examine between-group differences in the odds of hospitalization during the year following T2, we performed logistic regression analysis with respect to hospitalization (dichotomized), with T2 as the dependent variable, the treatment group as the independent variable, and hospitalization (dichotomized) one year prior to T1 as a confounder. Second, we determined difference scores with respect to the length of stay (days) at one year before and after treatment. Between-group differences were tested using the Mann–Whitney test. In our study protocol, we hypothesized that IMR-related improvement would be associated with the fidelity of IMR implementation (6). Therefore, we explored the impact of fidelity on outcomes. This analysis included all participants with fidelity scores who had attended at least ten IMR sessions, indicating consistent engagement with the IMR program. Participants were divided into high and moderate to- low fidelity groups, with mixed models used to assess differences in time effects between the control group and the two experimental subgroups. The control group was considered the reference group. The model components were comprised of the three measurement moments, the three subgroups (high fidelity,
RkJQdWJsaXNoZXIy MjY0ODMw