Predictors and Moderators of Post-Treatment Depression Severity After Hypnotherapy and Cognitive Behavioral Therapy in an RCT for Mild-to-Moderate Depression · Brockmann AL, Pfeiffer L, Batra A et al.

3. Results

As displayed in Table 2, in the total sample, the symptom severity at baseline, the duration of the current episode, global functioning, self-efficacy expectations, and quality of life regarding psychological and environmental domain significantly correlated with depressive symptoms after treatment. In patients receiving CBT, despite the symptoms at baseline, a lower school education, higher global functioning, better psychological quality of life, higher behavioral activation level, and lower social impairment, as well as higher self-efficacy expectations, correlated with lower symptoms in the outcome. In patients receiving HT, higher symptom levels at baseline and treatment with ADM were associated with higher symptoms after treatment.

3.2. Global Predictors for Treatment Outcome in the Total Sample

In the regression model for the total sample, the variable duration of the current episode, global functioning, self-efficacy expectations, and quality of life regarding psychological and environmental domain were entered in the second block after the MADRS at baseline (first block). Due to some missing data in questionnaires, the total sample for the regression analysis was n = 141. The MADRS at baseline (β = 0.28, t(137) = 3.71, p < 0.001) as well as self-efficacy (β = −0.34, t(137) = −4.29, p < 0.001) and the duration of the current episode (β = 0.17, t(137) = 2.28, p = 0.024) were identified as predictors for the outcome. The model explained 20% of the variance (ΔR2 = 0.205, F(3,137) = 12.98; p < 0.001); see Table 3.

In the regression analysis with MADRS at baseline as an additional variable (not entered in block 1), the same predictors could be identified (GSES t1: β = −0.32, t(137) = −4.29, p < 0.001, MADRS baseline: β = 0.28, t(137) = 3.71, p < 0.001, duration of current episode: β = 0.17, t(137) = 2.28, p = 0.024). Overall, 20% of the variance could be explained (ΔR2 = 0.205, F(3,137) = 12.997; p < 0.001). Depressive symptoms at baseline were therefore not the primary predictor for symptoms after treatment.

3.3. Treatment-Specific Predictors for Treatment Outcome

The variables self-efficacy, global functioning, school education, and environmental quality of life were entered in the regression analysis for CBT. Since there was a problem of multicollinearity resulting in a high correlation (r > 0.7) of the BADS total score with some of the subscales, only the BADS total score was entered in the regression analysis in block 2. The sample consisted of n = 75 patients. The MADRS at baseline (β = 0.31, t(73) = 2.92, p = 0.005) and self-efficacy (β = −0.27, t(73) = −2.52, p = 0.014) were identified as predictors for the outcome. The model explained 15.9% of the variance (ΔR2 = 0.159, F(2,73) = 8.07; p < 0.001); see Table 4.

In the second regression analysis with MADRS at baseline as an additional variable, other predictors were identified (WHOQOL-psychological: β = −0.33, t(73) = −3.08, p = 0.003; school education: β = 0.25, t(73) = 2.33, p < 0.023) but not the MADRS at baseline or the GSES. Here, 15.2% of the variance could be explained (ΔR2 = 0.152, F(2,73) = 7.74; p < 0.001). Given that the two predictors were identified in the stepwise regression model, the role of initial symptom severity has to be questioned for CBT patients in this study.

In the first regression model for HT patients, ADM, self-efficacy expectations, and quality of life regarding psychological and environmental domain were entered in the second block. Due to some missing data in questionnaires, the total sample for the regression analysis was n = 69. The MADRS at baseline (β = 0.26, t(64) = 2.50, p = 0.015), self-efficacy (β = −0.26, t(64) = −2.48, p = 0.016), quality of life regarding the environmental domain (β = −0.23, t(64) = −2.12, p = 0.038), and ADM (β = 0.38, t(64) = 3.70, p < 0.001) were identified as predictors for the outcome. The model explained 33.7% of the variance (ΔR2 = 0.337, F(4,64) = 9.65, p < 0.001); see Table 5.

In the second regression analysis with MADRS as an additional variable, the same predictors were identified (MADRS at baseline: β = 0.26, t(64) = 2.50, p < 0.015; GSES: β = −0.26, t(64) = −2.48, p = 0.016; WHOQOL-environmental: β = −0.23, t(64) = −2.21, p = 0.038; ADM: β = 0.38, t(64) = 3.70, p < 0.001). Overall, 33.7% of the variance could be explained (ΔR2 = 0.337, F(4,64) = 9.65; p < 0.001). Since the same variance could be explained with the same predictors as before, it can be assumed that the outcome was not primarily predicted by symptoms at baseline.

3.4. Moderators for Treatment Outcome in CBT and HT

For the moderator analysis, the variables ADM, BADS total score, and school education were selected. The overall model of the first moderator analysis was significant, F(3,148) = 3.36, p = 0.020, and explained a variance of 6.38% (R2 = 0.064). The results showed that ADM significantly moderated the effect between the treatment condition and the outcome (ΔR2 = 6.05%, F(1,148) = 9.57, p = 0.002, 95% CI [−14.918, −3.288]). Treatment with ADM was associated with lower outcome (or higher depressive symptoms after treatment) in HT compared to CBT. In the moderator analysis with the BADS total score, no effects were found (ΔR2 = 2.98%, F(3,144) = 1.721, p = 0.165, 95% CI [−0.2405, 0.0619]). Finally, the overall model with the variables school education and treatment condition was significant, F(3,148) = 3.68, p = 0.014, with a variance of 6.40% (R2 = 0.064) explained. The results showed a significant moderation effect of school education (ΔR2 = 6.39%, F(1,148) = 7.95, p = 0.006, 95% CI [1.3479, 7.6603]). Higher school education was associated with lower treatment outcomes for patients receiving CBT, while the opposite effect was observed for HT.