Supplemental Materials
Identifying Moderators of the Adherence-Outcome Relation in Cognitive Therapy for Depression
by K. Sasso et al., 2015, Journal of Consulting & Clinical Psychology
Online Supplement A
In order to aide interpretation of the three significant moderation effects identified, here we report the regression coefficients for all predictors included in these three models respectively (as illustrated in equation 2).
Table1. Regression Coefficients for Predictors in Primary Models Examining Three Significant Interaction Effects Identified
b / Standard Error / DF / t-value / 95% CICognitive Methods X Gender
BDI-II** / .97 / .02 / 155 / 39.11 / .92, 1.02Site / -.45 / .48 / 54 / -.93 / -1.42, .52
CM-Within* / -3.36 / 1.10 / 155 / -3.04 / -5.54, -1.18
Gender / -.26 / .48 / 54 / -.55 / -1.23, .70
CM-within X Gender* / 6.28 / 2.21 / 155 / 2.84 / 1.92, 10.63
Behavioral Methods X HAM-A
BDI-II** / .95 / .03 / 154 / 35.57 / .90, 1.01Site / -.57 / .49 / 53 / -1.18 / -1.55, .40
BH-Within / -1.11 / .87 / 154 / -1.28 / -2.82, .60
HAM-A / .05 / .04 / 53 / 1.30 / -.03, .13
BH-within X HAM-A* / -.57 / .19 / 154 / -3.07 / -.94, -.20
Behavioral Methods X MDEs
BDI-II** / .97 / .02 / 155 / 39.17 / .92, 1.01Site / -.53 / .47 / 54 / -1.13 / -1.48, .41
BH-Within / -.81 / .85 / 155 / -.95 / -2.48, .87
MDEs / -.06 / .13 / 54 / -.49 / -.31, .19
BH-within X MDEs* / 1.77 / .56 / 155 / 3.14 / .65, 2.89
Note: **p < .0001, *p < .01. CM-within = Within-patient variability in Cognitive Methods at the current session, HW = Homework; MDEs = prior major depressive episodes; HAM-A = pre-treatment anxiety severity as measured by the Hamilton Rating Scale for Anxiety, BDI-II = Beck Depression Inventory-II score at the current session
Online Supplement B
The correlations between moderators and within-patient adherence (i.e., our predictors) are by definition zero; however, to place our results in context we did examine the correlations among raw observed adherence scores and each moderator separately at each of the first four sessions (see table below). We then converted these correlation coefficients to z-scores using Fisher’s r to z transformations. We averaged these z-scores and converted those means back to r-values. These values, which reflect the average correlation of raw adherence scores and moderators over the first four sessions, are provided below. Overall we see very little evidence of any meaningful relationships, as almost all moderators showed no significant correlations with raw adherence scores at any of the four sessions.
Table 1: Average Correlations among Raw Adherence Scores and Putative Moderators Across Sessions
Age / Gender / HRSD / Prior MDEs / PD / HAM-ACognitive Methods / .06(0/4) / -.05(0/4) / -.01(0/4) / -.01(0/4) / -.22(1/4) / -.03(0/4)
Negotiating /Structuring / -.08(0/4) / -.09(0/4) / -.16(1/4) / .08(0/4) / -.04(0/4) / -.16(1/4)
Behavioral Methods/HW / -.01(0/4) / -.12(0/4) / -.11(0/4) / .15(0/4) / .02(0/4) / -.01(0/4)
Note. Fractions in superscript indicate at how many of the first four sessions the relationship was significant at p < .05. HW = Homework; HRSD = pre-treatment depressive symptoms as measured by the Hamilton Rating Scale for Depression; MDEs = major depressive episodes; HAM-A = pre-treatment anxiety severity as measured by the Hamilton Rating Scale for Anxiety, and; PD = Personality Disorder
Online Supplement C
In the primary moderation analyses reported, we identified patients for whom the positive effects of CM-within and BH-within (i.e., within-patient variation to Cognitive Methods and Behavioral/Methods Homework) were particularly strong. However, we must also consider the other patient groups identified by these analyses; the patients (i.e., men, those with more prior episodes, or those with lower anxiety) for whom adherence did not positively impact symptom change. Given that on average patients in our sample showed significant symptom improvement across the early sessions of CT examined here, and early change was not predicted by patient gender, pretreatment anxiety, or number of prior depressive episodes, other factors must account for symptom change in these patients. As therapists’ relative adherence to Cognitive vs. Behavioral Methods provides one possibility, we examined if either the co-occurrence of high within-patient adherence to BH and CM, or differences in such adherence, predicted next-session change, and if so, if these relations differed as a function of patients’ scores on the three moderators identified. To do so, we created: (1) a “Severity” score reflecting efforts to adhere to both BH and CM at a given session, and; (2) a “Directionality” score reflecting efforts to adhere more to BH relative to CM at a given session. Using the approach proposed by Essex and colleagues (2003), we first standardized BH-within and CM-within scores at each session, and then used the average of these standardized scores to reflect Severity, and the half difference of these two scores to reflect Directionality (Essex, Klein, Cho, & Kraemer, 2003). The resulting Severity and Directionality scores are, by definition, completely uncorrelated and both have means of zero; thus, they can be examined in the same model simultaneously without concern of colinearity.
When we examined the main effects of patients’ Directionality and Severity scores at the current session (while controlling for the main effect of site and patients’ current session BDI-II scores) on patients’ BDI-II scores at the following session using a single-level repeated measures regression model, the Directionality and Severity terms emerged as trend and significant predictors respectively (Directionality: t(155) = 1.83, b = .94, SE = .52, p = .07; Severity: t(155) = -2.65, b = -1.38, SE = .52, p = .01). Greater than expected within-patient adherence to both BH and CM at a given session predicted a greater than expected reduction in symptoms at the next-session and within-patient variability marked by greater adherence to BH than CM at a given session predicted a slower than expected degree of next-session symptom change at the level of a non-significant trend. We then examined if these relations varied as a function of patients’ gender, prior depressive episodes, and anxiety severity. To do so, we ran three separate repeated measures regression models for each of the three moderators; in each model patients’ next-session BDI-II scores were predicted from the main effects of patients’ Directionality and Severity scores at the current session, the main effect of the moderator, and the effects of both the Directionality by moderator and Severity by moderator interaction terms (while controlling for patients’ current session BDI-II scores and site). The following significant interactions emerged as predictors of next-session symptom change following a correction for multiple tests: First, we found a gender by BH-CM Severity interaction (t(153) = 2.64, b = 2.76, SE = 1.04, p = .01), such that higher than expected scores on both BH-within and CM-within at a given session predicted a greater than average reduction in next-session symptoms for women, but not for men (simple slope of Severity scores on symptom change for women: b = -2.66, SE = .67, t = -3.96, p < .001). Second, we found a BH v. CM Directionality by prior episodes interaction (t(153) = 2.33, b = .73, SE = .31, p = .02). For patients higher in prior episodes, a tendency to use greater CM than BH (as would be indicated by a lower Directionality score) at a given session predicted a greater than expected reduction in symptoms at the next-session (simple slope of Directionality on symptom change for those with 4 or more prior episodes: b= 2.09, SE = .76, t = 2.74, p = .009), whereas for patients low in prior episodes it was unrelated. We also found a significant prior episodes by BH-CM Severity interaction (t(153) = 2.33, b= .76, SE = .33, p= .02), such that higher than expected scores on both BH-within and CM-within at a given session predicted a greater than average reduction in next-session symptoms for those low in prior episodes (simple slope of Severity scores on symptom change for those with 1 or fewer prior episodes: b= -2.85, SE = .76, t = -3.77, p < .001). Lastly, we found a BH v. CM Directionality by HAM-A score interaction (t(152) = -3.16, b = -.33, SE = .11, p = .002). For patients lower in anxiety, a tendency to use greater CM than BH predicted larger than expected next-session symptom reductions (b= 2.70, SE = .76, t = 3.54, p < .001).
Reference
Essex, M. J., Klein, M. H., Cho, E., & Kraemer, H. C. (2003). Exposure to maternal depression and marital conflict: Sex differences in children's later mental health symptoms. Journal of the AmericanAcademy of Child & Adolescent Psychiatry, 42, 728-737.