Retention was 93% (106 of 114) at 12 months; 4 patients recurred or died, and 4 dropped out of the trial. Of the 106, 14 patients (12%) were intervention dropouts (most attended only one or two sessions, so process data were not available), but they remained in the trial and continued assessments. The 92 (81% of 114) patients receiving the intervention participated in an average of 22 of 26 sessions (85%), 14 (SD = 6) of the 18 intensive sessions and 6 (SD = 3) of the 8 maintenance sessions. Five participants did.not have utilization data because of delay in implementing the process measures and had to be excluded, which resulted in analyses with 87 patients. Preliminary analyses indicated there were no significant differences between cohorts of the intervention on any of the process, utilization,. or outcome measures described above (all ps > .40). Nevertheless, cohort was included as a control.
Descriptive Data and Analyses Testing Processes and Outcomes
In general, patients were very satisfied with the intervention content (M = 3.52, with 4 being highest, SD = 0.36; see ). The topics receiving the highest ratings (data not shown) were dietary information (M 3.90, SD = 0.36), relaxation (M = 3.70, SD = 0.58), and the stress conceptualization (M = 3.7, SD = 0;60). Even the topic with the lowest rating (communication with medical providers; M = 3.3, SD = 0.57) had a moderately satisfied rating. As evidenced by the Cohesion score (M = 7.8, with 10 being highest), women reported high involvement and reciprocated feelings of support. Patients' overall satisfaction with the intervention was significantly associated (r = .45), but did not overlap, with feelings of group cohesion.
Descriptive Statistics for Process, Utilization, and Outcome Measures
As robust effects of both satisfaction and cohesion on outcomes have been reported in the psychotherapy literature, all were tested. Longitudinal hierarchical linear modeling (HLM; Raudenbush & Bryk, 2002
) was used. HLM involves two levels. At the within-subject level, the outcome varies within participants over time as a function of a person-specific growth curve. At the between-subjects level, the person-specific change parameters vary randomly across participants as a function of the process variable (cohesion or satisfaction). HLM models considered outcomes at 4, 8, and 12-months. As process variables were assessed at 4 months only, data from the initial assessment were not included. The three assessment points representing time were coded such that the intercept reflected the level of the outcome variable at 4 months. Each model estimated time, process, and Process × Time effects. The time effect tested whether the outcome changed across the 4-, 8-, and 12-month assessments. The process effect tested whether the process variable covaried with the intercept (4-month outcome score). The Process × Time effect, if significant (p
≤ .05), indicated that the process variable covaried with the rate of change in the outcome. For interpretability, process variables were centered at the mean. We computed partial correlation coefficients using t
values and degrees of freedom to estimate effect size (Rosenthal, 1994
In the analyses, satisfaction was not associated with outcomes at 4 months (i.e., the intercept; all ps >.07) or the rate of change (i.e., the Satisfaction × Time interaction; ps >.06). In contrast, the relationship between cohesion and therapy outcome was robust. displays the fixed effects for time, cohesion, and the Cohesion × Time interaction for outcomes after cohort effects were controlled for. Time effects were significant for POMS TMD, KPS, and symptoms and signs (all ps ≤ .05). Cohesion was not associated with any of the outcomes at 4 months (i.e., the end of the intensive phase; all ps ≥.05), indicating that scores at 4 months did not covary with cohesion. However, significant Cohesion × Time effects were found for POMS TMD (p = .04), physical activity (p = .02), and performance status (KPS; p = .04) measures. The Cohesion × Time effect indicates that patients reporting greater personal involvement with and felt support from the group from the intensive phase also reported greater improvement (lower distress, more physical activity) and higher functional performance status into the maintenance phase (see for an example).
Fixed Effects for Hierarchical Linear Models Examining the Association Between Cohesion and Treatment Outcomes
Figure 2 Estimated trajectories of change in the patients' Profile of Mood States Total Mood Disturbance (POMS TMD) scores by level of cohesion. Whereas patients feeling less support from the group showed no change in the level of their emotional distress, those (more ...)
On the basis of the interaction findings, we conducted follow-up analyses to examine whether the associations between cohesion and positive outcomes could be related to (or a by-product of) an association between cohesion and more frequent use of the intervention techniques for these outcomes. We used the same HLM analytic model, and technique utilization at 4, 8, and 12 months was the outcome. Each model estimated time, cohesion, and Cohesion × Time effects while controlling for cohort. For all techniques, neither cohesion (all ps > .30) nor the interaction of cohesion and time (all ps > .14) was significant. In combination, this suggests that the cohesion interaction effects described above for the POMS, physical activity, and KPS outcomes were due to factors other than patients' differential use of the respective intervention strategies.
Analyses Testing Utilization of Intervention Techniques and Outcomes
Hierarchical linear models were used for utilization measures and their corresponding outcomes. Each model included the initial (baseline) value and 4-, 8-, and 12-month outcomes. For these analyses, the intercept reflected the level of the outcome variable at the initial assessment. The corresponding utilization variable, assessed at 4, 8, and 12 months, was included as a time-varying predictor; utilization at initial assessment was fixed to zero on the basis of the assumption that there was no technique use at that time and usage began with the intervention. This strategy allowed for the examination of discontinuous trajectories during the intensive (initial to 4 months) and maintenance phases (4 to 12 months). Cohort assignment was again controlled. Each model estimated time, utilization, and Utilization × Time effects. We computed partial correlation coefficients using t values and degrees of freedom to estimate effect size (Rosenthal, 1994
As recommended (Raudenbush & Bryk, 2002
), we first determined the general form of change that best fit the data. With four data points, linear and, potentially, quadratic models could be fit. According to a likelihood ratio test, if the fit of the quadratic model was not significantly better (α = .05), the linear .model was retained. If the Utilization × Quadratic Time interaction was not significant, it was dropped for parsimony (Cnaan, Laird, & Slasor, 1997
). displays the fixed effects for time, utilization, and the Utilization × Time interaction for each outcome.
Fixed Effects for Hierarchical Linear Models Examining the Association Between Utilization of Intervention Techniques and Treatment Outcomes
We first tested the association between utilization of the three strategies (stress conceptualization, relaxation, and problem solving) for stress–distress management and the POMS outcome. In each model, there was a significant time effect (ps < .05), indicating that POMS scores decreased over time, The stress conceptualization effect and the Stress Conceptualization × Time interaction effect were not significant. There was a significant effect for relaxation (p = .02), indicating that more frequent relaxation use was associated with a lower level of distress. The Relaxation × Time effect was not significant (p = .12). The problem solving effect was not significant (p = .19), but there was a significant Problem Solving × Time interaction (p = .02). POMS scores declined regardless of the level of problem solving use during the intensive phase. During maintenance, however, usage was associated with less improvement in POMS.
For the PSS Family outcome, the effects of time, social strategies use (both identifying and using social support and assertive communication) and the interaction were not significant. For the Food Habits Questionnaire outcome, time was not significant (p
= .08), but both dietary strategies use (reducing fat and increasing fiber; p
= .04) and the interaction (p
= .02) were significant. The interaction was graphed according to the strategies recommended by Preacher, Curran, and Bauer (2006)
. Three levels of utilization (monthly, weekly, and daily) were chosen for graphical representation (see ). The figure shows that more frequent dietary strategies use was associated with greater positive change in food habits, an effect most pronounced during the intensive phase but still evident during maintenance.
Figure 3 Estimated trajectories of change in the patients' Food Habits. Questionnaire scores by level of dietary strategies use. The findings suggest that more frequent dietary strategies use was associated with greater positive change in food habits, an effect (more ...)
As previously reported, the intervention did not result in a significant increase in exercise for the intervention group (p
= .08 for the Group × Time interaction; Andersen et al., 2004
). For the readers' information, however, we examined these data. The effects for time, use of energy expenditure techniques, and the Energy Expenditure × Time interaction were not significant for activity level outcome.
We tested the associations of physical functioning outcomes with four strategies–the stress conceptualization, relaxation, communication with medical providers, and energy expenditure. Time effects were observed for the KPS (ps < .03), indicating that patients' functional status improved with time. Effects were not observed for any of the strategies nor their interactions for the KPS outcome. In contrast, strategy use and interaction effects were observed across strategies on symptoms and signs. For this outcome, the quadratic form of change best fit the data. Stress conceptualization (p =. .001), relaxation (p = .004), medical provider (p = .002), and strategies for increasing physical activity (p = .004) were all associated with a reduction in symptoms and signs. Interaction effects of a similar form were observed with patients' use of the stress conceptualization (p = .02) and frequency of relaxation (p = .05; see ). In both cases, as chemotherapy began during the intensive phase, more frequent use of the conceptualization for understanding stress and more frequent relaxation practice were reported by those found to have the highest levels of signs and symptoms. However, during the maintenance phase, as cancer treatments ended, the level of strategy usage was related to the level of decline in symptoms and signs.
Figure 4 Estimated trajectories of change in the patients' reported frequency of their relaxation practice and the nurses' ratings of the patients' symptoms and signs. The findings suggest more frequent relaxation use by those with the highest levels of symptoms, (more ...)
The pattern of interaction effects for medical provider communication (p = .002) and increasing physical activity (p = .004) were of similar form. For illustration, displays the relationship between patients' reports of strategies to increase physical activity and nurses' evaluations of the patients' symptoms. As can be seen, the quadratic pattern of symptom change was manifested by the increase in symptoms as adjuvant cancer therapies began, and then symptoms declined as therapies ended and patients recovered. As symptoms increased, however, so too did patients' efforts to cope with them by increasing their activity level and communicating more frequently with their medical providers. As treatments ended, those patients who maintained their high level of activity (and provider communication) were evaluated as having the greatest reductions in symptomatology.
Figure 5 Estimated trajectories over time in the patients' reports of physical activity and the nurses' ratings of the patients' symptoms and signs. The findings suggest that the patients who made the most use of exercise strategies were those with the highest (more ...)
Finally, relevant to the above symptom effects are the analyses for dose intensity. Hierarchical linear models could not be used, as there was only one data point per drug per person (it was calculated once, when the regimen was completed), and the data were not normally distributed (the goal was for every patient to receive 100% of every drug). StatXact (Cytel Software Corp., 2004
) was used to compute permutation one-way analyses of variance (ANOVAs) with general scores (Gibbons, 1985
). This analysis uses exact permutational distributions rather than F
distributions and does not depend on assumptions of normality. Three utilization variables were hypothesized as relevant (see ). For each, the mean of the 4- and 8-month utilization scores (the period when chemotherapy was administered) was calculated. Via a median split, patients were classified as low (coded as 0) versus high (coded as 1) users of a strategy. We tested the null hypothesis that the low- and high-utilization groups would have identical dose distributions. Exact significance values were estimated via Monte Carlo sampling (1,000 tables sampled for each significance value). For analyses with significance values less than .05, post hoc analyses were conducted to identify the minimum level of technique use associated with improvements in dose intensity.
There was a significant association between dose intensity for the taxanes and exercise (T = 6.10, p = .003). Women who reported more frequent use of strategies to increase energy expenditure received a significantly higher proportion of taxanes compared to women who exercised less (or not at all). The actual dosage differences were substantial, with a 99% (SD = 3%) average dose intensity for women who reported a higher activity level, compared to 88% (SD = 11%) for those patients who reported lower activity levels. Post hoc analyses revealed that using strategies to increase energy expenditure four times per week was the minimum amount of use associated with increased dose intensity.