The Phase I sample included 785 patients who completed the preference questionnaire at baseline and attended at least one post randomization session (97.2% of the consented sample). Among these patients, 203 (25.9%) reported no treatment preference. Among the 582 participants who expressed preference, the majority preferred a combination of medication and psychotherapy (n=532, 91.4%) and the rest preferred medication only (n=50, 8.6%).
Descriptive information for the Phase I sample appears in . In Phase I, those with a treatment preference had experienced a greater number of years of MDD compared to those with no preference, t(380) = 2.72, p = .007, Cohen's d = .22. Those with a treatment preference also had a higher proportion of patients who reported prior ADM use compared to those with no preference, χ2(2) = 19.6, p < .001, odds ratio = 2.04. Those with a preference for medication only did not differ from those with a preference for combination in terms of proportion reporting prior ADM experience, χ2(2) = 1.34, p = .51.
Descriptive statistics of the sample.
Among those who reported prior use of ADM, those with and without a preference did not differ in their perceptions of best response to prior ADM, χ2(4) = .82, p = .93. However, those with a preference for medication only differed from those with a preference for combination in their perceptions of best response to prior ADM, χ2(4) = 30.48, p < .001. This finding appeared to be driven mainly by those with a preference for medication only having a higher proportion of patients reporting remission with prior ADM, odds ratio = 21.21, as well as a lower proportion of patients reporting marginal response to prior ADM, compared to the combination preference group, odds ratio = .55. In Phase I, preference groups did not differ on other descriptive characteristics listed in .
The Phase II sample included the 473 patients who completed the baseline preference questionnaire, did not remit during Phase I and were randomized to Phase II, and attended at least one post randomization Phase II session. Among them, 345 (72.9%) indicated a baseline treatment preference for combination, 105 (22.2%) had no preference, and 23 (4.9%) preferred medication only.
Associations Between Beliefs About Depression and Treatment Preference
Bivariate Pearson intercorrelations of the five belief statements were small or nonsignificant (all r's < +/− .22), suggesting that the five statements tapped relatively independent etiology beliefs. Belief items were not completed by 5% of the Phase I sample (4% of the medication only preference group; 6% of the combined preference group and 4% of the no preference group). A MANOVA used to compare patients' beliefs in the causes for depression (five items) by preference group (no treatment preference, medication preference, combined treatment preference) revealed a significant effect for preference group, F(10,1472) = 3.71, p < .001. Post hoc comparisons using a Bonferroni correction revealed that patients with a baseline preference for medication were significantly more likely to endorse the belief that their problems were due to a chemical imbalance than patients with a baseline preference for combination, mean difference = .48, p =.02, Cohen's d = .41 or with no baseline preference, mean difference = .59, p = .00, Cohen's d = .52. By contrast, patients preferring combination treatment were significantly more likely to attribute their problems to stressful life events than those preferring medication only, mean difference = .69, p < .001, Cohen's d = .59 or with no baseline treatment preference, mean difference = .37, p < .001, Cohen's d = .36. displays descriptive statistics for endorsement of the five beliefs by preference group.
Endorsement of five possible causes of depression.
Did Baseline Treatment Preference Predict Treatment Response?
Results of linear mixed effects models (LMM) appear in . Phase I HAM-D scores decreased significantly over time for patients regardless of baseline treatment preference. However, depressive symptoms of patients with no baseline treatment preference decreased more quickly than those reporting a preference, F(1, 616) = 8.01, p < .005. Based on LMM estimates, over the course of Phase 1, those with no preference decreased an average of 13.56 points (12 weeks × 1.13) compared to 11.16 points (12 weeks × .93) in those with a preference. In other words, symptom decrease in the preference group was only 82% of that in the no preference group. shows mean HAM-D scores for the no preference and any preference groups across Phase 1. Phase I HAM-D scores also decreased significantly across time for those with a preference for combination and medication only. However the time by preference group interaction was not significant, F(1, 457) = 1.28, p = .26, suggesting that the rate of symptom reduction did not differ between the two groups declaring a preference.
Linear mixed effects estimates for HAM-D scores among patients with and without a baseline treatment preference.
Mean HAM-D scores across Phase I for patients with and without a treatment preference.
Patients with no baseline treatment preference did not differ in HAM-D scores at entry to Phase II from those with a treatment preference, F(1, 457) = .77, p = .38. Scores decreased significantly from the beginning to the endpoint of Phase II for both groups, with no significant difference in rate of decrease between the groups, F(1, 424) = .23, p = .64. HAM-D scores at entry to Phase II did not significantly differ between patients preferring combination versus medication only treatment, F(1, 356) = .07, p = .79. Both groups demonstrated significant decreases in HAM-D scores across time but their rates of decrease did not significantly differ, F(1, 314) = 1.07, p = .30. Cell sizes for Phase II treatment arm by baseline treatment preference were 11 or less for those with a medication only preference, which precluded analyses of treatment preference by treatment arm interactions on treatment outcome.
Did Baseline Treatment Preference Predict Attrition?
The overall phase I treatment dropout rate was 20.9%; drop out rates by preference group were 38.0% among patients preferring medication only, 19.7% among those preferring combined treatment, and 19.7% among those with no preference. Chi-square tests indicated that those with no treatment preference did not differ from those with any treatment preference in attrition, χ2(1) = .23, p = .63, odds ratio = 1.10. However, patients who preferred medication were significantly more likely to drop out of Phase I than those preferring combination at baseline, χ2(1) = 9.09, p = .003, odds ratio = 2.50. In order to explore whether lack of response to preferred treatment might explain the counter intuitive finding that those who preferred medication only and were receiving medication only were more likely to drop out, we examined early HAM-D symptom response. Percent symptom change from baseline to Weeks 2 and 4 was compared among patients who did and did not complete Phase I. Weeks 2 and 4 were selected because nearly two-thirds of Phase I dropouts among patients preferring medication occurred prior to Week 6, and thus Weeks 2 and 4 provided the bulk of available data on subsequent dropouts. Among the medication only preference group at Week 2, no significant differences in percent symptom change emerged between patients who did and did not subsequently drop out, t(35) = −1.52, p = .14, M(SD)completers = 23.65(25.91), M(SD)noncompleters = 9.84(27.17). However, Week 4 comparisons using a Bonferroni corrected p-value of p < .025 revealed a trend: patients who dropped out tended to have less HAM-D symptom reduction at Week 4 than Phase I completers, t(32) = −2.08, p = .05, M(SD)completers = 40.51(28.73), M(SD)noncompleters = 17.95 (28.85), Cohen's d = .78. By contrast, among those with a combination preference, HAM-D symptom change did not differ by completion status at Week 2, t(412) = −.15, p = .88, M(SD)completers = 19.98(23.59), M(SD)noncompleters = 19.52(26.12), or Week 4, t(386) = .16, p = .88, M(SD)completers = 27.15(27.29), M(SD)noncompleters = 27.86(35.19).
To further explore determinants of attrition by preference, we tallied reasons for Phase I dropout by preference group (). This revealed that a large proportion of patients failed to return to clinic/lost contact. Cell sizes were very small for other attrition reasons, thus precluding statistical tests of differential reasons for dropout by preference group.
Reasons for phase I attrition among dropouts.
In Phase II, the overall treatment dropout rate was 11.6%; the dropout rate was 21.7% among patients with baseline treatment preference for medication only, 10.7% among those preferring combination, and 12.4% among no preference patients. Chi-square tests indicated no significant difference in attrition between patients preferring medication versus combination treatment, χ2(1) = 2.59, p = .11, odds ratio = 1.14, or between no treatment preference and preference for any treatment, χ2(1) = .07, p = .79, odds ratio = 1.09.