A total of 2,699 letters were mailed to eligible patients in the 4 Group Health primary care clinics, with 336 (12.4%) refusing to be interviewed and 312 (11.5%) patients who were ineligible or who couldn't be contacted. Patients agreeing to be interviewed (n = 2,051; 76.1%) did not differ in age or gender from patients refusing to be interviewed.
Of the 2,051 patients completing the screening, 272 (13.3%) were eligible for the persistence study and 694 (33.9%) were eligible for a parallel study on relapse prevention14
(based on having 0 to 3 DSM-IV symptoms and either recurrent depression or dysthymia). Of the 272 patients eligible for the persistence study, 176 (64.7%) received a baseline interview and 159 (90.3%) were successfully randomized to the persistence study. Ninety-six (35.3%) patients did not receive a baseline interview. Of these 96 patients, 30 refused, 30 did not have time to participate in an intervention program, 27 could not participate for other miscellaneous reasons, and 9 could not be contacted. Patients refusing baseline interview did not differ in age or gender from those agreeing to be interviewed.
Of the 694 patients eligible for the baseline interview for the relapse prevention study based on screening interview, 472 (68%) completed the interview and 69 (14.6%) were offered and accepted randomization to the persistence study based on an SCL of 1.5 or more. Patients refusing the baseline interview for the relapse prevention study (222 (8.1%)) did not differ in age or gender from those agreeing to be interviewed.
Of the 228 patients randomized, 209 (91.7%) completed the 1-month follow-up, 193 (84.6%) completed the 3-month follow-up, 192 (84.2%) completed the 6-month follow-up, and 171 (75%) completed the 28-month follow-up. There were no significant demographic or clinical differences between the patients who missed completing at least 1 follow-up and the patients completing all 4 follow-up interviews.
There were no significant differences between the 114 intervention and 114 usual-care patients on the following demographic variables, including age (I, 47.2 ± 14.0 years vs UC, 46.7 ± 13.4 years), percent with 1 or more years of college (I, 77.2% vs UC, 78.1%), percent employed full- or part-time (I, 72.6% vs UC, 64.9%), and percent Caucasian (I, 79.8% vs UC, 80.7%). There was a significant difference between intervention and control patients in the percent of female subjects (I, 67.5% vs UC, 81.6%, χ2 (1) = 5.20; P = .02), which was one of the covariates controlled for when analyzing differences in outcomes between the 2 groups.
There were also no significant differences between intervention and control patients on chronic disease score (I, 1,191 ± 979 vs UC, 1,368 ± 1,293), NEO Neuroticism score (I, 22.7 ± 5.5 vs UC, 23.0 ± 5.5), SCL-Depression score (I, 1.9 ± 0.5 vs UC, 1.9 ± 0.5), percent with recurrent depression (I, 76.3% vs UC, 83.3%), or percent with a history of dysthymia (I, 50.0% vs UC, 59.8%).
There were no significant differences in age, gender, randomization or strata groups, or baseline depression level between the 187 patients used in the cost and adherence analyses and the 41 patients who disenrolled at some point during the 30 months. There were no significant demographic or clinical differences between control and intervention patients in the cost and adherence analyses.
Effects of Intervention on SCL Depression Score
The longitudinal random regression model for SCL depression resulted in a significant 3-way interaction of time × treatment × strata (z = 2.16; P= .03). This was due to significant interactions of time × treatment (z = 1.94; P = .05) and time × strata (z = 3.16; P = .002). In this model, the main effects are not directly interpretable. Baseline depression, CDS, and NEO were significant covariates. On the basis of these results, we stratified by baseline severity in order to examine 28-month depression. The planned post hoc ANCOVAs for the 28-month SCL-Depression showed a significant treatment effect for patients in the moderate-severity strata (F1, 87 = 8.65; P = .004), with the adjusted mean for the control group (1.23 ± 0.62) significantly higher than that of the intervention group (0.88 ± 0.52) at the 28-month follow-up (see). In the high strata, treatment differences in SCL-Depression at 28 months were not statistically significant (F1, 51 = 0.02; P = .88] (see). The adjusted means were 1.16 ± 0.85 and 1.19 ± 0.72 for the intervention and control groups, respectively, at the 28-month follow-up.
SCL depression adjusted means over time for moderate strata: intervention versus controls.
SCL depression adjusted means over time for high strata: intervention versus controls.
Effects of Intervention on Sheehan Disability Scale
The longitudinal random regression model for Sheehan disability resulted in a significant 3-way interaction of time × treatment × strata (z = 3.08; P = .002). This was due to significant interactions of time × treatment (z = 2.06; P = .04) and time × strata (z = 3.69; P < .001), and treatment × strata (z = 2.41; P = .02). In this model, the main effects are not directly interpretable. CDS was the only significant covariate. On the basis of these results, we stratified by baseline severity in order to examine 28-month disability. The ANCOVAs for the Sheehan Disability Scale at the 28-month follow-up were not statistically significant. In the moderate strata, the control group reported more disability (adjusted mean = 3.58 ± 2.37) in comparison to the intervention group (adjusted mean = 3.09 ± 2.30) (F1, 87 = 1.21; P = .27). In the high strata, there were no significant differences between the intervention (adjusted mean = 3.41 ± 2.61) and control groups (adjusted mean = 3.20 ± 2.66) (F1, 51 = 0.09; P = .76).
Effects of Intervention on Dosage and Duration of Antidepressant Medications
The longitudinal ordinal random regression model examining patients meeting criteria for at least 90 or more days of an adequate dosage of antidepressants in each of five 6-month periods did not result in a significant 3-way interaction of time × treatment × strata (z = 1.79; P = .07). When this term was dropped, and the model was refit, the only significant 2-way interaction was time × treatment (z = 3.62; P < .001). In this model, the main effects of time and treatment are not directly interpretable. The strata main effect was very significant in the model (z = 4.49; P < .001). On the basis of this result, we stratified by baseline severity in order to examine 28-month adherence to an adequate dosage of medication. For the high-severity strata, intervention patients were significantly more likely to adhere to 90 days or more of antidepressants at the lowest dose consistent with Agency for Health Care Policy and Research guidelines compared to usual-care patients during the first and second 6-month blocks of time. In the high strata during the first 6 months, 72% (n = 24) of the intervention patients and 40% (n = 14) of the controls were adherent to an adequate dosage of medication (χ2 (1) = 8.23; P < .01). This trend was also seen in the second 6-month period: 70% (n = 23) of the intervention patients and 37% (n = 13) of the controls were adherent to an adequate dosage of medication (χ2(1) = 5.98; P < .05). For the moderate-severity strata, intervention patients were only more likely to adhere to 90 days or more of adequate dosage of antidepressants during the first 6-month block of time (76% of the intervention patients versus 46% of the controls, χ2(1) = 6.10; P < .05) Similar, but nonsignificant, trends were observed for the second 6-month block. For the other three 6-month periods, the percentages were very similar for the treatment groups in both strata.
Effects of Intervention on Costs
describes the unadjusted difference in health care costs between intervention and control patients. On the basis of linear regression models, there were no significant differences between intervention and control patients in total ambulatory costs (F1,180 = 0.77; P = .40), total health care costs (F1,180 = 0.91; P = .34), depression treatment costs (F1,173 = 2.65; P = .10), or non–depression-related outpatient costs (F1,180 = 0.11; P = .74). In the descriptive unadjusted data, intervention patients were found to have nonsignificant trends toward higher outpatient depression costs over the 30-month period, whereas controls had nonsignificant trends toward higher total nondepression costs and total outpatient costs. The incremental depression costs in the intervention group were primarily due to the increased costs of longer term use of serotonin reuptake inhibitors.
Unadjusted Annualized Cost Data for the Control and Intervention Groups (N = 187)*,†
Another cost of an intervention program for persistent depression would be the cost of instituting and maintaining a tracking system. We estimated that a program that would need to be developed by an HMO to screen patients 8 weeks after initiation of primary care physician antidepressant prescriptions would add approximately $67 per patient to the cost of the intervention.