Predictors of Antidepressant Prescription and Early Use Among Depressed Outpatients

The subjects were consecutive patients newly admitted to selected mental health outpatient clinics in Westchester County, N.Y. The 1996 U.S. census estimates that the county has 893,412 residents, of whom 73% are white and 26% are minority. The residents live in a broad range of urban, suburban, and rural communities, ranging from small rural villages to Yonkers, which is the third largest city in New York state according to the 1990 census estimate reported by the Westchester County Census Bureau. After consultation with the Westchester County Office of Mental Health, clinics were selected as sample sites on the basis of their location, to include rural and urban populations, and the patient population served, to include both minority and nonminority communities. Consecutively admitted patients were screened on site in six mental health facilities: one independently funded clinic, three county clinics, a psychiatric clinic in a general community hospital, and a university-based clinic. The subjects were between 18 and 64 years of age. Elderly persons (aged 65 years or older) were excluded from these analyses as the majority (79%) of elderly patients screened sought service at the university-based clinic program specializing in studies of antidepressant efficacy for geriatric patients.

The study used a two-stage sampling procedure to identify newly admitted patients with unipolar major depressive disorder. After the study was described to new patients, written informed consent was obtained for the screening. The phase I assessment consisted of administration of the Center for Epidemiologic Studies Depression Scale (CES-D Scale) (16), to determine the severity of depressive symptoms, and the Mini International Neuropsychiatric Interview (17), to assess the presence of the DSM-IV criteria for major depressive disorder. A CES-D Scale cutoff score of 16 or higher was used to identify cases of probable depression because of the reported high sensitivity (96%) of this score in psychiatric outpatients (18). The Mini International Neuropsychiatric Interview has high sensitivity (96%) and specificity (88%) in identifying major depressive disorder in psychiatric samples (17). A positive finding with either instrument was the entry criterion for the phase II assessment, which included administration of the Structured Clinical Interview for DSM-IV—Patient Edition (SCID-P) (19) to establish the diagnosis of major depressive disorder. Patients who screened positive for depression were contacted again, and written consent was obtained for participation in the phase II assessment and follow-up. The exclusion criteria included cognitive impairment as defined by a score of less than 24 on the Mini-Mental State (20), alcohol or substance abuse within the last month, and the presence of another axis I disorder.

Of the 815 newly admitted patients aged 18–64 years, 772 (95%) consented to the phase I assessment. Of those screened, 565 (73%) were positive for depressive symptoms. Of the 565 patients eligible for phase II interviews, 174 refused further follow-up, 55 could not be contacted again, and 336 (59%) consented to having the more extensive phase II diagnostic assessment. The phase II assessments were conducted by two research assistants and a clinical psychologist who were trained in the administration of the SCID-P and the Global Assessment of Functioning (21) with satisfactory interrater reliability (intraclass correlation coefficient=0.81). Over one-half (54%, 183 of 336) of the newly admitted patients who screened positive for depression met the criteria for major depressive disorder. One-quarter (23%, N=77) of those who screened positive for depression had bipolar or schizoaffective disorder, and one-quarter (23%, N=76) did not meet the criteria for an affective disorder. Only patients with the primary diagnosis of major depressive disorder or major depressive disorder with concurrent dysthymia were included in the follow-up group. Of the patients meeting these diagnostic criteria (N=183), 35 were excluded because of cognitive impairment or recent substance abuse, leaving 148 patients available for follow-up.

The subjects who met the inclusion criteria were evaluated with the structured interview guide for the 17-item Hamilton Depression Rating Scale (22, 23) to assess symptom severity. Information on duration of illness, suicidality, and prior episodes was gathered during administration of the SCID-P. The Global Assessment of Functioning Scale (21) was used to assess overall functioning. Concurrent medical illness was recorded by using the Chronic Disease Score (24), which assigns points for classes of medications prescribed for significant medical illnesses. While the presence of concurrent axis II personality disorders was not assessed, the 47-item version of the Inventory of Interpersonal Problems was administered and used as a screen for personality pathology (25). A cutoff score of 1.1 on the first three subscales of the Inventory of Interpersonal Problems was chosen to screen for interpersonal difficulties; this cutoff score had a reported sensitivity of 71% and specificity of 67% in outpatients with mood disorders (25).

Follow-up interviews were conducted 3 months after clinic admission. The patients were interviewed to determine the treatment recommended, services used, and current symptoms as indicated by the Mini International Neuropsychiatric Interview and the CES-D Scale. Interviews were completed for 84% of the 148 major depression patients (N=124). There were no differences in age, sex, ethnicity/race, depression severity, or overall functioning of the patients who were and were not able to be interviewed 3 months after admission. However, follow-up assessments were completed by a larger proportion of patients recruited from the university-based clinic (92%, 56 of 61) than patients recruited from the community outpatient clinics (78%, 50 of 64) (χ²=5.61, df=1, p=0.02).

Information on whether antidepressant treatment was recommended and on the dose and duration of treatment was collected from patient report and chart review. The average rate of concordance between the patient reports and chart records for whether medication was prescribed, across the university-based and county-funded mental health clinics, was 83%. When the patient report and chart notes were discrepant, the chart review data were used in the analyses.

Antidepressant use was recorded for both the 3 months before clinic intake and the 3-month follow-up period by using a modified version of the composite antidepressant treatment score described by Alexopoulos et al. (26). An objective of this study was to identify predictors of medication recommendation and factors that contribute to initiation of medication treatment. To this end, antidepressant use was defined as 1 week of medication use during the 3-month follow-up period. Recent use before admission was defined similarly, i.e., at least 1 week of use during the 3 months before the index admission. Adequacy of medication dose and its relation to outcome were not investigated.

All of the patients were between the ages of 18 and 64 years; the mean age was 38.93 years (SD=12.70). Other sociodemographic characteristics of the sample are summarized in table 1. Of the 124 patients, 31% belonged to ethnic minority groups, predominantly African American and Hispanic. The majority of patients reported being unemployed, and many reported being disabled.

Site of service, comorbid conditions, and history of hospitalization are presented in table 2. The patients’ mean score on the Global Assessment of Functioning Scale was 45.9 (SD=8.8), and their mean score on the Hamilton depression scale was 19.5 (SD=4.8). All of the followed patients met the DSM-IV criteria for major depressive disorder. More than one-quarter met the criteria for prior alcohol abuse but reported no use in the past month. One-fifth reported one or more previous psychiatric hospitalizations. Comparisons of the patients seen in the two types of clinics (mental health outpatient clinics versus the university-based clinic) revealed no differences in the baseline severity of depression, history of alcohol or substance abuse, or overall functioning as measured by the Global Assessment of Functioning Scale.

We conducted chi-square analyses of the dichotomous demographic and clinical variables to establish 1) the likelihood of receiving a recommendation for drug therapy and 2) the likelihood of initiating medication use when recommended. Differences between groups (e.g., medication recommended or not) in continuous measures were analyzed by using t tests. Variables found to be significant in bivariate comparisons at the 0.10 level or better were entered into a series of logistic regression analyses to build the best predictive model of group membership. To identify specific depressive symptoms associated with medication referral, we used t test comparisons of mean group scores on specific Hamilton depression scale items. To account for multiple comparisons in the analyses of Hamilton scale items, a Bonferroni correction was calculated and a significance level of p<0.002 was used (27). Two-tailed significance was used. Statistical analyses were conducted by means of SAS (28).

During the 3 months after admission, medication was recommended for the majority (71%, 88 of 124) of depressed persons seeking help from the mental health outpatient clinics. For most of these patients antidepressants were recommended during the initial evaluation, while for a few (N=6) medication was recommended later in treatment. Comparisons by type of clinic demonstrated that recommendations for medication were received by a larger proportion of patients seeking help at the university-based clinic (87%, N=53) than patients seen at the other clinics (57%, N=36) (χ²=12.10, df=1, p<0.001). As expected, a higher percentage of the overall clinical hours was provided by psychiatrists or psychiatric residents in the university-based clinic (67%) than in the other mental health facilities (7%–42%). In addition, the community-based clinics used only nonmedical staff to conduct the intake evaluations, whereas the evaluations at the university-based clinic were conducted by all disciplines.

Analyses of bivariate associations between patient sociodemographic characteristics and medication recommendation revealed that only ethnicity/race was associated with antidepressant recommendation: a larger proportion of nonminority patients (84%, N=72) than minority patients (45%, N=17) received recommendations for antidepressants (χ²=17.89, df=1, p<0.001). The number of minority patients (N=38) was too small for stratified analyses. Post hoc comparisons of the black, Hispanic, and white patients indicated that a larger proportion of white patients (84%, N=72) received a recommendation for antidepressant drugs than either the black patients (30%, N=6) (χ²=18.16, df=1, p<0.001) or Hispanic patients (56%, N=9) (χ²=4.67, df=1, p=0.03). There was no difference in recommendations for medication between the two minority groups (χ²=1.56, df=1, p=0.21). Recommendation for antidepressant medication was not associated with differences in gender, age, education, employment status, disability status, income, marital status, living arrangements, or method of payment (Medicaid, fee for service, managed care, or other insurance).

The two types of clinics, university based and community, differed in the association between ethnic/racial status and recommendation for antidepressant drugs. At the community sites, medication was significantly less likely to be recommended to minority patients; 74% of the nonminority patients (25 of 34) and only 38% of the minority patients (11 of 29) received recommendations for medication (χ²=6.71, df=1, p=0.01). At the university-based clinic, the difference was not significant: medication was recommended for 90% of the nonminority patients (47 of 52) and 67% of the minority patients (six of nine) (χ²=1.99, df=1, p=0.16). According to the data elicited by the SCID-P, there were no differences between minority and nonminority patients in the presence of specific depressive phenomena (such as melancholic or atypical symptoms), suicidality, age at onset, or duration of current episode. Exploratory comparisons of the minority and nonminority patients’ symptoms as shown by the Hamilton depression scale indicated that the minority patients reported greater weight loss (t=2.15, df=123, p=0.03) and the white patients reported greater disruption of work and/or activities. According to the SCID-P, there were no differences in clinical presentation (e.g., specific major depression criteria endorsed, presence of melancholic or atypical symptoms, suicidal ideation, age at onset, or duration of current episode). There were no other sociodemographic variables (e.g., gender, age, employment, self-reported disability, or marital status) that differed between the patients who did and did not receive recommendations for antidepressant medication.

The patients who received a recommendation for antidepressant treatment had higher overall scores on the Hamilton depression scale (mean=9.7, SD=5.0) than the patients who did not receive a recommendation for medication (mean=18.0, SD=3.6) (t=2.00, df=123, p=0.02). Medication recommendation was more likely among patients with a prior psychiatric hospitalization (88%, N=22) than among individuals who had never been hospitalized (67%, N=67) (χ²=3.13, df=1, p=0.08). Medication recommendation was not associated with differences in the subtype of depression (atypical, melancholic), suicidal ideation, duration of current illness, prior episodes of depression, or presence of a medical illness. Using the cutoff score of 1.1 on the Inventory of Interpersonal Problems, the screen for personality disorders, we found that patients with and without significant personality pathology were equally likely to receive recommendations for medication. Recent antidepressant treatment was a significant predictor of medication recommendation; 98% of the patients (39 of 40) with a recent history of prior antidepressant use (at least 1 week of use during the 3 months before intake) received a recommendation for antidepressant medication, compared to 60% of the patients without recent antidepressant use (50 of 84) (χ²=17.46, df=1, p<0.001).

Logistic regression was used to determine the independent effects of historical, sociodemographic, clinical, and site factors significant in bivariate comparisons at the p<0.10 level (table 3). White patients were almost three times as likely to receive a recommendation for antidepressants than minority patients, even after we controlled for severity of depression, site, and recent use of medication. Site of service remained a significant predictor of medication recommendation, such that patients with major depressive disorder who sought help at the university-based clinic were more than four times as likely to receive a medication recommendation than patients seen at other outpatient mental health clinics. To explore the differential treatment of minority patients by site, an interaction term (clinic type by ethnicity) was entered into the model. The term did not significantly contribute to the model. With depression severity controlled, the patients with recent antidepressant use were 18 times as likely to receive a medication recommendation than individuals without recent antidepressant use. When the Hamilton score was used as a continuous variable, overall depression severity continued to predict the likelihood of medication recommendation, such that each additional point on the scale was associated with a 1.15 odds of having medication recommended. Prior hospitalization did not contribute to the model. The four variables in the model accounted for 44% of the total variance.

The model created by the logistic regression analyses correctly predicted group membership (medication recommended or not recommended) for 82% of the patients, 92% of those who actually received medication recommendations (81 of 88) and 60% of those who did not receive medication recommendations (21 of 36).

To identify the depressive symptom profile of individuals who received a medication recommendation, we compared scores on the individual items of the Hamilton depression scale for these patients with scores for the patients who were not referred for medication. The patients referred for medication had significantly higher scores on the item for depressed mood than did the patients not referred for antidepressant treatment (t=3.84, df=123, p<0.001). This was the only symptom significant at the p=0.002 level or better; none of the other items approached significance. To examine whether depressed mood contributed more to the overall model than did the total Hamilton depression scale score, a separate logistic regression was conducted. Depressed mood was a more significant clinical predictor of medication recommendation than overall severity (odds ratio=3.80, χ²=6.78, df=1, p=0.009).

In the total sample of 124 patients with major depression who were followed for 3 months, 59% of the patients (N=73) took a prescribed antidepressant for at least 1 week during the 3-month follow-up period. Of the 88 patients for whom antidepressants were recommended, 83% (N=73) reported that they took an antidepressant medication for at least 1 week. Of the patients who took the recommended antidepressant, 86% (63 of 73) continued to take the medication for 4 weeks or longer. Only recent use of antidepressants was a significant predictor of actual medication use. All of the patients who reported antidepressant use before admission and received a recommendation for antidepressant medication took the recommended antidepressant, as compared to 70% of the individuals with no prior use of medication. The liberal definition of prior use contributes to the high use rate among individuals who received an antidepressant recommendation and may have limited our ability to identify other factors that contribute to use of the recommended antidepressant medication. There were no other clinical, sociodemographic (age, sex, payment type), or site characteristics associated with medication use.