Older Studies

Mulrow et al ¹⁰⁴ systematically reviewed the performance of screening tests for depression conducted between 1966 and February 1994. They identified 15 published and 4 unpublished articles that met their inclusion criteria, which required that the outcome status of at least 50% of the subjects be verified by an acceptable criterion standard examination. Eleven of these articles met our inclusion criteria as well and appear in Evidence Table 1.^{39,40,42,43,47,60,63,69,73,97,102}

To summarize performance, Mulrow et al ¹⁰⁴ calculated the average sensitivity and specificity for the included articles (based on the usual cut-points for each instrument) and constructed a summary receiver operating characteristics (ROC) curve. The overall sensitivity was 84% (95% confidence interval [CI], 79% to 89%), and overall specificity was 72% (95% CI, 67% to 77%). These values translate to a positive likelihood ratio (LR) of about 3 and a negative LR of 0.2. Results did not differ substantially based on the degree of verification bias. The included instruments were easy to administer and complete, and they had been written at either easy (third to fifth grade) or average (sixth to ninth grade) literacy levels.

General Primary Care Populations

We identified 23 newer articles that Mulrow et al had not included. Six of the 23 newer studies were conducted in primary care settings in nonelderly or mixed populations.^{36,51,61,62,65,71} Klinkman et al ⁵¹ found that the CES-D had a sensitivity of 81% and specificity of 72% for scores above 15, compared with a gold standard diagnosis based on a Structured Clinical (Diagnostic) Interview (SCID) for DSM-IIIR or -IV. Parkerson and Broadhead ⁶¹ found a similar level of performance for the Duke AD screener: 81% sensitivity and 64% specificity for scores greater than 30. Salokangas et al ⁶² found that The Depression Scale (DEPS) performed reasonably well (sensitivity, 74%; specificity, 85% for scores greater than 8). Bashir et al ³⁶ tested the GHQ in a random sample of British general practice attenders and found a sensitivity of 76% and a specificity of 74%. Steer et al ⁶⁵ reported that the BDI performed extremely well (sensitivity, 97%; specificity, 99%) against a less rigorous criterion standard, the mood module of the Primary Care Evaluation of Mental Disorders (PRIME-MD).

The study by Whooley et al ⁷¹ deserves special comment. They examined the performance of multiple screening tests, including the CES-D, BDI, and MOS, as well as a new two-item screener that included only questions about depressed mood and anhedonia, in a population of veterans (97% men) from an urgent care setting. The two-item screener (sensitivity, 96%; specificity, 57%; area under the ROC curve, 0.82) performed nearly as well as the CES-D and MOS (area under the ROC curve, 0.89 for each). Shorter versions of the CES-D and BDI also performed well.

Overall, these newer studies had sensitivity and specificity results similar to those found by Mulrow et al. ¹⁰⁴ Sensitivity with some of the newer short screeners was slightly improved, with specificity similar to that of older instruments.

Elderly Populations

Twelve newer studies (Evidence Table 1) specifically examined the performance of depression screening instruments in older adults, including 6 using the GDS, 3 using the Self Care-D, and 3 using the CES-D (Table 9). The age limits used to define "elderly" varied; 1 study included adults older than 50 years of age, another enrolled only those older than 75 years, and others fell in between. The settings included community-based recruitment, primary care clinics, geriatric assessment clinics, patients' homes, and a nursing home.

Table

Table 9. Screening Accuracy in Geriatric Populations.

Each of these screening instruments demonstrated relatively good test performance characteristics (Table 9), with sensitivities generally 80% to 95% and specificities of 70% to 85%. Each instrument showed modest variation between studies. In general, confidence intervals were not calculated for the sensitivity and specificity estimates, and few studies calculated area under the ROC curves. Two studies, Gerety et al ⁴⁵ and Lyness et al, ⁵⁶ compared the GDS and CES-D instruments; both found that the GDS performed better. In Gerety et al, ⁴⁵ the area under the ROC curve was 0.91 for the GDS and 0.85 for the CES-D. According to Lyness et al, ⁵⁶ each instrument had similar performance for major depression, but the GDS performed better for "minor" depression. None of the studies compared the Self Care-D with either the CES-D or the GDS.

Special Populations

We identified 5 studies of depression screening in special populations that met our inclusion criteria (Evidence Table 1). Geisser et al ⁴⁴ tested the CES-D in a pain clinic. The criterion standard was a clinical interview with a psychologist using DSM-IV criteria. They found a 33% prevalence, a sensitivity of 82%, and a specificity of 73% using a score of 27 or greater to define a positive screen.

Holcomb et al ⁴⁸ examined the performance of the BDI in an obstetrics and gynecology setting. They used the Diagnostic Interview Schedule (DIS) as a gold standard and found an 11% prevalence of current depression. A BDI score of 16 or greater had 83% sensitivity and 89% specificity for depression.

Irwin et al ⁵⁰ used the CES-D in a community-based sample of adults with known physical illness. They compared their screening results against the SCID as a criterion standard. Scores of greater than or equal to 4 had 99% sensitivity and 84% specificity for depression.

Leung et al (1998) studied the performance of the Zung SDS in Chinese family practice patients in Taiwan. ⁵³ This team reported that SDS scores of greater than or equal to 55 had 67% sensitivity and 90% specificity for depression when compared against a diagnosis by a physician using DSM-IV criteria.

Lustman et al (1997) examined the BDI in patients with diabetes, using the DIS as a criterion standard. ⁵⁵ The prevalence for major depression was 37%, and a BDI score greater than or equal to 13 had 85% sensitivity and 88% specificity.

Summary of Screening Accuracy in Adults

Several depression screening instruments appear to detect depression effectively. Recent research has shown that shorter screening tests, including simply asking 2 questions about depressed mood and anhedonia, appear to detect a large majority of depressed patients; in some cases, they perform better than the original instruments from which they had been derived.

In general, sensitivity results were good to excellent and specificity results were moderate to good; with commonly used cut-points, typical values were 80% to 90% for sensitivity and 70% to 85% for specificity. If the prevalence of major depression is estimated to be between 5% and 15% in primary care settings, the positive predictive value (probability of depression after a positive test) would be 25% to 50% (Table 10). Thus, more than half of patients who screen positive will be false positives for major depression. Some of these "false positives" may be patients with minor depression or dysthymia. People with positives screens require further diagnostic questioning before clinicians apply a diagnostic label and suggest a treatment plan.

Table

Table 10. Probability of Major Depression after a Positive Screening Test.

One problem with depression screening instruments is that continuous data (ie, scores on the instruments) are dichotomized into positive and negative results at an arbitrary cut-off value and then used to calculate sensitivity and specificity (as well as positive and negative likelihood ratios) for that cut-off. With this approach, valuable information is lost because all scores above the threshold are counted equally (similarly, all below the threshold are also treated the same).

Some studies in this report partially overcome this problem by providing information on area under the ROC curve, which quantitates overall performance by producing a score between 0.50 (no information) and 1.0 (perfect information). An even more useful technique is to calculate stratum-specific likelihood ratios (SSLRs) for ranges of scores on an instrument. The SSLR for the result of the screen is multiplied by the pre-test odds to give the post-test odds. Furukawa et al ¹⁰⁶ calculated SSLRs for the CES-D using data from Japanese psychiatric hospitals and clinics. Scores of 0 to 29 were associated with an SSLR of 0.35; scores of 30 to 49 were associated with an SSLR of 2.3; and scores over 50 were associated with an SSLR of 11.7 (Table 11).

Table

Table 11. Stratum-Specific Likelihood Ratios (SSLRs) for CES-D.

Another difficulty in measuring the accuracy of screening instruments comes when trying to interpret specificity. Instruments used in some studies to detect major depression may count subjects with subsyndromal depressive illnesses as false positives. A true measure of specificity would count as false positives only those patients who are free from any significant depressive illness but who screened positive, because patients with subsyndromal illnesses may also benefit from treatment or more careful observation. Patients with other important and treatable disorders such as substance abuse, anxiety disorders, complicated grief reactions, or bipolar disorders may also be counted as false positives in some studies, but they might well be identified by the more careful and in-depth assessment that would presumably follow a positive screen. If, however, treatment for depression is initiated on only the basis of screening positive, then patients with other related illnesses may receive suboptimal care.

Using Risk Factors to Identify Patients with Depression

Because the prevalence of depression is only 5% to 10% in primary care settings, some experts have suggested that the presence of known risk factors for depression be used to determine who should or should not be screened -- a strategy of selective screening. Although, intuitively appealing, most common risk factors for depression perform relatively poorly in discriminating patients who are depressed from those who are not depressed. Conde et al ¹⁰⁷ demonstrated that most common risk factors have positive likelihood ratios (LR) between 1 and 2 and negative likelihood ratios between 0.5 and 1, suggesting low predictive ability (Table 12).

Table

Table 12. Diagnostic Value of Risk Factors for Major Depression.

Other factors, such as a previous history of depression or concurrent diagnosis of panic disorder or generalized anxiety disorder, have positive LRs greater than 10; their presence warrants further investigation for depression, perhaps including a diagnostic interview. Their absence, however, does not significantly change the likelihood of depression.

Depression screening tools have a positive LR of approximately 3 and a negative LR of 0.2, demonstrating that they perform better than most of the common demographic risk factors. Based on these data, a strategy of selective screening does not appear to be superior to simply performing (or asking the patient to perform) one of the brief screening tools. In patients with previous depression or a current anxiety or panic disorder, directly proceeding to a full diagnostic interview may be warranted instead of initial screening.

Beck Depression Inventory (BDI)

Two studies looked at performance of the BDI in outpatient samples referred for psychiatric care;^109,110 most subjects were adolescents. Sensitivity was 48%, 86%, and 89% with corresponding specificities of 87%, 82%, and 88%. Positive predictive values were high (63%, 83%, and 93%) because of the high prevalence of depression in these referred patients.

Three studies used the BDI in general school samples of adolescents. The largest study included 1,704 Oregon high school students and used a BDI of ?11 for females and ?15 for males to assign a diagnosis of current depression (according to DSM-III criteria). ¹¹¹ Sensitivity was 84%; specificity, 81%. Positive predictive value was 10% and negative predictive value 99.5%. A small sample of 49 adolescents from a school population was a part of a study using the BDI to identify DSM-III major depression. ¹¹² Using a cut-off of 16, the investigators reported 100% sensitivity and 93% specificity for the BDI. Prevalence of depression was 10% (5/49 adolescents); positive predictive value was 61%. The third study of adolescent students used a BDI of ?16 to assess lifetime history of DSM-III major depression and dysthymia. ¹¹³ For depression, sensitivity and specificity were 77% and 65%, respectively. Prevalence of depression was 4%; positive predictive value was 8%. For dysthymia, sensitivity and specificity were 71% and 64%, respectively. Prevalence of depression was 5%; positive predictive value was 10%.

Finally, the only study conducted in a general primary care setting used a version of the BDI, the Beck Depression Inventory for Primary Care (BDI-PC) to assess major depression during 100 adolescent health maintenance examinations. ¹¹⁴ A BDI-PC cut-off of 4 yielded a sensitivity of 91%, a specificity of 91%, and a positive predictive value of 56% for the population with a high prevalence of 11%.

Center for Epidemiological Studies - Depression Scale (CES-D)

The CES-D is a 20-item scale developed for adults. The CES-D in children did not correlate well with the Children's Depression Inventory (CDI) and did not discriminate depressed and nondepressed patients adequately for use in children. ¹¹⁵

Two studies have described CES-D screening accuracy for depression in large school-based samples of adolescents. Roberts et al ¹¹¹ looked at CES-D scores in the Oregon sample that also used the BDI. Investigators applied a cut-off of 22 for males and 24 for females to identify current depression (DSM-III criteria) in 1,704 adolescents. Sensitivity was 84%; specificity was 75%; and positive predictive value was 8%.

Garrison et al ¹¹⁶ used a subsample of 332 students identified in a larger survey of adolescents in the Southeastern United States. Using various cut-off points, the researchers found that optimal screening characteristics for depression occurred at a cut-off point of 12 for males and 22 for females. For males, sensitivity was 85%, specificity was 49%, and positive predictive value was 13%. For females, sensitivity was 83%, specificity was 77%, and positive predictive value was 25%. Screening performance of the CES-D was also assessed for dysthymia using a cut-off of 16 for males and 20 for females. For males, sensitivity was 75%, specificity was 67%, and positive predictive value was 14%. For females, sensitivity was 100%, specificity was 67%, and positive predictive value was 8%.

The CES-D also has a version for children, the CES-DC. In 1 study of the CES-DC using a cut-off of 15, Fendrich et al ¹¹⁷ found the CES-DC to have a sensitivity of 71% and a specificity of 57%.

Other Screening Instruments

In a population of adolescents referred for psychiatric care, Angold et al ¹¹⁸ tested the Short Mood and Feeling Questionnaire in a mixed sample of 173 children and adolescents. They used the DISC as a criterion standard and found sensitivity of 70% and specificity of 85%.

Several other screening instruments have been used in children and adolescents, but most have not been used to screen a primary care sample of pediatric patients. These other tests include the Children's Depression Inventory (CDI), Child Depression Scale (CDS), Children's Self-report Rating Scale (CSRS), Depression Self-Rating Scale (DSRS), and Reynolds Adolescent Depression Scale (RADS). Studies of these scales have reported validation in psychiatric inpatient and referred samples, and so these instruments may be useful in some settings. However, the studies either do not report data in primary care populations or do not describe test performance results to address use as general screening tools. ¹¹⁹

The Pediatric Symptom Checklist (PSC) and Child Behavior Checklist (CBCL) have been shown to be feasible to implement in primary care practice and have relatively good sensitivity and specificity as a general screen of mental health needs. These tests may increase awareness of unrecognized psychosocial problems; however, they do not appear to perform well in identifying specific individual diagnosis such as depression.^120,121

Special Populations

Children with comorbid psychopathology or chronic medical illness and other pediatric subpopulations have been reported to have a higher prevalence of depressive disorders than the general population. Special populations may be candidates for targeted screening, but few studies report screening accuracy results. Sensitivity and specificity in psychiatric inpatient or outpatient groups are generally similar to the results presented above, although predictive values will be higher because of the higher prevalence of depression. ¹²²

One study in a population of chronically ill pediatric patients (an important subset of pediatric patients with higher prevalence of depressive disorders) evaluated test performance of the CDI, the PSC, and the CBCL. ¹²³ The authors found a high prevalence of depression and mental disorders and relatively good specificity of the measures at detecting depression, anxiety, or both (78% to 96%). They concluded, however, that low sensitivity of the tests (26% to 55%) limited their clinical usefulness for this patient population. Other, better performing depression scales have not been tested in children with chronic illnesses.

Summary of Screening Accuracy in Pediatric Populations

The existing literature suggests that screening instruments for depression in adolescents that have been tested in community or primary care settings perform reasonably well. They produce sensitivity values ranging from 75% to 100% and specificity values from 70% to 90%, values similar to those found in adults, although there are fewer studies and fewer total subjects. Fewer data are available for children. The prevalence of disease and the positive predictive value in children are quite low, but the values rise in adolescents. Like adults, those who screen positive should undergo a more rigorous diagnostic interview before being labeled as depressed.

Pharmacologic Interventions

Details about pharmacological treatment studies that met our inclusion criteria can be found in Evidence Table 2 (Appendix D). The discussion below is presented first for large-scale reviews (for which data have not been provided in Evidence Tables) and then for other studies; for the latter, the 6 entries in the Evidence Table (which cover 8 publications) are presented in reverse chronological order.

Psychotherapy Interventions

Evidence Table 3 (Appendix D) presents information on 13 studies of psychotherapy (covering 15 publications); the entries appear in alphabetical order. We present the discussion below in terms of studies on major depression, minor depression, dysthymia, and/or other depressive conditions.

Major Depression

Eleven of the 13 studies of psychotherapy involved patients with major depressive disorders (Evidence Table 3). The five studies that also included medication trials are described with respect to the medication efficacy in the previous section; the outcomes of psychotherapy are described below. As shown in Table 14, the more effective interventions tended to have a more highly structured intervention than is typically the case; that is, the more effective approaches were well formulated, limited in time, and standardized in application, and they tended to have clearly defined goals and stages. Only 6 studies used intention-to-treat approaches.^{74,75,78,85,88,89} The studies are reviewed below in the order of decreasing magnitude of effect and decreasing stringency of outcome measures (eg, recovery is more stringent than reduction in depressive severity).

Table

Table 14. Studies of Psychotherapy in Patients with Major Depressive Disorders.

Mynors-Wallis and colleagues ⁷⁷ compared 6 treatment sessions of well-structured PS therapy, guided by a treatment manual and provided by either an experienced psychiatrist or trained GPs, against usual care. As noted earlier, all groups (including the pharmacologic arm) had 3.5 hours of contact time. No intention-to-treat analysis was done. At 12 weeks, 60% of those in the PS group had recovered compared to 27% of those in the usual care arm.

Holden et al ⁸⁴ compared counseling by health visitors to usual care in a trial involving women with postpartum major or minor depression. The health visitors had limited training and provided 8 weekly sessions of an unstructured, supportive intervention of at least 30 minutes duration. The therapy was not administered according to any standardized manual or approach. Approximately two-thirds of each group had patients with major depression at the start of the trial. No intention-to-treat analysis was performed: 55 women were randomized; of these, 50 completed the trial and were included in the results. At 13 weeks, 69% of the health visitor group and 38% of the usual care group had neither major or minor depression as assessed by Research Diagnostic Criteria. The results did not distinguish between major and minor depression.

Scott et al ⁸⁹ compared six 30-minute cognitive therapy sessions to usual care. No manual was used, but the treatment was relatively well structured and a random sample of psychotherapy tapes were reviewed to ensure quality. In an intention-to-treat analysis, at 7 weeks 62% of the group randomized to cognitive therapy had recovered, as had 33.3% of those with usual care. Follow-up was also assessed at 58 weeks in a treatment-completer analysis, and the psychotherapy arm had significantly lower depressive severity (HAM-D=6.1) than the usual care arm (HAM-D=10.7). Of note was the large attrition rate at 1-year follow up (16/28 in cognitive therapy group, 8/24 in usual care group).

Katon et al ⁸⁵ evaluated a brief CBT intervention as part of a multi-faceted primary care intervention for major or minor depression that included on-site education and consultation for physicians about antidepressant and behavioral treatment of depression. Analysis was intention-to-treat. The psychotherapy intervention was geared toward improving medication adherence and consisted of 4 to 6 meetings with a psychologist for a total of 2.5 to 3.5 hours plus 4 telephone contacts. Outcomes for patients involved in this program were compared to outcomes for patients receiving usual care for the same conditions. For major depression, 70.4% of those receiving the multi-faceted intervention involving CBT had a greater than 50% decrease in depressive severity at 4 months compared to 42.3% of those in the usual care group. The effect size was smaller for minor depression (66.7% improved with therapy, 52.8% with usual care) and did not reach statistical significance.

Mynors-Wallis and colleagues ⁷⁵ compared 6 sessions of well-structured PS therapy by a trained GP to 6 sessions by a trained nurse, to antidepressant medication alone, and to a combination of the medication and PS therapy. Therapy was provided in either the patient's home or the local health center. The first PS sessions lasted 1 hour; subsequent sessions lasted 30 minutes. Patients receiving PS therapy alone had a mean number of 4.6 treatment sessions (2.8 hours total contact time); those receiving combination treatment had a mean number of 5.2 PS treatment sessions (3.1 hours contact in addition to medication management time).

After 3 months of treatment, 51% of the PS-GP group and 54% of the PS-nurse group had recovered (HAM-D <7), compared to 67% of the medication alone group and 60% of the combination group. At 1-year follow-up, 62% of the PS-GP group had recovered, as had 56% of the PS-nurse group, 56% of the medication alone group, and 66% of the combination group. As described before, the 4 groups did not differ significantly in terms of rate of recovery, suggesting that combination treatment for routine depressive illness in primary care is no more effective than a single intervention, and that outcomes will not differ between PS therapy delivered by a trained GP and that delivered by a trained nurse. These findings are in contrast to recent research in specialty settings suggesting benefit for combination in certain situations, such as preventing recurrence of depression in a geriatric psychiatry setting ¹⁴⁰ and in treating chronic depression in an outpatient psychiatry setting. ¹⁴¹

Schulberg et al ⁷⁸ compared 16 weeks of IPT delivered by doctoral-level, experienced therapists using a well-structured, standardized protocol to usual care. In an intention-to-treat analysis, 46% of those randomized to the IPT group recovered as did 18% of the usual care group.

Ross and Scott ⁸⁸ tested individual cognitive therapy (consisting of 12 sessions lasting 45 minutes over 3 months) or group cognitive therapy (12 sessions lasting 90 minutes over 3 months) to usual care. All treatment was delivered by the same experienced social worker; it is unclear if the treatment was structured. All groups appeared to improve. Following the 3-month intervention period, those receiving cognitive therapy appeared to have significantly greater reductions in depressive severity than usual care (32% reduction on HAM-D vs 17%, P <0.01; intention-to-treat analysis). The individual and group forms of treatment did not differ significantly. For the subset of patients who had been assessed 12 months after completing treatment, benefits appeared to be maintained, although no usual care group was available for comparison.

The Appleby et al ⁷⁴ study did not distinguish between patients who developed major or minor depression postpartum. Those receiving 6 sessions of minimally structured CBT totaling 3.5 hours by a nonspecialist with minimal training experienced a 64% decrease in the HAM-D score at 12 weeks compared to a 57.7% decrease for those receiving a single, 1-hour CBT session from a nonspecialist. These results were slightly less robust than the pharmacologic intervention. Again, significance was not reported for the intention-to-treat analysis. For the patients completing treatment (30% attrition), HAM-D scores decreased by 76% for those with 6 sessions, a significantly greater decrease than the 66% drop for the 1-session group.

Scott and Freeman ⁸¹ compared cognitive therapy delivered by a psychologist or supportive counseling delivered by a social worker to usual care. Neither the cognitive treatment nor the counseling was provided according to a formal manual or otherwise clearly structured. Analysis was not intention to treat. Over the 16-week course, the cognitive intervention averaged nearly 7.75 hours and the social work counseling more than 12 hours, compared to less than 1 hour by the general practitioners. At 16 weeks there was no difference in percentage recovered between the cognitive therapy group and usual care (41% vs 48%), but the social work group (72%) produced substantially higher rates of recovery.

Teasdale et al ⁹⁰ compared up to 20 one-hour sessions of cognitive therapy (mean 15.2 hours) delivered by doctoral-trained, experienced psychologists to usual care for patients with major depression in a primary care setting. The investigators ensured adequacy of treatment by tape review and did not employ a structured manual. Analysis was not done on the basis of intention to treat. Immediately post-treatment, patients in the therapy group averaged a greater change in depressive severity on the BDI than did the usual care group (22 point decrease vs 11.5 point decrease, P <0.01). This benefit was not apparent at follow up three months after completing treatment. Of note, contact time for the therapy group was substantially greater than for usual care.

Blackburn and colleagues ⁸³ compared the outcomes for patients receiving either a pharmacologic intervention (the TCA amitriptyline) or only cognitive psychotherapy to outcomes for patients receiving combined cognitive psychotherapy and amitriptyline; all patients had a diagnosis of major depression. Psychologists performed 12 to 20 sessions of therapy; no manual was used and the degree to which the treatment was structured is unclear. An intention-to-treat analysis was not done, and allocation to therapists was not randomized. Subjects in all 3 groups showed benefit, but patients in the cognitive therapy group and the combined treatment group tended to have a greater decrease in depressive severity than the amitriptyline-only group. Specifically, using a more than 50% decrease in depressive severity immediately post-treatment as the outcome of interest, 81.8% of the combined group, 72.7% of the cognitive therapy group, and 55% of the medication group achieved that outcome (overall chi-square test was not significant). The cognitive therapy and combined groups appeared to have substantially more visits than the medication-only group. Attrition during the trial was 27%.

Educational and Quality Improvement Interventions

Five studies examined health care delivery strategies that did not directly involve traditional medication or psychotherapeutic interventions.^129,132,133 Katon and colleagues ¹²⁹ tested a "Collaborative Care" model that included patient education, on-site consultation for patients, active collaboration with primary care physicians, and increased frequency and intensity of primary care visits. At 4 months, significantly more patients with major depression who received care through the collaborative care model had a greater than 50% decrease in depressive severity than did patients on usual care (74.4% vs 43.8%). The authors reported no significant difference for patients with minor depression (60% vs 67.9%).

Llewellyn-Jones and colleagues ¹³² tested a "Shared Care" model for "depressed" patients involving caregiver education, health education and promotion for patients, and improved communication between general practitioners and staff at a single elderly residential care facility. Their design examined control and intervention groups in a serial fashion. The intervention was "population-based" in that it was targeted to the entire living facility. Participation was variable: only 62% of either study group had general practitioners who attended the provider education program. The intervention itself was relatively inexpensive. Compared with patients in the control group, patients in the multi-faceted intervention group had a significantly greater reduction in their GDS scores (by 1.87 points) and were more likely to move to a "less depressed" state (45% vs 31%).

Peveler et al ¹³³ tested the benefits of 2 sessions of counseling about antidepressant medication adherence, or the provision of an information leaflet about adherence, versus usual care in a population with "depressive illness." No difference in depressive symptoms as measured by the Hospital Anxiety and Depression Scale ¹⁴² was found between treatment groups overall. However, among patients with major depression who received higher doses of medication, those in the counseled group had significantly lower final depression scores than those with usual care (4.0 vs 5.9).

Katzelnick et al ¹³¹ compared the benefits of a systematic, primary care-based depression treatment program for depressed "high utilizers" not in active treatment. This depression management program (DMP) consisted of patient education materials, physician education programs, telephone-based treatment coordination, and antidepressant medication treatment initiated and managed by the patients' primary care physician. Those receiving the DMP were compared to a usual care arm in an intention-to-treat analysis. The DMP group was significantly more likely to fill 3 or more antidepressant prescriptions in the first 6 months (69.3% vs 18.5%, vs, P< 0.001) and had significantly greater improvement in HAM-D depressive severity scores at 1 year (−9.2 vs −5.6, P< 0.001), with this benefit beginning by 6 weeks into the study. Additionally, at 1 year, intervention patients were more improved on mental health, social functioning, and general health self-report measures (P< 0.05 for each domain). Of note, mean visits counts in the DMP increased by 1.6 visits, whereas mean visits counts decreased in the usual care group by 2.0 visits (P=0.02).

Simon et al ¹³⁰ compared a program of feedback only and 1 of feedback plus care management to usual care in primary care patients with recently diagnosed depressive illness. The feedback-only intervention consisted of feedback and algorithmic recommendations to doctors at 8 and 16 weeks based on data from computerized records of pharmacy and visits. The feedback plus care management group additionally provided to patients 2 later telephone monitoring contacts (at 8 and 16 weeks), which were followed by more sophisticated feedback to the doctor based on information received during the phone call.

In an intention-to-treat analysis compared to usual care, the care management group had a higher probability of receiving at least moderate doses of antidepressants (OR, 1.00; 95% CI, 1.23 to 3.22). The care management group also had a significantly higher probability of showing a 50% decrease in depression severity (OR, 2.22; 95% CI, 1.31 to 3.75) and a significantly lower probability of persistent major depression (OR, 0.45; 95% CI, 0.24 to 0.86) at 6 months. Meanwhile, relative to usual care, the feedback-only group showed no difference on receiving at least moderate doses of antidepressants (data not provided), the probability of a 50% decrease in severity (OR, 1.12; 95% CI, 0.73 to 1.73), or the probability of major depression at follow-up (OR, 0.89; 95% CI, 0.55 to 1.46).

Conclusions about Therapies for Adults

Effective treatments for depressive illness in primary care are available. Antidepressant medications for major depression are clearly effective compared with placebo. Most of these results have come from structured efficacy trials with selected populations, although more recent studies using usual-care comparison groups and real-world settings have produced similar effects.^76-78

Antidepressant interventions for dysthymia are probably effective in primary care patients; although only 2 studies have been performed in primary care settings, evidence from multiple sites (inpatient psychiatric hospitals, outpatient psychiatric clinics, primary care practices, and the community) show a similar magnitude of effect. The evidence regarding the benefit of antidepressant medication for minor depression is limited. The 1 trial addressing this question (the Collaborative Care model, ¹²⁹ in which improved medication prescription and adherence was part of the intervention) did not find a statistically significant benefit with antidepressants, but it may have been underpowered to detect a modest but clinically important effect (10% to 15%).

Tricyclic agents and newer agents (including SSRIs) have similar efficacy. The newer agents, however, have fewer side effects and are less likely to have side effects that lead to drop-out. Total drop-out rates, however, did not differ. Of note, the 1 effectiveness study (which most closely represented actual practice in primary care by allowing naturalistic follow-up and management by primary care physicians) ¹³⁹ found that the drop-out rates for the tricyclic-treated patient were much higher than those for the SSRI-treated patients. For patients with major depression, greater side effects lead to significantly higher drop-out rates from treatment, although similar drop-out rates were not noted for patients with dysthymia.

Psychotherapeutic interventions appear as effective as antidepressant interventions for major depression, with a similar magnitude of effect. In general, the more effective psychotherapeutic interventions had greater structure to their treatments. Relative to pharmacologic interventions, psychotherapeutic interventions were clearly more time intensive. Four studies used between 4 and 6 sessions totaling 3 to 4 hours for their interventions.^74,85,89,135

Evidence on the effectiveness of psychotherapeutic intervention for patients with minor depression is limited, although the results of 2 studies using well-structured interventions suggest potential benefit.^86,87 No evidence exists concerning the use of psychotherapy alone for dysthymia.

Few studies have examined the effect of combining medications and psychotherapy. Two studies involving combined treatments did not find a significant incremental benefit when compared to a single active intervention.^74,83 However, 2 recent trials in psychiatry clinic settings suggest that combination therapy may improve long-term outcomes.^140,141

Preventing Depression

One method of reducing the impact of depression is to treat risk factors and symptoms before they lead to a full episode of major depression. Some studies, described below, provide limited evidence on this approach for children and adolescents (eg, intervening with children with subclinical depression or providing assistance with coping skills for children at risk of depression).

Jaycox et al ¹⁴³ reported reduction in depressive symptoms in the Penn Prevention Program, a prospective cohort study of 142 children ages 10 to 13 years. They used CBT to teach coping strategies to 69 "at-risk" children in a treatment group. At-risk children were selected based on depressive symptoms and reports of parental conflicts. The treatment group was compared to 73 control children who did not receive any intervention. Children were not randomized to intervention. Outcomes were assessed after 6 months using the Children's Depression Inventory (CDI). In the treatment group, the percentage of children who were moderately depressed (CDI >15) decreased significantly from 24% to 15% (P<0.05); in the control group the change in percentage of depressed subjects was not significant (24% to 23%, P=0.36). Based on self-reported depressive symptoms in the 6 months following the intervention, 23% of the children in the treatment group and 44% of the control group reported moderate depressive symptoms (P<0.05).

Clarke et al ¹⁴⁴ were able to demonstrate positive results in adolescents with depressive symptoms at risk for developing a DSM-IIIR-defined episode of depression. The intervention consisted of assessment of symptomatology by the CES-D with subsequent K-SADS diagnosis of depression or dysthymia. The investigators randomly assigned 172 adolescents with subclinical depression to a usual-care control group or an after-school cognitive psychotherapy group. Total incidence of major depression or dysthymia during follow up was 18 of 70 children (25.7%) in the control group and 8 of 55 (14.5%) for the intervention group.

Lamb et al ¹⁴⁵ conducted a school-based program designed to promote coping among rural adolescents with depressive symptoms. The study surveyed 222 students ages 14 to 19 years and identified a subgroup of subjects with moderate to high Reynolds Adolescent Depression Scale (RADS) scores who could be randomly assigned to treatment or control groups. The treatment consisted of 8 weeks of group sessions using coping techniques and role-playing tasks. Four students dropped out of the treatment group; 1 left the control group. The investigators found that 87% of the intervention group and 61% of the control group improved on RADS scores. These results were significant for females (P=0.032) but not for males.

Psychotherapy

Psychotherapy has been the mainstay of treatment for children diagnosed with depression. Various forms of psychotherapy and counseling have been used. CBT is the method that has been studied most rigorously and been shown to be effective.

Reinecke et al ¹⁴⁶ recently reviewed evidence on CBT in a systematic review and meta-analysis. The authors identified 6 controlled clinical trials with 14 post-treatment control comparisons and 10 follow-up control comparisons covering 217 subjects.^147-151 All studies were conducted in adolescents ages 10 to 19 years. All but 1 study recruited subjects in schools and used group therapy sessions. The interventions lasted 5 to 8 weeks and included 6 to 14 sessions with follow-up periods of 1 to 3 months. Outcomes were based on different depression scales.

The overall pooled effect size (a measure of change in standard deviations) at post-treatment was -1.02 (95% CI, −1.23 to −0.81) and for follow-up data −0.61 (95% CI, −0.88 to −0.35). Negative effect size scores indicated a decrease in combined depression measures and improvement of symptoms in terms of standard deviations. Thus, CBT appears to be effective in reducing depressive symptoms among adolescents. Treatment gains seem to be maintained after completion of therapy. The results of this meta-analysis were consistent with other meta-analyses of psychotherapy for depression in children and adolescents.^152,153

Pharmacotherapy

Additional Considerations

This review has attempted to describe generally the identification and management of children who present in primary care, but special patient populations should be considered. Gender, age, and ethnicity are important variables in existing studies that may limit generalizability of results. Many of the above studies did not have large numbers of minorities or patients of lower socioeconomic status. Most of the positive studies and results are based on adolescents and older children. Aside from case reports and series, very few data are available about interventions in young children. Individual characteristics should be considered before the results on any larger population are generalized or applied to a specific patient.

Finally, children with poor health and chronic illnesses have been reported to have higher rates of depression and mental health problems. It is very important to consider depression and comorbid effects on chronic medical conditions in terms of adherence to medical treatment, functionality, and outcomes. However, in pediatric patients with chronic illness, screening tools for depression appear to lack sensitivity and predictive value and thus cannot be recommended for routine use. ¹²³ In addition, studies are not yet sufficient to document treatment effectiveness in these patients.

Conclusions for Children and Adolescents

Data on prevention of depressive disorders in school and community settings provide support for intervention on selected youths with depressive symptoms, although no studies have described this type of intervention in primary care settings. The approach most relevant to primary care involves early recognition of depressed patients, proper identification and diagnosis, and facilitation of effective treatment.

Treatment of depression in adolescents with CBT or SSRIs appears to be effective. Whether these results can be generalized to primary care settings or to children is unclear. TCAs are not effective for treatment of depression in children and adolescents. The comparative efficacy of psychotherapy alone, medications alone, or combined treatments in children or adolescents is unknown.

Effect of Screening on Recognition and Diagnosis of Depression

Seven studies examined the effect of screening, compared to usual care, for the diagnosis of depression (Table 15a and 15b). Moore et al, ⁹⁸ Linn and Yager, ⁹⁵ and Magruder-Habib et al ⁹⁷ all used the Zung SDS screener. Callahan et al^91,92 and Williams et al ¹⁰² used the CES-D. Dowrick used the BDI. ⁹³ Whooley et al ⁸² used the GDS. In each study, the detection of depression was assessed by chart audit.

Table

Table 15a. Studies Examining the Effect of Screening and Feedback.

Table

Table 15b. Summary of the Effect of Screening and Feedback on Rates of Diagnosis.

Moore et al ⁹⁸ screened consecutive patients, 20 to 60 years of age, at a university-based family medicine residency program. All patients were asked to self-administer the Zung SDS. The intervention patients' providers received feedback about SDS results greater than 50; providers of patients who scored below 50 and of all control patients simply received notice that their patients had been screened. No attempt was made to confirm the diagnosis using a criterion standard. Of 212 subjects in the trial, 96 scored above 50 (45%). Recognition of depression, defined by any notation in the chart, was 56% for cases in the intervention group (28/50) and 22% in the control group (10/46). The difference between intervention and control groups was similar for "severe depression," but rates of detection in both groups were higher (73% vs 37%). Effect on treatment rate and outcomes was not described.

Linn and Yager, ⁹⁵ in testing the self-administered Zung SDS, randomized 150 consecutive new patients from a primary care clinic to feedback or no feedback. They found that patients assigned to the feedback group were more likely to have depression diagnosed (29% vs 8%) than the no-feedback patients, but they did not employ any criterion standard.

Magruder-Habib et al ⁹⁷ screened 800 Veterans Administration primary care clinic patients for depression. Research assistants administered the Zung SDS and used the DIS to confirm diagnosis with DSM-III criteria. Patients with SDS scores greater than 75 were excluded from randomization. The 100 patients who screened positive and met DSM-III criteria for major depression were then randomized to feedback or usual care. Those patients whose physicians received feedback were 3 times as likely to be accurately identified as depressed at the outset than were those whose clinicians had not received such feedback (25% vs 8%). At 1-year follow-up, 42% of the intervention patients, but only 21% of the controls, had been recognized as depressed.

Callahan et al^91,92 conducted an RCT of feedback from screening plus targeted educational information and treatment recommendations for patients over age 60 years in an academic primary care setting that served a low-income population. Potential subjects were screened by research assistants using the CES-D and HAM-D depression scales. Those patients scoring above the threshold for diagnosis were eligible to be randomized. Randomization was by physician, with certain clinic sessions randomly assigned to the intervention and others to control. All physicians received an educational talk at baseline.

Two articles appear to report results from this study. The first article, based on a 175-patient sample, found that patients in the intervention group were more likely than the control group to have a new notation of depression in their charts (32% vs 12%). ⁹¹ In the second paper, additional analyses on a larger sample size (n=222) found higher rates for documentation of depression (87% vs 40%). ⁹²

Williams et al ¹⁰² used the CES-D or a single question about depressed mood to examine the effect of feedback to providers for adult primary care patients. Most patients were able to complete the single question (90%) and the CES-D (54%) without assistance. The presence or absence of depression was later confirmed using the DIS and DSM-IIIR criteria. ¹⁰² Current depression was defined as either meeting the DSM-IIIR criteria for major depression or dysthymia or having minor depression (depressed mood or anhedonia plus 1 to 3 additional DSM-IIIR symptoms). Based on chart reviews, current depression was recognized in 39% of patients whose providers received feedback from screening and in 29% of controls. This difference of 10% in the rate of recognition did not reach statistical significance.

Dowrick ⁹³ randomized 116 patients who were initially rated "not depressed" by their usual general practitioners but had BDI scores greater than 14. Feedback was provided 1 week after the visit in which screening took place and was noted in the chart for subsequent visits. The study was powered to detect a 30% difference in the level of diagnosis after feedback. There was a higher level of depression diagnosis at 1 year in the feedback group (35% vs 21%; OR for detection, 2.10; 95% CI, 0.84 to 5.28), but the difference did not reach statistical significance.

Whooley et al ⁸² randomized primary care clinics to screening with physician feedback versus no screening or feedback for patients over age 65 years screened with the GDS. No criterion standard was applied. They found no difference in the rate of diagnosis of depression at 2 years.

In conclusion, feedback of screening results to providers increases the recognition of depression, especially major depression, by a factor of 2 to 3 in all cases except for the trial by Whooley et al. ⁸² The absolute increases in the diagnosis of depression range from 10% to 47%, with larger differences for major depression. Recognition and diagnosis of minor depression, when assessed, were generally low in both intervention and control groups.

Effect of Screening on Treatment of Depressed Patients

The effect of feedback of screening results on the proportion of depressed patients who receive treatment was examined in 7 studies (Tables 15a and 15c). Treatment generally included prescription of pharmacologic antidepressant therapy or referral to mental health services. Most studies evaluated treatment by chart audit; some used pharmacy databases. Actual patient adherence was not directly measured.

In contrast to recognition and diagnosis, the effect on rates of treatment was mixed. In 3 studies (Linn and Yager; ⁹³ Dowrick; ⁹⁵ Williams et al ¹⁰² ), the documented rates of treatment were nearly equal in the intervention and control groups (Table 15c). Other studies, however, found improvements in the rate of treatment, with increases in the prescription of antidepressant medication more common than changes in mental health referrals. Callahan et al,^91,92 using a stepped program of treatment recommendations in addition to the feedback, found a difference of 17% to 18% in the initiation of a treatment plan and an increase in 12% for the rate of antidepressant prescription (P=0.01). Magruder-Habib et al ⁹⁷ found an initial difference of 24% in the rate of treatment, although at 1 year it declined to a difference of 14% (56% vs 42%). The Williams et al ¹⁰² study also did not find an overall difference in treatment.

Table

Table 15c. Summary of the Effect of Screening and Feedback on Rates of Treatment.

Wells et al ¹⁰¹ studied the effect of combining screening and a quality improvement program for depression treatment in 46 primary care clinics and measured its impact on treatment and outcomes of depression. Patients were enrolled if they screened positive on a 2-question screener. Patients received the Composite International Diagnostic Interview (CIDI) criterion standard examination, but participation was not based on its results. Randomization was at the level of the practice, and the intervention included feedback on the results of the 2-item screener. Intervention practices also received educational materials and assistance with quality improvement in treatment initiation and maintenance plus access to nurse-led medication follow-up or to cognitive-behavioral therapy. The investigators screened 27,000 patients, identified 3,918 as potentially eligible, and randomized 1,356 patients. Subjects were followed for 12 months. The proportion of patients receiving appropriate treatment was increased in the intervention group at 6 months (50.9% vs 39.7%) and at 12 months (59% vs 50%, P=0.006).

Effect of Screening on Depression Outcomes

The effect of screening and feedback on depression outcomes was measured in 8 studies (Tables 15a and 15d).

Table

Table 15d. Summary of the Effect of Screening and Feedback on Rates of Patient Outcomes.

Johnstone and Goldberg ⁹⁴ applied the GHQ to 1,093 primary care patients and identified 119 cases of depression. These 119 subjects were randomly assigned to feedback of the results to the physician or to usual care. The investigators found no difference in mean GHQ scores at 12-month follow-up, but they did see a larger improvement with feedback among the subset of subjects with severe depression. For all patients, the mean duration of the first episode of depression and the total amount of time depressed were decreased by approximately 2 months (P<0.01).

Zung and King ¹⁰³ screened 499 patients at a single private physician's practice. Of the 60 who screened positive, 49 were confirmed to have major depression using DSM-III criteria and were randomized to feedback and treatment with the benzodiazepine alprazolam (n=23) or to usual care (n=26). Four weeks later, outcome data were available for 20 patients in each group. The feedback and treatment group was more likely than controls to improve by at least 12 points on retesting with the Zung scale (66% vs 35%, P <0.05).

In Callahan et al^91,92 no improvements in HAM-D score emerged among those who received feedback of screening results.

Reifler et al ¹⁰⁰ used the SDDS-PC, followed by a depression-specific diagnostic module, in 358 primary care patients. The 186 intervention patients had a lower mean number of visits than the 172 controls (3.7 vs 5.3, P=0.06), but other outcomes including SF-36 or SDS scores did not differ.

Lewis et al ⁹⁶ used the GHQ and a computer-based diagnostic tool (PROQSY) to examine the effect of feedback of positive scores on outcomes in low-income primary care patients in London. Compared with GHQ scores for controls at 6 weeks, GHQ scores were lower for patients whose providers received feedback on the PROQSY results but not for those who received only GHQ results. The differences were attenuated and nonsignificant at 6-month follow-up.

Williams et al ¹⁰² found a statistically nonsignificant difference of 9% in the proportion of subjects still depressed at 3 months. The rate of recovery (patients with 1 or no DSM-IIIR criteria), however, was higher in the intervention than control groups (48% and 27%, respectively; P <0.05).

Whooley et al ⁸² found little difference in the proportion of patients depressed on the GDS after 24 months of follow-up: 42% for intervention patients and 50% for controls (P=0.3).

Wells et al ¹⁰¹ found statistically significant increases in the proportion of intervention patients (intervention practices received feedback of screening results and a quality improvement intervention) who were not depressed at 6 and 12 months and in the rate of job retention. Based on CES-D scores, intervention subjects were less likely to be depressed at 6 months than controls (55% vs 64%, P=0.001) and at 1 year (55% vs 61%, P=0.04). Among patients initially employed, 90% were still working, as compared with 85% of controls. ¹⁰¹

Based on the results of Wells et al ¹⁰¹ , approximately 10 to 12 patients identified as being depressed by screening would need to be treated to produce 1 additional remission. Twenty patients would need to be treated to preserve 1 patient's job. If depression is present in 5% to 10% of primary care patients, 100 to 200 patients would need to be screened to produce 1 additional remission at 6 months.

Conclusions about Screening and Feedback

In summary, multiple studies have examined the effect of providing feedback of depression screening results to providers in primary care. The rate of detection and diagnosis of depression, based mainly on chart reviews or the completion of a study-specific form, increased by 10% to 47% in the 6 studies reporting this outcome. The effect on treatment was more variable. Four of the 8 studies reporting this outcome found small, nonsignificant increases in the proportion of patients treated for depression.^93,95,102 Magruder-Habib et al ⁹⁷ found a much larger increase (24%), and Callahan et al ⁹¹ noted increases in antidepressant prescribing but not referral for counseling or psychiatric care. Wells et al also noted a 10% increase in appropriate treatment, which was statistically significant.

The effects of depression screening on clinical outcome of depression were also mixed. Two small, older trials found large improvements in major depression.^94,103 Two larger, well-designed trials found moderate improvements (9%) in remission from depression in a population with a mixed set of diagnoses.^101,102 Four other studies found small or no improvements in outcomes.^82,91,92,100

Thus, although the effect of screening on diagnosis appears robust, improvements in more distal variables such as treatment and outcomes are not as consistent or as large. Translating the increased rates of detection with screening into improved outcomes may require that particular attention be paid to initiation and maintenance of effective therapy, perhaps in the form of a quality improvement effort or other programs systematically designed to provide appropriate care.

Demonstrating improvements in clinical outcomes (as measured by the proportion still depressed, for example) requires large samples. Studies with smaller sample sizes may be unable to demonstrate statistically significant results despite finding clinically significant differences in recovery.

Major depression appears more responsive to intervention with screening and feedback than minor depression, although the Wells et al ¹⁰¹ study suggests that outcomes can be improved for all subjects with sufficient attention to treatment. The appropriate outcome measure for minor depression differs from major depression, so failure to demonstrate changes in the proportion of patients depressed may not be a fair test for patients with subsyndromal illnesses.

Screening Outcomes for Children and Adolescents

No studies have examined the overarching question of treatment outcomes for children or adolescents identified by primary care providers using targeted screening or clinical suspicion. A large part of the literature focuses on development of screening measures and reliability testing; it does not provide information to assess screening accuracy or sample a general ambulatory population that generalizes to primary care settings. No randomized trials in children or adolescents evaluate the effects of screening for depression on outcomes of recognition, diagnosis, or treatment. No studies in pediatric patients have linked an initial screening assessment for depression with subsequent treatment and demonstrated improved patient outcomes as a direct result of screening. Some studies have shown that screening instruments, especially the relatively brief general measures such as the CBCL and PSC, may increase recognition of mental disorders and referrals; however, there is no evidence that these general screens of psychopathology can improve outcome of depressed children or adolescents.

Brief screens for depression, such as versions of the BDI and CES-D, have been used in children and adolescents. However, their predictive value in general populations with relatively low prevalence of depression may limit their effectiveness and usefulness as a screen for all pediatric primary care patients. Targeted screening or use of measurement instruments on patients with suspected psychiatric disorder can improve diagnostic accuracy, but whether selective screening produces improved outcomes compared to usual care remains untested.

In addition to specific measures of depression, 2 general instruments that seek to identify psychosocial issues have been extensively researched and implemented in primary care.