Key Question 1: Is there direct evidence that screening for depression among adults and older adults in primary care reduces morbidity and/or mortality?

Summary of findings. One fair-quality randomized controlled trial of primary care patients identified by the previous systematic review¹⁰⁸ compared screening’s impact with a non-screened usual care group (Table 3).¹⁰ Nine hundred sixty-nine patients were randomized, 863 of whom completed a post-visit interview. A subset of the randomized patients from one of the two sites, over-sampling those with depressive symptomatology at baseline, were re-assessed after 3-months.

Table 3

Summary of results for key question 1: study examining health outcomes of a screening program in a primary care setting.

At follow-up, screened patients who were depressed at baseline were more likely to be in complete remission than unscreened depressed patients (≤1 symptom of depression in 48 percent of those screened compared to 27 percent of those not screened, p<0.05). However, when the entire follow-up sample was considered (not merely those who were depressed at baseline), screened and unscreened samples did not differ in the proportion of patients meeting DSM-III-R criteria for a depressive disorder nor in the average change in number of depression symptoms at follow-up. The study’s follow-up procedures may have limited their power to detect group differences and introduced bias. This study was also at risk for intervention contamination because providers saw patients in both study conditions. (See Appendix F Table 1 for key elements of the trial and Appendix G Table 1 for detailed information)

Study details. Table 4 lists brief information regarding depression screening and interview instruments described in this and subsequent sections. Williams and colleagues¹⁰ conducted the only study that compared a screened sample to a non-screened sample, which is the optimal comparison for looking at the effects of mass screening. Participants were recruited at two sites: a VA general internal medicine clinic in San Antonio, Texas and a university-affiliated general internal medicine clinic in Washington, DC. Patients were randomized either to usual care, to complete a brief (one item) depression screener, or to complete the Center for Epidemiologic Studies-Depression Scale (CESD) before a scheduled appointment. In both screening groups, the results were placed in the patient’s chart on a bright orange form, regardless of whether the screening test was positive or negative. Providers saw both intervention and control participants and therefore contamination is possible. All providers were given a guide for managing depression in primary care and a continuing education session on diagnosing depression. No other depression care support was provided.

Table 4

Depression symptom rating scales and diagnostic interview tools.

After the baseline visit with the provider, research staff attempted to interview all randomized (n=969) participants by phone to determine whether they met DSM-III-R criteria for depression; 863 (89.1 percent) completed the post-visit interview. Three months after the index visit, they attempted to contact 230 participants (26.7 percent, 230/863), which included all patients at one study site who met DSM-III-R criteria for a depressive disorder at the post-visit phone interview (n=101), and a random sample at the same site (n=129) who did not meet criteria for depression, over-sampling those who had depressive symptoms without meeting DSM-III-R criteria for MDD, Dysthymia, or Minor Depression. The completion rate for the 3-month follow-up was 94 percent. Because only a subgroup of the non-depressed patients was followed from only one study site, the health outcomes for this study cannot be considered randomized comparisons. Data were not presented on the baseline comparability between the follow-up sample and the original sample, or between the intervention and control groups among those in the follow-up sample, so we can not be as certain that follow-up group differences represent the intervention’s effect.

According to the table of baseline characteristics, participants were primarily middle-aged females of Hispanic background and were largely low-income, with only 24 percent of the sample reporting annual incomes of ≥$16,800 at that time (late 1990s). They also had high rates of positive screening tests (41 percent on the single-item screener and 33 percent on the CESD) and confirmed DSM-III-R depression diagnoses (13 percent).

This study found mixed results. Among patients who were depressed at baseline, patients in the two screening groups were more likely to have fully recovered from depression (i.e., reported ≤ one depression symptom) at 3-month follow-up than those who were not screened (48 vs. 27 percent, p<0.05). Among the entire follow-up sample, however, the proportion meeting DSM-III-R criteria for depression was similar in the combined screening groups (37 percent) and in the usual care group (46 percent) at three month follow-up (p=0.19), though they lacked adequate power to detect a population-level impact (n=218). Also, after controlling for baseline severity of depression (which differed between the screened and usual care groups in the full randomized sample), the mean reduction in DSM-III-R symptom counts was similar for the two groups (1.6 in screened vs. 1.5 in unscreened after controlling for baseline severity, p=0.21). While screening improved health outcomes in depressed patients, the effect was not large enough to create group differences at the level of the full follow-up sample. Since the follow-up sample included a considerably larger proportion of people who were depressed or symptomatic than the full sample, the effect would be even weaker in a full sample of primary care patients.

Key Question 1a: What is the impact of clinician feedback of screening test results (with or without additional care management support) on depression response and remission in screen-detected depressed patients receiving primary care?

Summary of findings. Two good-quality^109,110 and six fair-quality^111–116 randomized controlled trials reported providing the impact of giving depression screening results to clinicians on health outcomes in screened populations (Table 5). Four of these studies involved general adult populations (N=1,908)^{109,110,113,114} and four focused on older adults (N=1,443).^{111,112,115,116} Four of these were also included in the previous review. ^109–112 All studies randomized and/or enrolled patients who screened positive on at least one depression screening instrument which was often administered in the clinic waiting room. Although all patients were screened, feedback of screening test results was only given to providers of patients who screened positive (Appendix G Table 1).

Table 5

Summary of results for key question 1a: studies examining health outcomes of screening results feedback among screen-identified depressed patients in primary care.

In general adult populations, four trials screened a total of 38,843 primary care patients to detect 1,908 depressed adult patients (Table 5). Bergus and colleagues conducted a small, fair-quality RCT in a rural setting which did not support the effectiveness of screening programs with no depression care supports beyond simple feedback of screening results.¹¹³ Another small, fair-quality RCT improved depressive symptomatology, but had a highly selected participant sample because they only enrolled screened adults with newly-detected depression (i.e., patients already known to be depressed were excluded from the trial, as were those who were actively seeking treatment for depression).¹¹⁴ In this trial clinicians were given a detailed depression treatment protocol during the visit that included a recommended follow-up schedule and educational materials for the patient, and also received logistical support from other staff for scheduling follow-up visits and facilitating referrals. Two trials with considerably higher intensity interventions involving depression care by other staff were effective in improving depression outcomes,^109,110 particularly for adults with newly detected depression. These trials included such elements as intensive clinician and office support staff training, support staff or specialty mental health provider participation in ongoing depression care, and multiple follow-up contacts. See Table 6 for a summary of intervention elements included in each study. It was impossible to determine the degree to which the screening components contributed to the positive effects in the studies that also included additional depression care supports beyond simple screening and related interventions targeting the primary care provider.

Table 6

Summary of depression care support elements provided in programs of depression screening with feedback of results to providers.

In screening focused on older adults, four fairly large-scale trials screened 12,432 primary care patients to identify a total of 1,443 depressed older adults to test the impact of screening feedback with some care supports on remission and symptom reduction. The only trial with a significant treatment effect was also the only trial that expanded the role of other staff to provide depression management functions (in this case, assessment and regular follow-up). None of the trials in older adult patients limited enrollment to patients with newly identified depression. Intervention and usual care groups all showed some improvement from baseline, but only one of the four interventions in these trials improved depression remission rates or symptoms beyond usual care. The trial conducted in the Netherlands,¹¹⁵ provided no extra care support beyond clinician training and provision of a treatment protocol consistent with Dutch depression care guidelines.¹¹⁷ Although this trial did find an increase in the proportion of patients treated with antidepressants, no differences were found in depression remission or symptom severity at 12 months. Another trial offered a psychoeducational group to patients in addition to screening and provider feedback,¹¹¹ but patient participation in the psychoeducational group was minimal. The intervention group in this trial showed similar rates of depression remission and improvement to those who were simply screened without further intervention. The intervention in a third trial¹¹² included individually-tailored treatment recommendations, educational materials, and three scheduled follow-up visits with the provider. Adherence with follow-up visits was not reported. These supports were ineffective in a medically indigent, largely black, older adult population with multiple medical and psychiatric co-morbidities.¹¹²

Finally, the one trial¹¹⁶ that did report an improvement in depression symptomatology among older patients who screened positive for depression involved the assistance of a case manager, who conducted an in-depth assessment and then referred the patient to primary or specialty care or to a multidisciplinary geriatric assessment team for further assessment. The case manager also provided patient education and follow-up. These results were reported in the subgroup of older adults screening positive for depression who were enrolled in a trial that attempted to identify patients with any of five high risk conditions. Therefore, this was not a randomized comparison, and the patient population was limited to patients who scored in a “high risk” range for a number of conditions. It was unclear if patients screening positive for depression only (i.e., none of the other four conditions screened for) could have been eligible for study inclusion. Thus, the generalizability of this trial to general primary care screening for depression in older adults may be limited.

Study details. All eight studies were conducted in outpatient primary care settings, seven in the United States and one in the Netherlands. Study settings included urban, rural, and indigent clinics, and two of the studies involved multiple geographic sites across of the US.^109,110 Four of the studies^{111,112,115,116} focused on screening older adults and will be discussed separately from the others. In all studies, the usual care participants were screened as well as those in the intervention groups, often in the waiting room prior to a scheduled visit, but results of the screening tests were not systematically given to the providers of usual care patients or used in their care. The primary outcomes of two studies involved subgroup analyses,^109,116 which were therefore non-randomized (though controlled) comparisons. We included these analyses in our review because they were conducted in the context of a well-designed and implemented randomized controlled trials and were likely planned a priori.¹⁰⁹

Studies focusing on general adult patients. Four studies focused on general adult populations. In a recent small (n=59) study in a rural setting,¹¹³ providers at two private health clinics were educated about the PHQ-9. Researchers subsequently screened consecutive clinic patients. Patients with a positive score on the PHQ-9 were randomized either to have their PHQ-9 results given to their providers or not. Providers of intervention participants were asked to review the completed PHQ-9. No other depression care support was provided. Providers saw both intervention and control participants and therefore contamination is possible.

After 6 months there were no statistically significant group differences in either the proportion of participants in remission or in change in PHQ-9 score. This was a small study, however, with only 59 randomized participants. Even with their fairly good follow-up rate of 86 percent, there was minimal power to detect even large differences. For example, they found that 52 percent of the intervention group achieved remission from depression at 6-month follow-up, compared with 38 percent of the control group. The power for this comparison was approximately 10 percent, given their sample size. In order to achieve statistical significance with 80 percent power with their sample size, approximately 80 percent of the intervention participants would have to be in remission, compared with 38 percent of the control group. Thus, this study was not adequately powered to detect differences that would likely be clinically significant. They also reported differences in PHQ-9 score change, which is likely to have more power than the binomial comparison, but they did not report the standard deviations that would have allowed us to analyze the power for this analysis. This study also included a fairly large proportion of participants who were already taking medications for depression or anxiety (38 percent), though they do not specify the proportion taking antidepressants.

In another small (n=61) study with a largely indigent population,¹¹⁴ Jarjoura and colleagues recruited patients from an internal medicine residency clinic who were either enrolled in Medicare or who were without private health insurance and had low income. Patients were excluded from this study if they were currently receiving treatment for any mental health problem (including depression), were seeking help for mental health problems, or reported suicidal ideation on the screening test. While there was a high background level of depression in the population, with positive screening test results for 45 percent of screened patients, only 9 percent of the screened patients were eligible for the study (i.e., had a positive screening test and were not already being treated for depression).

In the intervention condition, a screening nurse advised resident physicians of the positive screening results and provided a protocol outline asking the physician to: 1) explore symptoms with the patient and affirm screening results; 2) attempt to rule out physical conditions, medications, or other primary psychiatric diagnoses; 3) provide information, treatment, and follow-up at specified intervals; and 4) facilitate refer to behavioral treatment. Control participants who screened positive were informed they might have a problem with depression and that effective treatments were available. Participants were contacted at 6- and 12-months after the baseline visit and 90 percent completed at least one follow-up interview.

Researchers reported that intervention status predicted change in Beck Depression Inventory (BDI) in a mixed model linear regression. At six-month follow-up, average BDI score for the treatment group had improved by 7.6 points more than that of the control group (p<0.05, BDI range 0–63). Intervention effects were maintained at 12 months, with BDI score improvement in the intervention group 6.5 points greater than in the control group (p=0.03). As with the Bergus et al¹¹³ study, this was a small study that randomized individuals rather than providers (and was therefore vulnerable to contamination). This study, however, used an analysis technique likely to have more power than the Bergus et al analyses.

This study limited the sample to those who were not being treated for depression at baseline, thus eliminating patients who were diagnosed with depression but were untreated, under-treated, or resistant to treatment. The intervention in the Jarjoura et al¹¹⁴ study did provide elements beyond simple feedback to a provider, including a protocol and positive screening test results, but the program was not extremely extensive and involved a level of treatment that is feasible for many or most primary care settings. A total of 70 percent of the patients in the intervention condition were given depression care (24 percent received mental health specialty care, with or without antidepressants, and an additional 45 percent received antidepressants without specialty care), compared with only 15 percent in the usual care condition, none of whom received specialty mental health care. Thus, the combination of limiting the program to newly detected depression and moderate levels of support appeared to be sufficient to improve patients’ depression.

Two large-scale studies^109,110 included extensive interventions beyond screening. Both included study sites in many regions of the US and both used a two-step screening approach adapted from the Composite International Diagnostic Interview (CIDI), where depressive symptoms are only explored if the patient is positive on a two-item initial screener. One of the studies¹¹⁰ found significant group differences in depressive symptomatology and remission, while the other found a benefit only among the subgroup of newly detected patients with depression.¹⁰⁹

The study by Wells¹¹⁰ and colleagues randomized clinics to usual care or one of two extensive quality-improvement interventions for depression care. One intervention condition (QI-Meds, referred to as IG1 in Table 5) included screening; institutional monetary commitment; staff and clinician training (1- or 2-day workshops); clinician manuals; monthly training lectures; academic detailing; numerous materials for clinicians, staff, and patients; initial visit with nurse specialist for assessment, education, and discussion of patients preferences and goals; and trained nurse specialists for follow-up assessment and on-going support for medication adherence for those prescribed antidepressant medications. The other intervention condition (QI-Therapy, referred to as IG2 in Table 5) included all of the same QI elements as the QI-Meds condition except that it included trained therapists to provide manualized CBT and reduced co-pay for those referred for psychotherapy, rather than nurse specialists to support medication adherence. Providers in both intervention groups were expected to provide the treatment they deemed most appropriate for each patient, so patients in both groups could have received antidepressants or mental health specialty care. For example, referrals to mental health specialists ranged from 63 percent to 89 percent in the QI-therapy clinics and between 17 and 45 percent in the QI-Meds clinics. Structured follow-up sessions with a nurse medication specialist on the phone or in person averaged 1.8 sessions for patients in the QI-Therapy and 5.1 for QI-Meds patients.

Medical directors of the usual care clinics were mailed the AHRQ depression practice guidelines with quick reference guides for distribution to clinicians. Patients in control clinics were screened and patients were told that while they could inform their providers of the screening test results, the study staff would not send the results to their providers. Patients who screened positive on the CIDI were enrolled in the interventions and followed up.

At follow-up, the proportion of participants in either treatment arm who were still positive on the CIDI 2-item screener was 40 percent at 6 months and 42 percent at 12 months, while 50 percent of usual care participants still scored positive on the CIDI at 6 months and 51 percent at 12 months (p=0.001 for group differences at 6 months, p=0.005 at 12 months). At 5-year follow-up, program benefits were sustained for the QI-therapy group, which had positive CIDI screening scores of 36 percent of the intervention participants compared with 44 percent of the usual care participants (p=0.05). The difference between the control and QI-meds group was not significant at 5-year follow-up (38 vs. 44 percent positive, p=0.08).¹¹⁸ These results provide good evidence for the effectiveness of their program. It is impossible to determine, however, what role, if any, the screening component played in the success of their program, which included so many elements. Further, while this intervention was proven feasible for primary care settings, it involved significant institutional commitment and may not reflect the care that would be found currently in most settings.

The Rost et al study¹⁰⁹ was designed for practices without ready access to mental health specialty care and also contained extensive care support elements beyond screening. Researchers cluster-randomized clinics to usual care or an intervention in which office staff recruited, screened, and enrolled participants who screened positive prior to a clinic visit. If the physician confirmed the depression diagnosis, the participant was scheduled for a return visit with the physician and to meet with the nurse specialist in one week. The nurse specialist reassessed the patient’s level of depression, discussed treatment options and preferences, and asked the participant to complete a homework assignment. Participants completed up to eight additional sessions following the same pattern, either by phone or a face-to-face visit. Nurse documentation logs indicated that the nurses intervened in 92.5 percent of the patients in the intervention condition, contacting these patients an average of 5.2 times. Physicians and nurses in the intervention clinics were also trained in depression assessment and treatment and had access to free 24-hour consultation for patient-specific questions.

Although there was a significant overall treatment effect,¹¹⁹ outcome data were only presented in two strata: patients who were not already being treated for depression (“newly identified”), and patients who were already or had recently been treated for depression (“recently treated”). Among the newly identified, CESD scores dropped an average of 21.7 points in the intervention group and only 13.5 points in the control group (p<0.05). In the recently treated, CESD scores dropped in both groups and did not differ between groups. At 2-year follow-up, significantly fewer of the newly identified patients in the intervention group met CESD screening criteria for depression (26 percent), compared with the control group (59 percent) (p<0.05). As with the Wells study, the effect of the screening component cannot be isolated.

Studies focusing on older adult patients. Four fair-quality, cluster-randomized RCTs focused on older adults.^111,112,115 A 2006 study conducted in the Netherlands randomized 34 general practices (GP) to intervention or usual care conditions.¹¹⁵ Providers in the intervention practices attended a 4-hour training session on diagnosis and treatment of late-life depression, with a treatment protocol based on Dutch treatment guidelines.¹¹⁷ In all practices, patients aged 55 or older were screened for depression with the Geriatric Depression Scale (GDS), and those scoring five or greater underwent a diagnostic evaluation using the Primary Care Evaluation of Mental Disorders (PRIME-MD). Usual care providers were not informed of screening test or PRIME-MD results. The intervention providers were informed of screening results and conducted the PRIME-MD interview themselves to determine MDD diagnosis. Patients with MDD diagnoses who were not currently using antidepressants were invited to participate in the study.

Fifteen percent of the 3,937 patients screened scored five or greater on the GDS, and 53 percent of the 339 consenting to the diagnostic interview were diagnosed with MDD and enrolled in the study. At 12-month follow-up, the groups did not differ on either proportion depressed or on symptom severity scores. Fifty-seven percent of the intervention group participants and 52 percent of the usual care participants still met criteria for depression at the 12-month follow-up.

Whooley and colleagues¹¹¹ randomized HMO primary care clinics to either an intervention or control condition. The intervention condition involved screening, feedback, treatment guidelines, and training for providers and group psycho-educational classes offered to patients and families. The control condition also involved screening and the clinics received the same one-hour training session. Patients were screened with the Geriatric Depression Scale (GDS) and 14 percent of the participants screened positive for depression and were enrolled in the study, with 20 percent receiving antidepressants during the previous year. Although this study included group classes, only 12 percent of the patients attended them. Follow-up after 2 years revealed improvements in both groups with no differences between the intervention and control groups.

Callahan and colleagues¹¹² randomized clinical practices to usual care or to a program screening for depression, dementia, and alcoholism. Providers in the intervention group received a letter for each of their study patients screening positive on two separate depression instruments. These letters contained the results of the second screening test, a list of medications that may cause depressive symptoms, and treatment recommendations. This letter was placed in the patient’s chart, along with handouts for the patient. The provider was encouraged to schedule three appointments with the patient over the course of the next three months. Completion rates for visits were not reported. Providers in the control group were given no information or screening results, and additional appointments were left to the discretion of the provider. Providers in both groups were given an educational session on late-life depression and completed surveys after baseline visits of study patients. Sixteen percent of the patients in this study screened positive on the first screener, the Center of Epidemiologic Studies Depression Scale (CESD), and of these, 28.6 percent completed and were positive on the Hamilton Depression Rating Scale (HAMD). No group differences were found at 6 or 9 months follow-up, with very few patients in either group (12 to 13 percent) achieving remission. The patient population was an indigent, urban, largely Black group of seniors with co-morbid medical and psychiatric illnesses and significant functional disability at baseline.

Rubenstein and colleagues¹¹⁶ conducted a non-randomized controlled trial in which patients in two VA clinical practices aged 65 or older were screened via a postal survey for five different common geriatric conditions (depression, falls/balance problems, urinary incontinence, memory loss, and functional impairment). Those scoring in the “high risk” range (indicating impairment on four or more of the ten items on the screening instrument) were invited to enroll in the trial. Participants at the clinical practice assigned to the control group received usual care, though if a serious condition was identified then the patient’s primary care provider was notified. Participants in the clinical practice assigned to the intervention group were contacted by a case manager, who performed a more thorough assessment over the telephone. After completing the assessment, the case manager provided referrals to primary or specialty care providers, as needed, or for further evaluation by a multidisciplinary geriatric assessment team. The case manager also provided patient education, a written summary of recommendations, scheduled appointments, and conducted follow-up calls (one-month post-assessment and then quarterly). Overall, 42 percent of those returning questionnaires scored as “high-risk”, 79 percent (n=792) of these enrolled in the study, and 41 percent of the enrolled scored in the depressed range on the GDS (45.8 percent of those completing the GDS).

Among the entire enrolled population, including those who were not depressed at baseline, no differences were seen in change in depressive symptomatology over time. Among the 206 participants who were above the cut-off for likely depression on the GDS at baseline and completed the 12-month assessment, those in the intervention group showed a greater decline in depression severity after one year. GDS scores of participants in the intervention group dropped by an average of 3.7 points, compared with an average decline of 2.7 points among control participants (p=0.05). However, only 72.5 percent of the sample completed the GDS at one-year, so despite the high follow-up rate for the study overall, the follow-up rate for this instrument was considerably lower. Also, we do not know if the intervention and control participants in the subgroup of participants meeting depression criteria at baseline were similar to each other, nor do we know what the follow-up rate was among these baseline depressed.

All studies of older adults were well-designed, fairly large-scale studies with apparently similar rates of background depression in their populations. Two studies focused additional supports on the treating provider in the form of education, screening, and feedback, with other elements such as repeat visits with the provider (Callahan) or outside psycho-educational classes for patients.¹¹¹ One utilized case managers rather than focusing on improving the care provided by the primary care clinician.

Key Question 2: What are the harms of screening for depressive disorders in adults and older adults?

We did not find any studies that included adverse events of screening.

Key Question 3: Is antidepressant and/or psychotherapy treatment of older depressed adults effective in improving health outcomes?

Summary of findings. We found three good-quality systematic reviews in older adults^120–122 that each included meta-analyses (Table 7). One examined the efficacy of antidepressants,¹²¹ one examined the efficacy of psychotherapy,¹²² and one examined both types of treatments.¹²⁰ Two of these reviews^120,121 found that antidepressants were effective in treating depressed older adults, and approximately doubled the odds of remission compared with placebo controls, with OR of 2.03 (CI: 1.67, 2.46) for patients with major and minor depression; 2.13 (CI: 1.61, 2.86) for the community-dwelling subset; and 2.27 (CI: 1.72, 2.94) in the subset with MDD. Two good-quality meta-analyses^120,123 of psychotherapy effectiveness concluded that psychotherapy is effective in treating depression in older adults. Depressed older adults treated with psychotherapy were roughly two-and-a-half times more likely to achieve remission than those who were not treated with psychotherapy, with the ORs estimated at 2.47 (CI: 1.76, 3.47) and 2.63 (CI: 1.96, 3.53). Effect sizes estimates range from 0.72 to 1.09.

Table 7

Summary of the evidence for KQ3: treatment efficacy in older adults.

Study details - antidepressants. Numerous systematic reviews examining antidepressant efficacy in older adults have been published since the end of the previous USPSTF report’s search window. We limited our inclusion to the two most recent good-quality systematic reviews that also included meta-analyses — a Cochrane review published in 2000¹²¹ and another review published in 2006.¹²⁰ There was considerable overlap among the two included reviews, though they had somewhat different inclusion criteria and search windows and therefore different sets of included trials. Another good-quality review of newer antidepressant efficacy in adults was also published in 2000⁷⁵ and discussed efficacy in older adults separately. This study, however, overlapped substantially with the Cochrane review, and since the Cochrane review provided more detail, we excluded this other 2000 review.

The most recently published meta-analysis¹²⁰ examined 62 pharmacotherapy studies, 32 psychotherapy studies, and five studies examining both interventions in depressed older adults. This good-quality review included studies involving the full range of depressive disorders, including minor depression, dysthymia, and MDD. These authors included non-randomized comparative studies, although 89 percent of the included studies were randomized trials. All studies included a control condition and no restrictions were set with regard to setting (e.g., inpatient, outpatient, residential). In addition to providing overall results for all depressive disorders, they also reported the results of trials focusing exclusively on MDD separately. The authors combined all agents to determine overall effectiveness of antidepressants and combined all therapeutic modalities to determine overall effectiveness of psychotherapy. Of the 77 comparisons of clinician-reported outcomes, the specific agents examined included roughly the same number of reports for SSRIs (21/77) and TCAs (22/77), and over one third of the drugs were categorized as “other,” many of which were non-SSRI second generation medications.

This review found an overall effect size of antidepressants of 0.69 (CI 0.57, 0.81, k=77 comparisons) for clinician-rated depression (most commonly the HAM-D) and 0.62 (CI 0.45, 0.79 k=28 comparisons) for self-rated depression. Although they do not report the absolute proportion of patients in the treatment and control groups whose depression remitted, the authors do report that treated participants were approximately twice as likely as control participants to go into remission (OR 2.03, CI: 1.67, 2.46). For studies focusing exclusively on MDD, clinician-rated depression had an effect size of 0.79 (CI: 0.64, 0.95, k=39 comparisons).

This review reported effect sizes separately for different classes of antidepressant agents, though they did not limit this analysis to studies focused on MDD. The authors report effect sizes ranging from 0.48 (SSRIs) to 0.93 (TCAs) for clinician-rated improvement, and report effect sizes ranging from 0.22 (SSRIs) to 0.83 (TCAs) for self-rated improvement.

The 2000 Cochrane Review¹²¹ included randomized, placebo-controlled trials using antidepressants in the treatment of depression in older adults. The authors included studies of major and minor depression recruited from outpatient, community, institutional, and inpatient sources. They included 23 placebo-controlled, double-blind RCTs involving SSRIs, other second generation antidepressants, TCAs, and MAOIs, 17 of which were included in the meta-analysis. Eleven of the trials included in the meta-analysis were TCAs, and three of the trials involved second generation medications (fluoxetine and mirtazapine). Separate analyses were conducted for each of the medication categories, and separate analyses were conducted to determine the percent recovered at the end of the trial, continuous HAM-D scores, other observer rating scales, and discontinuation rates. For this review we limited our focus to recovery rates and HAM-D scores. Most of these studies measured treatment outcome at 3 to 8 weeks. This Cochrane review reported the number, proportion, and odds ratios of participants who were not recovered from depression at trial’s end. We converted these numbers to report those who recovered from depression for ease of interpretation and comparison with other meta-analyses.

The most commonly reported outcome was the proportion of patients still depressed at the end of the study, which we converted to percent no longer depressed. Pooled fixed effects odds ratios were estimated for each medication type. A total of 365 patients treated with SSRIs were compared with 372 placebo controls in two trials with estimated OR of 1.96 (CI: 1.39, 2.78). One of these trials also reported differences in HAM-D change scores of 1.7 points in favor of the treatment group (CI:0.54, 2.60). Trials of other second generation antidepressants examined mirtazepine and two other medications not approved for use in the U.S. (minaprine, and medifoxamine), estimating an overall OR of 1.92 (CI: 1.07, 3.45) in 102 patients treated with active agents and 96 placebo controls in two trials. No HAM-D results were reported for either of these trials. For TCAs, 245 patients treated with TCAs were compared with 223 placebo controls in ten trials and had an estimated OR of 3.12 (CI: 2.13, 4.76). One trial reported that treatment participants improved on the HAM-D by 9.6 points (CI: 9.4, 13.8) more than the control participants. Analyses of MAOIs included a total of 58 patients treated with MAOIs and 63 with placebo in two trials, and found an OR of 5.88 (CI: 2.56, 14.28). No HAM-D results were reported for MAOIs. In terms of recovery rates, these studies found 28 to 49 percent of patients recovering from depression among antidepressant users and only 10 to 25 percent recovering in the placebo control groups.

In addition to overall results by medication type, they report meta-analysis results separately for outpatient/community recruitment, which are of greatest relevance to this review. Antidepressants were more effective than placebo, with 36 percent of antidepressant users recovered at the end of the study compared with 21 percent of the placebo patients, estimated OR of 2.13 (CI: 1.61, 2.86).

Although they did include some studies of hospitalized and institutionalized patients, most of these trials excluded a large proportion of those screened for inclusion due to comorbid physical illness or high levels of depression severity. Thus, the results of these analyses are best generalized to relatively healthy elders with mild- to moderate- depression severity, and only to very short-term outcomes.

Study details- psychotherapy. As with antidepressant therapy, several systematic reviews have been published examining psychotherapy efficacy in older adults since the end of the previous USPSTF report’s search window. We limited our inclusion to the two most comprehensive recent good-quality systematic reviews of psychotherapy for depression in older adults, one by Pinquart and colleagues¹²⁰ and the other by Cuijpers and colleages¹²², both published in 2006. While there was substantial overlap between the two reviews, their inclusion and exclusion criteria led to somewhat different bodies of evidence and we therefore included both reviews.

The Pinquart review did not report search dates, but appeared to have covered the literature through 2004. It included studies in which participants had a median age of 60 or greater, required a placebo or no-treatment control group, and included non-randomized comparisons, for a total of 32 trials. They report results separately for clinician-rated and self-report outcomes. The Cuijpers review searched an additional year (through 2005), included trials of participants aged 50 or greater, included trials that compared two active treatments as well as those that included control groups, and only included randomized comparisons, for a total of 25 trials. Pinquart and colleagues specifically reported including non-English language studies. No mention of language restriction was reported in the Cuijpers report, and no non-English language studies were among the list of included trials. Most of the 13 studies included in Pinquart but not in Cuijpers appeared to be non-randomized controlled trials and non-English language trials. Most of the 12 trials included in Cuijpers but not in Pinquart were either comparative effectiveness trials (i.e., they did not include a no-treatment control group) or were published after 2004.

The systematic review and meta-analysis by Pinquart and colleagues¹²⁰ found an overall effect size of 1.09 (CI: 0.91, 1.26, k=35 comparisons) for clinician-rated depression severity (usually using the HAM-D) and 0.83 (CI: 0.69, 0.98, k=52 comparisons) for self-rated depression. Twenty-six of the comparisons of clinician-rated depression involved some form of cognitive behavioral therapy. The other comparisons involved problem-solving, reminiscence, interpersonal, rapid eye movement, or life review therapy. Likewise, more then three-fourths of the comparisons involving self-reported outcomes used cognitive behavioral therapy. While they do not report the absolute proportion of patients in the treatment and control groups whose depression remitted, they do report that treated participants were more than twice as likely to go into remission (OR 2.47, CI: 1.76, 3.47). For studies focusing exclusively on MDD (excluding dysthymia and minor depression), clinician-rated depression had an effect size of 0.96 (CI: 0.69, 1.23, k=16 comparisons).

The review by Cuijpers and colleagues¹²² reported an effects size of 0.72 (CI: 0.59, 0.85) for psychological treatment of depression in older adults, among trials that included a control condition (k=21 comparisons). This is within the 95 percent confidence interval of the Pinquart study for self-reported depression. This review did not report clinician- and self-reported symptomatology separately, but instead averaged these two measures if a trial reported both types of outcomes, and combined both types of outcomes in a single meta-analysis. Thus, this effect size reflects a blend of clinician report (which tends to show larger effects) and self-report (which tends to show smaller effects). Half of the comparisons in this review involved cognitive behavioral therapy, and the remaining involved behavior therapy alone, reminiscence or life-review, interpersonal, problem solving, or “other” therapies. Eleven studies in the Cuijpers review included remission rates, usually based on falling below a pre-specified cut-off on a continuous depression measure. They calculated an overall OR of 2.63 (CI: 1.96, 3.53), which is well within the 95 percent confidence interval for remission in the Pinquart review. Thus, these two reviews reported very consistent results for remission, but the average degree symptom improvement was somewhat smaller in the Cuijpers review than the Pinquart review, and this may in part be due to the method of analyzing the outcomes. It is also possible that randomized trials result in smaller group differences than non-randomized controlled trials, though the data were not presented that would allow us to verify this hypothesis.

Key Question 4: What are the adverse effects of antidepressant treatment (particularly SSRIs and other second-generation drugs) for depression in adults and older adults?

Summary of findings. We examined serious adverse effects associated with antidepressant treatment including: suicide-related events (completed suicide, serious self-harm or attempted suicide, suicidal ideation) and serious psychiatric events, including hospitalization. Since adverse effects—typically nausea, dizziness, diarrhea, headache, sexual dysfunction, or insomnia—occur in 61 percent of patients during clinical trials of antidepressant medications,^57,92 we examined rates of early discontinuation, particularly discontinuations due to adverse effects. For older adults, we also considered evidence of serious medical events (e.g., upper gastrointestinal bleeding) associated with SSRIs and other second-generation antidepressant use.

Suicide-related event rates were most commonly reported per person receiving treatment, without consideration of exposure time. Event rates per 10,000 persons for three suicide-related outcomes are reported in Table 8: for completed suicide, for suicidal behaviors (usually defined to include suicide attempts, preparatory acts, or nonfatal, serious self-harm), and for suicidal behaviors and ideation combined. Table 9 reports risk differences in these outcomes for patients with MDD.

Table 8

Summary of rate of suicide and related behavior or ideation.

Table 9

Risk differences (per 10,000) for suicide and suicidality among patients with major depressive disorder.

For completed suicide, none of the seven meta-analyses of short-term trials in adults being treated either for MDD or for any psychiatric indications supplied clear evidence that use of second-generation antidepressants—or of SSRIs in particular—significantly increased rates of completed suicide for those on antidepressant treatment compared with placebo. However, despite the large sample sizes, there were very few suicides (7 to 43 total suicides among treated and control patients per review) which limited power to detect these rare events; using reported or calculated OR, these results are compatible with no increase in short-term risk, some protective effect, or some increased risk. Most meta-analyses estimated suicide rates ranging from 3.8 to 8.8 per 10,000 antidepressant-treated adults, compared with 2.3 to 9.3 per 10,000 placebo-treated patients. Two studies’ data represented outliers. One FDA report estimated 1.8 completed suicides per 10,000 antidepressant-treated MDD patients (compared with 0.67 per 10,000 placebo-treated MDD patients);^124,125 this estimate is markedly lower than other studies and may reflect methodological differences (e.g., in how suicides and related events were categorized, among other differences). And, the much higher estimates of suicide rates in both antidepressant and placebo-treated patients in another review¹²⁶ could reflect the suicide risk among more severely depressed patients (since the majority of studies in this review employed active controls rather than placebos) as well as quality concerns with this review.

Three fair or good quality large observational studies reported on suicide with antidepressant treatment in a total of 383,796 patients from a large US HMO and from GP practices in the UK.^127–129 Among the two highest quality studies reporting a 6–8 month follow-up duration, crude suicide rates were 4.7 and 4.8 per 10,000 persons treated primarily with second-generation antidepressants, with slightly higher rates reported among children and adults under 30 years. These studies also indicate higher risk for suicide death among men compared with women. Although these observational studies do not give us comparative information for people who were not taking antidepressants, they give credence to the estimate of approximately 4/10,000 suicide cases among antidepressant users found most consistently in the meta-analyses of short-term trial data.

For suicidal behaviors (defined differently across studies, but usually including suicide attempts, preparatory acts, or serious self-harm), results from five meta-analyses did not show significant differences in the odds of suicidal behaviors in adults treated with antidepressants compared with placebo, with several exceptions. One fair quality systematic review indicated an increased odds of suicidal behaviors in adults of all ages being treated with SSRIs for any indication (OR 2.70; CI: 1.22, 6.97);¹³⁰ this report was limited to published studies only and did not have clear adverse event ascertainment for the majority of patients. In an FDA review of regulatory data of placebo-controlled trials, odds of suicidal behavior were approximately doubled in adults under age 25 taking second-generation antidepressants for all psychiatric disorders (OR 2.31; CI: 1.02, 5.64);¹²⁵ in contrast, the odds of suicidal behaviors was not changed among middle-aged adults, and was greatly reduced in older adults on second-generation antidepressants (OR 0.06, CI: 0.01, 0.58).¹²⁴

The highest odds of non-fatal suicidal behavior was reported in adults of all ages being treated for MDD with paroxetine compared to placebo (OR 6.70; CI: 1.1, 149.4), with most events (8/11) occurring in those aged 18–29 years.¹³¹ The NNT-harm for this estimate is 373 (CI: 208, 1818).

Two good quality observational studies suggest that, in contrast to a higher risk of suicide deaths in men, there were no sex differences in risks of self-harm, but there were age-related differences.^128,129 Suicide attempts were significantly greater in younger persons (under aged 18 years),¹²⁹ with a higher rate of self-harm in those aged 19–30.¹²⁸ The highest risk for suicidal behaviors occurred in the month prior to treatment initiation and the first month of treatment.¹²⁹ Rate of suicidal behaviors in real-world practice situations was similar to trial rates for one study¹²⁹ but was substantially higher in the other;¹²⁸ this higher rate could reflect real differences or perhaps represents study differences in definitions (and perhaps ascertainment).

For suicidal ideation, three meta-analyses used a combined endpoint (suicidal ideation or behavior) and found no differences between antidepressant and placebo-treated patients, except for a reduction in older adults treated with second-generation antidepressants for all psychiatric conditions (OR 0.39;CI: 0.18, 0.78).¹²⁵

For serious psychiatric events, we did not find any existing systematic reviews and found very limited reliable primary evidence to estimate mania precipitation or to distinguish other uncommon but important psychiatric side-effects from suicide-related behaviors.

For tolerability, we found eight systematic reviews that reported overall discontinuation rates and discontinuation due to adverse effects as measures of the impact of adverse effects associated with antidepressants,^{72,75,90–95} and two large cohort or uncontrolled treatment trials.^46,96 Early treatment discontinuation ranged from 16 to 29 percent in meta-analyses of antidepressant trials in primary care patients with depression, with a best estimate of 20 to 23 percent in “real-world” trials of primary care. Early discontinuations due to adverse effects were lower (5 to 12 percent). Patients aged 55 and older had higher discontinuation rates overall (27 to 36 percent) and for adverse effects (17 to 22 percent). With longer follow-up, adverse event discontinuation rates increased, particularly in those who switched or augmented medications due to lack of efficacy or intolerable side-effects.

As reported above, older adults were at lower risk for suicide-related harms during antidepressant treatment. For those over 65 years, antidepressant treatment was protective for both suicidal behaviors (OR 0.06; CI: 0.01, 0.58) and combined suicidal behaviors and ideation (OR 0.37; CI: 0.18, 0.76). For serious medical events in older adults, we found a fair-quality systematic review of six large observational studies from Denmark, Canada, the UK, and Holland examining bleeding risk in these SSRI’s.¹³² Among 26,005 Danish patients aged 16 years and older (almost half aged 60 or more years) risk for hospitalization for upper gastrointestinal (UGI) bleeding was increased compared with non-users during periods of current SSRI use only, with an excess risk of 3.1 per 1000 treatment years. In 317,824 Canadian patients 65 years and older on antidepressants, risk of hospitalization for UGI bleeding increased greatly with age, from 4.1 hospitalizations per 1000 person-years of SSRI treatment in those aged 65 to 70 years to 12.3 hospitalizations per 1000 person-years in octogenarians. Excess hospitalizations for UGI bleeding were increased five-fold (33.2 per 1,000 treatment years) in persons with prior UGI bleeding. In some, ^133–135 but not all,¹³⁶ studies, odds of UGI bleeding among SSRI users were further increased at least two-to-three fold when SSRI users were also taking NSAIDs, with a lesser risk associated with co-use of aspirin or other anticoagulant medications.

Summary of Evidence Quality

Table 10 summarizes the overall quality of evidence according to USPSTF criteria and other considerations about the body of evidence for each of the key questions addressed in this review.

Table 10

Summary of Evidence.