Depression among Asian-American Adults in the Community: Systematic Review and Meta-Analysis

Objectives In this systematic review, we provide an overview of the literature on depression among Asian-Americans and explore the possible variations in depression prevalence estimates by methodological and demographic factors. Methods Six databases were used to identify studies reporting a prevalence estimate for depression in Asian-American adults in non-clinical settings. Meta-analysis was used to calculate pooled estimates of rates of depression by assessment type. Statistical heterogeneity was assessed for subgroup analyses by gender, age, ethnicity, and other participant characteristics. Results A total of 58 studies met the review criteria (n = 21.731 Asian-American adults). Heterogeneity across the studies was considerably high. The prevalence of major depression assessed via standardized clinical interviews ranged between 4.5% and 11.3%. Meta-analyses revealed comparable estimated prevalence rates of depression as measured by the Center for Epidemiologic Studies Depression Scale (35.6%, 95% CI 27.6%–43.7%) and the Geriatric Depression Scale (33.1%, 95% CI 14.9%–51.3%). Estimates varied by Asian racial/ethnic group and other participant characteristics. Estimates of depression among special populations, which included maternity, caregivers, and homosexuals, were significantly higher than estimates obtained from other samples (58.8% vs 29.3%, p = .003). Estimates of depression among Korean and Filipino-Americans were similar (33.3%-34.4%); however, the estimates were twice as high as those for Chinese-Americans (15.7%; p = .012 for Korean, p = .049 for Filipino). Conclusion There appears to be wide variability in the prevalence rates of depression among Asian-Americans in the US. Practitioners and researchers who serve Asian-American adults need to be sensitive to the potential diversity of the expression of depression and treatment-seeking across Asian-American subgroups. Public health policies to increase Asian-American access to mental health care, including increased screening, are necessary. Further work is needed to determine whether strategies to reduce depression among specific Asian racial/ethnic groups is warranted.


Introduction
Beck Depression Inventory (BDI), Hamilton Depression Inventory (HDI), Patient Health Questionnaire (PHQ), and Geriatric Depression Scale (GDS)]. A recent review of depression among AA attributed variation in prevalence to the instruments used to measure depression and by the context of each study [12]. AA who suffer from MDD may not report sadness or depressed mood as their primary complaint, and thus may be less likely to meet clinical criteria when using a tool such as the CIDI [12]. The most commonly used tool to assess depression among AA is the CESD. In this particular ethnic group, a high rate of somatic symptoms of depression-such as changes in appetite, headaches, backaches, stomachaches, insomnia, or fatigue-has been reported using the CESD [12].
The purpose of this study was to: systematically review the estimates of depression prevalence among various ethnic subgroups of AA in community samples, derive synthesized estimates of depression, and examine possible gender, ethnicity, other participant characteristics (e.g., age, parents for teenagers, maternity, homosexuals), and methodological factors associated with variation in these estimates. By providing an estimate of depression prevalence among AA in the community, the results of this study would be beneficial to the development of policies that aim to reduce mental health disparities in the US.

Methods
The protocol and data extraction forms were designed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis: the PRISMA Statement [17,18]. The following six databases were systematically searched using a comparable search strategy, with adapted terms for each database: PubMed, MEDLINE (OVID), CINAHL (EBSCOHost), Psy-chINFO (OVID), Web of Science (Web of Knowledge), and the Cochrane Library. Keywords were "Asian-Americans" and "depression/depressive symptoms", and any matched subject headings or MeSH terms. For example, the search strategy in Inclusion criteria were as follows: full-text, published, peer-reviewed, English-language studies conducted in the US, target population of adults (! age 18), a reported prevalence level for depression, and conducted in a community setting. Studies published during the past 10 years (from 2004 to 2014) were included to provide a current prevalence of depression. Studies that did not provide a prevalence estimate or sufficient information from which a prevalence could be calculated, as well as those conducted in clinical care settings, were excluded.
Retrieved articles were exported to a reference manager and duplicates were hand-searched and removed. Two reviewers (H.J.K. and K.T.) independently reviewed all titles. After the first title review, the two reviewers independently reviewed selected abstracts. Full-text articles were selected when the two reviewers agreed that the article met the inclusion and exclusion criteria. H.J.K conducted the primary data extraction. All articles were examined independently by a second reviewer (K.T.). Inter-reviewer disagreement was minimal, and inconsistencies were discussed and resolved. References for all the articles were also scanned (citation tracking) for further relevant source papers, and similar procedures were used to include or exclude them.
Risk of bias in reporting depression prevalence in individual studies was assessed using the Agency for Healthcare Research & Quality (AHRQ) [19]. These criteria include sample size, methods for selecting participants, methods for measuring exposure variables, methods to deal with any design-specific issues such as recall or interviewer bias, and analytical methods to control for confounding factors. Because we focused only on sample characteristics and prevalence of depression for this review, four items: sample size (!500 vs. <500), sampling methods (random vs. convienient), participation rate (reported vs. unreported or <50%), and eligibility criteria (provided vs. unprovided) were reviewed for each article. Total scores ranged from 0 to 4, with a smaller number indicating a higher risk of bias.
We performed meta-analyses to calculate pooled estimates of depression prevalence by method of measurement with Stata (version13.1, StataCorp LP, College Station, TX). Studies that used the standard questionnaire items and cut-points for each measure were included in the meta analysis. Heterogeneity-differences in the prevalence of depression across studiesarises when there are clinical or methodological differences between studies (i.e., participants characteristics, outcomes, or study design). This information is important in meta-analysis because the presence of heterogeneity can influence the pooled prevalence; high heterogeneity may produce misleading results [20]. Statistical heterogeneity was assessed using I 2 , with thresholds of ! 25%, ! 50%, and ! 75% indicating low, moderate, and high heterogeneity, respectively [21]. To investigate heterogeneity, we performed subgroup analyses by gender (% of women), age, ethnicity, and other participant characteristics. Random-effects metaregression models (DerSimonian & Laird, with 95% CIs) were conducted to explore the subgroup differences with sample size as a covariate. We also conducted sensitivity analyses for prevalence of depression by including studies that used modified questionnaires or alternative cut-points, and by excluding studies with high risk of bias or with unreported participation rates. Egger's tests of publication bias were performed to assess publication bias due to preferential publication of small studies reporting high prevalence estimates [22].

Results
The literature search yielded 1,555 articles. An additional 62 articles were found through Google Scholar and relevant bibliographies. After removal of duplicates, the title screening process identified 263 potentially eligible studies (Fig 1). After abstracts were screened using the inclusion and exclusion criteria, 58 studies remained for qualitative synthesis (a total sample of 21,731 adults) .
The prevalence of depression and study characteristics of the selected articles are presented in Table 1. Multiple instruments have been used to assess depression. The most commonly used measure was the CESD, with 22 studies using this tool. The majority of the studies used the 20-item CESD and a score of 16 to classify depression. Jang and colleagues prefer a short form consisting of 10 items (with a cut-point of 10 demarcating depression), and these were included in the meta-analyses because a previous study indicated this is comparable to a score of 16 on the full version [81]. Three studies used modified versions of CESD, and one study used a different cut-point. Seven studies used the PHQ; one of the studies used only 2 items, while the others used 9 items in their assessment. Five studies used the GDS, 4 studies used the Hopkins Symptom Check List (HSCL), 2 studies used the BDI, and one study used the Postpartum Depression Screening Scale (PPDS). One study used screening stem questions of Fresno-CIDI, summed the symptoms that were under each criteria, and generated an estimate of lifetime depressive symptoms.
In total, 14 studies were found to use a depression measure based on standardized clinical criteria (Table 1); however, these studies were not included in the meta-analyses because 13 of the studies used the same database, the National Latino and Asian American Study (NLAAS). The other study by Hwang & Myers [67] reported lifetime Major Depressive Episode (MDE) with DSM-IV criteria which was different from other studies; therefore, pooled prevalence of depression measured by standardized clinical approaches were not estimated. In regard to the risk of bias of individual studies, nearly half of the studies, 28 out of 58, had a score of 1, which indicates a high risk of bias. Specifically, the sample size for 39 studies was less than 500; 38 studies used non-random/convinient sampling methods. Several of the studies sampled from a restricted geographical area, while others represented multiple sites (Table 1)  studies were limited to Koreans, and 8 studies were limited to Chinese. Two studies each for Japanese, Taiwanese, and Vietnamese, and 1 study each for Filipino, and Indian were found.

Prevalence of depression
The prevalence rate of depression ranged between 2.6% and 71.0% in individual studies ( Table 1). The summary of the prevalence rate of depression by the various tools used to assess it is presented in Table 2. The meta-analytic pooled prevalence rate of depression (Fig 2) using the CESD was estimated to be 35.6% (95% CI 27.6%, 43.7%), with high heterogeneity (I 2 = 97.9%). The prevalence rate of depression measured by the GDS, with a threshold of 11, was 33.1% (95% CI 14.9%, 51.3%), with high heterogeneity (I 2 = 95.4%). The prevalence rate of depression measured with the PHQ-9, with a threshold of 5 indicating mild depression, was 26.9% (95% CI 20.2%, 33.7%), with high heterogeneity (I 2 = 93.5%). A relatively low prevalence

Sensitivity and subgroup analysis
The results of sensitivity and subgroup analyses of the pooled prevalence rates of depression by measurement are shown in Table 3. There was no significant pattern in the sensitivity analyses. Excluding studies with unreported participation rate or The prevalence rates of depression did not differ statistically by the percentage of women or between adults and elderly; subgroup effects based on gender and age did not explain the high statistical heterogeneity found in the primary analysis; however, the pooled prevalence rate of depression measured by CESD of special populations, which included maternity, caregivers, and homosexuals, was 58.8% (95% CI 47.2%, 70.4%). This prevalence rate was significantly higher than that of the estimate based on other groups (p = .003). Group differences among different ethnic groups were identified. The pooled prevalence rate of depression measured by CESD, excluding the special populations-maternity, homosexual, and caregivers-for Koreans was 33.3% (95% CI 27.5%, 39.1%) and this was significantly higher than that for Chinese (15.7%, 95% CI 6.5%, 24.9%, p = .012). For Filipinos, the prevalence rate was estimated to be 34.4% (95% CI 23.2%, 45.6%), and this was also significantly higher than that for Chinese (p = .049). The prevalence rate of depression was estimated to be 20.4% (95% CI 6.9%, 34.0%) for Japanese, which was not significantly different than that for Chinese.

Discussion
This is the first study to estimate the prevalence rate of depression among AA adults in the community using synthesized data obtained from a systematic review of the literature published in the past 10 years. There was variation in the pooled estimates of prevalence by the type of measure used to assess depression, but we found that depression is highly prevalent in AA. The pooled estimates from this review indicate that 35.6% of AA have depressive symptoms as assessed by the CESD, 33.1% by the GDS and 26.9% by the PHQ. These estimates are comparable to those found among patients with chronic disease [82]. Studies indicate that 4.5% to 11.3% of adult AA in the community meet the criteria for major depression, but the majority of estimates come from the same dataset. The differences of magnitude between estimates obtained from symptom screening versus standardized clinical approaches were expected, and this result is consistent with a similar review of depression in a different population [82]. However, attention should be given to the wide gap between the two estimates. As stated above, it has been reported that AA are less likely to be diagnosed with a mental disorder [12]. Further, it has been documented that AA with depression are less likely to access any depression treatment or to receive adequate mental health care compared to non-Hispanic Whites [10,12]. This is also related to lower rates of detection and treatment of depression, which may lead to a worse prognosis. Given that depressive symptoms are highly prevalent among AA in the US, greater effort should be given to establishing public health policy programs that increase access to mental health care, including adequate screening and a referral system. We also suggest studies linking this issue to health insurance coverage.
We found no gender effect on the prevalence rate of depression from this review. Inconsistent findings have been reported in previous AA depression studies. While no male-female differences were seen in a study by Yeung et al. [83], depressive symptoms have been found to be higher among Asian women than men [84]. Our review also did not find any age group differences. These results may be influenced by the complex and diverse sample characteristics of each study in this review. Specifically, subgroup analysis revealed that the prevalence rate of depression of special population studies that included maternity, homosexual, and caregivers were significantly higher than other groups. Similar findings have been found in non-AA populations such that lesbian, gay, and bisexual groups have higher rates of depression [85]. Depression of perinatal women has been studied and the importance of depression screening for this group has been addressed previously [86]. A recent prospective study found that spousal caregivers of persons with dementia have a high risk of developing a depressive disorder [87]. These findings suggest that some populations could be prioritized in public mental health interventions to prevent and screen for the occurrence of depression. This may also apply to underserved ethnic minorities, including AA, as significant health disparities persist in diverse communities across the US.
Another factor that explains the heterogeneity of the prevalence rate of depression would be ethnicity. The prevalence rate of depression measured by CESD from the studies excluding the special population-maternity, homosexual, and caregivers-was higher among Korean and Filipino subgroups, but not in Japanese, compared to Chinese. The emotional characteristics of Koreans-the feeling of regret regarding neglect of children or parents that would be labeled guilt or shame-has been reported to be associated with depressive symptoms in Koreans [88]. Interestingly, according to a cross-national comparison study of depressive symptoms between Japanese and Whites, Japanese respondents tend to have lower mean scores on the CESD than Whites [89]. The prevalence rate of diagnostic major depression among Chinese-Americans has been found to be higher [75] than the average among the total sample of the NLAAS [78]. Our findings in this review suggest that differences in the prevalence rates of depression exist among Asian ethnicity groups. Heterogeneity of AA ethnic groups in sociohistorical, cultural, economic, and political characteristics has previously been reported [12]. Further studies providing information about the subgroups of AA in this regard are required to build evidence to develop strategies for preventing or reducing depression in AA.

Limitations
The high estimated prevalence rate of depression in AA from this review should be interpreted with caution. Heterogeneity across the studies was considerably high, and the samples were very diverse in age and context. In addition, information related to depression in this population, such as immigration status, length of stay in the US, English language proficiency, and other psychological status (e.g., acculturation, racial discrimination) [12,14,90], was not synthesized in this review because of lack of information or heterogeneity of such information across the studies. Significant publication bias was found using the Egger's test. Small studies with low prevalence may be less likely to be published than studies reporting a high prevalence of depression. Generally, the studies in this review had a high risk of bias in reporting prevalence of depression. Most studies included in the meta-analysis used non-random sampling methods; thus, it is difficult to generalize the results to the total population of AA in the US. However, the problem of sampling AA due to the geographical distribution of Asians in the US has been noted. Outside of the major states of California and New York, obtaining satisfactory samples of AA using random sampling techniques is challenging [12]. This study fills a gap in the literature by providing an estimated aggregated prevalence rate of depression in AA using studies that selected samples from non-clinical settings from relatively diverse areas in the US. Non-standardization of methods, such as the measure used to assess depression across studies also detracts from the findings. For example, the CESD (developed in 1977), the most commonly used tool, does not cover current diagnostic DSM criteria of depression, and contains items such as the perceptions of others, talkativeness, or comparisons with others that are not necessarily related to depression [91].

Conclusion
We have systematically reviewed estimates of the prevalence rate of depression among noninstitutionalized AA and found a wide range of heterogeneity in depression estimates among AA adults. Specific subgroups of the population, such as homosexual, caregiver, and perinatal woman, were more likely to be depressed. Possible variations by ethnicity were also noticed. Practitioners and researchers who serve AA adults need to be sensitive to the potential diversity of depression expression and treatment-seeking across AA subgroups, and pay further attention to better recognize those suffering from depression. Studies examining health insurance coverage and access to medical care for AA are required to provide the evidence needed to establish more effective public health or public policy programs to better recognize of depression, and to increase access to mental health care for AA. Such studies and policy programs may ultimately lead to a decrease in the variation in the prevalence of depression among ethnic minorities and curb treatment disparities.