Psychosocial Interventions for Perinatal Common Mental Disorders Delivered by Providers Who Are Not Mental Health Specialists in Low- and Middle-Income Countries: A Systematic Review and Meta-Analysis

In a systematic review and meta-analysis, Kelly Clarke and colleagues examine the effect of psychosocial interventions delivered by non–mental health specialists for perinatal common mental disorders in low- and middle-income countries. Please see later in the article for the Editors' Summary


Introduction
Common mental disorders, defined as depressive, anxiety, and somatic disorders, are a major cause of disability among women during the perinatal period, and may have consequences for children's growth and development [1][2][3][4]. In low-and lower middle-income countries an estimated 16% (95% CI 15.0%, 16.8%) of women suffer from these disorders in pregnancy, and around 20% (95% CI 19.2%, 20.6%) in the postnatal period [5]. To date, most reviews of interventions for perinatal common mental disorders (PCMDs) have focused on interventions for depression, and on evidence from high-income countries [6][7][8][9][10]. Their results may not be generalizable to low-resource settings, where specialists and financial resources for mental health care are scarce [11][12][13]. In these settings, the World Health Organization Mental Health Gap Action Programme recommends a costeffective package of interventions to treat depression that includes antidepressant, psychoeducation, and problem-solving therapies [14]. A recent meta-analysis showed that interventions for PCMDs in low-and middle-income countries are effective (effect Size [ES] 20.38; 95% CI 20. 56, 20.21), with benefits for children's health and cognitive development, and for the quality of mother-infant interactions [15]. The findings from this review, though useful, are limited by the diversity of interventions included and high statistical heterogeneity (I 2 = 79.9%). Effects of different intervention types and statistical heterogeneity were not fully investigated.
We have conducted a systematic review and meta-analysis of interventions for PCMDs in low-and middle-income countries that address the limitations of previous reviews. We include interventions for all PCMDs since depression and anxiety often coexist, and subcategories of common mental disorder may lack conceptual validity in some cultures [16][17][18]. We focus on psychosocial interventions (i.e., non-pharmacological interventions to influence thoughts, behaviors, skills, and associated feelings), given concerns about the safety of pharmacotherapy during the perinatal period and because access to psychotropic drugs and trained personnel to prescribe them can be limited in low-resource settings [19][20][21][22]. We also focus on interventions delivered by providers without specialized mental health training (''non-mental health specialists'') in community and primary care settings because of the lack of mental health professionals in low-and middle-income countries, and to address calls for integration of mental health interventions into existing community and maternal and child health programs [23,24]. We investigate the effects of these interventions based on the type of intervention, timing, and delivery mode, in order to make practical policy recommendations.

Methods
We conducted the systematic review in accordance with the 2009 PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement (Text S1) [25]. The protocol was finalized prior to conducting the systematic review and metaanalysis (Protocol S1). The review was not registered with PROSPERO or any other database.

Criteria for Trials Considered for the Review
Study types and origins. We considered published and unpublished, randomized and non-randomized controlled trials. Publication dates were not restricted, but only trials written in English, French, or Spanish were included. We restricted the review to trials conducted in low-and middle-income countries according to World Bank country classifications at the time of the search [26]. We included studies from mainland China because it is a middle-income country. However, we excluded studies from Taiwan and Hong Kong Special Administrative Region because economic conditions and health infrastructure in these regions of China are comparable with those of highincome settings.
Participants. We included trials that enrolled pregnant or postnatal women (#12 mo after delivery), or women who were not pregnant at recruitment but became pregnant during the trial.
Interventions. We considered preventive and treatment interventions involving a psychological or social component, delivered prior to pregnancy, during pregnancy, and/or postnatally. We included interventions delivered by non-mental health specialists, including lay persons (i.e., those without any health training), health workers and health volunteers (with some health training), and nurses and doctors with no specialized mental health training. We excluded interventions delivered by psychiatrists, psychologists (undergraduate or postgraduate level), and psychosocial workers, as these practitioners are not commonly available in low-and middle-income settings. We considered interventions in community settings (e.g., villages) and, knowing that the antenatal period is the time when most women are likely to come into contact with health services, the most commonly accessible health provider of antenatal care for their location (e.g., health posts, primary care centers, and hospitals).
Types of outcome measures. We included antenatal and postnatal PCMDs since a large proportion of PCMDs identified in the postnatal period are also present during pregnancy [27][28][29]. There is no consistent definition of the perinatal period in the psychiatric literature, so we adopted a working definition of pregnancy plus the first 12 mo after birth, in line with a number of trials [30][31][32]. We included trials measuring depressive, anxiety, panic, and somatic disorders, as well as perinatal psychological distress as a proxy measure of PCMDs [1]. We considered trials where outcomes were defined and measured using structured clinical interviews, such as the Clinical Interview Schedule-Revised [33], or validated screening questionnaires, for example, the Edinburgh Postnatal Depression Scale [34], the Kessler 10-Item Scale [35], and the 12-item General Health Questionnaire [36]. Outcomes included in the review were binary categorizations indicating the presence or absence of a PCMD (''caseness''), and reduction of symptoms of PCMDs as a continuous outcome.

Search Methods for Identifying Trials
Between 5 and 7 July 2013 we searched the following online bibliographic databases for trials that met the inclusion criteria detailed above: Medline, Embase, Cumulative Index to Nursing and Allied Health Literature, the Cochrane Library, PsycINFO, Web of Science, Scopus, Popline, Maternity and Infant Care, and the Global Health Library. We searched for unpublished completed or ongoing trials in the World Health Organization International Clinical Trials Registry. Customized search strategies were developed for each database. We used controlled vocabulary (e.g., MeSH terms) and search filters to identify randomized controlled trials, and trials from low-and middleincome countries where these were available. Our search of Embase (via OVID), including the exact search terms, is included in Text S2. The search mainly identified journal articles, but also reports, conference proceedings, and theses. We contacted experts in the field to identify further relevant trials, specifically unpublished or ongoing trials.

Data Collection and Analysis
Trial selection and data extraction. We removed duplicates and articles not written in English, French, or Spanish, and reviewed the abstracts of the remaining articles. Trials of interventions were retained, and observational studies excluded. We searched reviews of appropriate interventions for relevant citations. We contacted authors to request full articles where they were not available online, as well as further details of interventions as required. One reviewer (K. C.) independently screened full articles that appeared to meet the search criteria to assess the trial setting and design. We resolved any uncertainty about the inclusion of specific trials through discussions between reviewers, and documented reasons for exclusion.
Using a spreadsheet, two reviewers (K. C. and A. P.) independently recorded the following data for included trials: date of extraction, source reference and type, authors, publication year, article title, source of funding, trial design and methods, study setting and population, details about interventions and control conditions, participant inclusion and exclusion criteria and characteristics, sample size, definitions of PCMDs, screening tools, timing of assessments, variables adjusted for in the analyses, and results.
Assessment of methodological quality and small study effects. We did not exclude papers from the systematic review on the basis of methodological quality but assessed risk of bias for each study in terms of sequence generation, allocation concealment, blinding, incomplete outcome data, and selective reporting, using the Cochrane Risk of Bias Tool [37]. We defined trials at high risk of bias as those found to be at high risk or unclear risk of bias across five or more bias domains. Trials at low risk of bias were defined as those using adequate sequence generation and allocation concealment methods [37]. In reality these definitions were arbitrary, and studies may lie anywhere along the continuum from ''free of bias'' to ''undoubtedly biased'' [38]. We assessed potential small study effects using a funnel plot and the Egger test [38]. However, we attempted to limit small study effects by searching the World Health Organization International Clinical Trials Registry, and by asking expert informants about unpublished and ongoing trials.
Data synthesis and statistical analysis. We identified more than six studies that were not at high risk of bias and were comparable in terms of intervention content and study population [39]. We therefore conducted a meta-analysis to assess effects of psychosocial interventions versus usual care. We used the main outcomes reported in each publication, adjusted for clustering, baseline differences, and other covariates where appropriate. We conducted separate meta-analyses for binary and continuous outcomes. Odds ratios (ORs) were pooled for trials reporting binary outcomes. Where studies reported binary outcomes from both clinical interviews and screening questionnaires, we selected the former as the superior measure of PCMDs. One study reported a categorical outcome for the presence of PCMD (none/ mild, moderate, or severe) [40]. Data were therefore reanalyzed to calculate a binary outcome (none/mild versus moderate/severe) using the same methods reported in the publication. For continuous outcomes, standardized mean differences were calculated because different screening questionnaires were used to report the outcome [38]. We estimated statistical heterogeneity using the I 2 statistic and calculated confidence intervals around these estimates [41,42]. We used a random effects model to account for unexplained heterogeneity and because we assumed that the effects being estimated in the different trials were not identical [38]. We planned to exclude trials at high risk of bias from the meta-analysis (although all trials were included in the systematic review), and to conduct one sensitivity analysis including only trials at low risk of bias [38] and another including only results from the last follow-up assessment in each trial. We conducted a further post hoc sensitivity analysis excluding a study that was not peer-reviewed. We planned to conduct the following subgroup analyses: psychological interventions versus usual care, health promotion interventions versus usual care, group-based interventions versus usual care, individual-based interventions versus usual care, and combined (group-and individual-based) interventions versus usual care. We carried out two further post hoc subgroup analyses: antenatal interventions versus usual care, and antenatal and postnatal interventions versus usual care.
We wanted to compare treatment and preventive approaches versus usual care, but this was not possible because only one of the retrieved studies was a trial of a treatment intervention. In order to maximize power for subgroup analyses, we pooled results from all trials by converting ORs to ESs-comparable with the standardized mean difference-where studies did not report a continuous outcome [43]. In order to examine differences in ES between intervention subgroups we conducted a series of univariable random effects meta-regression analyses [44].
All data analyses were conducted with Stata (version 12.1) using metan and metareg commands.

Results
The database search identified 6,177 abstracts, which we screened according to the process outlined in Figure 1. We also identified five trials through personal communication with researchers. Two abstracts were unavailable online through University of London or British Library accounts. We were unable to obtain these abstracts through colleagues working in Asian institutions and could not locate the authors' e-mail addresses to contact them directly [45,46]. We screened 37 fulltext articles, and trials excluded at this stage are shown in Table  S1, with reasons for exclusion. Six ongoing trials, including one conducted in a low-income country, were not included in the review but are described in Table S2. In total, 11 trials were included in the review and are described in Table 1. Results from one trial are reported in two separate publications [47,48].
Target populations. Table 1 details components of interventions included in the review. We defined treatment interventions as those that targeted women diagnosed with a PCMD, and preventive interventions as those sampling from the general population, women at risk of developing a PCMD, and women with symptoms but not meeting the full criteria for a PCMD [58]. We identified only one treatment intervention [56].
In ten trials, participants were recruited during pregnancy [47,[49][50][51][52][53][54][55][56][57]. In Tripathy et al., all mothers who had delivered in the study area were interviewed approximately 6 wk after childbirth [40]. The intervention (participatory women's groups) was delivered at a community level: 18% of participants attended groups in the first year, rising to 55% in the third year. Overall for the included studies, participants' initial exposure to the intervention may therefore have occurred prior to or during pregnancy, in the postnatal period, or not at all.
The duration of interventions ranged from 4 wk [49] to 20 mo [40]. Where appropriate, we compared their intensity by calculating the number of scheduled contact events (group or individual) per month (Table 1). Of six trials, the least intensive intervention involved two group sessions plus a follow-up telephone call over a period of 9.5 mo [47]. The most intensive intervention involved three group exercise sessions per week for 3 mo [53].

Intervention Content
The treatment intervention and three of the ten preventive interventions involved psychological components [47,49,51,56]. Psychological interventions were defined as interventions incorporating a structured and explicitly psychological approach, such as cognitive behavior therapy (CBT) or interpersonal therapy (IPT).
Seven trials tested health promotion interventions [40,50,[52][53][54][55]57]. Health promotion approaches were defined by the absence of a structured and explicitly psychological approach, and incorporation of one of the following components: communication techniques to positively influence individuals and communities; education to improve knowledge and skills conducive to health; sharing of common experiences or problems and social support; creation of better environments to promote healthier living; community development and mobilization to address health problems; or advocacy and health policy development [59].
Two health promotion interventions involved educational workshops and/or home visits, specifically focusing on motherinfant interactions and attachment [52,55]. One intervention was a participatory learning and action cycle to improve maternal and newborn health, through women's groups [40]. Groups were also used to deliver an antenatal exercise program incorporating motivating techniques, including support by a physiotherapist, exercise with other women, and music [53]. Two interventions used home visits to communicate information to participants about topics including perinatal health care, nutrition, and mother-infant interaction [50,57]. One of these interventions promoted infant gender equality and had a strong emphasis on listening to participants [57]. Home visits were used in another intervention to disseminate information about pregnancy and delivery to participants and their chosen ''support persons'' [54].
Details of care received by control groups are included in Table 1.

Intervention Delivery Mode and Personnel
Psychological interventions were delivered by health workers [56], lay persons (''mentor mothers'' [51]), and doctors or midwives [47,49]. Three out of four psychological interventions were predominantly delivered in a group context [47,49,51]; one psychological intervention was delivered during individual home visits [56].
Health promotion interventions were delivered to groups and individuals by community health workers, social workers, physiotherapists, obstetric nurses, and lay women.

Methodological Quality of Trials and Risk of Bias
All but one of the studies had been peer-reviewed [57]. Most used self-report measures validated in the study population, and ten used a measure validated in the country in which they were conducted [40,47,[49][50][51][52][53][54][55]57]. One trial in Pakistan used the Hamilton Depression Rating Scale, which had not been formally validated in this context but which was translated, culturally adapted, and administered by experienced mental health professionals [56]. Statistical analysis in three trials included in the intervention did not take account of clustering [47,49,53].
With regards to completeness of follow-up data, the information provided was adequate in nine trials [40,47,49,50,52,53,[55][56][57] and inadequate in one trial [51]; in another trial, reasons for loss to follow-up were not discussed [54]. Two trials reported high attrition rates: 24% at 12 mo [52] and 55.6% at 6 mo [51]. We were unable to assess selective reporting (defined as the occurrence of one of the following: not all of a study's prespecified outcomes reported, primary outcomes not prespecified, outcomes incompletely reported, or key outcomes expected to be reported not reported) in the majority of trials for which the study protocol was not available.

Intervention Effects: Meta-Analysis
Out of the 11 trials that met the inclusion criteria, ten had useable outcomes for 18,738 participants [40,47,49,50,[52][53][54][55][56][57]. One trial found to be at high risk of bias (which had not used adequate sequence generation and allocation concealment methods) was excluded from the meta-analysis to reduce the impact of bias on the results [38,51]. Figures 2 and 3 show the pooled effects of any intervention (ten in total) versus usual care, for dichotomous ( Figure 2) and continuous outcomes (Figure 3), after intervention. There was evidence that interventions delivered by non-mental health specialists compared to usual perinatal care were associated with a reduction in PCMD symptoms (ES 20.34; 95% CI 20.53, 20.16) and caseness (OR 0.59; 95% CI 0.26, 0.92) immediately after the intervention. Heterogeneity was high (I 2 = 84.1% and 79.3%, respectively) and statistically significant.

Comparison 1: All Interventions versus Usual Care
We conducted sensitivity analyses excluding the study that was not peer-reviewed [57], and using binary and continuous outcomes associated with the final assessment, as opposed to the assessment immediately after the intervention. These analyses resulted in similar ESs. We also performed a sensitivity analysis using studies with low risk of bias and found that the ES was reduced for PCMD symptoms and caseness (ES 20.19; 95% CI 20.36, 20.02; OR 0.61; 95% CI 0.22, 1.01). Statistical heterogeneity was not significantly reduced in any of these sensitivity analyses.
In order to pool results from all ten trials of psychosocial interventions, we converted ORs to ESs where trials did not report a continuous outcome. The pooled ES of converted and unconverted outcomes was significant (ES 20.27; 95% CI 20.42, 20.13). A funnel plot of these outcomes was broadly symmetric ( Figure S1), though the power to detect small study effects with this method was low given the small number of trials included in the meta-analysis. The Egger test provided no evidence of small study bias on PCMD symptoms (p = 0.205).

Comparison 2: Interventions by Type versus Usual Care
We conducted a subgroup analysis to assess whether ESs differed by intervention type (Figure 4). This analysis was important because heterogeneity was high in the main comparison (Comparison 1), and subgroup analyses can provide explanations for heterogeneity. Health promotion interventions for PCMDs were evaluated in seven trials with a total of 17,401 participants, and these interventions were beneficial compared to usual care (ES 20.15

Comparison 3: Interventions by Delivery Method versus Usual Care
We also conducted subgroup analyses to examine whether ESs differed by intervention delivery method ( Figure 5

Comparison 4: Interventions by Timing versus Usual Care
We found that interventions delivered during pregnancy and postnatally had a significant overall effect compared to usual care (n = 15,816; ES 20.26; 95% CI 20.42, 20.10), whereas those delivered only during pregnancy did not (n = 2,555; ES 20.46 95% 20.94, 0.01) ( Figure 6). Only one trial evaluated an intervention restricted to the postnatal period [55]. Intervention timing was not associated with ES in a meta-regression analysis (b 0.16; 95% CI 20.16, 0.49).

Discussion
Our results show there is promise for psychosocial interventions delivered by non-mental health specialists for PCMDs in middleincome countries, and corroborate findings from a previous metaanalysis [15]. We identified a group of trials distinct from this previous meta-analysis through exclusion of trials that did not meet our inclusion criteria [32,[60][61][62][63], exclusion of a pilot study [64] of a trial that we included [52], and inclusion of recent [50,53] and additional [51,54] trials. In both meta-analyses, the lack of trials from low-income countries is striking, and research to determine the feasibility and effectiveness of delivering such interventions in these countries is urgently needed.

Study Limitations
Our findings are exploratory and should be interpreted with caution for several reasons. First, only ten trials were included, some of which were associated with an unclear risk of bias. The small number of trials made it difficult to assess small study effects, which, if present, may have led to overestimation of the true effect of interventions. Statistical analysis in three trials included in the meta-analysis did not take account of clustering [47,49,53]. The exclusion of these trials in the sensitivity analysis that included only trials at low risk of bias reduced the overall ES for PCMD symptoms and caseness, suggesting that trials that did not take account of clustering may have received more weight in the metaanalysis than is appropriate. Second, interventions differed in terms of participants, timing, setting, personnel, duration, and delivery mode, and meta-analyses showed high levels of statistical heterogeneity. However, the overall impact of psychosocial interventions on PCMDs was clear, and heterogeneity was reduced in subgroup analyses of psychological and health promotion interventions. Third, we excluded trials reported in all languages other than English, French, or Spanish. Fourth, the comparison group in most trials was usual perinatal care, which, in many settings, is likely to amount to no care. Beneficial effects of interventions in these trials are therefore not surprising, and future trials should consider more active comparison groups to control for nonspecific effects of contact with health workers and for ethical reasons. Finally, trials included in the meta-analysis were all from middle-income countries, and most were from Asia. The generalizability of the study findings for low-income and non-Asian countries is therefore limited.

Addressing PCMDs through Psychological Intervention
Our results suggest that psychological interventions for PCMDs are effective. Because we identified only three trials of psychological interventions, it is not possible to recommend one form of psychological therapy over another. However, meta-analyses combining trials from high-income countries and low-and middle-income countries have shown that CBT-based interventions are effective in reducing levels of PCMDs [7,9]. Moreover, a meta-analysis of psychological interventions for general adult depression and anxiety disorders in low-and middle-income countries found that CBT-based interventions had significantly larger ESs than interventions incorporating other therapies [65]. IPT-based interventions have also shown promise in resourceconstrained settings: two trials in rural Uganda showed strong benefits of group IPT interventions delivered by non-mental health specialists for treating general depression in adults and adolescents [66,67].
Despite these apparent benefits of psychological interventions for common mental disorders, the interventions must be adapted for individual contexts [68]. For example, where strong gender inequalities exist, it may be unrealistic to expect a psychological intervention to empower women in a way that they are individually able to negotiate for change in their lives. Also, there may be stigma associated with participation in an intervention explicitly for mental illness. The Thinking Healthy CBT intervention for depressed Pakistani mothers addressed these contextual factors by using infant health to mobilize family members to improve conditions for the infant's mother, and by integrating the intervention into an existing community health program [56].
We included one psychological intervention to treat participants with established PCMDs and two to prevent PCMDs. Psychological interventions may be more human-resource intensive than other interventions, since they require qualified trainers and supervisors, as well as multiple sessions to build rapport between the participant and ''therapist.'' Delivering preventive psychological interventions to all pregnant women or new mothers is unlikely to be cost-effective, particularly in remote rural contexts without access to mental health care. Further data on the sustainability and affordability of these programs is therefore required.

Health Promotion Interventions: Addressing Determinants of PCMDs
Although psychological interventions were associated with a significantly larger ES, we found that health promotion interventions also reduced symptoms of PCMDs compared to usual care. Health promotion interventions were diverse but had two common components: sharing information and developing skills to enhance perinatal health-though not specifically perinatal mental health-and giving women an opportunity to share concerns and feelings and receive social support in the context of a group or individually. Tripathy et al. hypothesized that their participatory intervention with women's groups also developed problem-solving skills and empowered women in their communities; this may account for the positive result reported in this trial in its final year [40]. Evidence from qualitative studies suggests that women with common mental disorders do not consider themselves to be ill but attribute their symptoms to social difficulties [69][70][71]. More social and less individual-focused interventions involving health promotion approaches may therefore be more acceptable.
Although health promotion interventions did not directly address mental illness, they did address determinants of PCMDs, such as poor maternal health, infant mortality, and lack of social support. Numerous general community-based interventions in low-and middle-income countries have successfully addressed risk factors for PCMDs, for example, domestic violence [72], poor access to maternal health care [73], and neonatal mortality [74,75]. More explicit recognition of women's mental health as both a mediator and consequence of these outcomes may increase the effectiveness of such interventions (both in terms of improving women's mental health as well as other targets including reduction in domestic violence), and future trials should consider incorporating a mental health outcome.
We were unable to carry out a subgroup analysis of treatment versus preventive interventions because only one treatment intervention was identified [56]. However, all seven health promotion interventions adopted a preventive approach. In the context of low-and middle-income countries, preventive interventions have several advantages over treatment interventions. First, the chance of detecting PCMDs in low-and middle-income countries is low if access to health care is low in the perinatal period. An intervention involving assessment of mental health status prior to participation may therefore be unrealistic. Preventive interventions are not necessarily dependent on detection of mental illness. Second, some of the effects of PCMDs on infants are thought to begin within the first few months after birth [76]. Delayed diagnosis and treatment could therefore lead to early disruption of the mother-infant relationship, as well as an extended period of distress for the mother. A preventive intervention might avoid these harmful effects. Third, training and supervising personnel to deliver psychological or pharmacological treatment interventions may be more laborious and costly than training personnel to deliver preventive interventions addressing social determinants of PCMDs. Finally, preventive interventions that reduce population levels of domestic violence, poverty, and reproductive ill health that perpetuate mental illness are likely to have a long-term impact on the prevalence of PCMDs.

Delivery of Psychosocial Interventions
Although delivery method was not associated with ES, we found evidence that interventions delivered through groups reduced symptoms of PCMDs compared to usual care, and that groupbased interventions were associated with a larger ES than individual interventions. The ''one-to-many'' approach employed by group interventions is attractive in resource-constrained settings and where it is more culturally appropriate for women to come together to discuss their problems rather than having one-to-one discussions with a health professional. Previous meta-analyses that included subgroup analyses of group interventions for PCMDs in high-income countries reported inconsistent results: one study reported no overall reduction in postnatal depressive symptoms; another study showed a significant effect of group interventions compared to individual interventions [7,8]. Such contradictory results have led some authors to question the efficacy of psychological group interventions for mothers with young children [77]. In contrast, three meta-analyses of trials from high-income countries all reported that individual interventions reduced levels of PCMD compared to usual care [7][8][9]. The fact that interventions incorporating both group and individual components did not have an impact on PCMDs warrants further investigation. However, only two trials were included in this subgroup, and the finding should therefore be interpreted with caution.

Onset, Duration, and Intensity of Intervention
We found that interventions delivered during pregnancy and postnatally were associated with a reduction in symptoms of PCMDs compared to usual care. The fact that interventions restricted to pregnancy had no significant effect on PCMDs compared to usual care suggests that intervention in the postnatal period is important. In support of this, a meta-analysis of trials from high-income countries showed that psychosocial interventions delivered postnatally prevented postnatal depression compared to usual care, whereas those beginning antenatally and continuing postnatally had no effect [8]. Postnatal psychosocial interventions may be more beneficial because women rely on social support and emotional resilience in the postnatal period to care for a newborn, recover from childbirth, and resume their daily routines. Interventions addressing anxieties around childbirth and perinatal health may be more appropriate for pregnant women.
In the current review the duration and intensity of interventions was variable but did not appear to be correlated with ES. There is little evidence in the literature for an optimum, or even minimum, number or frequency of sessions, although findings from a metaanalysis of trials in high-income countries indicated that interventions involving a single contact event do not prevent postnatal depression, whereas interventions with multiple contact events are efficacious [8].

Personnel
A recent review of possible packages of care for depression in low-and middle-income countries included routine screening for detection of depression, psychoeducation, and problem-solving [78]. This meta-analysis and other key trials of interventions for general common mental disorders provide some evidence that community health workers or lay workers can deliver these nonpharmacological interventions [67,[79][80][81]. Advantages of working with these cadres are that interventions can be delivered at the community level and in areas without access to mental health care. However, community health workers are already indispensable in the provision of perinatal care, family planning, health education, HIV/AIDs care, and immunization programs. Their existing workload may limit their availability for mental health interventions. Referral of severe mental illness must also be considered, and nesting interventions in existing health care services where specialist care and pharmacotherapy can be provided is one potential strategy. Trials of interventions integrated into primary care settings in India and Chile have reported promising results [32,81,82]. Further consideration is needed to adapt existing mental health care packages for PCMDs. For example, routine screening for PCMDs has been demonstrated to be not costeffective in high-income countries, and could overwhelm weak health systems in low-and middle-income countries [83].

Conclusions
Evidence supports the implementation of psychosocial interventions for PCMDs delivered by non-mental health specialists in middle-income countries. We found stronger evidence for the efficacy of psychological interventions, compared to health promotion interventions. More research is needed to evaluate the impact of such interventions in low-income countries, as well as research to compare treatment and preventive approaches, and antenatal versus postnatal interventions.

Supporting Information
Figure S1 Assessment of small study effects on ES of psychosocial interventions for PCMDs.

(TIF)
Protocol S1 Protocol for a systematic review and metaanalysis of psychosocial interventions for PCMDs delivered by non-mental health specialists in low-and middle-income countries.
(DOC) Text S1 PRISMA checklist of items to be reported in a systematic review or meta-analysis.

(DOC)
Text S2 Search terms for Embase via OVID.

(DOC)
Editors' Summary Background. Perinatal common mental health disorders are among the most common health problems in pregnancy and the postpartum period. In low-and middle-income countries, about 16% of women during pregnancy and about 20% of women in the postpartum period will suffer from a perinatal common mental health disorder. These disorders, including depression and anxiety, are a major cause of disability in women and have been linked to young children under their care being underweight and stunted.
Why Was This Study Done? While research shows that both pharmacological (e.g., antidepressants or anti-anxiety medications) and non-pharmacological (e.g., psychotherapy, education, or health promotion) interventions are effective for preventing and treating perinatal common mental disorders, most of this research took place in high-income countries. These findings may not be applicable in lowresource settings, where there is limited access to mental health care providers such as psychiatrists and psychologists, and to medications. Thus, non-pharmacological interventions delivered by providers who are not mental health specialists may be important as ways to treat perinatal common mental health disorders in these types of settings. In this study the researchers systematically reviewed research estimating the effectiveness of non-pharmacological interventions for perinatal common mental disorders that were delivered by providers who were not mental health specialists (including health workers, lay persons, and doctors or midwives) in low-and middle-income countries. The researchers also used meta-analysis and meta-regression-statistical methods that are used to combine the results from multiple studies-to estimate the relative effects of these interventions on mental health symptoms.
What Did the Researchers Do and Find? The researchers searched multiple databases using key search terms to identify randomized and non-randomized clinical trials. Using specific criteria, the researchers retrieved and assessed 37 full papers, of which 11 met the criteria for their systematic review. Seven of these studies were from upper middle-income countries (China, South Africa, Columbia, Mexico, Argentina, Cuba, and Brazil), and four trials were from the lower middle-income countries of Pakistan and India, but there were no trials from low-income countries. The researchers assessed the quality of the selected studies, and one study was excluded from meta-analysis because of poor quality. Combining results from the ten remaining studies, the researchers found that compared to usual perinatal care (which in most cases included no mental health care), interventions delivered by a providers who were not mental health specialists were associated with an overall reduction in mental health symptoms and the likelihood of being diagnosed with a mental health disorder. The researchers then performed additional analyses to assess relative effects by intervention type, timing, and delivery mode. They observed that both psychological interventions, such as psychotherapy and cognitive behavioral therapy, and health promotion interventions that were less focused on mental health led to significant improvement in mental health symptoms, but psychological interventions were associated with greater effects than health promotion interventions. Interventions delivered both during pregnancy and postnatally were associated with significant benefits when compared to usual care; however, when interventions were delivered during pregnancy only, the benefits were not significantly greater than usual care. When investigating mode of delivery, the researchers observed that both group and individual interventions were associated with improvements in symptoms.
What Do These Findings Mean? These findings indicate that non-pharmacological interventions delivered by providers who are not mental health specialists could be useful for reducing symptoms of perinatal mental health disorders in middle-income countries. However, these findings should be interpreted with caution given that they are based on a small number of studies with a large amount of variation in the study designs, settings, timing, personnel, duration, and whether the intervention was delivered to a group, individually, or both. Furthermore, when the researchers excluded studies of the lowest quality, the observed benefits of these interventions were smaller, indicating that this analysis may overestimate the true effect of interventions. Nevertheless, the findings do provide support for the use of non-pharmacological interventions, delivered by non-specialists, for perinatal mental health disorders. Further studies should be undertaken in low-income countries.