Effectiveness of cognitive behavioural therapy-based interventions for maternal perinatal depression: a systematic review and meta-analysis

Abstract

Background

Depression during the perinatal period (during pregnancy and the year after childbirth) is common and associated with a range of negative effects for mothers, infants, family members, and wider society. Although existing evidence suggests cognitive behavioral therapy (CBT) based interventions are effective for perinatal depression, less is known about the effect of CBT-based interventions on important secondary outcomes, and a number of potential clinical and methodological moderators have not been examined.

Methods

A systematic review and meta-analysis primarily examined the effectiveness of CBT-based interventions for perinatal depression on symptoms of depression. Secondary aims examined the effectiveness of CBT-based interventions for perinatal depression on symptoms of anxiety, stress, parenting, perceived social support, and perceived parental competence; and explored clinical and methodological moderators potentially associated with effectiveness. A systematic search of electronic databases and other sources was performed up to November 2021. We included randomized controlled trials comparing CBT-based interventions for perinatal depression with control conditions allowing for the isolation of the effects of CBT.

Results

In total, 31 studies (5291 participants) were included in the systematic review and 26 studies (4658 participants) were included in the meta-analysis. The overall effect size was medium (hedges g = − 0.53 [95% CI − 0.65 to − 0.40]); with high heterogeneity. Significant effects were also found for anxiety, individual stress, and perceived social support, however few studies examined secondary outcomes. Subgroup analysis identified type of control, type of CBT, and type of health professional as significant moderators of the main effect (symptoms of depression). Some concerns of risk of bias were present in the majority of studies and one study had a high risk of bias.

Conclusions

CBT-based interventions for depression during the perinatal period appear effective, however results should be interpreted with caution given high levels of heterogeneity and low quality of included studies. There is a need to further investigate possibly important clinical moderators of effect, including the type of health professional delivering interventions. Further, results indicate a need to establish a minimum core data set to improve the consistency of secondary outcome collection across trials and to design and conduct trials with longer-term follow-up periods.

Trial registration

CRD42020152254.

Introduction

Perinatal depression (PND) is a common mental health difficulty experienced during pregnancy and/or after childbirth, with global pooled prevalence rates estimated at 11.9% [1]. The impact of PND is significant for the mother, the infant, family members, and wider society. Negative effects for the mother include poor quality of life [2, 3], anxiety and stress [4, 5], and risk of death to the mother in the severest cases [6]. PND is also associated with difficulties in social relationships, partner relationships, and sexuality [6]. Critically, PND can have negative effects on the infant’s social, cognitive and emotional development, persisting into late childhood and adolescence [5]. This effect is mediated both directly via exposure to chronically elevated maternal cortisol during pregnancy, or indirectly, via the relationship between the parent and infant and parenting practices [7,8,9,10,11]. Mothers with PND are less likely to demonstrate sensitive and responsive interactions with their infants, and are more likely to report difficulties breastfeeding, establishing sleep routines, and attending vaccination appointments [12]. Psychological interventions, and investment in perinatal mental health services, are recommended for mothers with PND and other perinatal mental health disorders [13]. Importantly, mothers with PND report preferences for psychological support over medication, especially with concerns about the effects of medication on the infant [14]. However, despite these recommendations, gaps in mental health care provision in the perinatal period remain [15, 16].

Recent reviews have concluded that psychological interventions are probably effective for PND. However, some previous reviews have been broad in scope, for example including any type of psychological intervention (e.g., interpersonal psychotherapy (IPT), mindfulness and psycho education) [17, 18] or including both prevention and treatment interventions [19] and thus are highly heterogeneous [18]. Other reviews have been narrow in scope, for example focusing on interventions delivered in the postnatal period only [20, 21] or on specific delivery modalities (e.g., internet-administered interventions) [22, 23]. To the best of our knowledge, there is no recent systematic review and meta-analysis specifically focusing on the evidence-base for CBT-based interventions. Conducting a systematic review and meta-analysis focusing on CBT-based interventions may potentially reduce the high levels of heterogeneity present in more “broad scope” reviews [17,18,19]. Reducing high levels of clinical heterogeneity may also facilitate an exploration of a number of novel moderators, for example, the potential effect of the type of health professional delivering intervention and including parenting intervention components [24].

Additionally, there has been a large increase in the number of published randomized controlled treatment trials (RCTs) of CBT-based interventions for PND since the last review that focused on CBT-based interventions for PND was published [25]. Further, existing reviews of psychological interventions for PND have been criticised for being of low methodological quality and a need to conduct reviews of higher quality and avoid biases associated with cumulated evidence from individual trials of low methodological quality has been highlighted [17]. For example, existing reviews of CBT-based interventions for PND have included studies with non-randomized designs [25, 26], potentially resulting in biased estimations of effect [27].

Another limitation of the existing evidence base is that a number of important secondary outcomes have been largely unexamined [24], for example, anxiety, stress (individual and perceived parenting), parenting (e.g., sensitivity/responsiveness), perceived social support, and perceived parental competence. Given high comorbidity rates of PND and anxiety [4], the impact of stress on both mothers and the infant [28, 29], the association between PND and parenting difficulties [7], and poor social support [30], it is suggested that PND interventions should also try to improve these important outcomes. However, existing reviews have not adequately addressed the effect of CBT-based interventions on these secondary outcomes.

Finally, the increase in RCTs of CBT-based interventions for PND presents an opportunity to investigate a number of potential clinical and methodological moderators of intervention effectiveness. Whilst previous reviews [25] have examined some important clinical moderators (e.g., time point of intervention pregnancy/postnatal, and type of CBT intervention), there are a number of moderators potentially associated with effectiveness yet to be investigated. First, the potential moderating effect of the severity of depression at baseline has not been examined in reviews of CBT-based interventions for PND, despite evidence suggesting the effectiveness of CBT-based interventions for depression may vary by baseline severity [31]. Second, the relative effectiveness of the method of intervention delivery is important to investigate given recommendations that maternal mental health services should provide flexible and accessible intervention delivery formats to overcome multiple barriers to access experienced by mothers [32,33,34]. Third, little is known about the moderating effect of type of healthcare professional delivering the intervention (i.e., mental health provider or non-specialist provider). Given the global treatment gaps for PND, with up to 90% of mothers not receiving treatment in low- and middle-income countries (LMIC) [35], the provision of interventions by non-mental health specialist providers [36], may represent a solution to help close the treatment gap, should they be demonstrated to be effective. Finally, to date existing reviews of CBT-based interventions for PND have not examined the inclusion of parenting components as a moderator. Examining the potential effect of including intervention components targeting parenting is fundamental given the association between parenting difficulties, PND, and negative infant outcomes [7, 10, 11].

Given the aforementioned gaps in the current evidence base, an updated systematic review and meta-analysis of CBT-based interventions for PND is warranted. This review seeks to overcome the aforementioned limitations of previous reviews by: (1) attempting to reduce clinical heterogeneity by only including CBT-based interventions and excluding third wave CBT interventions and preventative interventions; (2) restricting study inclusion to RCTs whereby allocation and concealment procedures were determined to have a low risk of bias [24] and examine study quality as a potential moderator; and (3) investigating a number of clinical and methodological moderators potentially associated with effectiveness that are currently neglected in the literature.

The objectives are threefold:

• 1. To examine the effectiveness of CBT-based interventions for PND on symptoms of depression and depression diagnosis.

• 2. To examine the effectiveness of CBT-based interventions for PND on secondary outcome measures including: anxiety; stress (individual and perceived parenting stress); parenting (e.g., sensitivity/responsiveness); perceived social support; and perceived parental competence.

• 3. To investigate clinical and methodological moderators potentially associated with effectiveness.

Method

The review protocol is published [24] and registered in PROSPERO (CRD 42020152254). Methods are informed by Cochrane guidance [37], the Centre for Reviews and Dissemination guidance [38], and reporting follows the PRISMA 2021 statement [39] (Additional file 1).

Eligibility criteria

Population

Adult women (aged ≥16 years) with a diagnosis of PND, for example, Diagnostic and Statistical Manual of Mental Disorders (DSM) IV [40] or V [41] and/or reporting depression symptomatology within the perinatal period (from pregnancy to 12 months postnatal) using a validated tool (e.g., Edinburgh Postnatal Depression Scale (EPDS)) [42]. No limits were placed on depression severity given the variability in outcome measures and cut off scores across studies [43]. However, studies specifically designed to target populations referred to as “at risk” of depression were excluded.

Interventions

Eligible interventions explicitly targeted PND using CBT-based interventions, including standalone behavioral activation (BA) or problem-solving based interventions. CBT-based interventions were defined as interventions focusing on evaluating, challenging, and modifying dysfunctional beliefs [44], for example adopting treatment protocols in accordance with Beck’s manual [45]. Third wave CBT interventions such as mindfulness were excluded. Standalone BA interventions eligible for inclusion were defined as interventions targeting reductions in behavioral avoidance and increasing positively reinforcing activities [46], including pleasant activity scheduling [47, 48] and contextual BA models [49, 50]. Standalone problem-solving interventions eligible for inclusion were defined as interventions including a definition of personal problems, generation of multiple solutions to each problem, selection of the best solution, development of a systematic plan for this solution, and evaluation of the solution [51]. Eligible problem-solving intervention subtypes [44] included extended problem-solving therapy [52, 53], brief problem-solving therapy [54], and self-examination therapy [55].

No limitations were placed on the health professional group supporting or delivering the intervention, the clinical setting of intervention delivery, or method of intervention delivery. Following existing guidance [56, 57], self-help interventions were categorised as self-administered (no support provided in the use of the intervention), minimal contact (regular overview of materials in the provision of check-ins), and guided (regular support sessions provided to discuss progress and any process issues experienced using the materials). Interventions targeting a problem other than PND (e.g. bipolar affective disorder) or the prevention of PND in at-risk, but not currently symptomatic mothers were excluded.

Comparators

Eligible control conditions included: (1) no-treatment control; (2) wait-list control (WLC); (3) treatment-as-usual (TAU); (4) non-specific factors component control; (5) specific factors component control; and (6) active comparator, based on standard definitions [58]. Only trial designs allowing for the isolation of the effects of CBT were included [59].

Outcomes

Eligible studies used self-report or proxy/clinician administered standardized measures of depression or PND. Only studies using measures of depression with at least acceptable internal consistency (Cronbach’s alpha ≥0.70) and test-retest reliability (Cronbach’s alpha or correlation ≥0.70), as reported in outcome measurement validation studies, were included (Additional file 2). Secondary outcomes were self-report measures of: (1) anxiety; (2) individual stress; (3) perceived parental stress; (4) parenting (e.g., sensitivity/responsiveness); (5) perceived social support; and (6) parental competence. Observational parenting (e.g., sensitivity /responsiveness) measures (e.g., video tapes assessed with mind-mindedness coding manual (Meins & Fernyhough: Mind-mindedness coding manual, Version 2.2., unpublished) were also included.

Study designs

Only RCTs were included, with non-randomized and uncontrolled designs excluded. RCTs with randomization procedures explicitly not randomly allocated and/or with sequences not explicitly concealed (high risk of bias, in line with the Cochrane Collaboration’s Risk of Bias tool 2.0 (RoB 2.0)) [60] were excluded, in accordance with previous reviews, [57, 61] to minimize the risk of an inflated overall effect size resulting from the inclusion of low-quality studies [62].

Literature search and study selection

Electronic searches

Eligible studies in English and Swedish were identified through a comprehensive electronic database search (ASSIA; CENTRAL; CINAHL; EMBASE; ISI Web of Science; MEDLINE; Prospero; PsycINFO; SCOPUS; and SweMed+), clinical trials registers (www.ClinicalTrials.gov and www.who.int/trialsearch/) and conference proceedings (BIOSIS Previews; Conference Proceedings Citation Index, Health Management Consortium and Web of Science with Conference Proceedings). Grey literature was identified using OpenGrey, ProQuest, and DiVA (publishing database for Scandinavian universities). Databases were searched using medical subject headings (MeSH) and text words in the title and abstract. An example of the search terms used are provided: (postpartum OR post-partum OR antepartum OR ante-partum OR partum OR prepartum OR pre-partum OR intrapartum OR intra-partum OR peripartum OR peri-partum OR postnatal OR post-natal OR perinatal OR peri-natal OR antenatal OR ante-natal OR prenatal OR pre-natal OR pregnant OR pregnancy OR pregnancies OR puerper* OR maternal OR trimester OR impregnated OR gravid* OR multigravid* OR primigravid* OR parity OR obstetric OR gestation OR “in utero” OR maternity OR partus OR obstetrical) AND (depression OR depressed OR depressive OR “low mood” OR mood OR distress OR wellbeing OR “well-being” OR emotion OR emotional OR melanchol* OR affect OR affective OR dysphori* OR dysthymia OR alexithymia) AND (cognitive OR behaviour OR behavioural OR behaviour OR behavioural OR cognitive behavio* OR “behavioural activation” OR “behavioral activation” OR “problem solving” OR ccbt OR icbt OR “cognitive restructuring” OR “cognitive reframing” OR “activity scheduling”) AND (therapy OR therapies OR psychotherapy OR intervention OR management OR “program evaluation” OR program OR programs OR programme OR programmes OR group OR course OR online OR internet OR web OR “web-based” OR phone OR telephone OR skype OR “e-therapy” OR etherapy OR “computer assisted” OR “internet intervention” OR computer OR computerised OR computerized OR mobile OR tablet OR smartphone OR “internet administered” OR “e-mental health” OR “m-mental health” OR Ehealth OR “e-health” OR “e-intervention”) AND (“randomized controlled trial” OR “randomized control trial” OR RCT OR controlled OR randomised OR randomized OR randomisation OR randomization OR “random assignment” OR “random allocation” OR random OR randomly OR control OR feasibility OR pilot OR “comparative study” OR “follow up” OR meta-analysis OR “meta analysis” OR review). The exact search terms used for each electronic database search can be found in Additional file 3. 

The search strategy was developed following PRESS Peer Review Guidelines [63] (Additional file 4). All databases were searched from inception until November 2021. Searches for relevant dissertations were conducted; however, full dissertations (Additional file 5) were not reviewed and studies identified in languages other than English and Swedish (Additional file 6) were not included due to time and funding limitations.

Hand searches

Forward citation searches were conducted using Google forward citation chasing [64] and reference lists were hand searched for all included studies. Studies identified in relevant secondary evidence reports (e.g. relevant systematic reviews and meta-analyses) were also reviewed. Study selection was managed using Endnote referencing management software (Version, 9) and Microsoft Access 2016. Study duplicates across electronic searches were removed. Disagreements regarding inclusion were discussed between two reviewers (DP & OB), with a third (JW) or fourth (HOM) reviewer consulted when needed to reach consensus. Two independent reviewers conducted a wide screen of study titles and abstracts, followed by full paper checks of potentially eligible studies. Studies were excluded if they did not clearly meet the outlined PICOS criteria (Additional file 7). Authors were contacted by email in the event of missing data, with a follow up email sent if there was no response within two weeks.

Data extraction

Two reviewers independently extracted data from included studies and data was managed using Microsoft Excel 2016. Data extraction included: (1) study characteristics; (2) participant characteristics; (3) intervention characteristics; (4) study outcome measurements; and (5) participant flow. Discrepancies were discussed between the two reviewers (DP & OB), with a third reviewer (JW) consulted if consensus was not reached.

Risk of bias assessment

Methodological quality of the primary studies was assessed using RoB 2.0 [60]. Reviewers assessed risk of bias independently across the following domains: (1) randomization; (2) allocation to intervention; (3) adherence to intervention; (4) handling of missing outcome data; (5) measurement of outcome; and (6) selection of the reported results. Overall risk of bias was rated as “low”, “some concerns”, and “high” for each domain both across and within studies. Ratings were compared, discrepancies discussed, and consensus reached with a third reviewer (JW) where necessary. Rate of retention was set at 80% (for the primary time point at or closest to 6 months) as opposed to 95% suggested by the tool, as a cut-off of 80% is recommended elsewhere to separate high and low quality RCTs [65].

Statistical analysis

Measures of intervention effect

Meta-analysis was performed using Comprehensive Meta-Analysis version 3 [66]. Post-intervention between-group standardized mean effect sizes were calculated separately for primary (depression) and secondary outcomes (anxiety, individual stress, perceived parental stress, parenting, perceived social support, and parental competence) using Hedge’s g [67]. Incidence of major depressive disorder post-intervention was calculated using Odds Ratio (OR) alongside 95% confidence intervals (CIs) [68]. A primary end point ≤6 months post-intervention was adopted to minimise elevated effect sizes associated with short term follow up [69]. A random effects model [70] was adopted based on the expectation of large heterogeneity arising as a consequence of wide variations in the clinical and methodological parameters between studies [71].

Cochran’s Q statistic was used to examine the presence of heterogeneity [70], I² was used to measure the proportion of total variability due to between-study heterogeneity and the prediction interval was used as an index of dispersion of the population [72]. I² values are interpreted as low (above 25%), moderate (above 50%), and high (above 75%). On one occasion [73], a study included two CBT-based interventions delivered by different health professionals (nurses and psychologists), therefore comparisons were analysed separately, with the control condition sample size halved in each comparison. Where possible, intention-to-treat data was used, with completer data used when not available.

Sensitivity analysis

Sensitivity analyses for the overall effect size of the primary outcome measurement (depression) were conducted by temporary removal of: (1) each study individually from the overall analysis; (2) small studies (n ≤ 20 across conditions); and (3) studies with high attrition (≥30% in at least one arm), with the effect size recalculated.

Sources of possible bias

For outcome measures with at least 10 studies, funnel plot asymmetry was examined for sources of possible bias (e.g., publication bias, language bias, inclusion of small studies with poor methodological quality, and heterogeneity) [74, 75]. An estimated effect size taking biases into account was calculated using the trim and fill procedure [76].

Moderator analysis

Moderator analysis of associations between clinical and methodological moderators on the effect size for the primary outcome of depression were examined:

• Risk of bias (low vs. some concerns vs. high).

• Type of comparator (no-treatment control vs. WLC vs. TAU vs. non-specific factors component control vs. specific factors component control vs. active comparator).

• Length of follow up (short: post-intervention - less than 3 months vs. medium: 3–6 months vs. long: 7–11 months vs. extended:12 months+).

• Severity of depression at baseline (severe vs. moderate vs. mild), calculated using baseline mean scores and clinical cut offs for each depression measure.

• Type of CBT intervention (CBT vs. BA vs. problem solving).

• Interventions including additional social components (yes vs. no). Social components were defined as structured activities to improve social support e.g., partner session(s) or networking and communication skill building.

• Interventions including parenting intervention components (yes vs. no). Parenting intervention components were defined as including specific support in relation to the parent-infant relationship, for example specific sessions with a therapist, video feedback, or self-help materials (e.g., video interaction guidance) [77].

• Method of delivery (Individual ‘high intensity’ e.g., traditional CBT delivered by a trained psychological therapist workforce, typically weekly 60-minute sessions over at least a 10-week period [78] vs. group vs. guided or minimal contact self-help vs. self-administered self-help [56, 78, 79].

• Time point of intervention (prenatal vs. postnatal).

• Health professional delivering intervention (nursing professionals vs. social workers vs. psychologists vs. junior mental health workers vs. peers).

A random effects model was adopted, with Q reported as a measure of heterogeneity and I² used to measure the proportion of total variability due to between-study heterogeneity [80]. Consistent with previous meta-analyses [56, 81], the alpha level was set at ≤.10 in the event of there being a low number of available comparisons with respect to moderator analyses. Under these circumstances differences are reported as a trend in the data.

Protocol amendments

The following amendments were made to the published protocol: (1) the health professional delivering the interventions moderator was grouped into “non-specialist providers” e.g., peers and community workers, “health providers” e.g., nurses and midwives and “mental health providers” e.g., psychological wellbeing practitioners and clinical psychologists due to large variation in intervention providers; (2) a post-hoc moderator analysis was conducted comparing studies from LMIC and high income countries according to the World Bank classification [82]; and (3) the original protocol included a thematic synthesis of qualitative data to describe the acceptability of CBT-based PND interventions however due to the volume of qualitative studies eligible for inclusion being greater than anticipated, results are reported separately.

Results

Study selection

A total of 17,452 studies were identified via electronic databases with 262 potential clinical trials identified through searching www.ClinicalTrials.gov and www.who.int/trialsearch/. A further 40 possible studies were identified through reference and citation checking and contact with experts in the field. Following duplicate removal, the search strategy yielded 10,193 records. A total of 111 full text articles were assessed for eligibility and of these, 31 studies were eligible and included in the narrative synthesis (Additional file 8) and 26 studies provided enough data for inclusion in the meta-analysis (see Fig. 1). For references to excluded studies, see Additional file 9 and for references to included studies, see Additional file 10.

Fig. 1

Prisma flow diagram of the inclusion of studies

Study characteristics

A total of 5291 women were randomized in the 31 studies included in the narrative synthesis. Study characteristics are presented in Table 1 and intervention characteristics are presented in Table 2. Studies were conducted across 11 countries with 16.1% (5/31) conducted in LMICs. Thirty studies were published in academic journals and one was an unpublished report. The majority of studies provided information of funding sources, and ethical approval, with two studies providing no funding information [83, 84] and two providing no ethical information [85, 86].

Table 1 Study characteristics of studies included in the systematic review

Table 2 Intervention characteristics of studies included in the systematic review

Participants

In total, 38,059 women were screened across the 31 included studies with 5291 randomised, yielding an overall inclusion rate of 13.9% (5291/38059). Women’s mean age was 28.5 (SD 6.0, range 16–42 years) across the 25 studies reporting mean age. Interventions started during pregnancy in 12 studies, and during the postnatal period in 19 studies. In 15 studies, women were required to meet diagnostic criteria for MDD to be included, as measured by either the Structured Clinical Interview for DSM-IV (SCID; 8 studies); the Mini International Neuropsychiatric Interview (MINI; 3 studies); the Clinical Interview Schedule-Revised (CIS-R; 2 studies); or the Composite International Diagnostic Interview (CIDI; 2 studies). In the remaining 16 studies women needed to meet a standardised clinical cut off on a validated self-report measure of depression, including the EPDS (10 studies, cut offs ranging from ≥10 to ≥13) the BDI-II (3 studies, cut offs ranging from ≥11 and ≥ 14), and the PHQ-9 (3 studies, cut off ≥10) to be included.

Included studies described the cultural identities of women in diverse ways, with studies reporting ethnicity, race, or in some cases country of birth (Table 1). Average household income was reported using different currencies and was not suitable for synthesis, however three studies targeted low-income women [86, 106, 112], with one study specifically targeting minority ethnic women with low incomes [86]. In total, 28 studies reported severity of depression at baseline, with overall severity mild in two studies, moderate in 20 studies, and severe in six studies (Additional files 11 and 12).

Intervention

In the 31 studies, 32 CBT-based interventions were compared with a control group. Intervention characteristics of studies included in the systematic review are presented in Table 2. The majority of the interventions were based on CBT (24 interventions), others were labelled as standalone BA (5 interventions), standalone problem solving (2 interventions), or labelled as combined BA and problem solving (1 intervention). Methods of delivery included individual high intensity (17 interventions), group (8 interventions), guided or minimal contact self-help (6 interventions), and self-administered self-help (1 intervention). In the 25 interventions delivered in person/via the telephone, the number of sessions ranged from five to sixteen. In group interventions, group sizes ranged from 4 to 20 over 1–12 sessions. In total, five self-help interventions were delivered online and two were delivered by a workbook. Social support components were included in 12 interventions and parenting components included in two interventions.

Thirty-one interventions were supported or delivered by a variety of providers including health care providers such as nurses (8 interventions), mental health providers such as clinical psychologists, psychological therapists, and psychological practitioners (18 interventions), non-specialist providers such as peers (4 interventions) and mixed providers, consisting of health and mental health providers (2 interventions). Twenty-two interventions were delivered in person in a range of settings including in clinics (16 interventions), home (4 interventions), and mixed clinic/home settings (2 interventions).

Meta-analysis results for primary outcomes

CBT-based interventions for PND resulted in a medium effect size, Hedge’s g = − 0.53 (95% CI = − 0.65 to − 0.40; z = − 8.02; p < .001) for depression symptoms (26 studies, 27 comparisons, n = 4658) using a random effects model. Effect sizes and 95% CIs of the studies are shown in Fig. 2. Estimates of between-study variance were high and statistically significant (p < .001, Q = 77.0) and I² = 66.25. High heterogeneity is reflected in the prediction intervals, which indicated that the true effect size falls in the interval − 1.05 to 0.00 (Additional file 13).

Fig. 2

Effect sizes (Hedges g) and 95% confidence intervals for depression with time point of ≤6 months post-treatment

Sensitivity analyses

Sensitivity analyses revealed no significant change in effect size or p-value when temporarily removing each study from the analysis. Sensitivity analysis for studies with a small sample size was not possible as no studies met the a priori criteria of n ≤ 20 across conditions. In the 25 studies reporting attrition data, 9 studies (10 comparisons) with high attrition rate (≥ 30% in at least one arm) yielded a medium effect size, Hedge’s g = − 0.67 (95% CI = − 0.94 to − 0.41) and 15 studies (15 comparisons) with a low attrition rate yielded a small effect size, Hedge’s g = − 0.49; (95% CI = − 0.69 to − 0.32). Post-hoc analysis of the 9 studies (10 comparisons) with high attrition, revealed that the four studies using completer data only, yielded a large effect size, Hedge’s g = − 1.01 (95% CI = − 1.74 to − 0.26) and five studies (6 comparisons) using imputed data yielded a medium effect size, Hedge’s g = − 0.54 (95% CI = − 0.66 to − 0.42).

Sources of possible bias

The funnel plot and Egger’s test did not suggest significant publication bias and the trim and fill procedure suggested 0 studies were missing to the right of the mean effect, with an imputed point estimate Hedge’s g = − 0.44; (95% CI = − 0.51 to − 0.38), (Additional file 14).

Moderator analysis

Moderator analysis revealed three significant moderators on the overall effect size for depression; type of control [Q = 26.44, df = 3, p < .001], type of CBT intervention [Q = 9.50, df = 3, p = 0.02], and type of health professional delivering the intervention [Q = 23.19, df = 3, p < .001] (see Table 3). With respect to type of control, studies with a WLC or TAU yielded significantly larger effect sizes (p = < .001) than those using an active control or enhanced TAU. Studies using problem solving or CBT yielded significantly larger effects sizes (p = 0.02) than studies using BA or BA plus problem solving. Studies with interventions delivered by mental health providers and mixed providers (e.g., mental health provider and/or a health provider) yielded significantly larger effect sizes (p = < .001) than interventions delivered by only health or non-specialist providers. A trend was found for length of follow up (p = 0.09), with a short length of follow up yielding larger effect sizes than studies with medium or long follow ups. The remaining moderators including: risk of bias; severity of depression at baseline; inclusion of social components; inclusion of parenting components; method of delivery; point of intervention; and LMIC setting were all non-significant.

Table 3 Moderator analysis

Risk of bias

The majority of studies were rated as “some concern” of risk of bias (25 studies) and one study had a “high” risk of bias (Additional file 15).

Meta-analysis results for secondary outcomes

Meta-analysis for secondary outcomes where data was available (13 studies, 14 comparisons, n = 1689) are reported in Table 4.

Table 4 Meta-analysis for secondary outcomes

Anxiety

A small effect size was found for anxiety, Hedge’s g = − 0.44 (95% CI = − 0.55 to − 0.33; z = − 7.76; p < .001). Analysis of heterogeneity was non-significant (p = 0.77, Q = 9.02, I² = 0.00). Trim and fill analysis suggested 0 studies were missing to the right of the mean effect, with an imputed point estimate Hedge’s g = − 0.44; (95% CI = − 0.55 to − 0.33) (Additional file 16).

Stress and social support

A medium effect size was found for individual stress, Hedge’s g = − 0.56 (95% CI = − 0.80 to − 0.32; z = − 4.49; p < .001). Analysis of heterogeneity was not significant [p < .001, Q = 2.99, I² =0.00]. A small effect size was found for social support, Hedge’s g = 0.25 (95% CI = 0.14 to 0.36; z = − 4.46; p < .001). Analysis of heterogeneity was not significant (p = 0.32, Q = 5.88, I² = 14.95).

Parenting

No significant effects were found for perceived parental stress, Hedge’s g = − 0.16 (95% CI = − 0.77 to 0.45; z = − 0.51; p = 0.61). A large effect size was found for self-reported parenting, Hedge’s g = 0.94 (95% CI = − 0.01 to 1.88; z = 1.96; p = 0.05). Only one study reported outcomes for parental competence with a large effect size for CBT-based interventions for perinatal depression on parental competence Hedge’s g = 0.87; (95% CI = 0.26, 1.49; z = 2.78; p < 0.05) [86]. No studies reported observational parenting outcome measures.

Incidence of major depressive disorder

Five studies measured incidence of major depressive disorder post intervention, yielding a significant effect, OR = 0.21 (95% CI = 0.07 to 0.61, Z = -2.90, p < .05).

Discussion

Results of this systematic review and meta-analysis provide some support for the effectiveness of CBT-based interventions for PND, are in line with previous meta-analyses in the area [17, 18, 25], and findings remained significant after sensitivity analyses. Notably, effects were present in study populations that included women from a range of cultural and socio-economic backgrounds, including LMIC settings. However, studies with higher rates of attrition yielded significantly larger effect sizes than studies with smaller rates of attrition, especially when examining completer data only. There was a trend for studies with a shorter length of follow up to yield larger effect sizes than studies with medium or long-term follow ups. Only a small number of studies included important secondary outcome measures. Despite limiting the meta-analysis to studies of CBT-based interventions for PND, heterogeneity was high and similar to meta-analyses including a range of psychological interventions for the population [18], suggesting caution in interpreting these findings.

Moderator analysis indicated several significant moderators that may warrant further investigation. Consistent with previous research, studies with a WLC yielded significantly larger effect sizes than other control conditions [113,114,115]. Studies adopting a TAU control condition also yielded significantly larger effect sizes than those adopting an active control or enhanced TAU. Importantly, TAU has been identified as highly heterogeneous, with the effects of psychological interventions for adult depression found to differ, especially across countries [116]. Present results may indicate that perinatal women may not be receiving appropriate mental health treatment within usual care settings, a finding consistent with research demonstrating that access to evidence-based mental health treatment for women in the perinatal period is low, for example, within the United States, up to 85% of women with PND are estimated to remain untreated [117] and up to 90% of mothers in low- and middle-income countries do not receive treatment [35].

Type of professional delivering/supporting the intervention was also a significant moderator, with higher effect sizes found when the intervention was delivered by mental health and mixed providers (e.g., mental health provider and health provider teams) than interventions delivered only by health providers (e.g., midwives, health visitors, community nurse) or non-specialist providers (e.g., peer supporters). To the best of our knowledge, the moderating effect of type of healthcare professional delivering/supporting the intervention has not been previously examined within a PND population, and these findings require further investigation in future trials. Other research suggests guidance provided by certified psychotherapists versus general practitioners or medical doctors specialised in mental health may be associated with higher levels of adherence to internet-administered CBT for adult depression, but not overall symptom reduction [118]. Conversely, present findings indicate a need for interventions to be delivered by mental health providers, or within mixed health and mental health provider teams. There is a need for future research examining the impact of managing competing professional roles and identities [119, 120] on PND treatment delivery and what barriers health professionals may experience delivering perinatal mental health interventions [121, 122]. Future research may also wish to examine how PND treatment fidelity [123] may vary across professional groups to further inform training and supervision. Finally, a small effect size for non-specialist providers, indicates a need for enhancing training and supervision provided to develop and maintain competencies in line with the new competency framework for peer support workers [124].

Interestingly, type of CBT intervention was also a significant moderator, with larger effect sizes associated with CBT and problem-solving interventions versus BA interventions. Although this is a departure from previous meta-analyses of CBT intervention types for depression both within [125] and outside the perinatal period [126], it is important to note that three of the five BA studies included were delivered by health or non-specialist providers, suggesting intervention type and type of provider may have been confounded. Finally, we found a trend for higher depression severity at baseline to be associated with larger effect sizes. These results provide reassurance to clinicians treating women with more severe symptoms of depression and are consistent with reviews for adult depression [127].

Importantly, method of delivery was not a significant moderator of effect, and is supported by a mounting body of evidence from meta-analytic studies showing that method of delivery does not affect treatment effectiveness [18, 25, 79]. Similarly, there were no differences in the effect of the intervention on depression symptoms based on point of intervention, suggesting that interventions delivered either during pregnancy or postnatally are effective. We also found no evidence to suggest the inclusion of social or parenting interventions components were associated with effectiveness. However, it is important to note that in the 12 studies including some form of social component, components ranged from the provision of an informational brochure for a family member, work on communication skills, to including a family member in part of the treatment. Additionally, only two studies included parenting components, and given the association between parenting difficulties, PND, and negative infant outcomes [7, 10, 11] initial findings point strongly to the need for more research investigating CBT-based interventions that incorporate parenting components [11].

Finally, only a minority of studies in this review measured secondary outcomes and there was a lack of consistency regarding secondary outcomes adopted across trials. CBT-based interventions for PND produced large effects on self-reported parenting and parenting competence, moderate effect sizes on individual stress, and small effects on anxiety symptoms and perceived social support. No significant effects were found on perceived parental stress outcomes. Future CBT-based perinatal depression interventions would benefit from proactively addressing the problem of secondary outcome measurement, for example through the development of a minimum core data set for studies of interventions for perinatal depression [128].

Strengths and limitations

Strengths include: (1) literature search and study selection, data extraction, risk of bias assessment was completed by two independent reviewers; (2) a comprehensive search strategy, including electronic databases, grey literature, clinical trial registers, hand searching, and following PRESS Peer Review Guidelines [63] was conducted; (3) excluding RCTs with high risk of bias concerning randomization procedures, may have reduced the risk of an inflated effect size and increased the methodological quality of included studies; and (4) examination of important secondary outcomes and methodological and clinical moderators.

Limitations include: (1) limiting studies to those in English or Swedish which may have introduced language bias; (2) the small number of studies included in the moderator analysis means analysis was underpowered [129] and provides only correlational data with the potential for confounding between moderator variables [130]; (3) the methodological quality of included studies was low, with 96% of studies reporting “some concerns” and one study reporting a “high” risk of bias, however, the RoB 2.0 may not be suitable for psychotherapy outcome research [131] given challenges associated with blinding of participants and treatment providers [132]; (5) the potential dependency of effects was managed using a simplistic approach [133] and analysed potentially dependent effects as if they were non-independent, increasing the risk of Type I errors [134], an adoption of a three-level meta-analytic model may have been more appropriate [134, 135]; and (6) due to heterogeneity in the depression outcome measures adopted by included studies, severity of depression at baseline (severe vs. moderate vs. mild) was calculated using baseline mean scores and clinical cut offs for each depression measure, with moderator analysis performed. However, using mean depression scores to calculate severity of depression at baseline may reduce the variance of severity included in the meta-analysis, limiting our ability to make more definitive conclusions concerning the potential moderating effect of the severity of depression at baseline. The moderating effect of severity of depression at baseline could be further examined by conducting an individual patient data meta-analyses and examining the interaction between baseline severity and treatment effect using multilevel linear regression.

Conclusion

Findings from this meta-analysis demonstrate that CBT-based interventions for perinatal depression are effective both during pregnancy and the postnatal period for symptoms of depression. However, results should be interpreted with caution given high levels of heterogeneity and low quality of included studies. Results indicated that whilst it is important to increase access to PND interventions, caution should be exercised when utilizing healthcare and non-specialist providers without proper mechanisms in place to facilitate training, enhance fidelity, and avoid reduction of the power of interventions. Further, there is a need to conduct future research to examining factors such as the role of training, supervision, and treatment fidelity on treatment outcomes across different healthcare providers. Findings also highlight a need to integrate evidence-based parenting components into CBT-based interventions for PND, to establish a minimum core data set to improve the consistency of secondary outcome collection across trials, and conduct trials with longer-term follow-up periods.