Is cognitive behavioral therapy a better choice for women with postnatal depression? A systematic review and meta-analysis

The present study evaluated the combined effectiveness of cognitive behavioral therapy (CBT) for postnatal depression. A systematic search was conducted across databases including PubMed, Embase, and the Cochrane library to identify the randomized controlled trials (RCTs) that assessing CBT versus control for postnatal depression until March 2017. Data was extracted by two reviewers, independently. The Review Manager 5.3 and Stata 11.0 were used to calculate the synthesized effect of CBT on depression, and anxiety. A total of 20 RCTs involving 3623 participants were included. The results of meta-analysis showed that CBT was associated with a better Edinburgh Postnatal Depression Scale (EPDS) than control in short-term (mean difference = -2.86, 95% CI: -4.41–-1.31; P<0.05) and long-term (mean difference = -1.68, 95% CI: -1.81–1.56; P<0.05). CBT also improved short-term (mean difference = -6.30, 95% CI: -11.32–-1.28; P<0.05) and long-term (mean difference = -4.31, 95% CI: -6.92–-1.70; P<0.05) Beck Depression Inventory (BDI). Subgroup analysis based on intervention types showed that in-home and telephone-based therapy exhibited significant reductions in EPDS scores (P<0.05 for all). CBT significantly improved the short-term [odds ratio (OR) = 6.57, 95% CI: 1.84–23.48; P<0.05] and long-term (OR = 2.00, 95% CI: 1.61–2.48; P<0.05) depressive symptomatology as compared to control. CBT also reduced the score of Depression Anxiety Stress Scales (DASS), though without significance. In conclusion, CBT effectively improved the symptoms and progression of postnatal depression.


Introduction
Pregnancy is usually accompanied with various psychological conditions such as depression, even after childbirth [1]. As reported by the world health organization (WHO), depression is a leading cause of disability worldwide [2]. Depression is rendered to increase the disease burden in women with respect to all health conditions [3]. Depression negatively affects physical and social functions and is a major risk factor for suicide [4]. Anxiety and depression are the most common symptoms in the complex psychological changes during pregnancy [5][6][7]. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 applied during the search. The primary language of eligible studies was English; however, articles in Chinese or other languages were also considered for inclusion as necessary. The manually searching of target journals was performed to obtain potential eligible studies and the corresponding authors of included studies were contacted if there were missing key data.

Inclusion and exclusion criteria
Inclusion criteria: Patients: The cohort of patients included women, who gave birth to an infant in the past year, with formal diagnosis of PND or with the risk of PND. Comparison: Each trial was comprised of two groups, one received CBT, and the other received usual care, home visiting, waitlist control, or other conventional treatment. Outcomes: The mean depression score and its associated standard deviation before and after intervention in the control and experimental groups were extracted and recorded[common depression score scale includes Edinburgh Postnatal Depression Scale (EPDS) and Beck Depression Inventory (BDI-II)] [9,25]. The improvement rates of PND after treatments were also extracted. Study design: the forms of CBT included internet-based therapy, telephone-based therapy, in-home therapy, and mindfulness-based cognitive behavioral therapy. Therefore, if the interventions of the included studies belonged to one of the therapies mentioned above, the RCTs would be considered for inclusion.
Exclusion criteria: Observational studies, animal studies, reviews, and other types of non-RCTs were excluded. Articles without sufficient information or data were also discarded after contacting key authors.

Data extraction
The baseline characteristics of patients and target data of outcomes were extracted from eligible studies by two reviewers, independently. These information were about study authors, published year, the number of cases, average age, sex ratio, diagnostic criteria of PND, and baseline history of depression. Primary data were interventions, symptoms of anxiety and depression, improvements rates, different types of depression score scale, diagnostic accuracy statistics, relapse rate of PND. Secondary outcome was the improvement in quality of life. When there was discrepancy about the extracted data, it would be resolved by discussing with the third reviewer.

Quality assessment of included studies
RCTs were included only if they fulfilled the inclusion criteria and the following quality assessment. The quality of included studies was evaluated according to the criteria provided by the Cochrane Handbook for Systematic Reviews of Interventions [26]. The quality elements such as selection bias, blinding bias, incomplete outcome data, selective reporting bias, and other bias were assessed. A third reviewer resolved the disagreement in case of controversy. The final quality of each study was defined as high-risk, unclear risk or low-risk of bias, according to the handbook mentioned above.

Statistical analysis
This meta-analysis was conducted using the RevMan 5.3 software provided by the Collaboration Network [26] and the Stata 11.0 software (StataCorp LLC). Data were represented as standardized mean difference (SMD) and its 95% confidence interval (95% CI) [2]. The chisquared test (presented as I 2 ) was used to analyze the heterogeneity of the included studies (P<0.1 was set as the test level), and the degree of heterogeneity was judged based on the value of I 2 . If I 2 was >50%, it was considered a high-risk of heterogeneity. Then, the subgroup analysis was introduced to evaluate the clinical homogeneity. If a high-risk of heterogeneity as absent in the analysis, the fixed effect model was used, otherwise, the random-effects model was applied. The sensitivity analysis was conducted when a high-risk of bias was indicated. The odds ratio (OR) with a 95% confidence interval (95% CI) was used to present the dichotomous data. The continuous variables were analyzed in term of mean ± standard deviation. For continuous data, the weighted mean difference (WMD) was used if the measurement unit of continuous variables was identical, or, the standardized mean difference (SMD) was used. Funnel plots and the Egger's and Begg's tests were used to assess the publication bias. A P<0.05 was considered statistically significant for combined effectiveness assessment.

Literature research
The search results of the RCTs by the two evaluators were highly consistent. According to the search strategy and data collection methods, a total of 702 articles published from 1993-2017 were retrieved. After screening the titles and abstracts, 628 articles were abandoned due to the irrelevant topic or research purposes, animal study, drug treatment, or improper evaluation of efficiencies and outcomes. After further reading the full text, 54 studies were excluded because of non-RCTs, reviews, and reduplication. Finally, 20 studies [15,20,21,[27][28][29][30][31][32][33][34][35][36][37][38][39][40][41][42][43] (Fig 1) were included. These studies involved 3623 participants, comprising of 1923 cases in the psychological intervention group and 1700 in the control group. These studies were performed worldwide and published in English.

General information and characteristics of included studies
Patients. Among these 3623 participants, 1923 cases were in the intervention group, and the remaining 1700 cases were in the control group. The age of this population ranged from 18 to 37 years. Most of the participants were primiparous and married. The definition and criteria of PND varied among the included studies. EPDS, BDI, Beck Anxiety Inventory (BAI), and Hamilton Rating Scale were used in various studies for defining PND. The baseline characteristics of the eligible studies are listed in Table 1.

Interventions
All studies had intervention and control arms. Among these studies, internet-based therapy, telephone-based therapy, in-home therapy, and mindfulness therapy were used as the intervention for PND patients. These therapies were carried out by psychologists, or trained nurses, or researchers. The duration of treatment ranged from 1 to 12 sessions. On the other hand, traditional treatments such as medication, counseling service, collaborative care, and other forms of therapies were used in the control groups.

Quality assessment of included studies
Although randomization was reported in all included RCTs, few studies [27,33,39,42] reported specific method of randomization. It was impossible to set blind for patients due to the theory of the intervention, however, some studies [27,39] reported that blind was applied in the study design. We regarded high-performance bias in these trials. Due to the critical requirements or missing data of the allocation concealment, most of the included studies were considered as high-risk or unclear risk of allocation concealment. Overall, according to the risk assessment criteria of the Cochrane Handbook, more than 60% of the studies were considered as low-risk of bias, approximately 30% as unclear risk, and less than 10% as highrisk of bias (Figs 2 and 3).

Effectiveness of interventions PND symptoms.
A total of 13 studies reported the data pertaining the improvement of depressive symptoms after interventions as assessed by either EPDS or BDI. Heterogeneity detection results did not suggest a significant heterogeneity (I 2 <40%) for long-term evaluation, and the fixed effects model was introduced for this part of the meta-analysis. After CBT, the scores of EPDS in the intervention group were significantly lower than those in the control group, and the differences in both short-and long-term effects were statistically significant (short-term: SMD = -2.86, 95% CI: -4.41--1.31; P<0.05, and long-term: SMD = -1.68, 95%  [36] were the main sources of heterogeneity (S1 Fig). The study of Fathi-Ashtiani et al. [30] was considered as the primary source of heterogeneity for pooled analysis of short-term BDI (S2 Fig). After removing this study, the I 2 decreased from 91 to 58%. Other indicators of depression and/or anxiety were also evaluated, although only three RCTs reported data regarding BAI and DASS. No significant heterogeneity was observed as indicated by the I 2 %, and thus, the fixed effects model was used for DASS analysis, but not for BAI (I 2 >50%). The results of meta-analysis showed that the scores of BAI in the intervention group were lower than those in the control group; however, the differences of short-term (SMD = -4.64, 95% CI: -12.62-3.33; P = 0.25) and long-term (SMD = -0.81, 95% CI: -3.22-1.59; P = 0.51) were not statistically significant. On the other hand, the combined results showed that the scores of DASS in the intervention group were lower than those in the control group, and the difference of DASS-stress (SMD = -5.28, 95% CI: -8.25--2.32; P<0.01) was statistically significant, except for DASS-anxiety (SMD = -1.14, 95% CI: -3.64-1.36; P = 0.37). The detailed results are illustrated in Fig 6.

Subgroup analysis based on the types of intervention
As several types of therapies were used among these included studies, we performed a subgroup analysis to determine the most optimal intervention that exerted the best effect on relieving symptoms of depression with respect to EPDS. The results showed that in-homebased CBT was significantly superior in improving depression in short-term (SMD = -0.97, 95% CI: -1.81--0.13; P<0.01) and long-term (SMD = -0.31, 95% CI: -0.57--0.05; P<0.01) as compared to the control group. Internet-based and telephone-based therapies were also efficient; however, the results were not convincing owing to the limited number of included studies (S3 and S4 Figs).

Improvement of postnatal depression after treatments
Ten RCTs reported women under the diagnostic criteria after interventions. As there was significant heterogeneity (I 2 >50%), the random effects model was introduced during this part of the meta-analysis. As shown in Fig 7, after CBT, the number of women cured of PND in the intervention group was significantly higher than those in the control group, the differences of short-term (OR = 6.57, 95% CI: 1.84-23.48; P<0.05) and long-term (OR = 2.00, 95% CI: 1.61-2.48; P<0.05) were both statistically significant.

Publication bias
The publication bias of the included studies in EPDS was evaluated using the inverted funnel plots and quantified using the Egger's test and Begg's test.

Discussion
Meta-analysis is an important method for evidence-based medicine [26]. It refers to the systematic, objective, and comprehensive quantitative analysis of results from multiple independent studies with identical objectives by appropriate statistical methods. In this study, 20 RCTs were included in the meta-analysis. The results showed that CBT was clinically more effective than conventional therapies, with a reduced number of PND patients and improved symptoms of the disease. Notably, multiple therapies including CBT and anti-depressive medication were Recently, several meta-analyses [2,9,19,22] assessed the reliable and effective therapy for PND, and the results were not consistent. In 2013, Scope et al. [44] performed a meta-analysis to evaluate whether CBT could serve as an evidence-based treatment for women with PND as compared to other used packages of care. They identified 7 eligible studies for the final metaanalysis, which showed that the CBT group could be effective in reducing depression when compared with the usual care, routine primary care, or waiting list groups. However, all the included studies were not RCTs, and the varied definitions of the CBT component of the studies suggested that the results should be interpreted cautiously. Morrell et al. [9] systematically evaluated the clinical effectiveness and cost-effectiveness of postnatal intervention in women to prevent PND. The meta-analysis designated CBT as the most beneficial intervention. The study concluded that cognitive behavioral therapy was cost-effective with respect to traditional care, although much of the evidence was inconclusive. The present study focused primarily on the cost-efficiency of CBT; the clinical efficiency of the methods has not been assessed systematically. Furthermore, Lau et al. [19] performed another meta-analysis to evaluate the efficacy of the therapist-supported internet-based CBT on postpartum women. After searching, a total of 8 RCTs involving 1523 participants across six nations were included. They found that this therapy could significantly improve stress, anxiety, and depressive symptoms in postpartum depressive women as compared to those of the control group. Though the results were positive, the therapist-supported internet-based cognitive behavior therapy was only one specific type of the cognitive behavior therapy. Although these studies determined that CBT was useful in improving the depressive symptoms among women with postnatal depression. However, these meta-analyses did not systematically assess the efficiencies of different types of CBT in treating postpartum women as compared to control groups.
Our meta-analysis found that the CBT, overall or in different forms, had significant benefits in treating women with PND. The overall results of our meta-analysis showed that the CBT exerted good short-term and long-term effects on PND. After the administration of intervention, the symptoms of PND patients could be significantly improved immediately after treatment within few weeks and maintained for >3 months. Non-guided counseling and usual therapy had only a good short-term effect on PND. At the end of the intervention, these methods could significantly improve the symptoms; however, the long-term efficacy was poor. Inhome-based, internet-based, and telephone-based exhibited a good short-term effect on PND, and the long-term efficacy remains to be further elucidated. The sub-group analyses showed that the internet-based therapy, telephone-based therapy, in-home therapy, mindfulness therapy, and other forms of CBT were efficacious in relieving symptoms and reducing the number of PND patients. These findings indicated that CBT could exert its therapeutic role in various forms of delivery, enabling the participants to select appropriate therapeutic pathway, thereby increasing the treatment compliance.
This meta-analysis aimed to systematically evaluate the efficiencies of CBT in treating new mothers with PND based on the latest clinical evidence. In the current study, the sub-group analysis was introduced to determine the efficacy of different interventions and forms of therapy delivery. The outcomes of symptoms improvement and short-and long-term effects were also analyzed in our meta-analysis, further strengthening the findings of this meta-analysis. However, there were several limitations within our meta-analysis. First, the number and samples of included RCTs were relatively small, especially in the sub-group analysis. Second, heterogeneity was observed among included studies with respects to study design, treatments, indicators of outcomes, and level of detail of the follow-up. The relapse rate of PND differed between studies. They reported different duration and improvements of depression symptoms. The period of follow-up was different among the included studies, which might over-or under-estimate the actual effectiveness of CBT. Third, the language of eligible studies was primarily limited to English, which might lead to the risk of selection bias. Nevertheless, the results of this meta-analysis showed that CBT was effective in treating PND, both in improving symptoms and reducing the number of patients suffering from the disorder.

Conclusion
In conclusion, the cognitive behavioral therapy is sufficient to relieve the psychological symptoms of postnatal depression and effectively improve the quality of life in new mothers, resulting in a reduced prevalence of postnatal depression. However, our findings are limited due to the uneven quality, the differences in the diagnostic criteria, evaluation methods, and followup of the included studies. Thus, rigorous high-quality randomized controlled clinical trials are essential to substantiate the present conclusion.