Neonatal Outcomes in Women With Untreated Antenatal Depression Compared With Women Without Depression: A Systematic Review and Meta-analysis

Importance  Despite the prevalence of antenatal depression and the fact that only one-third of pregnant women with depression consider it acceptable to take antidepressants, the effect of untreated depression on neonatal outcomes remains to be addressed thoroughly.

Objective  To undertake a systematic review and meta-analysis to understand the effect of untreated depression on neonatal outcomes.

Data Sources  We executed our search strategy, with emphasis on its exhaustiveness, in MEDLINE, EMBASE, PsycINFO, Cumulative Index to Nursing and Allied Health, Cochrane Central Register of Controlled Trials, and Web of Science. The search was conducted in July, 2015.

Study Selection  We included randomized and nonrandomized studies that examined neonatal outcomes in women with depression receiving neither pharmacological nor nonpharmacological treatment compared with women without depression.

Data Extraction and Synthesis  Two reviewers independently screened titles and abstracts, assessed full-text articles, extracted data, and assessed their quality using a modified version of the Newcastle-Ottawa Scale. We pooled data using random-effects meta-analyses, quantified heterogeneity using the I² statistic, and explored it with subgroup analyses by type of assessment of depression, severity, reported conflicts of interest, and study quality.

Main Outcomes and Measures  Primary outcomes were preterm birth before 37 weeks and before 32 weeks, small and large for gestational age, low birth weight, and neonatal intensive care unit admission.

Results  Of the 6646 titles initially identified, 23 studies met inclusion criteria, all observational, with a total of 25 663 women. Untreated depression was associated with significantly increased risks of preterm birth (odds ratio [OR], 1.56; 95% CI, 1.25-1.94; 14 studies; I², 39%) and low birth weight (OR, 1.96; 95% CI, 1.24-3.10; 8 studies; I², 48%), with a trend toward higher risks for exposure to more severe depression. While the odds of preterm birth more than doubled in studies reporting conflicts of interest (OR, 2.50; 95% CI, 1.70-3.67; 5 studies; I², 0%), studies not reporting such conflicts showed more moderate results (OR, 1.34; 95% CI, 1.08-1.66; 9 studies; I², 30%).

Conclusions and Relevance  Our results contrast with what is, to our knowledge, the only previous systematic review that examined the question of untreated depression because we found significant risks of 2 key perinatal outcomes, preterm birth and low birth weight. These are important results for pregnant women and clinicians to take into account in the decision-making process around depression treatment.

Because only one-third of pregnant women with depression would consider taking antidepressants as an acceptable treatment option,¹ it is critical to understand pregnancy outcomes with untreated depression for clinicians to be able to adequately counsel and support women who choose this course. Antenatal depression is a prevalent problem, affecting 5% to 15% of pregnant women.² It also has high public health significance because depression has been associated with an increased risk of preterm birth and low birth weight,^3,4 which are 2 leading causes of mortality and morbidity in infants.^5,6

The use of antidepressant medications during pregnancy has been increasing in the last few decades both in Europe and in the United States, with 3% to 8% of women being prescribed or having used antidepressants during pregnancy.^7-10 However, since 2010, several meta-analyses examining antidepressant use during pregnancy found significantly higher risks of preterm birth and low birth weight in women with depression taking antidepressants compared with either women without depression or women with untreated depression.^11-13 Moreover, in a meta-analysis by Huang et al,¹¹ the risks associated with antidepressant use were higher when the comparison group was women with untreated depression (odds ratio [OR], 2.85; 95% CI, 2.00-4.07) than when the comparison group was women who were not depressed (OR, 1.88; 95% CI. 1.50-2.27). These results highlight the complexity of this topic and the need to better understand the risks in untreated women.

In previous systematic reviews of risks of neonatal outcomes associated with antenatal depression,^3,4 most of the included studies did not control for the potential confounding effect of antidepressant use³; as such, the estimations of the risks of untreated depression might be biased, resulting in an overestimation of adverse perinatal outcomes such as preterm birth and low birth weight. Therefore, we hypothesized that the risk of adverse infant outcomes would be lower after rigorously excluding the potential confounding effect of antidepressants. The results of a subgroup analysis³ partially support this hypothesis because the odds of preterm birth were not only lower but also not statistically significant in the studies of women with short or no exposure to antidepressants. Additionally, it is surprising that the influence of financial conflicts of interest with direct or indirect funding by pharmacological companies has not been explored in previous reviews, given that a statistically significant association between industry sponsorship and proindustry conclusions is well known.^14-16

Our objective was to address the limitations of existing reviews by undertaking a meta-analysis of randomized and nonrandomized studies to determine neonatal risks associated with untreated antenatal depression, using strict inclusion criteria to remove any potential effect of antidepressant medication.

Box Section Ref ID

We published our protocol in the Prospective Register of Systematic Reviews database (registration number: CRD42015007455).

We executed our search strategy in 6 electronic databases (MEDLINE, EMBASE, and PsycINFO [all through OVID] and Cumulative Index to Nursing and Allied Health, Cochrane Central Register of Controlled Trials, and Web of Science) using both controlled vocabulary and free text terms, developed in consultation with an experienced research librarian, with no restriction by publication date. The full electronic search strategy for all databases can be accessed online (http://www.crd.york.ac.uk/PROSPEROFILES/16038_STRATEGY_20150016.pdf).

We performed several complementary steps to overcome challenges in identifying studies assessing untreated depression, ie, women with untreated depression are usually a comparison group in intervention studies and not always clearly described in the abstracts. To be confident in the exhaustiveness of our literature search: first, besides keywords for untreated depression, we also included terms for pharmacological and nonpharmacological therapies in our main search strategy. Second, in an effort to capture studies assessing the risks of antenatal depression that did not state the presence or absence of any treatment in the title or abstract, we complemented the main search strategy with a broader, less specific one not including any intervention term. Using this search strategy, we considered only studies published in 2010 or later, the date of the literature searches of the 2 most recent systematic reviews,^3,4 owing to the very large number of results obtained without time limits in such a nonspecific search. Third, we identified studies and reviews closely related to the topic under study during the titles and abstracts screening stage. We read their full text and examined all potentially relevant references. Finally, we screened the full text of primary studies in systematic reviews of antidepressant effects because 2 control groups (pregnant women with untreated depression and pregnant women without depression) were sometimes used but not explicitly stated in the abstract.

We included randomized and nonrandomized studies reporting the risk of adverse neonatal outcomes in pregnant women with untreated depression compared with pregnant women without depression. We excluded case reports, case series, reviews, conference abstracts, non–peer-reviewed literature, and non–English language studies (owing to potential bias of a single translator not familiar with the area).

We included studies that assessed depression using either a clinical interview/diagnosis or a screening tool or scale at any time during pregnancy. We excluded studies including and not stratifying outcomes by multiple pregnancies owing to worse outcomes in twins and higher-order multiples, women experiencing domestic violence, or with other reported comorbid psychiatric diseases (eg, anxiety or bipolar disorders).

The absence of any pharmacological treatment for depression had to be explicitly stated in the article or its effects adjusted for in the analyses. Studies that reported no changes in the results after redoing the analyses without antidepressant users were included (and their effect also assessed in a sensitivity analysis). However, we did not include studies that reported only that the conclusions did not change (in terms of accepting or rejecting the null hypothesis) after redoing the analyses because this could still be the case even with considerable differences in the effect estimates. We also excluded studies in which participants received nonpharmacological treatments.

Primary outcomes were preterm birth before 37 weeks and before 32 weeks, small for gestational age (birth weight <10th percentile for sex and gestation), large for gestational age (birth weight >90th percentile for sex and gestation), low birth weight (<2500 g), and neonatal intensive care unit admission. Secondary outcomes were birth weight less than the third and fifth percentiles and greater than the 95th and 97th percentiles for sex and gestation; high birth weight (macrosomia, >4000 g and >4500 g); and gestational age and birth weight (continuous data).

Two reviewers, 1 of them always being the first author, screened the titles and abstracts, assessed the full text of the potentially eligible studies, and extracted data of included studies using a piloted data extraction form. A third reviewer was available as an adjudicator if disagreements could not be settled by discussion. When additional information regarding the inclusion and exclusion criteria were required, we contacted the authors, all of whom replied to our inquiries. Data extracted included information regarding the authors’ affiliations and conflicts of interest (direct or indirect funding by or links to pharmacological companies), study characteristics (design, sample, and assessment of depression), and outcomes of interest (both crude and adjusted). The same reviewers assessed the methodological quality (risk of bias) using the Cochrane Risk of Bias Tool for randomized trials and a modified version of the Newcastle-Ottawa Scale¹⁷ for observational studies.

We pooled ORs (for binary outcomes) and mean differences (for continuous outcomes) using random-effects meta-analyses. Adjusted and nonadjusted effect estimates were pooled separately. Heterogeneity was quantified using the I² statistic.¹⁸ Publication bias was assessed using Duval and Tweedie’s trim and fill method.¹⁹

To explore heterogeneity, we had planned to do a number of subgroup analyses, but owing to the relatively small number of studies located, we decided to limit them to assessment of depression (clinical diagnosis or interview vs self-administered questionnaire), depression severity (moderate vs severe), declared conflicts of interest (reported direct or indirect funding by or links to pharmacological companies vs not reported or stated none), and study quality (high or acceptable, defined as 5 or more points in the Newcastle-Ottawa Scale vs low, defined as 4 or less points). We did a post hoc subgroup analysis of term gestation (sample limited vs not limited to term infants) because 2500 g, the approximate average at 35 weeks’ gestation, is appropriate for many preterm infants but is the equivalent of small for gestational age at 37 weeks and older, which is term. To explore the effect of newer studies in our pooled effects, we did further post hoc subgroup analyses by publication year (before and after 2010, which was the year the previous systematic reviews executed their search strategy). The effect of including studies that reported only “no changes after excluding antidepressant users from their analyses” was assessed in sensitivity analyses.

Of the 6646 studies initially identified, 2367 duplicates were deleted, leaving 4279 for titles and abstracts screening, resulting in 347 full-text articles for assessment. Of these, 23^20-42 met our inclusion criteria (Figure 1), involving 25 440 women. All but 1 study (which assessed depression retrospectively) were prospective observational studies.

Nine studies (40%) defined depression using a clinical diagnosis or interview. Although half of these studies (5) used a DSM-IV diagnosis of major depression disorder, other DSM-IV criteria, such as dysthymia or both major and minor depression, were also used, and 1 study used the definition of moderate depression of the International Classification of Diseases, Tenth Revision. Among the 14 studies using a self-administered questionnaire to measure depression, the most commonly used scale was the Center for Epidemiologic Studies Depression Scale (6 studies), usually with a cutoff value of 16, followed by the Edinburgh Postnatal Depression Scale (3 studies) and the Beck Depression Inventory (2 studies; Table 1). Through inclusion and exclusion criteria, studies excluded some potential confounding variables, ie, illicit drug use in some studies and chronic diseases in others (Table 1).

Seven studies (30%) reported conflicts of interest (direct or indirect funding by or links to pharmacological companies), 8 studies (35%) stated that there were no conflicts of interest, and conflicts of interest were not reported in the rest of the studies (8 studies, 35%). Sixteen studies (70%) had a quality score equal to or above our cutoff value of 5, indicating an acceptable or high methodological quality. The mean (SD) intraclass correlation coefficient between raters was 0.64 (0.30), which is considered good agreement.⁴³ Three studies reported our outcomes of interest in a metric that could not be pooled with the rest of the studies and were therefore not included in the meta-analyses^38,40,41 (Table 1). Two more studies could not be pooled in any meta-analysis because only gestational age and birth weight of term infants (continuous data) were reported.^39,42

Pregnant women with untreated depression had a significantly increased risk of both preterm birth (<37 weeks; OR, 1.56; 95% CI, 1.25-1.94; 14 studies; I², 39%) and low birth weight (<2500 g; OR, 1.96; 95% CI, 1.24-3.10; 8 studies; I², 48%) compared with women without depression. Only 1 study reported small for gestational age (birth weight <10%; OR, 1.37; 95% CI, 1.10-1.70), 2 studies reported neonatal intensive care unit admission (OR, 1.12; 95% CI, 0.40-3.15; I², 0%), and none reported either preterm birth before 32 weeks or large for gestational age (birth weight >90%, Table 2; eFigures 1-3 in the Supplement).

Only 1 study reported birth weight greater than 4500 g without finding significant differences between the groups. There were no statistically significant differences between women with depression and women without depression in gestational age (continuous data, 10 studies) and no clinically significant differences in birth weight (mean difference, –84 g; 95% CI, –153 g to –15 g; 11 studies; I², 77%; Table 2; eFigures 4 and 5 in the Supplement).

Although the funnel plot suggested potential publication bias for preterm birth before 37 weeks, the results did not change significantly after 3 studies were inputted using the trim and fill method (adjusted OR, 1.47; 95% CI, 1.17-1.85). None of the other outcomes were suspicious for publication bias.

For preterm birth before 37 weeks, results were not statistically different among subgroups based on assessment of depression by a clinical diagnosis/interview vs self-administered questionnaire or quality of the study (Figure 2). There appeared to be a trend toward an increased risk of preterm birth in women with more severe depression, although the differences were not statistically significant (Figure 2). There were significant differences between the results of studies reporting conflicts of interest (OR, 2.50; 95% CI, 1.70-3.67; 5 studies; I², 0%) and those that did not (OR, 1.34; 95% CI, 1.08-1.66; 9 studies; I², 30%). This difference remained after redoing the analyses without low-quality studies and does not seem to be explained by severity of depression.

For low birth weight (<2500 g), our subgroup analyses found a significant increased risk in term infants, which would be approximately the equivalent of being small for gestational age (<10th percentile). Our results also suggested that there are significant differences (P = .06) between the results of high- or acceptable-quality studies (OR, 2.39; 95% CI, 1.72-3.30; 5 studies; I², 0%) and low-quality studies (OR, 0.89; 95% CI, 0.33-2.35; 3 studies; I², 36%). There was also a trend toward significant differences between the studies reporting conflicts of interest (OR, 1.66; 95% CI, 0.98-2.79; 6 studies; I², 53%) and those that did not (OR, 3.76; 95% CI, 1.69-8.37; 2 studies; I², 0%) (P = .09), although it disappeared when removing low-quality studies (Figure 2).

Few studies reported adjusted effect estimates, although each study excluded a number of potential confounding variables through their population inclusion and exclusion criteria. Although pooling these adjusted effect estimates yielded different results than pooling the nonadjusted effect sizes of all the studies, pooling the nonadjusted effect sizes of the same subset of studies resulted in almost identical adjusted and crude results. For example, 3 studies^22,23,32 reported both adjusted and nonadjusted effect estimates for preterm birth before 37 weeks. Pooling these 3 adjusted effect estimates resulted in an adjusted OR of 1.03 (95% CI, 0.64-1.64), which is significantly different from pooling the nonadjusted effect estimates of all 14 studies reporting this outcome. However, if the nonadjusted effect estimates of only those 3 studies were pooled together, the results were almost identical (nonadjusted OR, 1.08; 95% CI, 0.74-1.57) to the pooled adjusted estimates. Therefore, the differences in the results of the meta-analyses of adjusted values (not shown) are mostly owing to the differences in the subset of studies pooled instead of the use of adjusted data.

One study³⁰ applied the DSM-IV criteria for depression retrospectively after delivery. Repeating the analyses excluding this study yielded almost identical results for almost all outcomes and subgroup analyses. The only outcome where the results were not almost identical was in the subgroup analysis for low birth weight by study quality, in which the pooled effect size of subgroup with low quality changed to an OR of 0.63 (95% CI, 0.29-1.34), a statistically significant difference (P = .001) from the high or acceptable quality group (OR, 2.39; 95% CI, 1.72-3.30).

Two studies^25,31 included a small proportion of antidepressant users (and multiple gestations in 1 of them²⁵), although their authors reported no changes after excluding these cases. Excluding these studies from our analyses yielded similar results (not shown).

In our systematic review, we found that pregnant women with depression who were not receiving any treatment for their depression had significantly increased infant risks compared with pregnant women without depression, specifically with regard to preterm birth and small infant size (whether defined as low birth weight overall or restricted to term infants or birth weight <10%). We also found a trend toward higher risks with more severe depression. The odds of preterm birth in studies with authors reporting conflicts of interest (ie, received pharmaceutical support) were significantly higher than in studies not reporting such conflicts. This difference was not explained by either differences in depression severity or study quality and remains to be fully understood.

We had hypothesized that we would find lower risks of preterm birth and small infant size than existing systematic reviews, but this generally was not the case (OR, 1.56; 95% CI, 1.25-1.94 and OR, 1.96; 95% CI, 1.24-3.11; respectively). Previous systematic reviews found that depression treated with antidepressants was associated with significantly increased risks of preterm birth (ORs ranging from 1.44; 95% CI, 1.34-1.56 to 1.69; 95% CI, 1.52-1.88)^11-13 and low birth weight (OR, 1.44; 95% CI, 1.34-1.56).¹¹ The 2 other existing systematic reviews on depression,^3,4 which included studies potentially confounded by antidepressant use (because in these primary studies, women taking antidepressants were not excluded), found ORs ranging from 1.13 to 1.37 for preterm birth and from 1.18 to 1.21 for low birth weight. However, despite rigorously excluding the potential confounding effect of antidepressant use, we did not find lower risks of either preterm birth or low birth weight. This stands in direct contrast to what is, to our knowledge, the only previous systematic review attempting to examine this in a subgroup analysis of studies of women with no or short exposures to antidepressants and that found no significant increase in either preterm birth or low birth weight.³ This might have inadvertently conveyed a message that not using antidepressant medications could remove these risks. We explored this contradiction, and it could not be attributed to the newer studies published after the search dates of the previous systematic reviews^3,4 (2010) because a lower pooled effect was seen in the more recent studies.

Our results highlight the risks of untreated depression during pregnancy, although they cannot be used as an argument in favor of antidepressant use because evidence shows that women treated with antidepressants have risks of similar magnitude. Nonpharmacological therapies might be more acceptable to women, but there is still a lack of evidence regarding their effect on preterm birth and low birth weight.^44-46 However, these therapies might not be an effective option for treating more severe depression, which in turn appears to have higher risks than more moderate cases in our subgroup analyses.

The main strength of our systematic review was its strict inclusion criteria to make sure that we obtained results that were not confounded by the use of antidepressant medications. Further strengths include an exhaustive literature search, which allowed us to include several studies not included in previous systematic reviews; the assessment of the risk of low birth weight separately in studies limiting and not limiting their sample to term infants only, which is more clinically meaningful because this is approximately the equivalent of being small for gestational age; and a consideration of depression severity. Finally, we are not aware of any other systematic review on the topic that explored the effect of conflicts of interest. We are unable to explain why they seem to affect preterm birth but not low birth weight (after excluding low-quality studies). However, these are preliminary findings that need further exploration.

Our study has several limitations. First, the necessarily strict exclusion criteria might have filtered out studies with more detailed reporting or an assessment of certain exclusion variables that might be present but not measured or reported in other included studies. Second, more than half of the included studies lacked a rigorous diagnostic assessment of depression, using only screening tools instead. Although we found no significant differences between studies that used a clinical diagnosis and those that did not, it is not possible to know whether the symptoms measured constitute a major depressive episode or the trajectory of the symptoms. Third, there is no consensus on the best method to assess study quality in observational studies. We used a modified version of a previously validated and frequently used scale, the Newcastle-Ottawa Scale.¹⁷ Fourth, journals’ requirements for reporting conflicts of interest have changed over time and vary widely. Therefore, the definitions used in this review (which only took into account the conflicts and affiliations reported in the publication) are likely to be of limited precision. Finally, there were several other important confounding variables whose effects could not be taken into account owing to a lack of reporting of adjusted data in most studies.

Taking a rigorous approach to understand the effect of untreated depression on pregnancy, we found increased risks of preterm birth and small infant size, in contrast to what is, to our knowledge, the only previous systematic review exploring untreated depression in a subgroup analysis, which found no increased risks. Moreover, we excluded the potential effect of comorbid anxiety and explored the potential effect of conflicts of interest. Our findings have important clinical implications for pregnant women and health care professionals because they suggest the need for more surveillance for preterm birth and small infant size, key perinatal outcomes in women with untreated depression. Our results also suggest methodological directions that future systematic reviews could take including examining the effect of reported conflicts of interest and potential confounding variables.

Corresponding Author: Alexander Jarde, PhD, Department of Obstetrics and Gynecology, McMaster University, 1280 Main St W, Hamilton, ON L8S 4K1, Canada (jarde@mcmaster.ca).

Submitted for Publication: January 15, 2016; final revision received March 21, 2016; accepted March 24, 2016.

Published Online: June 8, 2016. doi:10.1001/jamapsychiatry.2016.0934.

Author Contributions: Drs Jarde and McDonald had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Jarde, Kingston, Giallo, MacQueen, Giglia, McDonald.

Acquisition, analysis, or interpretation of data: Jarde, Morais, Kingston, Gaillo, Giglia, Beyene, Wang, McDonald.

Drafting of the manuscript: Jarde.

Critical revision of the manuscript for important intellectual content: Jarde, Morais, Kingston, Gaillo, MacQueen, Giglia, Beyene, Wang, McDonald.

Statistical analysis: Jarde, Beyene.

Administrative, technical, or material support: McDonald.

Study supervision: Beyene.

Conflict of Interest Disclosures: Dr MacQueen has received funding to speak or consult for Lundbeck, Lilly, Pfizer, and Janssen in the past 3 years. No other disclosures are reported.

Funding/Support: Dr Beyene holds the John D. Cameron Endowed Chair in the Genetic Determinants of Chronic Diseases, Department of Clinical Epidemiology and Biostatistics, McMaster University. Dr McDonald is supported by Canadian Institutes of Health Research Tier II Canada Research Chair Sponsor Award 950-229920.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We thank Neera Bhatnagar (BSc, MLIS, McMaster University) for her help developing the search strategy and Julie Yu (BHSc_c, McMaster University) and Sugee Korale Liyanage (BHSc_c, McMaster University) for their administrative support of the project. No compensation was received from a funding sponsor for their contributions.