Risk of cardiovascular malformations after exposure to paroxetine in pregnancy : meta-analysis

Sciences Citation Index (SSCI), King’s Fund, DH-Data, CINAHL, Allied and Complementary Medicine Database (AMED) and British Nursing Index (BNI). Combinations of the terms ‘SSRI’, ‘selective serotonin reuptake inhibitor(s)’, ‘SRI’, ‘serotonin reuptake inhibitors’, ‘paroxetine’, ‘pregnancy’, ‘congenital malformation(s)’, ‘congenital defect(s)’, ‘cardiovascular malformation(s)’, ‘cardiac defect(s)’, ‘cardiovascular defect(s)’, ‘fetal malformation(s)’ and ‘fetal anomalies’ were used for the search. The search was restricted to articles published in English but there was no exclusion on the basis of country, ethical approval, etc. No grey literature was searched for this review. Each abstract/title and article was scrutinised by two of the authors (N.P. and R.P.) and the differences between them were resolved by consensus. Relevant articles were hand-searched for cross-references. The GlaxoSmithKline website was searched for recent data on paroxetine. To exclude repetitive data-sets, only the study with the most updated data was taken up for analysis. A repeat data search was done in August 2012, after the first review of this article, and results were updated.

Aims and method To examine the association between the use of paroxetine during pregnancy and the risk of cardiovascular defects in the newborn.A systematic review of nine electronic databases was carried out and bibliographies were hand-searched for other relevant articles.Inclusion criteria for studies were the use of selective serotonin reuptake inhibitors in the first trimester of pregnancy, with separate data available for paroxetine and cardiovascular defects in newborn babies.A randomeffect model was used to combine the data.
Results A total of 11 studies were included in the analysis, concerning 4515 offspring who were exposed to paroxetine in the first trimester and 1 469 302 controls.In pooled analysis, paroxetine in the first trimester of pregnancy was slightly, but significantly, associated with a risk of cardiovascular malformations in the offspring (relative risk = 1.25, 95% CI 1.01-1.54).Separate analyses of case-control and cohort studies made this difference non-significant.
Clinical implications This meta-analysis supports current guidelines advising not to use paroxetine in early pregnancy.
Declaration of interest P.S. received a research grant as a principal investigator from Eli Lilly for a project that was completed about 6 months prior to his involvement in this study.

REVIEW ARTICLE
Risk of cardiovascular malformations after exposure to paroxetine in pregnancy: meta-analysis 2 control group of unexposed women available for comparison 3 as an outcome, separate data available for congenital cardiovascular defects in newborns, for instance conotruncal heart defects, septal heart defects, ventricular outflow tract obstruction.
Exclusion criteria were: 1 papers published on repeat data 2 studies with no control group for comparison 3 no cardiovascular defect in both study and control group.
Excluded studies are presented in online Table DS1.
The modified QUOROM Flow Chart 10 (Fig. 1) was used to show the study search process.

Outcome measure
The outcome measure for this review was cardiovascular malformation in the newborn.

Data collection and analysis
We collected data from the studies that met the selection criteria.The quality of studies was assessed by criteria adapted from Centre for Reviews and Dissemination guidelines. 11Descriptive data were mainly expressed in actual numbers of exposed mothers and controls.Where exact numbers were not available, frequencies were changed into actual numbers (described odds ratios (ORs) were used to resolve doubts).Results were presented in terms of risk ratio (RR) with 95% confidence intervals.A funnel plot was used to assess publication bias and heterogeneity among studies was analysed by the w 2 -test.A random-effect model was applied to combine the data.Subgroup analysis was carried out for cohort and case-control studies separately.Sensitivity analysis was carried out by the sequential removal of studies with maximum weight.Data analysis was performed with Review Manager (RevMan 5.0) for Windows.A checklist recommended by the Meta-analysis of Observational Studies in Epidemiology (MOOSE) group 12 was used.

Quality analysis
As shown in Table 1, the studies that met the selection criteria were from all grades except grade B and the lowest grade E on the Centre for Reviews and Dissemination hierarchy of observational studies. 11

Publication bias
The funnel plot (Fig. 2) shows the relative absence of smallsample sized studies which showed teratogenic effect of paroxetine.In trim-and-fill analysis, three studies on the left side of the plot were trimmed, but the adjusted risk ratio for the main analysis remained significant (RR = 1.23, 95% CI 1.05-1.42).

Pooled results
Paroxetine use in the first trimester of pregnancy was found to be significantly associated with cardiovascular malformations, compared with unexposed controls (RR = 1.25, 95% CI 1.01-1.54)(Fig. 3).

Sensitivity analysis
In sequential removal of studies with maximum effect sizes, the difference between paroxetine and the unexposed control remained significant after excluding the studies by Alwan et al 9

Discussion
The validity of meta-analysis of observational studies has always been debated, as observational studies are more prone to biases when compared with the gold-standard randomised controlled trials. 22However, a meta-analysis of observational studies seems justified for assessing the teratogenic effect of medications used during pregnancy because experimental studies cannot be conducted and large samples are required to observe rare events such as specific congenital malformations.In recognition of the limitations of meta-analysis of observational studies, we applied a random-effect model (rather than a fixed-effect model) to combine the results, as it can be applied irrespective of the level of heterogeneity of studies.
Combining case-control and cohort studies is a wellrecognised practice in meta-analysis of epidemiological studies, 12,23 although we also carried out a subgroup analysis for case-control and cohort studies separately.Further, we performed a sensitivity analysis to assess the robustness of results.For quality analysis of the studies, the key components of design were considered, as this method has been found to be more appropriate for meta-analysis of observational studies. 12In general, the study met the requirements of the MOOSE guidelines. 12lthough more than half of the identified studies were excluded from the analysis, most of them presented repeat data; thus, the combined results can be taken as a fair representation of the identified studies.There may be some doubts as to the reliability of actual numbers, as in some studies numbers were extrapolated from the frequencies and odds ratios; however, this should not affect the results considerably bearing in mind the large size of the collective sample.The apparent discrepancy between sample size and weight for each study (Fig. 1) corroborates the fact that in meta-analysis, weight given to a particular study depends not only on the sample size, but also on the variance of the data.
Underrepresentation of positive studies with small sample size in publication bias analysis could be a reflection of Type II error, a likely outcome in view of the rarity of the

Paroxetine
Unexposed controls Study or subgroup Alwan et al 9 Bakker et al 13 Berard et al 8 Davis et al 14 Diav-Citrin et al 15 Einarson et al 16   Louik et al 18 Malm et al 19 Reis & Kallen 17 Vial et al 21 Wogelius et al 6 Total (95% CI) Total events Heterogeneity: t = 0.03; w 2 = 14.34, d.f.= 10 (P = 0.16); I 2 = 30% Test for overall eect: Z = 2.03 (P = 0.04)  occurrence of cardiovascular defects.The trim-and-fill analysis only confirmed the limitation of this method, as it does not take into account the reasons for funnel plot asymmetry other than publication bias.Our meta-analysis, based on largest collective data sample so far, suggests that offspring of women who are exposed to paroxetine in the first trimester of pregnancy are at a small but significant increased risk of cardiovascular malformations.However, subgroup analysis and sensitivity analysis shows the fragility of this association.It is also possible that the borderline significant results of our meta-analysis could disappear, if the crude numbers used for the combined analysis were adjusted for various confounders such as maternal age, race, smoking, medical comorbidities, concomitant use of possible teratogens, etc.
Results of our meta-analysis fall in line with two other meta-analyses. 24,25O'Brien et al 24 separately analysed three case-control (n = 30 247) and six cohort (n = 66 409) studies and they did not find any significant association of cardiac malformation with paroxetine exposure.On the other hand, meta-analysis by Wurst et al 25 combined ten cohort and four case-control studies (n = 109 958) and found an increased prevalence of cardiac defects with first-trimester paroxetine use (OR = 1.46, 95% CI 1.17-1.82).Whether it is the large sample size which overcomes Type II error and exposes the teratogenic potential of paroxetine or too much heterogeneity (for the sake of large sample size) that brings spurious association remains debatable.In future, an analysis with large but more homogeneous data might provide the answer.In the meantime, our meta-analysis suggests that there is a possibility that exposure to paroxetine could be significantly associated with cardiovascular malformations and in that sense it supports the existing guidelines, 4,26 which advise avoiding paroxetine use in early pregnancy.

About the authors
Nitesh Painuly is consultant psychiatrist, Derbyshire Healthcare NHS Foundation Trust, Derby, UK; Ritu Painuly is specialty registrar (obstetrics and gynaecology), Royal Derby Hospital, Derby, UK; Reinhard Heun is consultant psychiatrist, Derbyshire Healthcare NHS Foundation Trust, Derby, UK; Pratap Sharan is professor of psychiatry, Department of Psychiatry, All India Institute of Medical Sciences, New Delhi, India.

Fig 1
Fig 1 Modified QUORON flow chart 10 describing the search process.
BMI, body mass index; NSAIDs, non-steroidal anti-inflammatory drugs; SSRIs, selective serotonin reuptake inhibitors.a. Adapted from Centre for Reviews and Dissemination guidelines: 11 A (highest quality): cohort (prospective study) with concurrent controls, B: cohort (prospective study) with historical controls, C: cohort (retrospective study) with concurrent controls, D: case-control (retrospective) study, E: observational study without control groups or large differences from comparisons between times and/or places.

Fig 2
Fig 2 Funnel plot of studies included in the meta-analysis.RR, risk ratio; SE, standard error.

Fig 3
Fig 3 Risk of cardiovascular malformations with first-trimester use of paroxetine in comparison with unexposed controls (forest plot).M-H, Mantel-Haenszel method.

Table 1
Characteristics of included studies continued Table 1 Characteristics of included studies (continued)