Maternal use of selective serotonin re-uptake inhibitors in early pregnancy and infant congenital malformations

Authors


Abstract

BACKGROUND:

Maternal use of selective serotonin re-uptake inhibitors (SSRIs) has recently been associated with an increased risk for certain malformations.

METHODS:

Using the Swedish Medical Birth Register, we identified women who had reported the use of SSRIs in early pregnancy and studied their infants, born between July 1, 1995 and the end of 2004. Congenital malformations were identified from that register, from the Register of Congenital Malformations, and from the Hospital Discharge Register. The effect of drug exposure was studied after adjustment for a number of identified maternal characteristics that could act as confounders.

RESULTS:

We identified 6,481 women who reported the use of SSRIs in early pregnancy and their 6,555 infants. There was no general increase in malformation risk. An increased risk for cystic kidneys was seen, but this was based on only nine malformed infants, and the pathology varied between these cases. An in-depth study of cardiovascular defects identified an association between such defects and notably ventricular and atrial septum defects and maternal use of paroxetine but not other SSRIs. No support for a postulated association between SSRI use and infant craniostenosis or omphalocele was found.

CONCLUSIONS:

Use of SSRIs in early pregnancy does not seem to be a major risk factor for infant malformations. The association between paroxetine use and infant cardiovascular defects may be a result of multiple testing, but is supported by other studies. Birth Defects Research (Part A) 2007. © 2007 Wiley-Liss, Inc.

INTRODUCTION

Many articles have been published on the possible effect of maternal use of selective serotonin re-uptake inhibitors (SSRIs) on pregnancy outcome (Hallberg and Sjöblom, 2005). Most stress is currently laid on the possibility that use of SSRIs during late pregnancy may cause unwanted effects on the neonate, whereas no teratogenic effect of SSRIs is usually thought to exist. An extensive review and discussion of studies on fluoxetine is available (Hines et al., 2004), which underlines the shortcomings of most studies thus far published.

Most studies on the effect of SSRIs in early pregnancy involved small cohorts and had insufficient power to detect anything but a very large teratogenic effect. There are two relatively large studies of SSRI exposure in early pregnancy: one (Goldstein et al., 1997) was based on reporting to the manufacturer of fluoxetine and contained information on 796 prospectively collected cases. No adequate control group existed, and the reporting of specific malformations was crude. The second study (Ericson et al., 1999) was an early report from the Swedish Medical Birth Register, which is also the basis of the present article.

Before 2005, no teratogenicity had been demonstrated for SSRIs. That year, however, five reports appeared that indicated that teratogenicity may occur, three of which were abstracts from meetings, one was a web publication from a drug company, and one was a letter to the editor. One report (Wogelius et al., 2005) described data from a Danish prescription register study and found an increased risk for congenital heart defects after maternal use of SSRIs, but the specific type of drugs used was not specified in the available abstract. A report from two teratogen information services (in Israel and Italy) suggested a specific association between maternal use of paroxetine and congenital heart defects (Diav-Citrin et al., 2005). This conclusion was based on five exposed cases and the OR was given as 3.46 (95% CI = 1.06–11.24). The third and largest report came from GlaxoSmithKline and is not yet published in a scientific journal, but is available on the following website (http://ctr.gast.co.uk/Summary/paroxetine/studylist.asp) and is summarized by Williams and Wooltorton (2005). By record linkage using the Ingenix Research Database, 704 women were identified who had used paroxetine in early pregnancy and 1,178 women were identified who had used other antidepressants (mainly fluoxetine). The adjusted OR for cardiovascular malformations was 2.26 (95% CI = 1.17–4.43). The majority of the cardiovascular defects were ventricular septum defects. In a Letter to the Editor (Källén and Otterblad Olausson, 2006) we published a preliminary report from the Swedish Medical Birth Register and found an OR of 2.22 (95% CI = 1.39–3.55) for any cardiovascular defect after maternal use of paroxetine (815 women, 17 cardiac defects).

In an abstract, Alwan et al. (2005) suggested an association between the use of SSRIs in early pregnancy and an increased risk for omphalocele and craniostenosis. The findings were based on retrospective exposure data.

These publications initiated a new detailed analysis of delivery outcome after maternal use of SSRIs in early pregnancy, using the Swedish Medical Birth Register. The data previously presented were extended with data from 2004. The main advantages of this data source are that the information on drug used was obtained in early pregnancy and therefore cannot be biased by the outcome, that many putative confounders can be identified and adjusted for, and that relatively large numbers can be identified. The use of this register and also other Swedish health registers also makes it possible to rather completely identify the presence of congenital malformations.

MATERIALS AND METHODS

The study is based on three Swedish health registers: the Medical Birth Register, the Register of Congenital Malformations, and the Hospital Discharge Register. All registers are kept by the National Board of Health and Welfare in Stockholm and patient identification is made using the unique personal identification number given to every Swedish citizen at birth or immigration and which is widely used in society and in all health care.

The Medical Birth Register contains information on the pregnancy, delivery, and the infant(s) born for nearly all deliveries in Sweden (a few percent are missing each year). A detailed description and evaluation is available at: http://www.sos.se/FULLTEXT/112/2003-112-3/2003-112-3.pdf.

The following data were used in the study:

  • Year of birth. The study was restricted to July 1, 1995 up to and including 2004. Adjustments were made by 1-year intervals.

  • Maternal age in 5-year groups (<20, 20–24, 25–29, etc.)

  • Parity (number of previously born children + 1) divided into four classes: 1, 2, 3, and ≥4.

  • Number of previous miscarriages. Information was obtained by midwife interview in early pregnancy at first antenatal care visit and was divided into none, 1, 2, and ≥3 previous miscarriages.

  • Maternal smoking in early pregnancy. Information was obtained by midwife interview at first antenatal care visit. Information was missing in ∼6% of cases. Smoking was categorized as none, <10 cigarettes per day, and ≥10 cigarettes per day.

  • Maternal prepregnancy weight and height from which BMI could be calculated. Information was obtained at the first antenatal care visit and was missing in ∼15% of women. BMI was divided into three groups: <19.8, 19.8–25.9 (normal), and ≥26.

  • Couple subfertility was estimated as years of unwanted childlessness, also reported at the first antenatal care visit. This was divided into six groups: none, 1, 2, 3, 4, or ≥5 years of unwanted childlessness.

  • Maternal use of drugs in early pregnancy up to the first antenatal visit (in 90% before the end of week 12). Data were available for infants born on July 1, 1995 and later. Drug information is stored as ATC codes (Anatomical Therapeutic Chemical classification system). Drug information was obtained from routine midwife interviews at the first antenatal care center visit (practically all pregnant Swedish women attend the free antenatal care visit system) using a standardized form, which is identical throughout the country. Data include names of drugs used, dosage, and time of use, but the latter two pieces of information are often incomplete. As the majority of the women initially attend the antenatal care service before the end of the first trimester, exposure data will basically refer to first-trimester exposures. Further details are given by Källén and Otterblad Olausson (2001).

Maternal country of birth was obtained by record linkage with the birth register of Statistics Sweden. Maternal education as linked from the Swedish Education Register is only available for births up to and including 2001 and refers to educational status in 2002.

Identification of congenital malformations among infants born was made by linking three health registers: the Medical Birth Register, where ICD diagnoses among the neonates are recorded, including codes for congenital malformations; the Register of Congenital Malformations, which is a surveillance register and contains more details than are available in the Medical Birth Register; and the Hospital Discharge Register, which contains discharge diagnoses of all hospitalized patients. Linkage between these three registers identifies a total of 4.7% of congenital malformations among infants born. All three registers were used up to the end of 2004. Data for some infants born in 2004 and hospitalized in 2005 will thus be missing. Details on the system of registration of congenital malformations can be found at: http://www.socialstyrelsen.se/Publicerat/2004/5120/2004-112-1.htm.

Statistical Analysis

Most analyses were made using the Mantel-Henszel procedure with adjustment for selected confounders, and risks were expressed as ORs. The 95% CIs were estimated with the Miettinens method. When the numbers were lower than expected, the observed number of exposed cases was compared with the expected number, estimated after adjustment for selected confounding factors. The observed over expected risk ratio (RR) and its 95% CI was calculated using exact Poisson distributions (SABER software).

RESULTS

Use of SSRIs in Early Pregnancy

A total of 6,481 women reported the use of any SSRI in early pregnancy, and these women had 6,555 infants (75 twin pairs). Table 1 shows the use of different SSRIs, alone or in combination with other SSRIs or with other antidepressants. Among all women using SSRIs, 6,256 (96.5%) reported the use of only one SSRI, 47 reported the use of two SSRIs (0.7%), and 179 reported the additional use of a non-SSRI antidepressant (2.8%). One of the latter women reported the use of two SSRIs and one non-SSRI. Table 1 shows that four SSRIs dominated: fluoxetine, citalopram, paroxetine, and sertraline. Among non-SSRIs, most were tricyclic, with clomipramine as the most common drug.

Table 1. Number of Women Using Specific SSRIs and Combinations of Different Antidepressants
Drug 1NoneDrug 2Total
CitalopramParoxetineSertralineFluoxamineEscitalopramNon-SSRI
  • *

    One of these women also used a non-SSRI.

Fluoxetine86094100036919
Citalopram2,57941002*632,658
Paroxetine90871027943
Sertraline1,80700471,854
Fluvoxamine360137
Escitalopram66470
Non-SSRI antidepressant drugs reported with SSRI use:
Clomipramine57     
Trimipramine2     
Amitryptiline21     
Nortryptiline4     
Moclobemide3     
Mianserin32     
Nefazodone3     
Mirtazapine27     
Venlafaxin30     
Reboxetin3     

Characteristics of Women Who Reported the Use of SSRIs in Early Pregnancy

The outcome of pregnancies of women who had used SSRIs in early pregnancy may differ from the expected outcome because of confounding from the women's characteristics. Table 2 shows an analysis of some such putative confounders. In this table, the OR estimates for maternal age, parity, smoking, and number of previous miscarriages were all adjusted for the other three variables and for year of birth. In the analysis of BMI, years of subfertility, and maternal country of birth, adjustment was made for year of birth, maternal age, parity, smoking, and ≥3 previous miscarriages versus <3 previous miscarriages.

Table 2. Some Maternal Characteristics of Women Reporting the Use of SSRI in Early Pregnancy
CharacteristicSSRI usePopulationOR95% CI
  1. For maternal age and parity, there is no natural reference group, which is why each group was compared with all other groups: the most common maternal age group varies with parity, for instance. Each variable is adjusted for all other variables listed in the table.

  2. Cigs: cigarettes.

Total number6,481860,215  
Maternal age
 <2012516,2820.770.64–0.92
 20–24849123,7970.860.79–0.92
 25–291,820290,0670.850.81–0.90
 30–342,119286,0671.010.96–1.07
 35–391,280121,6691.341.26–1.43
 40–4427021,4931.411.25–1.60
 45+188402.381.50–3.76
Parity
 13,043373,8591.321.25–1.39
 21,714309,8680.670.64–0.71
 31,063121,2131.071.00–1.15
 466155,2751.211.11–1.32
Smoking
 Unknown18555,958 
 None4,544704,9181.00Reference
 <10 cigs/day1,02767,4842.572.40–2.75
 10+ cigs/day72531,8554.033.74–4.35
 Any1,75299,3393.032.88–3.21
Previous miscarriage
 None5,052693,1781.00Reference
 11,052128,4841.000.93–1.07
 225727,8651.040.91–1.18
 ≥312010,6881.180.99–1.42
 Any1,429167,0371.010.95–1.07
Body mass index
 Unknown800132,003 
 <19.845964,6241.020.93–1.12
 19.8–25.93,314471,6061.00Reference
 26+1,908191,9821.241.16–1.32
Years of subfertility
 None6,062803,0371.00Reference
 110015,1060.790.63–0.99
 213316,6810.930.77–1.14
 3589,1600.760.57–1.01
 4405,3630.890.63–1.26
 5+8810,8680.830.61–1.06
 Any length of time41957,1780.840.75–0.94
Maternal country of birth
 Unknown668,335 
 Sweden5,721702,7111.00Reference
 Nordic countries14820,6540.800.69–0.94
 Other countries546128,5150.510.47–0.55

There is a clear-cut maternal age dependency with an increasing use of SSRIs with maternal age. The second parity has a lower rate of SSRI use than other parities, with the highest OR at the first parity. Smoking is strongly associated with the use of SSRIs. There is a very weak effect of previous miscarriages with a borderline significance for ≥3 miscarriages.

A high BMI is associated with an increased use of SSRIs, whereas there is only a weak negative association between years of subfertility and use of SSRIs. Women born outside Sweden (especially those born outside the Nordic countries) have a distinctly lower use or reporting of SSRI use than women born in Sweden.

Table 3 shows that maternal educational level affects the use of SSRIs; basically, there is decreased use with increased education. Unfortunately, information on maternal education is available for only ∼45% of the women (giving birth before 2002).

Table 3. Educational Level among Women Using SSRIs in Early Pregnancy
EducationPopulationSSRI useOR95% CI
  • Each variable is adjusted for year of birth, maternal age, parity, and smoking in early pregnancy. Deliveries up to and including 2001 (education level in 2002).

  • *

    Swedish “gymnasium” represents grades 10–13 and is equivalent to the upper secondary school in the United Kingdom and to senior high school in the United States.

Unknown18,723660.720.55–0.93
<9 years compulsory school10,549400.680.49–0.94
9–10 years compulsory school56,2275811.641.49–1.82
Swedish “gymnasium”*294,3881,4951.00Reference
<3 years post- “gymnasium”31,4081260.870.73–1.05
≥3 years post- “gymnasium”159,1955860.770.69–0.85
Postgraduate education3,07570.430.21–0.89
Total573,5652,901  

Some of these characteristics may affect the rate of congenital malformations and will be considered in the analysis. Maternal education has little effect on the risk for congenital malformations in Sweden and will not be considered further.

Table 4 shows associations between the use of SSRIs and the use of other drugs. Some drugs are used less often than expected: vitamins, including folic acid, progesterone/gestagens (nearly exclusively used in association with in vitro fertilization), minor analgesics, and cough medicines. One explanation for such associations may be that women who report the use of SSRIs have a tendency to neglect reporting other, less-significant drugs.

Table 4. Use of Other Drugs Together with SSRIs during Early Pregnancy
Drug groupNumber of womenOR95% CI
SSRIPopulation
  1. NSAID, nonsteroid antiinflammatory drugs; NVP, nausea and vomiting in pregnancy.

  2. ORs with 95% CIs comparing women who used SSRIs with other women with adjustment for year of birth, maternal age, parity, and smoking.

Drugs for stomach ulcer1916,5273.202.77–3.68
Insulin262,5851.160.78–1.71
Multivitamins15819,1180.770.65–0.90
Folic acid17821,2680.780.67–0.91
Oral contraceptives502,9341.901.44–2.51
Progesterone/gestagens93,4440.330.18–0.61
Ovulation stimulation92,1460.660.34–1.27
Thyroid hormones1648,9182.061.76–2.40
NSAID19414,7341.251.08–1.44
Opioids1203,7272.952.47–3.52
Minor analgesics49364,5850.830.76–0.92
Drugs for migraine512,3812.501.90–3.29
Anticonvulsants572,1842.922.21–3.76
Neuroleptics1492,5106.905.93–8.02
Sedatives4491,88430.225.7–30.2
Hypnotics3741,98318.416.8–20.1
Antiasthmatics32025,4511.501.34–1.68
Cough medicine204,0320.590.38–0.91
Antihistamines for NVP42935,3001.641.48–1.81
Other antihistamines15915,4741.261.08–1.48

Some drug categories are used more often with SSRIs than without them. The increased use of drugs moderate for stomach ulcer, oral contraceptives, thyroid hormones, NSAIDs (nonsteroid anti-inflammatory drugs), opioids, drugs for migraines, anticonvulsants, antiasthmatics, and antihistamines was moderate with ORs of 1.25–3.20. For neuroleptics, sedatives, and hypnotics, very high ORs were seen: 6.9–30.2 (Table 4).

In an analysis of the risk for congenital defects, only some of these drugs are of interest, namely those that are known or suspected teratogens. Among the weakly associated drugs, this is true for anticonvulsants and possibly true for NSAIDs. Not enough is known about possible teratogenic effects of neuroleptics, sedatives, or hypnotics. Folic acid has been thought to reduce the risk for congenital malformation.

Risk for Congenital Malformations

In the total population, 4.7% of the infants born are identified with a malformation; this figure also includes minor conditions of little clinical significance. The corresponding percentage after exposure to SSRIs is 4.1% and the adjusted OR (adjusted for year of birth, maternal age, parity, smoking, and ≥3 previous miscarriages versus <3 previous miscarriages) is 0.89, not significantly less than 1.0 (Table 5). There are no significant differences between the different SSRIs used, but the estimates for fluvoxamine and escitalopram have very large CIs.

Table 5. Presence of Congenital Malformations in Infants Whose Mothers Reported the Use of SSRIs in Early Pregnancy
DrugNo. of infants malformedTotalOR/RR*95% CI
  • Any congenital malformation is cited, irrespective of severity.

  • Adjustments were made for year of birth, maternal age, parity, smoking, and ≥3 previous miscarriages.

  • *

    Risk ratios (RRs) were used when the expected number was <10; 95% CIs were based on exact Poisson distributions.

Population41,233873,8761.00Reference
Any SSRI2696,5550.890.79–1.07
Fluoxetine369260.850.61–1.19
Citalopram1192,7010.940.78–1.13
Paroxetine469591.030.76–1.38
Sertraline671,9060.780.61–1.00
Fluvoxamine2381.050.13–3.80*
Escitalopram3720.910.19–2.66*

Specific types of congenital malformations are shown in Table 6. There is only one type of malformation with an RR that reaches statistical significance: cystic kidney, based on nine cases. Among them, two were infantile type polycystic kidneys (one exposed to citalopram and the other to sertraline), one was medullary cystic kidney (sponge kidney, exposed to paroxetine), one was cystic dysplasia (exposed to citalopram), and one was an unspecified cystic kidney (exposed to sertraline). The other four seemed to be secondary to urinary tract obstructions (one exposed to fluoxetine, one to citalopram, and two to sertraline). Within this group, apparently conditions of different natures were grouped.

Table 6. Specific Relatively Severe Congenital Malformations among Infants Whose Mothers Had Used SSRIs in Early Pregnancy
Congenital malformationNo. of infantsOR/RR*95% CI
SSRI usePopulation
  • Malformations in the absence of a chromosome anomaly.

  • Adjustments were made for year of birth, maternal age, parity, smoking, and ≥3 previous miscarriages.

  • *

    Risk ratios (RRs) were used when the expected number was <10; 95% CIs were based on exact Poisson distributions.

Any central nervous system malformation165,3310.850.51–1.39
Neural tube defects27070.490.06–1.78*
Eye malformations61,0880.930.82–2.07*
Severe ear malformations82,1390.540.27–1.07*
Any cardiac defect7811,3670.970.77–1.21
Ventral septal defect and/or atrial septal defect547,1591.100.84–1.44
Unspecified cardiac defect71,2111.040.42–2.15*
Orofacial cleft132,3730.800.46–1.39
Esophageal atresia34171.250.31–4.98*
Anal atresia24950.560.07–2.04*
Small gut atresia43321.630.44–4.18*
Pylorostenosis86931.550.76–3.14
Abdominal wall defect33161.160.24–3.40*
Diaphragmatic hernia22461.270.15–4.57*
Hypospadias263,1621.180.80–1.75
Kidney agenesis or hypoplasia32811.540.32–4.50*
Cystic kidney94153.501.60–6.65*
Poly- or syndactyly142,0421.090.61–1.98
Limb-reduction defects26710.440.05–1.59*
Craniostenosis66121.530.56–3.33*

Among the three infants with abdominal wall defects, two had gastroschisis and one had omphalocele.

As mentioned in the Introduction, special interest has been paid to the possible association between maternal use of SSRIs (especially paroxetine) and infant cardiovascular defects. The data in Table 7 demonstrate a difference between the four main types of SSRIs and the risk for a cardiovascular defect. Only for paroxetine is a significantly increased risk seen for any cardiac defect, and 13 of the 20 cases had ventricular or atrial septum defect, or both. In this analysis, adjustment was made only for year of birth, maternal age, parity, smoking, and ≥3 previous miscarriages versus <3 previous miscarriages. Women using SSRIs were shown more often than other women to have a high BMI (>26), less often were born outside Sweden, and less often were subfertile (see Table 2). The study was repeated for paroxetine and for all SSRIs except paroxetine. The analysis was then repeated excluding women who had used neuroleptics, sedatives, or hypnotics together with SSRIs and also excluding women who had reported the use of folic acid, NSAIDs, or anticonvulsants. For practical reasons, women in the total population who were exposed to these drugs could not be excluded. This should have slightly reduced the estimated ORs, but none of the drugs have a strong teratogenic effect (Källén and Otterblad Olausson, 2003). The results are shown in Table 8. Even though based on rather small numbers, the ORs for paroxetine, if anything, have increased.

Table 7. Cardiac Malformations among Infants Whose Mothers Used Different SSRI Drugs in Early Pregnancy
SSRI drugNumber of infantsOR/RR*95% CI
ObservedExpected
  • Malformations in the absence of chromosome anomaly.

  • Adjustments were made for year of birth, maternal age, parity, smoking, and ≥3 previous miscarriages.

  • *

    Risk ratios (RRs) were used when the expected number was <10; 95% CIs were based on exact Poisson distributions.

Any cardiac malformation
 Fluoxetine1211.11.090.62–1.92
 Cipramil2733.90.800.54–1.17
 Paroxetine2012.51.631.05–2.53
 Sertraline1722.30.760.47–1.23
VSD and/or ASD
 Fluoxetine66.80.880.32–1.92*
 Cipramil1920.40.930.59–1.47
 Paroxetine137.21.810.96–3.09*
 Sertraline1514.21.060.63–1.77
Unspecified cardiac defects
 Fluoxetine20.82.440.30–8.81*
 Cipramil23.00.660.08–2.38*
 Paroxetine21.11.830.22–6.63*
 Sertraline11.60.550.01–3.06*
Table 8. OR for Any Cardiovascular Defect and for Ventricular and/or Atrial Septum Defect (VSD + ASD) in Infants Whose Mothers Had Used Paroxetine or Other SSRI in Early Pregnancy
 Any cardiac defectVSD + ASD
  • Women with a BMI >26, born outside Sweden, or with reported subfertility (estimated as years of unwanted childlessness) were excluded in the first analysis, then also women using neuroleptics, sedatives, hypnotics, folic acid, NSAID, or anticonvulsants.

  • *

    Odds ratio (OR) or observed over expected ratio (RR) with 95% based on exact Poisson distributions when expected numbers were <10.

Excluding high BMI, non-Swedish, subfertile women
Paroxetine (n = 405)
 Observed number138
 Expected number4.92.6
 RR (95% CI)*2.63 (1.40–4.50)3.07 (1.32–6.04)
Other SSRI (n = 2,495)
 Observed number199
 Expected number29.716.0
 OR (95% CI)*0.64 (0.41–1.01)0.56 (0.29–1.08)
Excluding also women using certain other drugs
Paroxetine (n = 340)
 Observed number127
 Expected number4.12.2
 RR (95% CI)2.93 (1.52–5.13)3.23 (1.30–6.65)
Other SSRI (n = 2,082)
 Observed number155
 Expected number24.813.4
 OR (95% CI)0.61 (0.37–1.01)0.38 (0.16–0.87)

Except for the infants with ventricular septal defects and/or atrial septal defects, there were two cases of endocardial cushion defects (one with stenosis of pulmonary artery) and one case of tetralogy of Fallot. One infant had a coarctation of the aorta, one had a pulmonary artery stenosis, one had an unspecified defect in cardiac septa, and one had pulmonary valve stenosis.

DISCUSSION

The present study has some advantages and some disadvantages. The most important advantage is, except for its size, that the information on maternal drug use was obtained before anything was known about pregnancy outcome. Drug information was based on interviews conducted in early pregnancy, usually towards the end of the first trimester. Another advantage is the access to information on putative confounding factors.

There are also disadvantages. One is that reporting of drug use was certainly incomplete. Either the woman may not have wanted to report drug use (this could be especially relevant for drugs used for psychiatric conditions) or the midwife did not bother to or did not want to write down the information. If the woman attended antenatal care very early without SSRI use and then started using SSRIs before the end of the first trimester, the exposed woman will also be unknown. This means that some infants in the control group, regarded as unexposed, will actually have been exposed, but they will represent only a very small fraction of the control group and will only marginally affect risk estimates by biasing them towards unity.

In Sweden, practically all pregnant women have a second trimester ultrasound examination, but a late such examination (usually around week 32, i.e., after the upper limit for an induced abortion) is made only in some areas. It is possible that women who have used SSRIs are more likely to get such an examination than other women and that could explain the increased risk for severe kidney malformations and notably cystic kidneys. Another likely explanation to the finding is the multiple testing situation; many types of malformations have been studied and one type might quite likely be “statistically significant” just by chance.

The next problem is that little is known about actual dosage and timing of drug use. Most women use antidepressant drugs during an extended period and not as a temporary drug administration of single tablets. The exact timing of drug use is not known, and some women who reported the use of antidepressants may have used them outside the formative period of congenital malformations. Also, this will bias the risk estimates towards 1.0.

Nothing is known about the medical condition that was the reason for drug intake. This leaves the underlying disease as a confounder, which is very difficult to eliminate. There are in the literature a number of suggestions that stress and depression may negatively influence pregnancy outcome, notably preterm birth and perhaps intrauterine growth retardation, even in the absence of drug use (Hedegaard et al., 1996; Oberland et al., 2006). Little is known about such effects on the risk for congenital malformations even though stress has been implicated in the origin of orofacial clefts.

One way to take underlying disease into consideration is to compare the effects of different drugs used for similar conditions. This may not be 100% effective if there are differences in disease panorama that affect the choice between, for instance, paroxetine and other SSRIs.

The study tried to identify putative confounders that could affect the results. In order to be confounding, the variable must both covary with the exposure (drug use) and independently with the outcome (e.g., presence of congenital malformations). Women using SSRIs had a number of characteristics that differed from those of other women, and some of these may appear as confounders.

Women using SSRIs were older and were less often of parity 2 than other women. Both factors affect preterm birth and infant weight and (slightly) the risk of congenital malformations. They were adjusted for in the statistical analysis. The women also smoked more in early pregnancy than other women did, a factor that will be appear as a strong confounder in studies of preterm birth and intrauterine growth and also to some extent as a confounder in studies on congenital malformations (Källén, 2002). Adjustment was also made for maternal smoking in rather broad classes. This adjustment may not be complete for two reasons. The upper class is “smoking 10 cigarettes or more per day” and is thus open-ended. Women using SSRIs may be smoking more cigarettes per day than other women within each smoking class. Smoking refers to smoking in early pregnancy, which is relevant with respect to congenital malformations but may be insufficient in studies of preterm birth and so on; women using SSRIs may stop smoking during pregnancy less often than other women. There may thus be a residual confounding from smoking in that part of the analysis.

Women using SSRIs were found to have had ≥3 previous miscarriages more often than expected. This occurred in ∼2% of the women. It could be a confounder because repeated previous miscarriages may be associated with a poor pregnancy outcome. Adjustment for this variable was therefore made.

Subfertility, estimated as years of unwanted childlessness, was weakly and negatively associated with the use of SSRIs. Subfertility is a risk factor for poor pregnancy outcome, although not a very strong factor (Ghazi et al., 1991). There was a slightly higher use of SSRIs among women with a high BMI than among other women. This may also affect pregnancy outcome (Cedergren and Källén, 2003). Women born outside Sweden used SSRIs or reported SSRI use less often than did Swedish-born women. The risk for a cardiac defect may vary with ethnicity.

Concomitant drug use differed markedly between women using SSRIs and other women. The most remarkable difference concerned psychoactive drugs: neuroleptics, sedatives, and hypnotics, where the OR for drug use was from 6.8 to 30.2.

In general, there was no increase in the risk for congenital malformations among infants whose mothers had used SSRIs compared with other infants. When the analysis was restricted to cardiac defects, the pattern changed markedly: a significantly increased risk was seen after paroxetine use but not after the use of any other SSRI, and this difference is statistically significant. It is made up of an increased risk for ventricular and/or atrial septal defects. There were few serious cardiac defects after paroxetine exposure.

In the analysis of congenital malformation risk, adjustment was made for year of birth, maternal age, parity, and maternal smoking in early pregnancy. In order to investigate whether effects of subfertility, high BMI, or the mother being born outside Sweden could explain the increased OR for cardiovascular defects after maternal use of paroxetine, women with any one of these characteristics were excluded from the analysis: if anything, the ORs increased.

The finding of an increased risk for a cardiac defect after exposure to paroxetine could well be random as a consequence of multiple testing. The finding is, however, supported by the large GlaxoSmithKline study and a small study from two teratology information centers (Diav-Citrin et al., 2005). It may still be random, however.

Two further malformations have recently been associated with maternal use of SSRIs (Alwan et al., 2005): craniostenosis and omphalocele. The OR for omphalocele (based on a study of 161 infants with omphalocele) was 3.0 with a 95% CI of 1.4–6.1. The number of exposed cases was not given. The estimate for any body wall defect (based on 316 infants with such malformations) in the present study had an upper confidence limit of 3.40 and thus does not exclude the finding made by Alwan et al. It should be noted, however, that among the three infants with abdominal wall defects, only one had omphalocele and the other two had gastroschisis, which is quite a different malformation. The rate of omphalocele after SSRI use is thus only 1/6,555 = 1.5 per 10,000, which is the same as the population rate, but obviously the rate after SSRI use has a very large CI.

The finding of an association between maternal use of SSRI and craniostenosis only gets half-hearted support from the present data. In the study published by Alwan et al., an OR of 1.8 (95% CI = 1.0–3.2) was found, based on 372 cases, again without information on the number of exposed cases. The present data (based on 612 cases) give an OR of 1.53, not significant but with an upper confidence limit of 3.33. In a study (Källén and Robert-Gnansia, 2005) on maternal drug use and infant craniostenosis, the estimated OR from the Swedish data was 2.2, not significant, and no such case was found in the Central-East France register based on 235 craniosynostosis cases with known maternal drug use.

Even though a large number of women using SSRIs in early pregnancy have been studied, it is possible that an association with a rare malformation has not been noticed. The practical significance of such a tentative association is small.

In conclusion, most likely the use of SSRIs is not associated with any increased risk of a congenital malformation in the offspring with one probable exception: use of paroxetine may be associated with an increased risk for a cardiovascular defect, made up of an excess of infants with ventricular and/or atrial septal defects. This finding cannot be explained by confounding of those putative confounders that could be studied and appears to be drug specific, as no similar effects were seen with other SSRI use. It could well be a result of multiple testing, but the finding is supported by two other published studies. If possible, avoidance of the use of paroxetine in early pregnancy can be recommended, but if exposure has already occurred, the individual risk is small.

Ancillary