Selective serotonin reuptake inhibitors (SSRIs) are increasingly used in the treatment of depression and anxiety disorders. Currently, SSRIs appear to provide the best balance between efficacy and safety and are the preferred first-line treatment. Depression is common during pregnancy, and given the risks of untreated depression to mother and fetus,1,2 pharmacologic treatment is warranted, often with SSRIs.3 However, SSRIs cross the placenta, and a number of safety concerns have been raised.4 These include concerns about birth defects, adverse obstetrical and neonatal outcomes, and effects on cognitive and behavioral development in childhood.
Recently, autism spectrum disorders have been linked to maternal use of SSRIs during pregnancy. In a case–control study involving 298 children with autism spectrum disorders, the use of SSRIs by the mother during pregnancy was shown to be associated with a risk of autism spectrum disorder that was increased by a factor of 2.5 A causal association is plausible. Increased blood levels of the neurotransmitter serotonin have been observed in persons with autism spectrum disorders.6 Furthermore, this neurotransmitter appears to play an important role in early brain development, and manipulation of serotonin homeostasis can alter neuroanatomical and neurophysiological development and produce enduring behavioral changes in animal models.7-9 However, more research — including large observational studies involving humans — is needed to address a potential association. Using data from the Danish health registries, we conducted a nationwide cohort study of SSRI use during pregnancy and the risk of autism spectrum disorders in the offspring.
We used the nationwide Medical Birth Registry to construct a cohort of all live births in Denmark during the study period of January 1, 1996, through December 31, 2005.10 The Medical Birth Registry contains records of all births in Denmark, including information on the personal identification number (a 10-digit number assigned to all Danish residents and used in all nationwide registries11) of the parents and the newborn, the date of birth, whether the birth was a single or multiple birth, the gestational age, vital status, and other physical characteristics of the newborn. The study was approved by the Danish Data Protection Agency. Approval by an institutional review board and informed consent are not required for registry-based research in Denmark. The first author vouches for the accuracy and completeness of the data.
We estimated the beginning of pregnancy by subtracting the gestational age from the date of birth. The gestational age recorded in the Medical Birth Registry is based on the self-reported first day of the last menstrual period and in most cases is confirmed by prenatal ultrasonography. An estimated 93% of all pregnant women underwent an obstetrical ultrasound examination in 2000.12 We included only births with a known gestational age and further restricted the cohort to singleton births.
From the National Patient Register,13 we obtained information on a number of genetic conditions (the fragile X syndrome, tuberous sclerosis, Angelman's syndrome, Down's syndrome, DiGeorge's syndrome, neurofibromatosis, and the Prader–Willi syndrome) and on the congenital rubella syndrome. All these conditions are associated with an inherent increased risk of autism, and offspring with any of these conditions were excluded from the cohort.
The unique personal identification numbers for mother and child allowed us to link information on maternal use of SSRIs, diagnoses of autism spectrum disorders among offspring, and potential confounding variables relevant to the births in the cohort.
Exposure to SSRI Drugs
Information on SSRI prescriptions filled by women in the cohort was obtained from the National Prescription Registry.14 This register contains individual-level information on all prescriptions dispensed at Danish pharmacies since 1995. Each record includes the personal identification number of the patient, the date the prescription was filled, the type of drug according to the Anatomical Therapeutic Chemical (ATC) classification system, the number of daily doses specified in the prescription, and the number of packages obtained. We included prescriptions with the ATC code N06AB (selective serotonin reuptake inhibitors) that were filled during the period from 2 years before the beginning of the pregnancy until delivery. We used the date on which a prescription was filled to indicate initial use of the prescribed drug.
Autism Spectrum Disorders
Information on diagnoses of autism spectrum disorder during the study period was obtained from the Danish Psychiatric Central Register.15 This register includes diagnoses made and diagnostic codes assigned by child psychiatrists and contains information from psychiatric hospitals and psychiatric units (inpatient and outpatient ). The coding classification used during the study period was the International Classification of Diseases, 10th Revision (ICD-10). We classified autism spectrum disorders in two groups: autistic disorder (ICD-10 code F84.0) and other autism spectrum disorders (including atypical autism, Asperger's syndrome, and other or unspecified pervasive developmental disorder; ICD-10 codes F84.1, F84.5, F84.8, and F84.9).
We selected a priori a number of potential confounders that were plausible risk factors for either autism or SSRI use and for which information was available to us from the nationwide Danish health registries. From the Medical Birth Registry and the Danish Civil Registration System11 (the key national demographic registry in Denmark) we obtained information on the year of the birth and on maternal parity, age at the onset of pregnancy, country of origin, place of residence at the start of the pregnancy, and smoking status during pregnancy, and we linked this information to the cohort. From the Danish Psychiatric Central Register and the National Prescription Registry we obtained information on maternal psychiatric conditions and selected drugs other than SSRIs that were used during the pregnancy. The Danish Psychiatric Central Register includes diagnoses made in a psychiatric hospital or psychiatric unit (inpatient or outpatient) but does not include diagnoses made by a medical specialist in the primary care setting. Information on employment status and the mother's level of education was obtained from Statistics Denmark.16
We performed a survival analysis to follow up children from birth until January 1, 2010, or until the child reached 10 years of age, died or was lost to follow-up, or received a diagnosis of autism spectrum disorder — whichever came first.
The resulting follow-up times and numbers of diagnoses of autism (autism counts) were aggregated according to maternal use or nonuse of SSRIs during pregnancy. We used Poisson regression on the autism counts with the logarithm of the follow-up times as the offset term, assuming a Poisson distribution for the autism counts, to estimate incidence rate ratios and 95% confidence intervals so that we could compare the rates of autism spectrum disorder among the offspring of women who had been exposed to SSRIs during pregnancy with the rates among the offspring of women who had not been exposed to SSRIs during pregnancy.17 The regression analysis was performed with the use of the PROC GENMOD procedure in SAS software, version 9.1 (SAS Institute).
The date the prescription was filled was considered to be the date of exposure. The use of SSRIs during pregnancy was defined as use during the period 4 weeks before the beginning of the pregnancy until delivery. We further analyzed the use of SSRIs specifically in the first trimester. To evaluate the potential for confounding by indication (i.e., the possibility that women with depression or other indications for SSRI use would be more likely to have children with autism spectrum disorders), we also looked at the use of SSRIs during the period from 2 years until 6 months before the beginning of the pregnancy. This allowed us to identify a group of women who received SSRIs during that period but not during pregnancy and two groups of women who received SSRIs during pregnancy: those who received the drugs both before and during the pregnancy and those who did not receive the drugs before the pregnancy but did receive them during the pregnancy. In all the analyses, unexposed pregnancies were considered to be pregnancies in women who had no exposure to SSRIs from 2 years before the beginning of pregnancy through the end of the pregnancy.
We explored whether the association between the use of SSRIs during pregnancy and autism spectrum disorder differed in subgroups defined according to the child's age at the time of the follow-up assessment, the calendar period during which the follow-up assessment was performed, the type of SSRI used, previous psychiatric diagnoses in the mother, other drugs used during pregnancy, and the type of autism spectrum disorder in the offspring. We estimated crude rate ratios, rate ratios adjusted for age and calendar period, and fully adjusted rate ratios that included, in addition to age and calendar period, all the potential confounders listed above. No more than 5% of the values were missing for any of the potential confounders, and we therefore used simple imputation, in which missing values were replaced with the most common value.18 The implications of this strategy were explored in sensitivity analyses.
We identified 658,755 live births in Denmark during the period from January 1, 1996, through December 31, 2005. We then excluded births without a known gestational age (4016), multiple births (26,526), and births resulting in offspring with the congenital rubella syndrome or with genetic conditions that are associated with an inherent risk of autism (1338). This resulted in a final study cohort of 626,875 children (51.3% were boys). A total of 6068 mothers (1.0%) of the children in the cohort used SSRIs during pregnancy. Table 1
Characteristics of Mothers in a Cohort of 626,875 Live Births in Denmark, According to Status with Respect to SSRI Use during Pregnancy.
shows the characteristics of the mothers according to their status with respect to SSRI use during pregnancy. Parity and maternal age at the beginning of the pregnancy were similar in the group of women who received SSRIs and the group of women who did not receive SSRIs. As compared with women who did not use SSRIs during pregnancy, those who did were slightly more likely to have been born in Denmark and to be residing in Jutland or Funen; had a lower level of education and lower socioeconomic status; and were more likely to have psychiatric diagnoses, to use other drugs during pregnancy, and to smoke during pregnancy. Furthermore, the use of SSRIs during pregnancy became more prevalent over the course of the study period.
Autism Spectrum Disorder
During 5,057,282 person-years of follow-up, we identified 3892 cases of autism spectrum disorder (1603 cases of autistic disorder and 2289 cases of other autism spectrum disorders), yielding an incidence rate of 77.0 per 100,000 person-years. The median age at the diagnosis of autism spectrum disorder was 5.6 years (interquartile range, 4.1 to 7.5). During the follow-up period, 15,585 children emigrated from Denmark, 3157 died, and 387 were lost to follow-up (i.e., their personal identification numbers could no longer be found in the national registries). Table 2
Autism Spectrum Disorder in Offspring, According to Maternal Characteristics.
shows some of the risk factors for autism spectrum disorder among the offspring in this study. (Table S1 in the Supplementary Appendix, available with the full text of this article at NEJM.org, shows the association of autism spectrum disorder with other risk factors, including the age of the child, the calendar period, and maternal age at the time of delivery.) Strong associations were seen between autism spectrum disorder and maternal psychiatric diagnoses and the use of drugs other than SSRIs during pregnancy. These associations suggest possible confounding by indication, especially given that mothers with these characteristics were also more likely to use SSRIs during pregnancy (Table 1).
SSRI Use during Pregnancy and Risk of Autism Spectrum Disorder
Association between Period of SSRI Use and Autism Spectrum Disorders in Offspring.
shows the association between maternal use of SSRIs during pregnancy and the risk of autism spectrum disorder in the offspring. During 42,400 person-years of follow-up, we identified 52 cases of autism spectrum disorder among the offspring of women who were exposed to SSRIs during pregnancy (incidence rate, 122.6 per 100,000 person-years). For the comparison with no use of SSRIs before or during pregnancy, this rate corresponded to a crude rate ratio of 1.62 (95% confidence interval [CI], 1.23 to 2.13). However, in a fully adjusted analysis, the use of SSRIs during pregnancy (including use both before and during pregnancy and use only during pregnancy) was not associated with a significantly increased risk of autism spectrum disorder in the offspring (rate ratio 1.20; 95% CI, 0.90 to 1.61). Table S2 in the Supplementary Appendix shows the extent to which each potential confounder included in the fully adjusted models influenced the estimate. The fully adjusted rate ratio associated with the use of SSRIs in the first trimester was 1.35 (95% CI, 0.97 to 1.87), whereas the fully adjusted rate ratio in the subgroup of women who received SSRIs during pregnancy but not before pregnancy was 1.40 (95% CI, 0.92 to 2.13). These estimates should be compared with the fully adjusted rate ratio of 1.46 (95% CI, 1.17 to 1.81) in the subgroup of women who had used SSRIs before pregnancy but did not use these drugs during pregnancy.
Table S3 in the Supplementary Appendix shows the association between the use of SSRIs during pregnancy and the risk of autism spectrum disorder in offspring according to several prespecified characteristics. Although the point estimates suggested that in some subgroups of women the association differed from the fully adjusted rate ratio of 1.20, tests of homogeneity were consistent, with no significant differences.
Since we used a simple form of imputation for missing data, we tested the validity of this approach by assessing the association between SSRIs use during pregnancy and the risk of autism spectrum disorder in offspring in an analysis that was restricted to pregnancies for which complete information was available (574,020 pregnancies). This analysis yielded a fully adjusted rate ratio of 1.15 (95% CI, 0.85 to 1.56), which was similar to the results of our main analysis.
In a large population-based study of 626,875 live births, the use of SSRIs during pregnancy was not associated with a significantly increased risk of autism spectrum disorder in the offspring. Our study was prompted by experimental evidence implicating serotonin in autism6-9 and by recent work from Croen and colleagues, who conducted a case–control study using data from a health maintenance organization in Northern California.5 The researchers identified 298 children with autism spectrum disorder; the mothers of 20 of those children, had received antidepressants (15 had received SSRIs) during the year before the child's birth. This corresponded to an increase by a factor of 2 in the risk of autism spectrum disorder associated with the use of antidepressants during pregnancy. The risk was increased more with SSRIs than with other antidepressants, and the risk was increased by a factor of more than 3 with the use of SSRIs specifically in the first trimester. A case–control study conducted in Sweden showed an odds ratio of 1.65 (95% CI, 0.90 to 3.03) for the association between SSRI use during pregnancy and the risk of autism spectrum disorder in the offspring, on the basis of 14 exposed pregnancies. The study assessed self-reported SSRI use and could not take confounding by depression directly into account.19
Our study has a number of strengths. First, the Danish registries allowed the linkage between drug use during pregnancy and autism spectrum disorders in the offspring, so that we were able to determine the potential effect of the drugs later in childhood. Second, our study was conducted on a nationwide level, and therefore the cohort was large, with 626,875 live births, including 52 cases of autism spectrum disorder in offspring of women with SSRI exposure during pregnancy — more than three times the number of cases among exposed women in the study by Croen and colleagues or the Swedish study. Third, we used administrative health registry data in a historically prospective study design, with independent ascertainment of exposure and outcome, thus reducing the potential for selection and recall bias. Finally, the National Prescription Registry, from which we obtained data on exposure to SSRIs, is considered to be nearly complete14; all Danish pharmacies must report filled prescriptions to this registry for reimbursement purposes.
Our study also has a number of limitations. First, in our cohort, the prevalence of pregnancy-related use of SSRIs was 0.97%. In a survey of the automated databases of seven health plans across the United States during the period from 2001 through 2005, SSRIs were much more widely used during pregnancy, with a prevalence of 5.6%.3 Thus, our findings may not be generalizable to other countries. Second, in our study and similar studies relying on registry data, the date on which the prescription is filled is assumed to be the same as the date of initial use of the prescribed drug. This assumption can overestimate the prevalence of exposure during pregnancy and bias the observed results toward no effect.
Third, we used the Danish Psychiatric Central Register for case ascertainment. A previous study showed that 94% of 499 children registered with autism spectrum disorder diagnoses met the necessary diagnostic criteria in a chart review.20 The reported prevalence rates in previous Danish epidemiologic studies of autism spectrum disorder that used this register are consistent with rates reported in other, similar countries.21 The prevalence of autism spectrum disorder in our cohort was 0.62%. The Centers for Disease Control and Prevention recently estimated that the childhood prevalence of autism spectrum disorder in the United States is 1 case per 88 children (1.14%).22 It is worth noting that not all children in our study have been followed throughout childhood, and it is likely that some will receive a diagnosis of autism spectrum disorder at older ages. However, the date of diagnosis may differ considerably from the date on which symptoms are first noted. This discrepancy can introduce detection bias if autism spectrum disorders are diagnosed earlier in the offspring of women who were exposed to SSRIs than in the offspring of women who were not, such that the risk with SSRIs would be overestimated if autism spectrum disorders in children whose mothers were not exposed to SSRIs were not detected during the study period.
Finally, the main methodologic challenge of our study and similar studies is confounding by indication.23 If women with depression or other indications for SSRI use are more likely to have children with autism spectrum disorders, a false association between SSRI use and autism spectrum disorders will be present in an observational study. In unadjusted analyses, we did find a significantly increased risk of autism spectrum disorder in association with the use of SSRIs during pregnancy. In fully adjusted analyses, however, the risk was no longer significant. This result was primarily due to adjustment for a number of psychiatric diagnoses (Table S2 in the Supplementary Appendix), which is consistent with the presence of confounding by indication in the unadjusted analysis. We had access only to maternal psychiatric conditions diagnosed in psychiatric hospitals and psychiatric units (inpatient and outpatient), with no information on cases diagnosed in primary care settings. We introduced a period of SSRI use before pregnancy, which allowed us to evaluate whether maternal use of SSRIs before but not during pregnancy was associated with an increased risk of subsequent autism spectrum disorders in the offspring. The increased risk associated with this pattern of use similarly suggests that any risk associated with SSRI use during pregnancy may be related to the indications for its use rather than a causal effect.
The prevalence of autism spectrum disorders has been increasing in the past couple of decades.24 The causes of this increase have been hotly debated. Parallel increases in diagnostic practices, the availability of special health care services, public awareness, and suspected environmental risk factors (both intrauterine and postnatal) are often cited. With respect to environmental factors, the use of SSRIs has been increasing. In our study, the risk of autism spectrum disorders was not significantly increased among children whose mothers received SSRIs during pregnancy. The statistical power of the study allows us to rule out an increase in the relative risk of more than 61% with a high degree of certainty, but we cannot exclude the possibility of a smaller increase in risk. However, the increased risk associated with SSRI use before but not during pregnancy suggests that any risk associated with the use of SSRIs during pregnancy may be related to the indications for their use rather than an effect of these drugs. This highlights the potential effect of confounding by indication in our study and similar studies of SSRIs and the importance of being able to adequately take this confounding into account in the study design. As with all observational studies, the possible presence of residual and unmeasured confounding or ascertainment bias with respect to exposure and outcome adds to the imprecision of our estimates. Interpretation of our results should take this factor into account.