Meta-analyses of studies assessing the risk of cerebral palsy (CP) in children born as a result of in vitro fertilization (IVF). OR indicates odds ratio; CI, confidence interval; NC, naturally conceived.
Hvidtjørn D, Schieve L, Schendel D, Jacobsson B, Sværke C, Thorsen P. Cerebral Palsy, Autism Spectrum Disorders, and Developmental Delay in Children Born After Assisted ConceptionA Systematic Review and Meta-analysis. Arch Pediatr Adolesc Med. 2009;163(1):72-83. doi:10.1001/archpediatrics.2008.507
Copyright 2009 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2009
To assess the existing evidence of associations between assisted conception and cerebral palsy (CP), autism spectrum disorders (ASD), and developmental delay.
Forty-one studies identified in a systematical PubMed and Excerpta Medica Database (EMBASE) search for articles published from January 1, 1996, to April 1, 2008.
Studies written in English comparing children born after assisted conception with children born after natural conception assessing CP, ASD, and developmental delay, based on original data with a follow-up of 1 year or more.
In vitro fertilization (IVF) with or without intracytoplasmic sperm injection or ovulation induction with or without subsequent intrauterine insemination.
Main Outcome Measures
Cerebral palsy, ASD, and developmental delay.
Nine CP studies showed that children born after IVF had an increased risk of CP associated with preterm delivery. In our meta-analysis including 19 462 children exposed to IVF, we estimated a crude odds ratio of 2.18 (95% confidence interval, 1.71-2.77). Eight ASD studies and 30 studies on developmental delay showed inconsistent results. No studies assessed the risk of CP, ASD, or developmental delay in children born after ovulation induction exclusively.
Methodological problems were revealed in the identified studies, and the gaps in our knowledge about the long-term outcomes of children born after assisted conception are considerable, including a lack of information on the long-term consequences of ovulation induction. Possible associations with ASD and developmental delay need assessment in larger studies. Studies on assisted conception and CP from countries outside of Scandinavia are needed, including detailed information on time to pregnancy, underlying cause of infertility, and type of IVF treatment.
In this review, we examined studies on the long-term outcomes of assisted conception, defined as in vitro fertilization (IVF) with or without intracytoplasmic sperm injection (ICSI) or ovulation induction (OI) with or without subsequent intrauterine insemination (IUI).
In vitro fertilization treatment alone now accounts for an estimated 1% to 4% of births in European countries1 and 1% of US births.2 There is a particular obligation to evaluate treatments offered in health care systems to guarantee their safety.
In vitro fertilization is associated with adverse perinatal outcomes such as preterm delivery (PTD, <37 weeks of gestation) and low birth weight (LBW, <2500 g) because of the strong association between IVF and multiple pregnancies and because even IVF singletons have an increased risk of PTD and LBW compared with naturally conceived (NC) singletons.3 Likewise, OI leads to more multiple pregnancies than natural conception,1,4 and children born after OI also have an increased risk of PTD and LBW, as reported in most studies4- 8 but not all.9,10 Pregnancy with multiples, PTD, and LBW are strongly associated with a range of long-term child health problems3,11,12 including admission to neonatal intensive care units and prolonged hospitalization,13 vision impairment,14 and cerebral palsy (CP).15 Moderate associations between delivery of multiples, PTD, LBW, advanced parental age, and developmental disabilities such as autism spectrum disorders (ASD) have also been reported.16- 23 Advanced parental age is strongly associated with assisted conception. Additionally, possible prenatal hormonal disturbances in autism, eg, elevated levels of prenatal testosterone24 and lower levels of oxytocin,25 have been reported and may be linked to reproductive problems.
We conducted a systematic review of the current evidence regarding associations between assisted conception and severe long-term outcomes, specifically CP and ASD. We also reviewed general developmental delay outcomes, as these are often the initiating diagnoses or symptoms.
We searched PubMed using the Medical Subject Headings of the National Library of Medicine terms presented in Table 1. We limited our search to studies reporting human outcomes of assisted conception published from January 1, 1996, through March 31, 2008 in English and including children exposed to assisted conception (IVF, ICSI, IUI, or OI). We performed our search on April 10, 2008. The main outcome measures of interest were CP, ASD, and developmental delay.
A total of 130 articles met our initial search criteria. All abstracts were reviewed and 41 articles met the additional inclusion criteria of original data, follow-up time of 1 year or more, and a comparison group of unexposed NC children. Excluded articles are shown in an online appendix (http://www.nanea.dk/articles/hvidtjorn-2008-review). We examined the reference lists of all 41 articles eligible for full review, but did not identify any additional articles. We searched EMBASE using the search terms in Table 1 for CP and ASD, but did not identify any additional articles with original data. The initial screening of abstracts was conducted by an author (D.H.). Each of the articles meeting the final inclusion criteria were reviewed in full by 2 authors (D.H. and 1 coauthor; L.S., D.S., P.T., or B.J.).
Here, we present findings from the reviewed studies on CP, ASD, and developmental delay. For each outcome we discuss the main methodological strengths and limitations of the studies. Because of the elevated risk of multiple births in assisted conception, whenever possible we present the articles' findings for singletons and multiples separately, as well as all births combined, with or without adjusting for PTD.
We performed meta-analyses with fixed-effect models using the Mantel-Haenzel method and calculated nonadjusted summary estimates for the CP studies. Several studies included overlapping cohorts; for these we selected the study presenting the most detailed data (the necessary numbers to calculate a weighted summary). Stata software version 10.0 (Stata, College Station, Texas) was used.
We did not perform meta-analyses for the ASD studies reviewed because of the small number of studies within the specific study design types and potential methodological problems in the few studies similarly designed (see below). We did not perform meta-analyses with the developmental delay studies because a wide range of different measurements were used to assess cognitive, motor, and behavioral development. Moreover, even when 2 studies used the same instrument, child age at assessment was often variable, as were sample inclusion and/or exclusion criteria.
Of the 41 included studies, only 2 were case-control studies; the remainder were cohort studies. Nine studies assessed the risk of CP (Table 2) and 8 the risk of ASD (Table 3). Thirty studies assessed developmental delay based on various standardized scales (Table 4). Within all 3 outcomes there were overlapping study populations.
Results for CP are shown in Table 3. Of the 9 studies assessing the risk of CP in children born after IVF, only 1 was conducted outside of Scandinavia (a Croatian study of triplets).26 The Scandinavian studies were population-based, diminishing any selection bias. However, the Scandinavian countries are very similar in demographic factors, socioeconomic status, ethnicity (mainly white), and free access to health care (including fertility treatment), and this uniformity might limit extrapolation of the findings to populations of different ethnic profiles, demography, and health care systems.
Sample Size and Precision. The Scandinavian studies were based on data from registers comprising health status on the whole population and information on IVF treatments from all fertility clinics retrieved via the unique identification number given to each citizen, resulting in large cohorts reporting fairly precise risk estimates. Risk estimates were less precise in the strata of multiplicity because of reduced sample sizes and lower expected prevalence within the group of singletons only.
Exposure Data. One study assessed the risk of CP in triplets born after all types of assisted conception, identified through a single hospital.26 The remaining 8 studies evaluated children born after IVF using specific population-based data on exposure and identified either through an IVF register (in Denmark) or through all fertility clinics in the specific country (Sweden or Finland); they likely included practically all children born after IVF. However, no study specifically examined OI and only 1 study specified that pregnancies resulting from OI (OI children) were excluded from the comparison group of children born after natural conception.27 Thus for most studies, some misclassification of the unexposed group likely occurred, as OI children were counted as NC children. This would presumably lead to bias toward the null hypothesis, as OI is strongly associated with multiple pregnancies.28,29 In the studies of twins, Pinborg et al30,31 reported that 17.3% of pregnancies in the unexposed groups were the result of OI.
Outcome Data. Cerebral palsy was defined as a diagnosis of International Statistical Classification of Diseases, 10thRevision (ICD-10) code G80.0-G83.9, stated as “ICD-10 diagnosis CP,” or described in 1 study as “children diagnosed with CP by a pediatric neurologist.”26 Six studies obtained information about CP diagnoses from hospital discharge registers27,31- 35; the remaining 3 used records from rehabilitation centers26,36 or questionnaires confirmed by discharge registers.30 However, the validity and completeness of the CP diagnoses in hospital discharge registers have been questioned.37 Klemetti et al27 supplemented their information on CP diagnosis using registers reporting child care support. Only Strömberg et al36 obtained the CP diagnoses from the medical records of all disability centers in Sweden.
All studies of singletons and multiples combined found a statistically significant increase in the risk of CP in children born as the result of IVF (IVF children) compared with NC children.27,32- 34,36 However, a disproportionate number of the IVF children were multiples. Odds ratios (OR) ranged from 1.6 to 3.7 in analyses adjusting for various factors other than PTD. The strongest association between IVF and CP (OR, 3.7; 95% confidence interval [CI], 2.0-6.6) was reported by Strömberg et al.36 A meta-analysis (without overlapping study cohorts) included 19 462 IVF children and demonstrated an increased risk of CP in IVF children (OR, 2.18; 95% CI, 1.71-2.77) (Figure).
Some studies also presented results from analysis including PTD. Although Strömberg et al found a decrease in the magnitude of the association after adjusting for PTD, it remained strong and statistically significant (OR, 2.9; 95% CI, 1.4-6.0). Hvidtjørn et al32 found no association between IVF and CP after adjusting for PTD. In the study by Källen et al,33 the OR in a stratum of children carried to term was 0.88 (95% CI, 0.46-1.70), and Klemetti et al27 found that 88% of IVF children with CP were born preterm. Dissimilarities in analytic approach complicate an immediate comparison of these findings; however, it seems clear that the risk of CP in IVF children at least partially operates through PTD.
Only Källen et al33 had information on time to pregnancy (TTP) and reported no association between IVF and CP after adjusting for TTP.
There was a tendency toward an increased risk for CP in singletons born as a result of IVF (IVF singletons) compared with non-IVF singletons,27,32,35,36 but not all studies reached statistical significance.27,32 The 2 Danish studies used overlapping cohorts but had different analytical approaches; Lidegaard et al35 reported a crude rate ratio of 1.8 (P < .01) in contrast to the non–statistically significant finding by Hvidtjørn et al32 who found a hazard rate ratio of 1.28 (95% CI, 0.80-2.03) in an analysis adjusting for sex, parity, maternal age, and educational level. Strömberg et al36 reported an OR of 2.8 (95% CI, 1.3-5.8) after adjusting for sex, year of birth, and birth hospital. A meta-analysis (without overlapping study cohorts) comprised 12 191 IVF singletons and showed an increased risk of CP in IVF singletons (OR, 1.82; 95% CI, 1.31-2.52) (Figure).
While the effect estimates varied (OR range, 0.6-1.5) of the 5 studies examining CP in twins born as a result of IVF (IVF twins) vs non-IVF twins, confidence limits were wide and overlapping (1.0) in all studies.27,30- 32,36 In contrast, in the study on the risk of CP in triplets, assisted conception had a statistically significant protective effect.26
Data regarding ASD can be found in Table 4. All but 1 of the ASD studies (an Israeli case-control study)26 originated in Scandinavian countries; consequently, they have the same limitations regarding external validity as the CP studies.
Sample Size and Precision. The identified ASD studies covered the period from 1970 to 2001. During this period the prevalence of ASD apparently changed considerably from 4 to 5 per 10 000 children to 6 to 7 per 1000 children.38 It has been questioned whether this increase reflects a true rise in the prevalence of ASD or at least partly reflects changes in diagnostic criteria and increased medical and/or public awareness of these disabilities.39,40 Given that active monitoring of ASD was not in operation in most areas, the expected prevalence at a particular time and area is not known. If we apply a conservative estimate of about 3/1000 children for the time period of these ASD studies, it would require sample sizes of 5268 exposed and 15 804 unexposed to detect a risk of 2.0 in cohort studies with 80% power and 95% CI (estimations in Calculations in Epi Info; Centers for Disease Control, Atlanta, Georgia). Only 2 studies on multiples and singletons combined and 1 on singletons alone achieved this size.34,35
Exposure Data. For the cohort studies from Scandinavia, nondifferential misclassification of OI in the unexposed group was likely, as described above for the CP studies.
In the 1 case-control study, information on exposure was retrieved differently, namely from birth records. However, a validation study on information about infertility treatment in birth records indicated low sensitivity overall for fertility treatment reporting and potentially differential reporting. Higher-risk infants such as multiples were more likely than singletons to be correctly reported as conceived after infertility treatment (personal communication, L.S.; February 1, 2008).
Outcome Data. Most studies obtained information on outcome from hospital discharge registers27,31,34,35,41 and others used records from rehabilitation centers36,42 or questionnaires confirmed by discharge registers.30 Autism spectrum disorder was defined as a diagnosis of ICD-10 code F84.0, F84.1, F84.5, or F84.9 or ICD-8 and Diagnostic and Statistical Manual of MentalDisorders(Third or Fourth Edition) (DSM III-IV). Three of the studies used only infantile autism as the study outcome, 3 used ASD, and 2 used an even broader range of psychiatric diagnoses including ASD, complicating the comparison between the studies. Diagnosis was retrieved from Hospital Discharge Registers, from The Danish Psychiatric Central Research Register, and in 1 case from an autism treatment organization.42
Findings. Two studies evaluated children born after assisted conception in general,41,42 while the remaining 6 evaluated IVF using information from all of the fertility clinics in each country. Findings were inconsistent overall and when considering singletons and multiples separately. Only the study by Klemetti et al27 reported a statistically significant increased risk for a broad range of psychiatric disorders (F80-F98) including ASD in children born after IVF (OR, 1.68; 95% CI, 1.11-2.58); however, they did not provide results for ASD alone.
In contrast, Maimburg and Væth41 reported a protective effect between assisted conception and infantile autism in their case-control study, including 473 children with infantile autism and 473 matched control children (OR, 0.37; 95% CI, 0.14-0.98) adjusting for several factors including gestational age.
Developmental delay associations can be seen in Table 4. Most studies on developmental delay limited their study participants to children exposed to assisted conception at 1 or 2 fertility clinics, introducing the possibility of selection bias if these clinics were not representative of the entire population. Furthermore, bias might have resulted from differential participation because of substantial nonparticipation in some studies requiring individual examination of the included children. For example, in one of the largest international studies,43 participation rates varied across countries, with rates from 25% to 96% and differences of participation rates of exposed and nonexposed children up to 50%.
Perhaps the foremost problem with these studies was that 21 of them excluded children with risk factors such as multiplicity, PTD, and neonatal complications a priori to test the possible influence of IVF or ICSI on development apart from these complications. While this was certainly a reasonable method for the stated objective, it severely compromised the value of the studies to inform an increase in general developmental delay in the total population of children born after assisted conception, as one could argue that delivery of multiples and PTD are key factors in the causal pathway.
However, studies on developmental delay were conducted on several continents, ensuring representation of diverse populations.
Sample Size and Precision. Except for 1 population-based register study,33 the studies on developmental delay individually included between 43 and 999 children, with each study sample of insufficient size to identify relatively rare events such as CP or ASD. Many sample sizes were also insufficient to detect moderate differences in the broader child development measures.
Exposure and Outcome Data. Exposed children were identified at fertility clinics. Nearly all studies used parental questionnaires and/or some type of individual standardized examination of the child to assess development, but they used different measurement scales. The population-based study used hospital discharge registers and ICD codes.33 While ascertainment of both exposure and outcome were based on reliable sources and standardized measures, these strengths did not overcome the limitations of possible selection bias due to low participation rates in some studies and selection of specific low-risk children only.
Findings. We identified 30 studies assessing developmental delay in children born after assisted conception. Seventeen studies evaluated children born after ICSI solely, 13 also evaluated IVF, and 2 evaluated children born after assisted conception in general. Eight studies reported the number of children with CP in their cohorts and 2 stated the number of children with ASD. The numbers were too small for statistical estimation regarding these conditions, as would be expected in cohorts of fewer than 1000 children. In one study the children identified with CP or other adverse outcomes were excluded before further examination.44
Fourteen studies assessed motor development,43- 56 and 2 reported that children born as a result of ICSI (ICSI children) had a statistically significant higher risk of delayed motor development at 1 to 2 or 5 years of age, respectively.44,57 Eleven studies assessed behavioral development33,44,46,56,58- 64; 2 reported a statistically significant higher risk of delayed behavioral development in ICSI children aged 1 to 2 years44 and IVF children aged 9 to 10 years,58 respectively. The large population-based register study found an increased risk of behavioral problems in children born after IVF (OR, 1.74; 95% CI, 1.11-2.74).33 Nine studies assessed delay in cognitive development43,45,51,54,56- 58,65,66 and 1 study reported that ICSI children had a statistically significant lower risk of delayed cognitive development at 8 years of age, while another study reported that ICSI children had a statistically significant higher risk of delayed cognitive development at 5 to 8 years of age.65 Eleven studies assessed delay in mental development,44,46- 50,52,67- 71 and 1 found less adverse outcomes in twins born as a result of assisted conception71 while all other studies on developmental delay described non–statistically significant findings.
This systematic review included studies assessing the risks of CP, ASD, and developmental delay in children born after assisted conception. Owing to the size of the study cohorts and the concordant results, the CP studies offer persuasive evidence of an increased risk of CP in children born after IVF that is explained in part by an increased risk of PTD. Cerebral palsy is a lifelong condition and a heavy burden on the child, family, and health care system in terms of both personal and economic costs.72
In contrast, studies assessing the risk of ASD were inconsistent. This might have been owing to the aforementioned methodological problems in these studies or simply might have reflected a lack of association between assisted conception and ASD. As ASD seems to have a heterogenic etiology, the effect of weak associations will only be apparent in larger samples.40 Moreover, a possible association between assisted conception and ASD needs examination in studies covering recent time periods with more complete ASD reporting.
Studies on developmental delay following assisted conception mainly included small select groups of low-risk children, and they presented generally non–statistically significant results. Thus, data on larger samples of the full range of children conceived via assisted conception are needed.
We did not identify any studies assessing the risk of CP, ASD, or developmental delay in children born after OI specifically. The lack of evidence on the long-term risks of OI is of concern given associations shown in previous studies between OI and PTD, LBW, and delivery of multiples.1,4- 8
While the studies offer persuasive evidence of an association between IVF and CP, gaps remain in understanding this relationship. Because the increased risk of CP in children born after IVF seems to operate partly through the causal pathway of IVF, delivery of multiples, and PTD, the extent of the CP risk associated with IVF in a population will likely depend on the rate of IVF multiples. This rate is lower in Scandinavian countries that regulate the number of embryos transferred; eg, in Denmark in 2005 only 1.5% of IVF children were triplets and 32.6% were twins.73 In contrast, the rate of multiples born after IVF in the United States in 2004 was 50%, with a larger proportion of triplets and longer gestations.2
The etiology behind the risk of CP in IVF singletons remains unclear, but 2 Danish studies suggest that the phenomenon of vanishing embryos in early pregnancy might be part of the etiology. Hvidtjørn et al74 found that 3.9 of 1000 (95% CI, 2.2-5.5/1000) singletons born after transfer of more than 1 embryo had CP, similar to the proportion among twins born after the transfer of 2 embryos (4.4 of 1000 children; 95% CI, 1.9-6.9). Pinborg et al75 showed that the group of IVF singletons in which a coembryo had vanished before 22 weeks of gestational age had nearly twice the risk of CP (OR, 1.9; 95% CI, 0.7-5.2) compared with IVF singletons originating from pregnancies with only 1 fetus at 8 weeks of gestational age. These findings need further exploration.
We also still need to determine whether the increased risk of CP after IVF is associated with subfertility, type of subfertility, or any specific subtype of IVF. Subfertility has been associated with adverse pregnancy outcomes such as PTD and neonatal death,76,77 though not by Kapitejn et al.9 Ericson et al34 found an increased risk of hospitalization with increasing time to pregnancy. We found only 1 study that adjusted for time to pregnancy33 and, in doing so, the risk of CP disappeared. The type of subfertility was taken into account in 1 study only, revealing similar risks of CP within the different types, though this was based on small numbers.32
While animal studies report long-term adverse effects of urinary gonadotrophins compared with other treatment regimens used in controlled ovarian stimulation,78,79 only a few studies in this review compared different treatment types. Three studies evaluated possible differences between conventional IVF and ICSI31- 33 and found comparable risks in the 2 groups. One study compared the risk of CP in children born after use of fresh vs frozen embryos and reported a hazard rate ratio of 2.32 (95% CI, 0.80-6.76) in the latter group,32 but this was based on small numbers.
A comprehensive search in PubMed and EMBASE revealed 130 articles, of which 41 were eligible for review. The main reasons for exclusion of articles were (1) commentary and/or review, (2) different exposure or outcome, and (3) no NC comparison group. Excluded articles are shown in an online appendix. We consider the possible selection bias minimal, as none of the studies excluded fulfilled the a priori criteria.
In summary, this systematic review revealed important gaps in the evidence of long-term outcomes in children born after assisted conception. Possible associations between assisted conception and ASD need assessment in larger studies with well-defined outcomes. Studies on assisted conception and CP from countries outside of Scandinavia are needed as well as studies with detailed information on TTP, underlying causes of infertility, and types of IVF treatment. The long-term outcomes of OI must be addressed. Given the continually increased use of fertility treatments worldwide, studies addressing these very large gaps in the knowledge of the long-term health and development of children born after assisted conception are an important public health objective.
Correspondence: Dorte Hvidtjørn, MPH, NANEA, Institute of Public Health, Department of Epidemiology, University of Aarhus, Paludan-Müllers vej 17, 8000 Aarhus C, Denmark (firstname.lastname@example.org).
Accepted for Publication: May 29, 2008.
Author Contributions:Study concept and design: Hvidtjørn, Schieve, Schendel, and Jacobsson. Acquisition of data: Hvidtjørn, Schieve, and Jacobsson. Analysis and interpretation of data: Hvidtjørn, Schieve, Schendel, Jacobsson, Sværke, and Thorsen. Drafting of the manuscript: Hvidtjørn. Critical revision of the manuscript for important intellectual content: Hvidtjørn, Schieve, Schendel, Jacobsson, Sværke, and Thorsen. Statistical analysis: Schieve and Sværke. Obtained funding: Hvidtjørn. Administrative, technical, and material support: Hvidtjørn, Schendel, and Thorsen. Study supervision: Schieve, Schendel, Jacobsson, and Thorsen.
Financial Disclosure: None reported.
Funding/Support: The study was funded by the Danish Agency for Science, Technology, and Innovation; the University of Aarhus; the Elsass Foundation; the Health Insurance Foundation; the Augustinus Foundation; the Julie von Müllens Foundation; Direktør Jacob Madsen & Hustru Olga Madsens Fond; and the Aase and Ejnar Danielsen Foundation.
Disclaimer: The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the Centers for Disease Control and Prevention.
This article was corrected online for typographical errors on 1/26/2009.