If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
University Medical Center Groningen, Department of Paediatrics, Division of Developmental Neurology, University of Groningen, Hanzeplein 1, 9713 GZ Groningen, The Netherlands
Department of Department of Mathematics and Computer Science, Division of Stochastics W&I, Eindhoven University of Technology, De Rondom 70, 5612 AP Eindhoven, The Netherlands
University Medical Center Groningen, Department of Paediatrics, Division of Developmental Neurology, University of Groningen, Hanzeplein 1, 9713 GZ Groningen, The Netherlands
Department of Department of Mathematics and Computer Science, Division of Stochastics W&I, Eindhoven University of Technology, De Rondom 70, 5612 AP Eindhoven, The Netherlands
University Medical Center Groningen, Department of Paediatrics, Division of Developmental Neurology, University of Groningen, Hanzeplein 1, 9713 GZ Groningen, The Netherlands
Research on cognitive and behavioural development of children born after assisted conception is inconsistent. This prospective study aimed to explore underlying causal relationships between ovarian stimulation, in-vitro procedures, subfertility components and child cognition and behaviour. Participants were singletons born to subfertile couples after ovarian stimulation IVF (n = 63), modified natural cycle IVF (n = 53), natural conception (n = 79) and singletons born to fertile couples (reference group) (n = 98). At 4 years, cognition (Kaufmann-ABC-II; total IQ) and behaviour (Child Behavior Checklist; total problem T-score) were assessed. Causal inference search algorithms and structural equation modelling was applied to unravel causal mechanisms. Most children had typical cognitive and behavioural scores. No underlying causal effect was found between ovarian stimulation and the in-vitro procedure and outcome. Direct negative causal effects were found between severity of subfertility (time to pregnancy) and cognition and presence of subfertility and behaviour. Maternal age and maternal education acted as confounders. The study concludes that no causal effects were found between ovarian stimulation or in-vitro procedures and cognition and behaviour in childrenaged 4 years born to subfertile couples. Subfertility, especially severe subfertility, however, was associated with worse cognition and behaviour.
European IVF-Monitoring Consortium (EIM)European Society of Human Reproduction and Embryology Assisted reproductive technology in Europe, 2011: results generated from European registers by ESHRE.
). These perinatal adversities, in turn, are associated with neurodevelopmental disorders, such as attention deficit hyperactive disorder and learning disabilities (
Are fetal growth impairment and preterm birth causally related to child attention problems and ADHD? Evidence from a comparison between high-income and middle-income cohorts.
). Yet, it seems that assisted reproduction techniques are not associated with adverse cognitive and behavioural development during the first postnatal years (
). Although reassuring, it does not preclude an association with impaired development at later age. Although only subtle signs or even no symptoms are present at early childhood, children may still grow into neurodevelopmental deficits at older age, as it may take time for developmental disorders to emerge (
). In addition, over time, genetic effects on intelligence and behaviour may get increasingly expressed owing to the long-lasting interaction with social conditions (
The increased risk of complication observed in singleton pregnancies resulting from in-vitro fertilization (IVF) does not seem to be related to the IVF method itself.
Results of long-term studies on cognitive and behavioural development in children born via assisted reproduction techniques vary, partly owing to difficulties in distinguishing relationships between assisted reproduction techniques and underlying characteristics of subfertility, parents and child. Other factors that may explain the diversity in study outcomes may be the age at which children were studied (varying from 5–26 years), total sample sizes (varying from 69–45.557 children), sex ratios (the proportion of females varying from 37.5–52%) and the instruments used to assess outcome (among others the Teacher's Report Form, Stanford-Binet Intelligence Scale, Revised Amsterdam Child Intelligence Test, Iowa Tests of Basic Skills/Educational Development, Wechsler Intelligence Scale for Children-Revised). As a consequence, cognitive and behavioural outcome of children born via assisted reproduction techniques has been reported as similar (
Neuromotor development and mental health at 5.5 years of age of singletons born at term after intracytoplasmatic sperm injection ICSI: results of a prospective controlled single-blinded study in Germany.
Leiden Artificial Reproductive Techniques Follow-up Project Cognitive development of singletons born after intracytoplasmic sperm injection compared with in vitro fertilization and natural conception.
To investigate the influence of specific factors involving assisted conception on developmental outcome, we composed the Groningen assisted reproduction technique cohort. It consists of three groups of singletons born to subfertile couples: children born after conventional ovarian stimulation with IVF and intracytoplasmic sperm injection (ICSI), children born after modified natural cycle IVF–ICSI (MNC–IVF), and children born after natural conception (Sub–NC) (
). We previously reported on neurodevelopmental and cardiometabolic outcome of the Groningen assisted reproduction technique cohort until 4 years of age. Developmental outcome of the groups was similar up to the age of 2 years (
). A negative association was found between the duration of subfertility (TTP) and neurological condition at 2 and 4 years, suggesting that the severity of subfertility rather than its presence or the assisted reproduction technique components affect neurological outcome (
). Cognitive and behavioural development within the Groningen assisted reproduction technique cohort was studied up to the age of 2 years. We reported no differences in cognitive and behavioural outcome between the three study groups. Children of the three subfertile study groups, however, had higher scores on anxious and depressed behaviour than the children from a reference group born to fertile parents (
In a previous study at 4 years, we applied a causal inference approach to evaluate factors affecting anthropometrics and cardiovascular health. The study indicated direct positive effects of children born after OS–IVF and not of children born after MNC–IVF on outcomes, suggesting that ovarian stimulation was involved in worse cardiometabolic health (
Unlike traditional statistics, a causal inference approach is able to unravel underlying causal mechanisms and distinguish between confounders and intermediate effects. This makes it an appropriate tool to explore relationships between assisted reproduduction techniques and underlying characteristics of subfertility, parents and child. The primary aim of this study is to explore underlying causal relationships between ovarian stimulation, the in-vitro procedure and the combination of both and cognition and behaviour in 4-year-old singletons. Second, we aim to explore the underlying causal relationships between two aspects of subfertility and cognition and behaviour. We addressed the contribution of the presence of a history of subfertility by including a reference group of 4-year-old singletons born to fertile parents, and the duration of subfertility in terms of TTP, as a proxy for the severity of subfertility. We have chosen cognitive outcome as our primary outcome parameter, as cognition has a stronger neurobiological basis than behaviour (
Couples who achieved a singleton pregnancy after IVF–ICSI with a term date between March 2005 and December 2006 were recruited at the Department of Reproductive Medicine of the University Medical Center Groningen (UMCG) (
). This resulted in two groups: OS–IVF and MNC–IVF, in which medication use was minimal and in which the follicle that develops naturally to dominance is used for assisted conception (for details see
). Couples who were treated with donated or cryopreserved oocytes or embryos to achieve pregnancy were excluded. A third group (Sub–NC) was formed by naturally conceived children born to subfertile couples. These couples had tried to conceive for at least 1 year and eventually conceived naturally while on the waiting list for fertility evaluation or treatment at the UMCG. For details of the Groningen assisted reproduction technique cohort study see Figure 1 (
For the present study, a new retrospective reference group was recruited between December 2009 and February 2012 at six child-welfare centres in and around Groningen. All parents of 4-year-old singletons who visited the child welfare clinic for routine general health care were invited to participate. Children of couples who had tried to achieve pregnancy for more than 1 year or achieved pregnancy by any form of assisted conception were excluded.
Setting
Prenatal, perinatal and demographic information was gathered by means of standardized charts during the first follow-up assessment at about 2 weeks after term (
). Information on the causes and treatment of infertility was retrieved from medical records. The causes of subfertility were classified as ‘tubal pathology’ in case of abnormalities of the fallopian tubes, as ‘male factor’ in case of male infertility, as ‘other causes’ in case of endometriosis, cervical factor or hormonal cause and as ‘unknown cause’ in case of lack of a specific cause for subfertility. The TTP was defined as the interval between the start of timed unprotected intercourse or a previous pregnancy and conception, recorded in years and months and finally converted into decimal years. Note that, in case of spontaneous abortion, a new onset of TTP was defined. This explains why some participating subfertile couples in the present study had a TTP of less than 1 year (Table 1). In the fertile reference group, TTP was not recorded in months but in one-half-year periods (<0.5 year or 0.5–1.0 year).
Table 1Characteristics of participating parents and children.
Birth weight for gestational age is less than −2 SD compared with the Dutch reference population (Dutch reference tables, perinatal Registration Netherlands).
ICSI, intracytoplasmic sperm injection; MNC–IVF, infants born after modified natural cycle IVF or ICSI; OS–IVF: infants born after ovarian hyperstimulation IVF or ICSI; Sub-NC, naturally conceived infants born to subfertile parents. In the subfertile group COH–IVF, MNC–IVF and Sub–NC groups are taken together. TTP, time to pregnancy.
Mann–Whitney U tests, Student's t-test or Fisher's exact tests were carried out to investigate differences between groups.
* P < 0.05;
** P < 0.01;
*** P < 0.001.
a Birth weight for gestational age is less than −2 SD compared with the Dutch reference population (Dutch reference tables, perinatal Registration Netherlands).
b Signs of fetal distress denoted by meconium stained amniotic fluid, cardiotocographic signs, acidosis or all three.
c University education or vocational colleges.
d TTP was recorded in one-half-years. In case of a spontaneous abortion, the onset of TTP restarted, therefore TTP may be shorter than 1 year.
e Overall missing data: gestational age (n = 1), preterm (n = 1), birthweight (n = 4), low birthweight (n = 3) , small for gestational age (n = 1) , Apgar score 5 min <7 (n = 31), maternal age at conception (n = 1), paternal age at conception (n = 8), education level father (n = 8), time to pregnancy (n = 1).
Follow-up assessments were carried out by trained assessors supervised by a neurodevelopmental expert (MH-A), whowere blind to the mode of conception. Blinding was not possible for the reference group, as this group was recruited separately from the subfertile groups.
Measurements
Cognitive development was evaluated using the Kaufman Assessment Battery for Children, second edition (K-ABC-II,
). This standardized instrument measures in about 45 min cognitive and processing abilities in children aged between 3 and 18 years. Outcome is expressed in a total intelligence quotient (IQ) score and four IQ scale scores. In the present paper, only the total IQ score is used. Raw test scores are normalized into global scores (mean: 100; SD: 15). Reliability and validity of the K-ABC-II are good (
). The CBCL is a parental questionnaire to identify emotional and behavioural problems in children aged 1.5 years to 5 years, classified into problem scales. The sum of all questions results in the total problem scale score. Raw test scores are normalized into T-scores (mean: 50, SD: 10). Higher T-scores represent more problematic behaviour. The reliability and validity of the CBCL are good (
Fisher's exact tests, Mann–Whitney U-tests and Student's t-tests were carried out to investigate differences between groups using the IBM Statistical Package for the Social Sciences (SPSS), version 20 (IBM Corp., USA). P < 5% were considered statistically significant for differences between the subfertile group and the reference group. Bonferroni correction was applied for the comparisons between the assisted reproduciton technqiue study group, meaning that P < 1.7% was considered statistically significant.
Two separate sets of explorative analyses, each consisting of a dozen of explorations, using causal inference search algorithms were carried out. The first focused on assisted reproduction technique treatment effects (ovarian stimulation; in-vitro procedure; combination of both) on cognitive and behavioural outcome on the basis of the children of the subfertile couples of the Groningen assisted reproduction technique cohort. The second focused on subfertility effects (presence and severity of subfertility, defined as TTP >1 year and TTP as a continuous variable, recorded in one-half years, respectively) on outcome at 4 years on the basis of the assisted reproduction technique cohort children and the reference children born to fertile couples. The modelling takes into account causal relations of the ‘focus’ factors and other parental and child aspects.
Causal inference search algorithms result in causal models that are found to be compatible with the data, based on the theory of causal graphs. A causal model consists of a statistical model and a graph, in which the latter describes the causal relation between the variables and consists of vertices (the variables), connected by edges, which can be oriented by arrows. An oriented edge represents a direct causal effect between two connecting vertices. Conservative Peter-Clark (CPC), Greedy Equivalency Search (GES) and Conservative Fast Causal Inference (CFCI) algorithms were applied. For a concise overview about the terminology and theory of causal models and causal inference search algorithms, including references, please see refer to Supplementary material.
The resulting graphs per se represent an observationally equivalent class of causal models, i.e., other graphs from that class are equally likely to have generated the data. Model fit was calculated and compared between the found classes of models, by choosing one directed acyclic graph, representing such a class. We used the chi-square test, chi-squared test/degrees of freedom ratio and the Bayesian information criterion to compare models. A ratio of less than 2 suggests a good model fit. Additionally, structural equation modelling was applied to a graph that represents one of the better fitted classes of models to estimate effect size.
By comparing the various graphs resulting from the performed search algorithms, the most prominent and consistent direct effects could be distinguished. Additionally, effects that are consistently not found can be considered an indication of the absence of a causal effect. The explorative analyses were conducted using the freeware program TETRAD, version 403.10-6 (
The Medical Ethical Commission of the UMCG approved the study design on 7 July 2009 (reference number M09.074824). Parents provided written informed consent for study participation of their child.
Results
Participation and demographic characteristics
At the age of 4 years, 20 (9.3%) children were lost to follow-up (Figure 1). Intra-group characteristics for participants and non-participants were similar, except for a longer TTP in participating MNC-parents compared with non-participating MNC-parents (P = 0.021). Demographic characteristics of the three groups are listed in Table 1.
Parents of 215 eligible singleton reference children were invited to participate in the present follow-up study. Ninety-eight (46%) parents allowed their child to participate. Demographic characteristics such as sex, gestational age, first-born, maternal age and parental educational levels were similar for participants and non-participants (data not provided). Demographic characteristics of participants of the subfertile and reference groups are listed in Table 1.
Effect of ovarian stimulation and in-vitro procedure
Most of the children in the three study groups (n = 189 [98%]) had an IQ within the normal range (scores between 85 and 115), except for two OS–IVF children (IQ: 82 and 79) and two MNC–IVF children (IQ: 79 and 77). Most children in the assisted reproduction technique study groups (n = 186 [97%]) also had total problem T-scores within the normal range, except for two OS–IVF children and four Sub–NC children. The mean IQ and total problem T-scores of the three assisted reproduciton technique study groups are presented in Table 2. Univariable analyses indicated no differences between the assisted reproduction technique study groups in mean IQ and total problem T-scores (Table 1) (Bonferroni correction P < 0.017). Moreover, no differences were found between the groups on the secondary outcome parameters mean IQ subscale scores and internalizing and externalizing behaviour. The same accounts for other behavioural parameters measured with the CBCL (emotionally reactive, anxious/depressed, somatic complaints, withdrawn behaviour, sleep problems, attention problems and aggressive behaviour; data not shown). Also, ICSI and specific causes of subfertility were not associated with IQ and behaviour (data not shown).
Table 2Group differences in IQ and CBCL scores of participating children.
Significant mean differences in outcomes: P < 0.05 for differences between the subfertile group and the fertile reference group and P < 0.017 for differences between the ART study groups (Bonferroni correction applied).
Significant mean differences in outcomes: P < 0.05 for differences between the subfertile group and the fertile reference group and P < 0.017 for differences between the ART study groups (Bonferroni correction applied).
Sequential IQ
95.9(1.31)
1.2 [−3.0; 5.4]
0.5 [−3.5; 4.5]
99.6(1.18)
−3.2 [−6.0; −0.3]
Simultaneous IQ
115.5(1.59)
−1.5 [−6.3; 3.4]
−2.4 [−7.0; 2.2]
117.8(1.46)
−3.5 [−6.9; −0.0]
Learning IQ
98.3(1.26)
−2.6 [−6.7; 1.5]
−1.5 [−5.4; 2.4]
99.6(1.02)
−2.4 [−5.2; 0.3]
Knowledge IQ
112.7(1.12)
−6.0 [−10.3; −1.8]
−4.4 [−8.5; −0.4]
111.3(1.26)
−1.7 [−4.7; 1.4]
Behavioral development
Sub–NC
MNC–IVF
OS–IVF
Fertile reference group
Subfertile group
(n = 78)
(n = 52)
(n = 62)
(n = 96)
(n = 192)
CBCL total problem T-score
47.8(1.06)
−1.6 [−4.8; 1.5]
−0.5 [−3.4; 2.4]
44.4(0.99)
2.2 [−0.0; 4.5]
CBCL internalizing T-score
48.0(1.14)
−1.3 [−4.8; 2.3]
−3.0 [−6.3; 0.4]
45.0(1.04)
1.8 [−0.7; 4.2]
CBCL externalizing T-score
48.7(1.01)
−1.2 [−4.2; 1.9]
−0.5 [−3.4; 2.4]
46.2(0.95)
2.0 [−0.2; 4.2]
MNC–IVF, infants born after modified natural cycle IVF or ICSI; OS–IVF, infants born after ovarian hyperstimulation IVF or ICSI; Sub-NC, naturally conceived infants born to subfertile parents. In the subfertile group OS–IVF, MNC–IVF and Sub–NC groups are taken together. CBCL, Child Behaviour Checklist.
Two children of the subfertile group (2 Sub–NC) and one reference child had missing data on the K-ABC-II. Three children of the subfertile group (1 OS–IVF, 1 MNC–IVF and 1 Sub–NC) and two reference children had missing data on the CBCL questionnaire.
* Significant mean differences in outcomes: P < 0.05 for differences between the subfertile group and the fertile reference group and P < 0.017 for differences between the ART study groups (Bonferroni correction applied).
The causal graph that had the best model fit was the causal model found as a result of running the GES algorithm with a penalty discount of 0.5 (χ2 = 98.5, df = 94, P = 0.356, chi-squares/df = 1.048, suggesting a good model fit). This model indicated the absence of direct effects of OS–IVF or IVF on cognitive or behavioural outcome given the other variables in the model. A direct effect of OS–IVF or IVF was also absent in causal graphs resulting from other searches (data not provided).
Presence and severity of subfertility
Most of the children of the subfertile group (n = 189 [98%]) had an IQ within the normal range, except for the four children mentioned above. None of the reference children had IQ scores below average (n = 97 [100%]). Most of the children in both groups (subfertile group n = 186 [97%]; reference children n = 92 [96%]) also had total problem T-scores within the normal range, except for six children of the subfertile group and four reference children. The mean IQ and total problem T-scores of the subfertile group and the reference group are listed in Table 2. Univariable analyses indicated that children of the subfertile group had significantly lower mean total IQ scores compared with children of the reference group (P < 0.05 (Table 2). No differences were found between the groups in problem T-scores. Moreover, no differences were found between the groups on the secondary outcome parameters mean IQ subscale scores and internalizing and externalizing behaviour. The same accounts for other behavioural parameters measured with the CBCL (data not shown).
The causal graph that presented the class of found models with the best model fit as a result of the causal effect search algorithms is represented in Figure 2 (chi-squared = 84.7, df = 72, P = 0.145, chi-square/df = 1.176, suggesting a good model fit). It indicated that the presence of subfertility did not have a direct or indirect effect on the total IQ score. Similar results were found in resulting causal graphs from other searches (data not provided). The causal graph suggests that the total IQ score was directly affected by two factors: TTP and maternal smoking during pregnancy. A longer TTP was associated directly with a lower total IQ score. This effect was confounded by high maternal education through various paths: one path ran from maternal education via maternal and paternal age, via TTP to total IQ, whereas another path ran from maternal education via smoking during pregnancy to total IQ. The other direct effect on the child's total IQ score was the negative effect of maternal smoking during pregnancy, an effect that again was confounded by maternal education. Taking this complexity of the pathways into account and correcting for maternal education level, the direct, unconfounded effect of TTP on total IQ score is estimated at −0.712 (Figure 2).
Figure 2The underlying causal mechanism as found in the subfertile (OS–IVF, MNC–IVF and Sub–NC groups pooled) and fertile reference group regarding presence and severity of subfertility and cognitive and behavioural development. The causal graph representing the model with the best model fit of the causal inference search algorithms (Greedy Equivalency Search algorithm, penalty discount: 0.5; model fit: chi-squared = 84.7, df = 72, P = 0.145, chi-square/df = 1.176 and Bayesian information criterion score = −317.0). Each arrow in the graph is accompanied by two numbers, reflecting the size of the causal effect per unit change. The first number is the estimated regression weight and the second number between brackets is its corresponding standard error. A positive estimated regression weight in the graph means that higher values of the variable are associated with an increase in the value of the variable to which the arrow is pointing. Likewise, a negative estimated regression weight in the graph means that higher values of the variable are associated with a decrease in the value of the variable to which the arrow is pointing. Time to pregnancy was recorded in one-half-years. High educational level indicates university education or vocational colleges.
The causal graph suggests that the presence of subfertility had a direct adverse effect on the total problem T-score of behaviour (Figure 2). As such, subfertility status acted as mediator on three different paths running from high maternal educational level on total problem T-score. These paths ran either via maternal age at child conception, or via TTP, or via a combination of both, including an effect of paternal age at child conception (Figure 2). It also indicates that maternal age directly affected the total problem T-score: a higher maternal age was associated with less child behavioural problems. Also in this path, maternal age mediated the effect of high maternal educational level on total problem T-score. Subsequently, maternal age at conception and high maternal educational level are considered to be confounders for the effect of subfertility status on the total problem T-score. The unconfounded, direct effect of the presence of subfertility on the total problem T-score is estimated as follows: conditionally on maternal age at conception, high maternal educational level, TTP, or both, the direct effect of subfertility status is estimated to be 3.003 for less optimal behaviour.
The causal graph also suggests a relationship between the outcome parameters, cognition and behaviour: an increase in total IQ score of one point was associated with a decrease in total problem T-score with 0.120 points, indicating better behaviour in case of a more optimal cognitive development (Figure 2). According to the causal implications, TTP has an indirect (mediated by subfertility status and total IQ score) effect on more problematic behaviour (Figure 2).
Discussion
The present study indicates that in 4-year-old singletons born to subfertile couples direct or indirect causal effects between ovarian stimulation or the in-vitro procedure on the one hand and cognitive and behavioural outcome on the other hand are absent. A direct causal relationship, however, was found between the severity of subfertility (TTP) and cognition at age 4 years. Also, a direct negative causal relationship between the presence of subfertility and behaviour was found. Both effects were confounded by maternal age at child conception and maternal educational level. The TTP had a direct negative effect on the child's IQ score, whereas the presence of subfertility had a direct adverse effect on behaviour. Given that cognition and behaviour are directly related to one another (with behaviour being influenced by cognition), our results suggest that the severity of subfertility has an indirect, through subfertility status as well as through IQ, effect on the child's behavioural scores, i.e., was associated with more behavioural problems. The association between a higher IQ and less behavioural problems corresponds to the findings of
that a higher IQ may protect a child to develop behavioural problems in response to life stress events.
Our findings strengthen the impression that aspects of subfertility rather than components of assisted reproduction techniques are involved in determining cognitive and behavioural development of children born after IVF–ICSI. Our statistical approach also revealed important confounders for these effects, especially maternal age and education. In turn, the effect of the presence of subfertility on behavioural outcome was confounded by TTP. Our results imply that suffering from subfertility per se, especially from more severe subfertility, which in turn is affected by a higher maternal age and higher educational level, may negatively affect the child's cognitive and behavioural development. More studies, however, are needed to test the effects.
The fact that already known and generally expected mechanisms were also detected (Figure 2), underlines the general robustness and validity of our results. The graph with the best model fit of the GES search algorithm and also other graphs resulting from the GES searches with a good model fit showed similar mechanisms concerning the above-mentioned effects, indicating a certain consistency of the associaitons found between the variables. Similar effects of maternal age, maternal educational level and TTP on the outcomes were also found with the CPC search algorithm (model fit indices for the CPC-result with the best model fit: alpha value 0.15; model fit: chi-square = 72.1, df = 68, P = 0.344, chi-square/df = 1.060). The fact that similar mechanisms are revealed by two different algorithms in search approach and data assumptions, underline the validity of the found mechanisms. One might have expected to find direct effects of birth weight and gestational age on cognitive and behavioural development; however, our analyses did not result in such effects (Figure 2).
Birth weight is indeed associated with the total problem T-score via maternal age at conception, but this association disappeared after correction for maternal age. This is in line with our expectations: our groups contain few preterm and low birth weight infants. Therefore, it seems more probable that our analyses resulted in effects of subfertility and maternal factors on developmental outcome, rather than in effects of birth weight and gestational age.
The participating children in both the subfertile and the reference group had a relatively high IQ, which may be explained by the relatively large proportion of highly educated participating parents. The well-known positive association of maternal age and maternal education with child cognitive and behavioural development (
) is also present in our material. The graphs illustrate the complexity of the relationships: on the one hand higher maternal age and better maternal education are associated with higher IQ-scores and lower behavioural problem scores, but on the other hand they are also associated with subfertility and prolonged TTP which, in turn, are associated with lower IQ-scores and higher behavioural problem scores.
To the best of our knowledge, this is the first study in the context of neurodevelopment in children born via assisted reproduction techniques using causal inference search algorithms combined with structural equation modelling. It allowed us to distinguish between direct and indirect causal effects and to detect and correctly adjust for confounders at the same time. Our findings that the severity of subfertility plays a role in cognitive development is in line with
, who reported that longer TTP may be associated with a delay in achieving certain milestones, in particular those involved in cognitive and language development. In another study by
, however, a subfertility effect on child behaviour could not be demonstrated.
To intelligibly answer our research questions, we conducted two separate sets of explorative analyses: one focusing on the effects of assisted reproduction technqiue treatment, the other focusing on the effects of presence and severity of subfertility on developmental outcome. The first exploration can only be translated to its particular (smaller) subfertile population and its findings do not extend to the different population that was used for the second exploration. In the latter analysis a reference group was added, allowing for the in concert evaluation of the effect of the presence of subfertility and that of its severity.
The search algorithm strategy was an important strength of the study. Additional strengths of the study are the attrition rate of 9.3%, blinding of assessors to the mode of conception and its prospective design with three subfertile groups.
It must be realized that the statistical tools used in our study are explorative in nature and especially serve as indications for new research hypotheses, meaning that our results need to be interpreted with appropriate caution. This caution, however, is not restricted to search algorithms alone: applying multiple sets of multivariable regression analyses is in fact also explorative in nature. Because of the explorative nature of our analyses a post-hoc power calculation is irrelevant. Previously group sizes of the Groningen assisted reproduction technique cohort were based on neurological outcome at 18 months (
). As a general rule for a structural equation model, 10 or more observations for each variable may be considered reliable, a criterion that we have met.
A more general limitation of the study is related to the reference group. The group was recruited retrospectively, and consequently, we were not blinded to group status of the children. In addition, only about one-half of the eligible families agreed to participate. This is a well-known phenomenon in the field of child development, also in the field of assisted reproduction techniques (among others
Leiden Artificial Reproductive Techniques Follow-up Project Cognitive development of singletons born after intracytoplasmic sperm injection compared with in vitro fertilization and natural conception.
Neuromotor development and mental health at 5.5 years of age of singletons born at term after intracytoplasmatic sperm injection ICSI: results of a prospective controlled single-blinded study in Germany.
but unfortunately rarely reported. Previous studies reported that parents who are concerned about their child's health in general are more likely to participate in developmental studies (
). This means, however, that the associations reported in the present study may have been underestimated. The relatively low participation rate in the reference group may be an indicator for selection bias. An additional limitation is that we used American norms instead of Dutch norms. It is difficult to know how this may have confounded our results. We assume, however, that this effect is minor compared with the effect of parental education, which we took into account.
In conclusion, our study implies that suffering from subfertility per se, and especially from more severe subfertility, which by itself is affected by higher age and high educational level of the mother, is associated with less favourable cognitive and behavioural outcome of the child. Long-term follow-up of development and growth of children born from subfertile couples, in particular when assisted reproduction technqiues is applied, remains important, given the steadily increasing prevalence of subfertility in modern society.
Acknowledgement
We thank participating parents and children for their cooperation and enthusiasm during the assessments; Arend F Bos, MD, PhD, for his help in initiating the study and Maaike Haadsma, MD, PhD, for her help in including the participants; Rosan Aapkes, MD, Elise Bennink, MSc, Hanneke Geut, MD, Marjolein Jongbloed-Pereboom, MSc and Bertine de Vries, MSc, for support in collecting and organizing the data; Michiel Schrier, MSc, Linze Dijkstra BSc, and Loes de Weerd for technical assistance. The study was financially supported by the University Medical Center Groningen, grant number: 754510, the Junior Scientific Masterclass, the Postgraduate School Behavioural and Cognitive Neurosciences and the Cornelia Foundation, Groningen, The Netherlands.
Appendix. Supplementary material
The following is the supplementary data to this article:
An example of a directed acyclic graph (DAG) representing possible hypothetical causal relationships between the variables Maternal educational level, Smoking during pregnancy, Time to pregnancy, Child IQ score and Child behavioural problems.
Assuming the true underlying mechanism to be depicted in Figure S1, Child behavioural problems too is independent of Maternal educational level, Smoking during pregnancy and Time to pregnancy when conditioned on Child IQ score. Hence, the algorithm concludes that the three corresponding edges in Figure S2A can be removed, resulting in the undirected graph in Figure S2B. As Child IQ score is conditionally independent from Maternal educational level when simultaneously conditioning on both variables Smoking during pregnancy and Time to pregnancy according to the assumingly true underlying mechanism from Figure 1, the algorithm also knows to remove the edge between Child IQ score and Maternal educational level. A similar argument leads to the removal of the edge between Smoking during pregnancy and Time to pregnancy. All other edges in the graph remain, as those dependencies between the variables cannot be eliminated by conditioning on any subset of the other variables. These steps result in the so-called undirected dependency graph (Figure S2C), a graph similar to the graph in Figure I, but without direction, represented by the arrows.
The partially directed graph of the five known variables, representing the result of the search algorithms. The open circles at the end of the edges denote the fact that there could either be an arrow or an empty space (but we could not determine which, based on the data). Definite non-colliders are denoted in the graph by the drawn bars over or under a variable and two open circle ends. The definite non-collider Maternal educational level as represented by the pattern:
Smoking during pregnancy o − o Maternal educational level o − o Time to pregnancy represents one of the three following situations:
Smoking during pregnancy o→ Maternal educational level −o Time to pregnancy
Smoking during pregnancy o − Maternal educational level ←o Time to pregnancy
Smoking during pregnancy o − Maternal educational level −o Time to pregnancy
Note that the latter is indeed the one from the assumingly true underlying mechanism as depicted in Figure S1. Similarly, all other open circle ends of the definite non-colliders could be replaced, as long as no new colliders are introduced.
Neuromotor development and mental health at 5.5 years of age of singletons born at term after intracytoplasmatic sperm injection ICSI: results of a prospective controlled single-blinded study in Germany.
Are fetal growth impairment and preterm birth causally related to child attention problems and ADHD? Evidence from a comparison between high-income and middle-income cohorts.
The increased risk of complication observed in singleton pregnancies resulting from in-vitro fertilization (IVF) does not seem to be related to the IVF method itself.
Pamela Schendelaar is a medical doctor and researcher at the Institute of Developmental Neurology, Department of Paediatrics, The University Medical Center Groningen. In February 2015, she obtained her PhD at the University of Groningen with the thesis ‘Offspring of subfertile couples: neurodevelopmental outcome at preschool age’.
Article info
Publication history
Published online: September 09, 2016
Accepted:
August 23,
2016
Received in revised form:
August 21,
2016
Received:
February 13,
2016
Declaration: The authors report no financial or commercial conflicts of interest.
30504-1&id=rbmo1606-fig-0001.jpg){kind=link}
30504-1&id=rbmo1606-fig-0002.jpg){kind=link}