Pre-pregnancy body mass index, weight change during pregnancy, and risk of intellectual disability in children

Authors


Correspondence: Dr J Mann, University of South Carolina School of Medicine, 3209 Colonial Drive, Columbia, SC 29203, USA. Email joshua.mann@sc.edu

Abstract

Objective

This study investigated pre-pregnancy body mass index (BMI) and weight change in pregnancy as potential risk factors for intellectual disability (ID) in children.

Design

Retrospective cohort study.

Setting

South Carolina, USA.

Population

A total of 78 675 mother–child pairs, insured by the South Carolina Medicaid programme, born in the period 2004–2007.

Methods

We analysed South Carolina Medicaid data, linked to data from both the South Carolina Department of Education (DOE) and the South Carolina Department of Disabilities and Special Needs (DDSN). Maternal pre-pregnancy BMI and weight change during pregnancy were obtained from birth certificates. ID cases were identified from the three sources listed above. We used generalised estimating equation logistic regression models to model the odds of ID in children.

Main outcome measures

Identified as having ID in special education, DDSN, or Medicaid billing records.

Results

The risk of ID was greater in children of women with pre-pregnancy obesity, and the risk was greatest in children born to women with morbid obesity (OR 1.52, 95% CI 1.30–1.77 for ID of any severity; OR 1.73, 95% CI 1.23–2.45 for severe ID). Gestational weight change (gain or loss) was not significantly associated with odds of ID.

Conclusions

Pre-pregnancy obesity may be a modifiable risk factor for ID in children, although further study is needed to evaluate whether the association meets criteria for causation.

Introduction

Intellectual disability (ID) is a developmental disability characterised by ‘significant limitations both in intellectual functioning and in adaptive behavior, which covers a many everyday social and practical living skills’, with onset before 18 years of age.[1] Limitation in intellectual ability corresponds to an intelligence quotient (IQ) of <70, described as mild (IQ 55–70), moderate (IQ 40–55), or severe to profound (IQ of <40).[2] The prevalence of ID in the general population is estimated to be approximately 1%, and among those with ID 85% are characterised as having mild ID.[3, 4]

The causes of ID are extensive, and include disorders that substantially affect brain development and functioning. According to the timing of the insult, causes can be classified as prenatal, perinatal, and postnatal, or acquired during childhood. Causes of ID can also be classified as genetic or non-genetic. Genetic causes include chromosomal abnormalities (e.g. trisomy 21 and other trisomies), inherited genetic traits (e.g. fragile X syndrome), and single gene disorders (e.g. Prader–Willi syndrome).[5, 6] Important non-genetic causes include brain trauma, congenital infection, meningitis and encephalitis, environmental toxins or teratogens, radiation, and extreme childhood deprivation.[7] The cause of ID in particular cases is often unknown,[6, 8] and the proportion of cases with unknown cause is generally highest in mild ID.[9] Increasing maternal age, decreasing maternal education, multiple gestation, male sex, preterm birth, maternal infection, and pre-eclampsia have been identified as risk factors for ID of unknown cause.[9, 10]

Obesity (body mass index, BMI > 30) and overweight (BMI 25–29.9) are growing problems in the USA. According to the most recent National Health and Nutrition Examination Survey (NHANES) data for the year 2007–2008, 33.8% of American adults are obese and another 34.2% are overweight; 5.7% of American adults have morbid obesity (BMI > 40). The same data also revealed that over one-third of women aged 20–39 years in the USA are obese.[11, 12] Even women with normal pre-pregnancy BMI can be at risk of excessive weight gain during pregnancy. Some weight gain is recommended for all women (11.34–15.88 kg for women of normal weight at baseline, and less for women who are overweight or obese, according to the 2009 Institute of Medicine, IOM, guidelines).[13] However, approximately 40–60% of women experience gestational weight gain that exceeds the current IOM recommendations.[14, 15]

Pre-pregnancy obesity and excessive weight gain is associated with a wide range of adverse pregnancy outcomes in the mother and child, such as increased risk of pre-eclampsia, gestational diabetes, stillbirth, perinatal death, fetal macrosomia, and delivery by caesarean section.[16-19] Furthermore, maternal pre-pregnancy obesity is associated with an increased risk of having a child with a range of structural birth defects, including neural tube defects, spina bifida, and cardiovascular anomalies.[20-25]

Few studies have investigated the relationship of maternal pre-pregnancy obesity with children's ID or other neurodevelopmental disabilities. Dodds et al.[26] reported that a maternal pre-pregnancy weight of 90 kg (approximately 200 lb) or greater, and a pregnancy weight gain of 18 kg (approximately 40 lb) or greater were each significantly associated with a risk of autism.

Heikura et al.[27] followed two cohorts of Finnish children born in 1966 (n = 12 058) and 1985–1986 (n = 9432). Pre-pregnancy maternal obesity was found to be associated with an increased risk of child ID (OR 2.9) only in the 1985–1986 cohort. Low pre-pregnancy BMI (<20) was associated with ID (OR 2.1) in the 1966 birth cohort, but obesity was not significant.[27] The discrepancy in results between the two birth cohorts renders the study's findings inconclusive.

Another study in the USA found that children of women who had pre-pregnancy obesity had significantly lower general cognitive scores (β = −4.7, P = 0.001) and nonverbal development scores (β = −5.6, P = 0.003), as compared with children of mothers of normal weight.[28] No association was present between maternal weight gain during pregnancy and any of the cognitive (IQ), verbal, or nonverbal scores. The study was limited by its restriction to African American children from low-income families and a sample size of only 355.

Analysing data from two cohorts of children, one British and one Dutch, totalling approximately 7500 children, Brion et al.[29] reported a significant association between maternal overweight/obesity and reduced odds of being in a higher quintile for IQ (adjusted OR 0.84, P = 0.03). However, on the basis of a lack of consistent associations between maternal overweight/obesity and measures of specific aspects of cognitive ability, the authors concluded that the results were probably the result of unmeasured confounding factors.

Most recently, researchers in California conducted a case–control study that included 517 children with autism spectrum disorders, 172 children with another developmental disability, and 315 controls.[30] They found that maternal obesity was significantly associated with both autism spectrum disorders (OR 1.61) and developmental disability (OR 2.35). This study did not assess the impact of weight gain during pregnancy.

These studies indicate that there may be an association between maternal weight and ID in children, but additional studies are needed, especially well-powered studies that assess both pre-pregnancy BMI and weight change during pregnancy.

Methods

We conducted a retrospective cohort study to investigate pre-pregnancy BMI and maternal weight gain during pregnancy as potential risk factors for ID in their children. Our a priori hypotheses were that pre-pregnancy obesity and high levels of weight gain would be associated with increased odds of ID.

We obtained data for children from South Carolina Medicaid billing records and birth certificates for pregnancies and births that occurred during the period 2004–2007 (Medicaid purchases health services for low-income pregnant women and children), linked to data from the South Carolina Department of Education (DOE) and Department of Disabilities and Special Needs (DDSN). These linked data are available through the South Carolina Office of Research and Statistics (ORS), which maintains administrative data for state agencies at the individual level. At first contact with a state entity (often at birth, when the birth certificate is completed), each individual who has contact with one or more agencies of the State government is assigned a unique identification number. This universal cross-agency identification number permits linkage of individual data across different types of services. It also permits linkage of data for mothers and their children, through birth certificate records. For this study, a maternal match was identified for 94.1% of the babies who were part of the Medicaid birth cohort for the years of interest. The study protocol was reviewed by all of the agencies from whom data were requested and the University of South Carolina Institutional Review Board, which granted exempt status.

Case definition of intellectual disability

Children with ID were identified in the three data sources, updated to 31 December 2010. The first source was the Medicaid file that includes the International Classification of Diseases, Ninth Revision (ICD–9) diagnosis codes in clinical billing records. The ICD-9 codes for ID are 317 (mild), 318 (moderate, severe, or profound, depending on the specific fourth digit), and 319 (unspecified or unknown severity). The second source for identifying cases of ID was a data file from the DOE. Children in special education are placed in one of three categories: educable mentally handicapped (EMH, which corresponds to mild ID), trainable mentally handicapped (TMH, which corresponds to moderate to severe ID), and profoundly mentally handicapped (PMH, corresponding to profound ID). The final source for identifying cases of ID was enrollment in ID support services from DDSN. Eligibility for special education and DDSN ID services requires formal psychological assessment, including IQ testing. However, because of the young age of the cohort, we anticipated that the majority of cases would come from Medicaid or DDSN, rather than the school system, as children frequently do not enroll in public school until 5 years of age. Medicaid diagnoses of ID are based on the ICD-9 codes entered by healthcare providers (including psychologists, therapists, and physicians), without corroborating information about these cases. To be conservative and minimise the risk of wrongly identifying children as cases of ID, we defined a case as a child who was either enrolled in special education in a South Carolina public school, enrolled in DDSN services for ID, or diagnosed with ID in the Medicaid billing data on at least five different occasions. Children diagnosed with ID fewer than five times and not confirmed in DOE or DDSN data were completely excluded from the analyses.

We also created a variable that identified children who were diagnosed with moderate to profound ID in Medicaid billing data (ICD-9 code 318), were enrolled in TMH or PMH special education classes, or were receiving DDSN services (the last group was included because, according to personal communications with DDSN staff, a majority of DDSN clients have moderate to profound ID). We used this as a separate outcome in the modelling.

For cases not confirmed in DDSN or DOE records but with a Medicaid diagnosis of ID (any severity and severe), we performed sensitivity testing based on the number of times diagnosed and/or number of different providers making a diagnosis of ID. We examined multiple cut-offs, the most restrictive being at least 25 separate diagnoses of ID or being diagnosed by at least five different providers. For the sensitivity testing, we excluded children identified as having ID but not meeting the criteria for number of diagnoses or diagnosing clinicians.

Identification of independent variables and covariates

Pre-pregnancy BMI was calculated based on the pre-pregnancy height and weight measurements provided on the birth certificate: BMI (kg/m2) = (weight in kg)/(height in m)2.

Weight gain or loss was calculated as the difference between the mother's weight at delivery and the pre-pregnancy weight as reported on the birth certificate.

We controlled for a wide range of potential confounding factors using the birth certificate, Medicaid data, or both. Demographic information, obtained from birth certificates, included maternal age, race, ethnicity, and level of education. Information about maternal tobacco use, gonorrhea, chlamydia, syphilis, and intrapartum fever were also obtained from the birth certificate, as were the child's sex, gestational age (in completed weeks) at delivery, and birthweight. Women with hypertension and those with diabetes mellitus were identified based on a combination of ICD-9 codes and birth certificate information. Maternal epilepsy was identified using ICD-9 code 345 in the Medicaid data.

We also obtained information about whether each mother had ever been reported as having ID, so that we could control for the presence of this potential confounding factor. We searched the Medicaid billing data for mothers and identified all those who had ever been diagnosed with ID of any severity. We also identified those who had received DDSN services for ID, and among those who could be identified in South Carolina public school records (41.1%), we identified those who were ever enrolled in special education in EMH, TMH, or PMH placements. We then created a dichotomous indicator variable for maternal ID, which was included as a covariate in all of the regression models.

Exclusions

We excluded mothers with pre-pregnancy BMIs >51.31 or <17.6. We also excluded mothers with heights >178 cm or <146 cm. These outliers were suspected to result from data entry errors. The cut-off points used were the corresponding first and 99th percentiles for females aged 20 years and over from the NHANES for the US population in 2003–2006. The first and 99th percentiles were not published, so we obtained cut-off values via personal communication with staff from the Centers for Disease Control and Prevention (CDC)/National Center for Health Statistics.[31] We also excluded women who gained or lost more than 45 kg during pregnancy, out of concern that these extreme values may have represented data entry errors.

We excluded children who were not singleton births and those who were diagnosed with a known or likely cause of ID (whether ID was actually confirmed in the Medicaid, DOE, or DDSN records). The known/likely causes that were removed, based on birth certificate data, were trisomy 21, anencephaly, other brain anomalies, spina bifida, suspected chromosomal abnormalities with chromosomal analyses pending, and meconium aspiration syndrome. Medicaid diagnoses excluded were Prader–Willi syndrome, fragile X syndrome, fetal alcohol syndrome, congenital syphilis, congenital hypothyroidism, amino acid disorder, hydrocephalus, birth trauma, neonatal haemolytic disease, neonatal drug withdrawal, brain anomaly, traumatic brain injury, meningitis, encephalitis, congenital infections, neonatal infections, respiratory distress syndrome, birth asphyxia, and cardiac arrest. By excluding children with these known or likely causes of ID, we were able to evaluate maternal BMI and gestational weight gain or loss as potential risk factors for unexplained ID.

Intellectual disability cannot be diagnosed reliably in very young children. To ensure that children had a sufficient level of follow-up to be identified as having ID, we restricted analyses to children who met at least one of the following criteria:

  • enrollment in the South Carolina Medicaid programme until at least 3 years of age;
  • enrollment in the South Carolina public school system; or
  • identified as a case of ID in at least one of the three data sources.

Figure 1 displays the number of children removed from the analyses for each category of exclusion. After exclusions, 62.53% of the observations from the original cohort were used for analysis. Among them, 3.96% were cases of ID.

Figure 1.

Study exclusions.

Statistical modelling

Some of the women in our cohort gave birth to more than one baby during the study period. Results from multiple pregnancies in the same woman violate the assumption of independence of observations for logistic regression models. To account for such correlated data, we estimated the population-averaged logistic regression generalised estimating equations (GEE) models, which appropriately analyse longitudinal/repeated measures and other correlated data, even when the response is discrete.[32, 33] We specified an exchangeable structure (all observations have the same pairwise correlation) to model the correlation between the babies born to the same mother.

We anticipated that the assumption of a constant odds ratio across the range of some independent variables might not be reasonable. If so, the associations between the outcome and those predictors would not be linear in our logistic model. An approach that allows a nonlinear exposure–response association is to use fractional polynomials.[34] The fractional polynomials method gives a functional formula for the nonlinear term and produces useful exposure–response plots.[35-37] The functional form for an independent variable X is defined as inline image, with p1 and p2 taken from the set S = {−2, −1, −0.5, 0, 0.5, 1, 2, 3}, where X0 = log(X). The best functional form for X is the one with the highest likelihood.

Results

Characteristics of the cohort

The demographic and other characteristics of the 78 675 mother–child pairs included in our analyses are summarised in Table 1. There were 3113 children identified with ID. Only three cases of ID were identified in DDSN or school data, and all three were also diagnosed in the Medicaid file.

Table 1. Demographic and other characteristics (n = 78 675)
  No ID (n = 75 562) ID (n = 3113) P a
  1. For continuous variables, means (standard deviations) are reported, whereas for categorical variables, frequencies (percentages) are reported.

  2. a

    P values test the equivalence of means (Student's t-test) for continuous variables and equivalence of proportions (chi-square test) for categorical variables.

Infant characteristics
Sex
Male37 528 (49.67%)2141 (68.78%)<0.0001
Female38 034 (50.33%)972 (31.22%)
Gestational age (weeks)38.59 (1.52)38.30 (1.85)<0.0001
Birthweight (g)3210.3 (496.3)3146.1 (572.2)<0.0001
Maternal characteristics
Mean age23.71 (5.29)24.60 (5.65)<0.0001
Age category
18–3466352 (87.82%)2705 (86.89%)<0.0001
<185831 (7.72%)199 (6.39%)
>343374 (4.47%)209 (6.71%)
Education
<HS28 638 (38.01%)1312 (42.27%)<0.0001
≥HS46 714 (61.99%)1792 (57.73%)
Race
White37 617 (49.84%)1626 (52.35%)0.0164
Black36 761 (48.71%)1443 (46.46%)
Other1092 (1.45%)37 (1.19%)
Hispanic
Yes8294 (11.02%)389 (12.53%)0.0083
No66 998 (88.98%)2715 (87.47%)
Tobacco use
Yes13 874 (18.37%)629 (20.21%)0.0095
No61 646 (81.63%)2483 (79.79%)
Weight change (kg)11.70 (8.60)11.24 (8.69)0.0040
Pre-pregnancy BMI27.55 (6.74)28.14 (7.11)<0.0001
Intellectual disability
Yes2392 (3.17%)270 (8.67%)<0.0001
No73 170 (96.83%)2843 (91.33%)
Febrile at delivery
No74 890 (99.11%)3080 (98.94%)0.3219
Yes672 (0.89%)33 (1.06%)
Chlamydia during pregnancy
No70 927 (93.87%)2932 (94.19%)0.4659
Yes4635 (6.13%)181 (5.81%)
Gonorrhea during pregnancy
No74 730 (98.90%)3073 (98.72%)0.3369
Yes832 (1.10%)40 (1.28%)
Syphilis during pregnancy
No75 436 (99.83%)3107 (99.81%)0.7284
Yes126 (0.17%)6 (0.19%)
Epilepsy
No75 258 (99.60%)3084 (99.07%)<0.0001
Yes304 (0.40%)29 (0.93%)
Diabetes
No66 198 (87.61%)2693 (86.51%)0.0685
Yes9364 (12.39%)420 (13.49%)
Hypertension
No64 669 (85.58%)2590 (83.20%)0.0002
Yes10 893 (14.42%)523 (16.80%)

We compared the characteristics of children with ID against children without ID. As can be seen from Table 1, children with ID were more likely to be boys, had shorter gestational ages, and had lower birthweights. Maternal characteristics associated with an increased likelihood of child ID included older age, completion of <12 years of education, being white or Hispanic, smoking during pregnancy, higher pre-pregnancy BMI, less weight gain, intellectual disability, and epilepsy.

Pre-pregnancy BMI and odds of intellectual disability

We investigated the relationship between maternal pre-pregnancy BMI and odds of child ID, controlling for baby characteristics of sex, gestational age, and birthweight, and maternal characteristics of race, ethnicity, education level, tobacco use, weight change during pregnancy, epilepsy, and ID. We modelled both child ID of any severity and moderate to profound ID (henceforth termed ‘severe ID’) as our responses. We investigated the effect of pre-pregnancy BMI both as a continuous variable and as a categorical variable using commonly used BMI classifications. Degree of maternal weight gain, as a continuous variable, was the second key independent variable. Intuitively, we anticipated that the impact of antenatal weight change would be modified by pre-pregnancy BMI; that is, more weight gain would be beneficial for women who were underweight at baseline but harmful for women who were overweight or obese before becoming pregnant. Because of this potential effect modification, we first tested whether there was a significant interaction between pre-pregnancy BMI and weight change. We found that there was not a significant interaction between the two variables.

When we estimated the models with pre-pregnancy BMI and weight change as continuous variables, only BMI was statistically significant. For ID of any severity, each one-unit increase in pre-pregnancy BMI was associated with a 2% increase in the odds of ID (OR 1.02, 95% CI 1.01–1.02). For severe ID, the increase was slightly greater (OR 1.03, 95% CI 1.01–1.04).

The results of the models where pre-pregnancy BMI is categorised are presented in Table 2. BMI was classified as follows:

Table 2. Model results for ID and severe ID (pre-pregnancy BMI categorised)
 ID (3113 cases)Severe ID (579 cases)
OR (95% CI) P OR (95% CI) P
Underweight 1.08 (0.86–1.37)0.4901.31 (0.79, 2.17)0.298
Overweight 1.09 (0.99–1.20)0.0791.25 (1.00–1.56)0.049
Mild obesity 1.18 (1.06–1.33)0.0031.35 (1.05–1.74)0.021
Severe obesity 1.26 (1.09–1.45)0.0021.70 (1.26–2.29)0.001
Morbid obesity 1.52 (1.30–1.77)0.0001.83 (1.31–2.56)0.002
Maternal weight gain 1.000 (0.999–1.001)0.6901.001 (0.999–1.003)0.442
Female sex 0.44 (0.40–0.47)0.0000.29 (0.24–0.36)0.000
Gestational age Nonlinear0.001Nonlinear0.0051
Birthweight Nonlinear0.0000.86 (0.78–0.95)0.004
Education < 12 years 1.16 (1.06–1.26)0.0011.17 (0.97–1.40)0.107
Maternal race
Black versus white0.83 (0.76–0.90)0.0000.95 (0.79–1.16)0.634
Other versus white0.76 (0.54–1.07)0.1180.11 (0.01–0.89)0.039
Hispanic ethnicity 1.20 (1.06–1.37)0.0050.67 (0.47–0.96)0.029
Tobacco use 1.08 (0.97–1.19)0.1481.00 (0.80–1.25)0.993
Maternal intellectual disability 2.90 (2.52–3.34)0.0002.93 (2.17–3.96)0.000
Maternal age
<18 years0.80 (0.68–0.93)0.0050.54 (0.36–0.82)0.004
>34 years1.52 (1.30–1.76)0.0001.31 (0.91–1.89)0.144
Febrile at delivery 1.19 (0.82–1.72)0.3681.05 (0.43–2.59)0.915
Chlamydia during pregnancy 0.97 (0.83–1.14)0.7400.74 (0.50–1.09)0.130
Gonorrhea during pregnancy 1.20 (0.86–1.67)0.2871.23 (0.59–2.56)0.576
Syphilis during pregnancy 0.82 (0.33–2.05)0.6780.96 (0.14–6.66)0.964
Maternal epilepsy 1.97 (1.33–2.93)0.0012.07 (0.89–4.82)0.091
Maternal diabetes 1.01 (0.90–1.13)0.9001.00 (0.78–1.28)0.980
Maternal hypertension 1.03 (0.92–1.14)0.6351.00 (0.79–1.26)0.982
  • underweight (BMI < 18.5);
  • normal weight (BMI 18.5–24.99);
  • overweight (BMI 25–29.99);
  • mild obesity (BMI 30–34.99);
  • severe obesity (BMI 35–39.99);
  • morbid obesity (BMI 40 or greater).

Normal weight is the reference level.

Being underweight was not associated with increased odds of child ID, whereas there was a trend towards increased odds of ID in children of women who were overweight but not obese. Maternal obesity was significantly associated with risk of ID, and it appeared that a dose–response relationship was present. The strongest association was for children of women who were morbidly obese, compared with those who were of ideal weight. The odds of ID of any severity in children born to women with morbid obesity were increased by over 50% (OR 1.52, 95% CI 1.30–1.77), and the odds of severe ID were increased by more than 80% (OR 1.83, 95% CI 1.31–2.56). Again, weight change during pregnancy was not significantly associated with odds of ID. Other significant associations with ID can be seen in Table 2.

Table 3 summarises the results of our sensitivity analyses, based on the number of times a child was diagnosed with ID, or the number of different providers making the diagnosis. For each model, the odds ratio was adjusted by the full list of covariates. In every model, for both ID of any severity and for severe ID, children born to obese women were significantly more likely to have ID. The point estimates were consistently somewhat greater for severe ID than for ID of any severity, and they were also higher for women with severe or morbid obesity (with BMIs of 35 or greater) than for those with mild obesity.

Table 3. Adjusted odds ratios for BMI categories, sensitivity testing
 IDSevere ID
  1. Case definitions: 2—At least 5 Medicaid diagnoses or at least 2 different providers or DDSN or special education; 3—At least 10 Medicaid diagnoses or at least 3 different providers or DDSN or special education; 4—At least 15 Medicaid diagnoses or at least 4 different providers or DDSN or special eduction; 5—At least 25 Medicaid diagnosis or at least 5 different providers or DDSN or special education.

Case definition 2
Underweight1.10 (0.87–1.38)0.4321.31 (0.79–2.17)0.298
Overweight1.10 (1.00–1.21)0.0601.25 (1.00–1.56)0.049
Mild obesity1.20 (1.07–1.34)0.0011.35 (1.05–1.74)0.021
Severe obesity1.27 (1.11–1.47)0.0011.70 (1.26–2.29)0.001
Morbid obesity1.54 (1.32–1.80)0.0001.73 (1.23–2.45)0.002
Case definition 3
Underweight1.07 (0.84–1.37)0.5711.26 (0.73–2.15)0.406
Overweight1.12 (1.01–1.24)0.0301.21 (0.96–1.53)0.103
Mild obesity1.20 (1.07–1.35)0.0031.38 (1.06–1.80)0.017
Severe obesity1.27 (1.09–1.47)0.0021.73 (1.26–2.36)0.001
Morbid obesity1.55 (1.32–1.83)0.0001.61 (1.11–2.33)0.012
Case definition 4
Underweight1.07 (0.82–1.39)0.6201.22 (0.67–2.22)0.518
Overweight1.14 (1.02–1.26)0.0181.25 (0.97–1.61)0.091
Mild obesity1.25 (1.11–1.42)0.0001.50 (1.12–1.99)0.006
Severe obesity1.30 (1.11–1.52)0.0011.78 (1.26–2.51)0.001
Morbid obesity1.69 (1.42–2.00)0.0001.90 (1.29–2.81)0.001
Case definition 5
Underweight0.95 (0.70–1.29)0.7491.40 (0.73–2.69)0.315
Overweight1.16 (1.03–1.30)0.0161.21 (0.90–1.63)0.216
Mild obesity1.23 (1.07–1.41)0.0041.54 (1.11–2.14)0.010
Severe obesity1.29 (1.08–1.54)0.0051.63 (1.08–2.47)0.020
Morbid obesity1.76 (1.46–2.13)0.0002.08 (1.33–3.24)0.001

Discussion

To our knowledge, this is the largest study to investigate both pre-pregnancy BMI and gestational weight change as risk factors for ID in children. The key finding is that pre-pregnancy BMI is associated with significantly increased odds of ID in the children. The association is significant when controlling for a wide range of demographic and other characteristics. When we classified pre-pregnancy BMI by level of obesity, we found a marginally significant association for being overweight, with an increase of the strength of the association for increasing levels of obesity. These findings are important from a public health perspective given the recent, dramatic increase in the rates of obesity in the USA and other industrialised nations.[38]

We did not find a significant association between weight change (gain or loss) in pregnancy and odds of ID. This is consistent with the findings of Neggers et al., who reported that pre-pregnancy obesity but not weight gain during pregnancy was associated with lower psychomotor development scores in 355 African-American children.[28] Our study expands upon their finding by assessing diagnosed ID in a much larger sample that includes approximately 50% white children.

One pathway by which maternal obesity could increase the risk of ID in the child is fetal/neonatal macrosomia, which is more likely in the context of maternal obesity and is associated with an increased risk of a number of adverse pregnancy and birth outcomes, including stillbirth, birth asphyxia, and neonatal mortality. However, macrosomia does not explain the observed association between pre-pregnancy obesity and ID in our study, as the association was present despite controlling for birthweight.

Obesity in pregnancy has a number of metabolic and other effects, which could account for the association. These include insulin resistance/hyperinsulinaemia and systemic inflammation, as well as epigenetic effects on fetal DNA.[39] Obese pregnant women exhibit low-grade endotoxemia and increased systemic C-reactive protein and interleukin-6 concentrations, as well as increased gene expression for a number of pro-inflammatory cytokines in adipose tissue,[40] and they also exhibit increased macrophage accumulation and inflammation in the placenta.[41] Furthermore, there is evidence that prepubescent children of women who were obese during pregnancy were dramatically more likely to have detectable C-reactive protein levels than children of non-obese mothers, even after controlling for the child's BMI and other potential confounding factors.[42] There is convincing evidence that intrauterine inflammation and fetal systemic inflammation are associated with an increased risk of brain injury leading to neurodevelopmental disability, such as cerebral palsy.[43, 44] It is possible that the potential effects of increased inflammation in obese women, even in the absence of clinical signs or symptoms, is sufficient to significantly alter brain development in some children. Other metabolic and/or epigenetic effects of obesity may also have an impact on fetal brain development, but currently there is not sufficient evidence to speculate on what those effects may be. Other characteristics accompanying obesity, such as differences in dietary intake and physical activity patterns, may also play a role. Finally, obesity is known to alter levels of steroid hormones, including those involved in oocyte maturation and ovulation, and obesity is a known risk factor for infertility.[45] So it is possible that endocrine factors early in pregnancy play a role.

This study does have a number of limitations. First, it is limited to Medicaid-funded births in South Carolina. Medicaid is a health insurance programme for low-income families and funds approximately half of the births in South Carolina. Findings in the South Carolina Medicaid population may not be fully generalizable to middle and upper income families, or to other geographic areas outside of the southern USA.

A second limitation is our reliance on administrative data for the ascertainment of both independent and dependent variables. Pre-pregnancy and delivery height and weight have been recorded on South Carolina birth certificates since 2004. We relied upon this information for our analyses. Height and weight are recorded on birth certificates based on the mother's self-report, rather than on objective measurement. Therefore, they are subject to being misreported. Nevertheless, many recent studies have used birth certificate data for epidemiological studies related to maternal weight.[46-51] A large study of women with low incomes in Florida demonstrated that pre-pregnancy height and weight as recorded on birth certificates were generally reliable.[52] In a smaller, clinic-based study, Wright et al. found that gestational weight gain recorded in birth certificates was within 4.5 kg for 48% of women, whereas 52% had weight gain either over- or under-reported by at least 4.5 kg.[53] For the misreporting of BMI to have accounted for the findings of this study, a systematic relationship between maternal misreporting and the odds of child ID would be needed. Rothman describes how non-differential misclassification can result in a bias away from the null if the exposure variable is continuous (like BMI).[54] However, he further explains that changing to a dichotomous variable (such as obese versus not obese) generally alleviates this problem. In this case, the results of comparing each category of abnormal weight with women of normal weight resulted in the same findings as the models with BMI as a continuous variable. In addition, our decision to remove women with extreme values for pre-pregnancy weight or weight gain from the analyses also reduces the likelihood that erroneous weights accounted for the observed findings. We believe it is highly unlikely that the findings of this study were substantially impacted by the misreporting of BMI or gestational weight change.

For the outcome of ID we relied on Medicaid diagnoses, public school records, and receipt of services from the South Carolina DDSN. The public school system and DDSN require a formal psychological assessment, with IQ and adaptive function testing, before a child is eligible for special education classes or ID-related services. We do not have information on the factors underlying a diagnosis of ID in the Medicaid billing records.

The number of children enrolled in DDSN programmes for ID or in special education in public schools was exceedingly low. This is probably because of their young ages—all were born from January 2004 through December 2007: therefore, the maximum age was 6 years. We were unable to go further back than 2004 in our cohort because that is when South Carolina birth certificates began including pre-pregnancy height and weight. To determine how many of the cases of ID identified using Medicaid records ultimately wind up enrolling in DDSN services or special education, we would need to follow up with the analyses in a few years, when all of the children have had an opportunity to enroll in school and have had time for their learning difficulties to be identified.

Meanwhile, our decision to limit cases to those identified in the public school or DDSN records, or diagnosed with ID on at least five different occasions, minimises the likelihood of ‘false-positive’ diagnoses of ID in our case group. In addition, we conducted extensive sensitivity testing based on the number of times diagnosed with ID and the number of providers making the diagnosis. The association between obesity and ID was robust to the sensitivity testing, even when requiring that a child be diagnosed with ID at least 25 times or by at least five different healthcare providers in order to be considered a ‘case’. Such a robust finding indicates that the associations reported in this paper are highly unlikely to be attributable to misidentified ID.

One of the greatest strengths of this study is the large sample size, which provides ample power for the analyses. Our cohort included almost 80 000 children after exclusions, over 3100 of whom were identified as having ID. Another strength is our analytic method, which allowed us to model and interpret nonlinear relationships between pre-pregnancy BMI and weight change during pregnancy and the outcome ID. This is vital, as we would not necessarily anticipate a linear relationship between BMI/weight change and odds of ID. Our consideration of weight change during pregnancy is also an important strength, as it would be reasonable (although apparently incorrect) to anticipate that gaining too much weight during pregnancy would be similarly associated with ID as having pre-pregnancy obesity. Finally, we controlled for a number of potentially important factors (including demographic characteristics, gestational age, birthweight, and diagnoses of epilepsy and ID in mothers), so we can be confident that the association between obesity and ID does not result from confounding by those factors. We are not aware of any studies of maternal obesity and child ID that have adjusted for such a broad range of covariates. And we excluded children with birth defects, chromosomal conditions, or other diagnoses known to be causes of ID, so our findings are directly relevant for the approximately 50% of cases of ID that do not have known causes.

The findings regarding other factors associated with risk of ID (such as maternal ID and maternal epilepsy, lower gestational age, and birthweight) are consistent with what might be expected, which lends additional support to the general validity of ID diagnoses in the cohort. One notable exception is the higher rate of ID among white children compared with black children, as in the US ID is more prevalent among children from minority groups; on further examination, we found that before exclusions, ID was in fact more common among black children with ID, but that a larger number of ID cases were excluded from minority groups by the diagnosis of known causes of ID (data available from authors).

Despite its strengths, this is an associational study and does not establish a causal relationship between maternal obesity and child ID. Two key factors in establishing causation are consistency/replication of findings and the identification of one or more plausible mechanisms for causation.[55] Additional epidemiological study in a variety of geographic and socio-economic groups is needed to test the consistency of the association, and to assess the generalisability of our findings to the entire population of pregnant women and children. In addition, basic science and translational research are needed to provide a better understanding of the potential mechanism(s) whereby maternal obesity may impact fetal brain development, and (if the association is determined to be causal) to identify potential methods to reduce the effects of obesity on fetal brain development. Meanwhile, the observed link between pre-pregnancy obesity and risk of ID in offspring may be yet another impetus (in addition to the established associations between obesity and numerous other adverse health outcomes) to develop and implement effective approaches to counteract the obesity epidemic in the USA and other industrialised nations.

Conclusions

Pre-pregnancy obesity is associated with an increased risk of intellectual disability in children. If this finding is confirmed and determined to be causal, then efforts to reduce obesity in childbearing women have the potential to reduce the prevalence of ID in children.

Acknowledgements

The authors would like to thank Heather Kirby and other staff at the South Carolina Office of Research and Statistics for making available the data used in this study.

Disclosure of interests

The authors have no interests to declare.

Contribution to authorship

JM, SM, and JH are co-investigators. CP performed the majority of the data analysis. ZZ assisted in the literature review and contributed substantially to the writing of the article. The contributions made by all authors to the article are consistent with co-authorship.

Details of ethics approval

The study was granted ‘exempt’ status by the University of South Carolina Institutional Review Board.

Funding

This study was funded by the Health Resources and Services Administration, grant number R40MC21523 (to J.R.M., as principle investigator), ‘Maternal Obesity, Excessive Weight Gain, Diabetes Mellitus, and Hypertension during Pregnancy and Risk of Neuro-developmental Disability in Children’.

Ancillary