Keywords
autism - neurodevelopment - hyperemesis gravidarum - race - ethnicity
Autism spectrum disorder (ASD) affects 1 in 68 children in the United States[1] and its prevalence has dramatically increased over the last two decades.[2]
[3] It persists into adulthood and is a significant burden to families and society.[4]
[5] Although genetic and environmental factors have been suggested, though no specific
cause of the condition has been identified, and it is thought to be a complex disorder
of the developing brain with multifactorial etiology affecting the brain at critical
times during development. Potential risk factors include advanced parental age, white
race, male sex, and multiple births.[6]
[7]
[8] Antepartum and intrapartum conditions,[9]
[10] preterm birth,[11] small for gestational age birth,[12] having a sibling with autism,[13]
[14] and mood disorders in family members[15] have also been suggested to increase risk.
Hyperemesis gravidarum (HG), a severe form of nausea and vomiting, affects 0.3 to
4% of pregnant women[16]
[17]
[18] and can result in maternal weight loss, malnutrition, dehydration, and electrolyte/acid-base
imbalance and low birthweight and small-for-gestational age birth of infant morbidity.[16]
[19] Like autism, its pathoetiology is unknown but risk factors include family history,
HG in a prior pregnancy, non-white race, female fetus, multiple gestation, and low
body mass index.[16]
[20]
[21]
[22]
[23] Previous studies suggest that metabolic problems, such as undernutrition,[24]
[25] gestational diabetes mellitus (GDM),[9] and hypothyroidism[26] during early pregnancy may lead to neurodevelopmental conditions, such as schizophrenia
and ASD. Therefore, we hypothesized that HG is associated with increased ASD risk
in the children and that this association may be modified by child's sex, race/ethnicity,
timing of first in utero exposure, severity of the condition, and gestational age
at birth.
Materials and Methods
We conducted a retrospective longitudinal cohort study using medical records of pregnant
women and their children (n = 469,789) born between January 1, 1991 and December 31, 2014 at all Kaiser Permanente
Southern California (KPSC) hospital. To be eligible, children must have been born
to KPSC-members, be a live-singleton birth between 28 0/7 and 42 6/7 weeks of gestation and be a KPSC health plan member for ≥3 months between 2 and 17
years of age. The study cohort composition is shown in [Supplementary Appendix A: Supplementary Fig. S1] (available in the online version). Children's medical records were linked longitudinally
to biological mothers using unique identifiers. The linked medical records contain
information on maternal sociodemographic and behavioral characteristics, maternal
medical and obstetrical history, and complete child health care information. International
Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes
from inpatient and outpatient physician encounters, as well as laboratory results
and pharmacy records were used to ascertain the exposure and outcome of interest.
Validation of the linked data have been reported in detail elsewhere.[27] Institutional review board approval was obtained for this study.
Fig. 1 Overall (A) and sex-specific (B) ASD survival curve by child age. Mean age (standard deviation [SD]) at ASD diagnosis
as follows: no-HG total (5.15 [3.57]), HG total (3.87 [2.44]), no-HG male (5.12 [3.51]),
HG male (4.00 [2.48]), no-HG female (5.28 [3.83]), and HG female (3.41 [2.24]). ASD,
autism spectrum disorder; HG, hyperemesis gravidarum.
The primary exposure of interest was the clinical diagnosis of HG (ICD-9-CM code 643.0x
and 643.1x). To assess the effect of the severity of illness on ASD risk, we chose
an approach with indicators of a more severe form of the condition that likely requires
hospitalization. This approach used the following diagnoses and procedures: HG with
metabolic disturbance (ICD-9-CM codes 643.1x and 99.18), requiring nasogastric enteral
feeding (ICD-9-CM codes 96.07, and current procedural terminology [CPT] codes 43752
and 43760), and total parenteral nutrition (ICD-9-CM codes 96.07, 96.08, 96.6, 99.15,
and 99.18).
To ascertain a diagnosis of ASD, we used the Diagnostic and Statistical Manual of
Mental Disorders, Fourth Edition, Text Revision (DSM-IV-TR)[28] for any of the following conditions: autistic disorder, childhood disintegrative
disorder, Rett's disorder, Asperger's disorder, or pervasive developmental disorder,
not otherwise specified. Children, ages between 2 and 17 years, with at least one
documented DSM-IV-TR code for ASD on any two separate visits during the follow-up
period formed the ASD cases. Clinical experts validated the accuracy of these codes
through medical record review.[29] As per KPSC guidelines, last revised April 2013,[30] ASD services are covered by a plan (a child/adolescent psychiatrist, developmental/behavioral
pediatrician, child psychologist, or neurologist). The experts are required to perform
a series of behavioral and developmental surveillance at all well-child visits, as
early as 4 months of age. If there is a sufficient degree of suspicion, a modified
version of the checklist for autism in toddlers [M-CHAT][31] and developmental screening questionnaires for toddlers is completed (as early as
18 months of life). For each diagnosed child, a treatment plan is developed and provided
by a qualified provider in autism services.
Potential confounders/modifiers in this study include child's race/ethnicity and sex,
maternal age (<25, 25–34, and ≥35 years), education (<12, 12, and ≥13 years), median
family household income based on census tract of residence, timing of prenatal care
initiation, smoking during pregnancy (yes/no), gestational age at delivery based on
clinical estimates, and maternal comorbidities. Maternal and paternal race/ethnicity
defined the child's race/ethnicity. A child was categorized non-Hispanic white (white)
if born to non-Hispanic white mother and father. The same applies to non-Hispanic
black (black), Hispanic, and Asian/Pacific Islander race/ethnicity groups. The Other/Multiple
race/ethnicity category includes children born from interracial/interethnic relationship.
The electronic medical and pharmacy records contain detailed clinical and drug-dispensing
data on medications commonly used to treat mild form of HG (diphenhydramine, promethazine,
ondansetron, and metoclopramide), nutritional interventions (nasogastric enteral feeding
and total parenteral nutrition), and injection or infusion of electrolytes.
Follow-up of children started from the delivery date until the date of ASD diagnosis.
Censoring occurred on the earliest of the following dates: health plan disenrollment,
17th birthday, non-ASD related death, or end of study (December 31, 2016). We examined
the distributions of maternal and child characteristics by HG status using χ
2 tests. HG may recur between pregnancies and ASD also tends to recur in siblings;
such clustered observations are likely to be correlated. Therefore, to examine the
association between prenatal exposure to HG and ASD, we used marginal Cox's proportional
hazards model to account for within-cluster correlation. Confounding variables were
chosen a priori or if they resulted in p-value < 0.05.
Because the timing and severity of prenatal exposure, and the fact that child's sex
and gestational age at birth may influence ASD risk, we stratified the data for these
potential mediators ([Table 3]; [Supplementary Appendix B: Supplementary Fig. S2] [available in the online version]). To assess the impact of the severity of illness
on the association between prenatal exposure to HG and ASD, we compared the association
of HG with metabolic disturbances to those without. We also examined the effect of
HG treatments on the observed association for each level of disease-modifying therapy
after limiting the analysis to pregnancies (n = 190,126) with complete medication data available during the study period (2006–2014).
We performed the following sensitivity analyses ([Supplementary Appendix C: Supplementary Table S1] [available in the online version]) to evaluate the relative influence of each parameter
on the observed risk relationships: (1) exclude children with only one ASD diagnosis,
(2) exclude children with a history of congenital anomalies and developmental and
emotional comorbidities (mental retardation, developmental dyslexia, deficits in language
processing, conduct disorder, irritability, bipolar or anxiety disorders, and depression),
(3) exclude pregnancies with a history of medical and/or perinatal adversities including
gestational hypertension, diabetes, intrauterine growth restriction (IUGR), and low
birth weight, (4) limit the cohort to children aged ≥ 4 years because children aged
2 to 3 years may have shorter follow-up period to capture events, (5) limit the cohort
to children without genetic predisposition, (6) account for maternal comorbidities
([Supplementary Table S2] [available in the online version]), (7) account for “year of birth,” (8) account
for maternal prepregnancy body mass index (BMI, kg/m2) in the model, and (9) examine the potential effects of residual confounders on the
observed association using a range of E-values for the overall result and the lower
limits of their 95% confidence intervals (CI) observed in Cox's regression models.[32] The E-values were estimated using the online calculator for hazard ratios (HR) with
an outcome prevalence of <15%. Crude and adjusted HR and 95% CI were used to quantify
the magnitude of associations. The analyses were performed using SAS version 9.4 (SAS
Institute, Cary, NC).
Table 1
Distribution of maternal and child characteristics based on HG status
Characteristics
|
No HG n = 455,263 (%)
|
HG n = 14,526 (%)
|
p-Value
|
Maternal age (y)
|
|
|
<0.001
|
< 20
|
27,171 (6.0)
|
760 (5.2)
|
|
20–29
|
196,748 (43.2)
|
7,348 (50.6)
|
|
30–34
|
138,693 (30.5)
|
4,093 (28.2)
|
|
≥ 35
|
92,651 (20.4)
|
2,325 (16.0)
|
|
Maternal education (y)
|
|
|
<0.001
|
< 12
|
46,726 (10.3)
|
992 (6.8)
|
|
12
|
128,537 (28.2)
|
4,459 (30.7)
|
|
≥ 13
|
264,821 (58.2)
|
8,864 (61.0)
|
|
Missing
|
15,179 (3.3)
|
211 (1.5)
|
|
Median household income, USD
|
|
|
<0.001
|
< 30,000
|
45,228 (9.9)
|
1,105 (7.6)
|
|
30,000–49,999
|
156,488 (34.4)
|
4,914 (33.8)
|
|
50,000–69,999
|
135,396 (29.7)
|
4,432 (30.5)
|
|
70,000–89,999
|
69,182 (15.2)
|
2,450 (16.9)
|
|
≥ 90,000
|
46,127 (10.1)
|
1,569 (10.8)
|
|
Nulliparous
|
175,982 (38.7)
|
5,756 (39.6)
|
<0.001
|
Smoking during pregnancy
|
31,126 (6.8)
|
834 (5.7)
|
<0.001
|
Late/no initiation of prenatal care
|
45,935 (10.1)
|
766 (5.3)
|
<0.001
|
Gestational age <37 weeks
|
33,133 (7.3)
|
1,223 (8.4)
|
<0.001
|
Child's Race/Ethnicity
|
|
|
<0.001
|
Non-Hispanic White
|
98,328 (21.6)
|
2,336 (16.1)
|
|
Non-Hispanic Black
|
38,757 (8.5)
|
1,924 (13.2)
|
|
Hispanic
|
170,416 (37.4)
|
5,526 (38.0)
|
|
Asian/Pacific Islander
|
39,437 (8.7)
|
1,334 (9.2)
|
|
Other/Multiple
|
102,125 (22.4)
|
3,346 (23.0)
|
|
Child' sex
|
|
|
<0.001
|
Female
|
221,856 (48.7)
|
7,592 (52.3)
|
|
Male
|
233,407 (51.3)
|
6,934 (47.7)
|
|
Abbreviations: HG, Hyperemesis gravidarum; USD, United States dollar.
Table 2
Association between hyperemesis gravidarum and autism spectrum disorder
Hyperemesis Gravidarum status
|
Number
|
Incidence rate[a]
|
Hazard ratios (95% confidence Intervals)
|
Total births
|
ASD
|
|
Crude
|
Adjusted[b]
|
p-Value
|
No HG
|
455,263
|
8115
|
1.71
|
1.0 (ref.)
|
1.0 (ref.)
|
|
HG
|
14,526
|
332
|
2.87
|
1.47 (1.31–1.64)
|
1.53 (1.37–1.70)
|
<0.001
|
Trimester at first diagnosis
|
|
|
|
|
|
|
1st trimester (weeks of gestation)
|
11,505
|
269
|
2.94
|
1.50 (1.33–1.71)
|
1.58 (1.40–1.79)
|
<0.001
|
2nd trimester (weeks of gestation)
|
2,785
|
59
|
2.68
|
1.36 (1.05–1.76)
|
1.36 (1.05–1.75)
|
0.02
|
3rd trimester (weeks of gestation)
|
236
|
4
|
1.80
|
1.01 (0.38–2.69)
|
1.06 (0.40–2.83)
|
0.91
|
Severity of illness[c]
|
|
|
|
|
|
|
HG without metabolic disturbance
|
8,195
|
183
|
2.77
|
1.42 (1.23–1.64)
|
1.46 (1.26–1.69)
|
<0.001
|
HG with metabolic disturbances
|
6,331
|
149
|
3.00
|
1.53 (1.30–1.80)
|
1.63 (1.38–1.92)
|
<0.001
|
Abbreviations: ASD, autism spectrum disorders; HG, hyperemensis gravidarum.
a Incidence rate is shown per 1000 person-years.
b Hazard ratios (HR) were adjusted for maternal age, education, smoking during pregnancy,
parity, prenatal care, maternal psychosocial disorder, child's sex, race/ethnicity,
and year of diagnosis.
c Severe metabolic disturbances that require immediate clinical intervention (injection
or infusion of electrolytes, nasogastric enteral feeding, or total parenteral nutrition).
Table 3
Associations between hyperemesis gravidarum and autism spectrum disorder by gestational
age at birth and child's sex
Conditions
|
Incidence rates (IR)[a] and adjusted hazard ratios (HR)[b] for ASD
|
Preterm (n = 34,356)
|
Term (n = 435,433)
|
Female (n = 229,448)
|
Male (n = 240,341)
|
IR
|
HR (95% CI)
|
p-Value
|
IR
|
HR (95% CI)
|
p-Value
|
IR
|
HR (95% CI)
|
p-Value
|
IR
|
HR (95% CI)
|
p-Value
|
No HG (n = 455,263)
|
2.27
|
1.00 (ref.)
|
|
1.67
|
1.00 (ref.)
|
|
0.65
|
1.00 (Ref.)
|
|
2.73
|
1.00 (Ref.)
|
|
HG (n = 14,526)
|
4.82
|
1.98 (1.48–2.66)
|
<0.001
|
2.69
|
1.39 (1.23–1.56)
|
<0.001
|
1.20
|
1.62 (1.28–2.05)
|
<.001
|
4.74
|
1.50 (1.33–1.70)
|
<0.001
|
Timing of 1st HG diagnosis
|
1st trimester (n = 11,505)
|
4.73
|
1.95 (1.35–2.72)
|
<0.001
|
2.78
|
1.44 (1.26–1.64)
|
<0.001
|
1.21
|
1.64 (1.26–2.12)
|
<.001
|
4.94
|
1.57 (1.37–1.80)
|
<0.001
|
2nd trimester (n = 2785)
|
5.60
|
2.27 (1.28–4.02)
|
0.001
|
2.36
|
1.21 (0.91–1.61)
|
0.20
|
1.17
|
1.57 (0.91–2.72)
|
0.10
|
4.19
|
1.31 (0.98–1.75)
|
0.07
|
3rd trimester (n = 236)
|
0.00
|
–
|
–
|
1.97
|
1.19 (0.45–3.16)
|
0.73
|
0.89
|
1.36 (0.19–9.66)
|
0.76
|
2.72
|
0.99 (0.32–3.06)
|
0.98
|
Severity of illness
[c]
|
HG-no metabolic disturbance
|
4.57
|
1.85 (1.23–2.77)
|
0.003
|
2.62
|
1.35 (1.15–1.57)
|
<0.001
|
1.58
|
2.15 (1.64–2.81)
|
<0.001
|
4.09
|
1.28 (1.08–1.53)
|
0.01
|
HG + metabolic disturbance
|
5.10
|
2.16 (1.44–3.25)
|
<0.001
|
2.78
|
1.45 (1.22–1.73)
|
<0.001
|
0.68
|
0.93 (0.58–1.48)
|
0.76
|
5.62
|
1.82 (1.53–2.16)
|
<0.001
|
Abbreviations: ASD, autism spectrum disorders; CI, confidence interval; HG, hyperemesis
gravidarum..
a IR, incidence rate is shown per 1,000 person-years.
b HRs were adjusted for maternal age, education, smoking during pregnancy, perinatal
care, parity, maternal psychosocial disorder, child's race/ethnicity, and year of
diagnosis.
c Severe metabolic disturbances that require immediate clinical intervention (injection
or infusion of electrolytes, nasogastric enteral feeding, or total parenteral nutrition).
Results
We identified 469,789 children, aged between 2 and 17 years, who were born singleton
in all KPSC hospitals between January 1, 1991 and December 31, 2014 with at least
90 days active enrollment during the follow-up periods (1993–2016). During the study
period, 8,447 (1.8%) children were diagnosed with ASD. The mean age at first diagnosis
was 6 years (standard deviation [SD] = 3.3). The mean follow-up time was 10.4 years
(SD = 4.7) for children diagnosed with ASD.
Compared with women without a history of HG, women with a history of HG were more
likely to be 20 to 29 years of age, have completed ≥12 years of education, have a
high-household income, be nulliparous, and not smoke during pregnancy ([Table 1]). The rate of prenatal exposure to HG varied by child's race/ethnicity and sex.
Black and Asian/Pacific Islander children were more likely to be exposed to HG. Girls
were more likely than boys to be exposed. Exposed children were more likely to be
born preterm than unexposed children.
Compared with unexposed children (1.71/1,000 person-years), the overall ASD incidence
rate in exposed children was higher (2.87/1,000 person-years; adjusted HR: 1.53; 95%
CI: 1.37–1.70). This remained significant when HG diagnosis was made during the first
(HR: 1.58; 95% CI: 1.40–1.79) and second (HR: 1.36; 95% CI: 1.05–1.75) trimesters
of pregnancy ([Table 2]). Exposure to HG was associated with ASD risk, regardless of the severity of the
exposure. Limiting the cohort to children aged 4 to 17 years had no effect on our
findings.
The association between prenatal exposure to HG and childhood ASD was modified by
gestational age at birth ([Table 3]; [Supplementary Appendix B: Supplementary Fig. S2] [available in the online version]). In term-born children, prenatal exposure to
HG is associated with a 1.39-fold increased ASD risk (95% CI: 1.23–1.56), with the
greatest risk occurring after exposure in the first (HR: 1.44; 95% CI: 1.26–1.64)
and second (HR: 1.21; 95% CI: 0.91–1.61) trimesters, although the latter did not attain
statistical significance. In preterm children, ASD incidence rates were twice as common
in the exposed (4.82/1,000 person-years) compared with unexposed (2.27/1,000 person-years)
groups, with significantly increased risk during the first (HR: 1.95; 95% CI: 1.35–2.72)
and second (HR: 2.27; 95% CI: 1.28–4.02) trimesters. When further stratified by severity
of fetal exposure, risk did not differ substantially at preterm and term births. Analyses
stratified by child sex indicated that in both boys (HR: 1.50; 95% CI: 1.33–1.70)
and girls (HR: 1.62; 95%CI: 1.28–2.05), HG exposure was associated with ASD risk and
was the strongest when exposure occurred during the first two trimesters.
Our results indicated that commonly prescribed medication to treat mild forms of HG
did not significantly affect ASD risk ([Table 4]). On further examining the effect of immediate versus later in-hospital or outpatient
care on ASD risk, we observed no significant difference between women who were admitted
early versus late during their illness (HR: 1.24; 95% CI: 0.69–2.24; p = 0.47). However, we observed higher ASD risk (HR: 1.41; 95% CI: 1.05–1.90; p = 0.02) among children of women who received electrolyte replacement. In this sample
limited to recent birth years with complete data available, HG with metabolic disturbance
is still significantly associated with ASD risk (HR: 1.33; 95% CI: 1.11–1.60; p = 0.002); however, the magnitude of the effect is lower than that observed for the
total sample.
Table 4
Hyperemesis gravidarum interventions and autism spectrum disorder in children born
between 2006–2014
Treatment modalities
|
Counts
|
Incidence rate[a]
|
Hazard ratios (95% CI)
|
p-value
|
Total birth
|
ASD
|
Crude
|
Adjusted[b]
|
No-HG
|
179,602
|
3,555
|
3.16
|
1.00 (Ref.)
|
1.00 (Ref.)
|
(Ref.)[c]
|
HG without metabolic disturbance
|
5,538
|
117
|
3.12
|
1.07 (0.89–1.28)
|
1.13 (0.94–1.36)
|
0.20
|
No treatment
|
2,425
|
51
|
2.76
|
1.00 (Ref.)
|
1.00 (Ref.)
|
(Ref.)[c]
|
Diphenhydramine
|
16
|
1
|
9.09
|
2.45 (0.34–17.67)
|
2.53 (0.35–18.5)
|
0.36
|
Promethazine
|
1,067
|
15
|
2.18
|
0.93 (0.59–1.46)
|
0.96 (0.60–1.52)
|
0.85
|
Ondansetron
|
1,624
|
25
|
3.67
|
1.04 (0.72–1.49)
|
1.01 (0.70–1.46)
|
0.95
|
Metoclopramide
|
950
|
29
|
4.45
|
1.41 (0.93–2.14)
|
1.39 (0.91–2.13)
|
0.12
|
HG with metabolic disturbances[d]
|
4,986
|
121
|
3.97
|
1.24 (1.03–1.48)
|
1.33 (1.11–1.60)
|
0.002
|
Inpatient care for HG as principal diagnosis
|
4,020
|
95
|
3.85
|
1.20 (0.98–1.47)
|
1.29 (1.05–1.58)
|
0.02
|
Treatment started on the day of diagnosis
|
3,550
|
82
|
3.77
|
1.00 (Ref.)
|
1.00 (Ref.)
|
(Ref.)[c]
|
Treatment started at ≥1 day after diagnosis
|
470
|
13
|
4.44
|
1.19 (0.66–2.14)
|
1.24 (0.69–2.24)
|
0.47
|
Total enteral/parenteral nutrition
|
139
|
2
|
2.21
|
0.71 (0.18–2.85)
|
0.76 (0.19–3.06)
|
0.70
|
Electrolyte injection/infusion
|
1,798
|
44
|
4.26
|
1.27 (0.95–1.72)
|
1.41 (1.05–1.90)
|
0.02
|
Abbreviations: ASD, autism spectrum disorders; CI, confidence interval; HG, hyperemesis
gravidarum.
Note: Analysis was limited to pregnancies (2006–2014) with complete medication data
available in electronic medical records (n = 190,126).
a Incidence rate is shown per 1,000 person-years.
b Hazard ratios were adjusted for maternal age, education, smoking during pregnancy,
perinatal care, parity, maternal psychosocial disorder, child's race/ethnicity, and
year of diagnosis.
c (Ref.); reference for adjusted p-value test statistics for strata differences.
d HG with metabolic disturbances requiring medical interventions that included injection
or infusion of electrolytes, nasogastric enteral feeding, or total parenteral nutrition.
ASD diagnoses occurred at younger ages in children exposed to HG prenatally than in
those who were not ([Fig. 1A]). Although, ASD risk was elevated in both boys and girls regardless of the severity
of exposure, the relationship was strongest among girls ([Fig. 1B]).
The incidence of ASD diagnosis was higher for HG exposed children of all racial–ethnic
backgrounds. HG exposure was found to be significantly associated with ASD in whites
(HR: 1.70; 95% CI: 1.23–2.36) and Hispanics (HR: 1.76; 95% CI: 1.41–2.18), but not
in blacks (HR: 1.18; 95% CI: 0.77–1.81) and Asian/Pacific Islanders (HR:1.37; 95%
CI: 0.88–2.14).
The results of the sensitivity analyses after excluding children with only one ASD
diagnosis, a history of developmental and emotional comorbidities, congenital malformation,
excluding children aged <4 years, and accounting for “year of birth,” family predisposition
to ASD, and maternal medical and obstetrical comorbidities did not differ substantially
from the overall analysis ([Supplementary Appendix C: Supplementary Table S1] [available in the online version]). Lastly, a sensitivity analysis using E-value
to assess bias due to potential unmeasured confounders[32] gave an E-value of 2.43 (95% CI: 2.08–2.79) for the overall adjusted HR of 1.53
(95%CI: 1.37–1.70). Thus, a minimum risk ratio of 2.08 would be required between unmeasured
confounders and HG exposure, and between unmeasured confounders and ASD to explain
away the significant association between HG and child ASD.
Discussion
In this study, we demonstrated that in utero exposure to HG is associated with increased
risk of ASD that could not be explained by confounding factors. The magnitude of association
did not vary with the severity of exposure and was strongest when the exposure occurred
during the first two trimesters. Exposure to HG increases ASD risk, regardless of
term or preterm birth with greatest risk in the first two-thirds of pregnancy. We
also found that prenatal HG increased ASD risk in boys and girls and that there were
significantly increased risks for whites and Hispanics. Commonly prescribed medication
for mild forms of HG did not alter ASD risk.
We hypothesized that the effect of HG on ASD is highly dependent on the timing of
fetal exposure to HG because previous studies demonstrated that poor maternal nutrition
during early fetal life leads to long-term neurodevelopmental impairments.[9]
[33]
[34]
[35] However, our inability to detect HG as a cause of ASD in the third trimester may
also be due to greatly reduced rates of HG in that time. Moreover, adjusting for history
of prenatal psychosocial disorders, a factor linked to cognitive impairments in children,[36] did not affect the magnitude of association, suggesting an independent effect of
HG on ASD risk.
Although, there is increasing awareness of the fetal origins of adult disease, the
long-term consequences of HG for the children are understudied. Malnutrition during
pregnancy, even if limited to a few weeks or months, can lead to fetal nutrient deficiencies
at a vulnerable period in organ development, increasing risk for long-term adverse
health outcomes.[37] HG, as a cause of maternal malnutrition in early pregnancy, may have a profound
effect on fetal brain development either (1) directly, through metabolic imbalances
resulted from maternal malnutrition and dehydration,[38] or (2) indirectly, through hypoxemia and reduced nutritional supply to the fetus
by compromising placental function.[39]
[40]
Women experiencing HG often have poor nutritional intake and are dehydrated, creating
conditions that are functionally similar to famine and may affect development in a
similar manner. The best-known example is the Dutch Winter Hunger study, a natural
experiment that utilized data collected on women that were forcibly deprived of food
during the World War II. Caloric deprivation during the first two trimesters of pregnancy
of these women increased cognitive disabilities and schizophrenia risk in the offspring.[33] Other studies have shown this effect of undernutrition on cognitive impairment may
continue throughout life.[41] Our finding that HG during the first two trimesters increases the ASD risk is consistent
with these previous reports. However, it is unclear how, or to what extent neurodevelopment
is impaired by HG.
The placenta regulates the transfer of nutrients and oxygen from the mother to the
developing fetus and can adapt morphologically and functionally to optimize nutrients
and oxygen, promoting fetal organ development. Restriction of oxygen and protein clearly
affects placental structure and vascularity.[39] Animal[42]
[43] and human[9]
[10] studies suggest that compromised placental function (as found in preeclampsia, GDM,
and IUGR) is associated with ASD risk. Changes in placental function may also contribute
to ASD development. The placenta produces several neurodevelopmental factors, such
as nerve growth factor (NGF) and brain-derived neurotrophic factor (BDNF).[40] BDNF regulates dopaminergic neuron development whose functions are disrupted in
ASD and mental retardation.[44]
[45] Limiting the data to pregnancies without placental abnormalities resulted in little
effect on our overall findings. This suggests that HG may have specific effects on
placenta function that result in ASD or that placental functions are largely unaltered
but there are fewer nutrients to transport.
The KPSC health plan approach is comprehensive and more valid than parental reporting,
nonspecialist diagnosis, or case identification from records because all ASD diagnoses
are confirmed by a pediatric specialist. This limits the potential for patient misclassification.
Furthermore, the large size and diverse population provide opportunities to explore
how risks differ by the child's sex, race/ethnicity, gestational age or timing of
HG exposure. The findings add to the strong evidence that the environment that babies
are exposed to in the womb is a major cause of metabolic diseases in adults. We performed
several sensitivity analyses to evaluate the independent effect of prenatal exposure
to HG on the observed risk relationships. Although, the magnitude of the effect is
slightly lower than that observed for the full sample, we still found an association
between HG and ASD after limiting the sample to recent birth years (2008–2014) that
had complete data on maternal prepregnancy BMI and comorbidities ([Supplementary Appendix C] [available in the online version]).
This study has potential limitations. For example, diagnostic suspicion bias can occur
when exposed and unexposed patients differ in measurement (both the intensity and
diagnostic process) of health outcome statuses. Perinatal exposure to maternal smoking
has been linked to behavioral and cognitive impairments in children[46] and we relied on self-reported smoking during pregnancy. However, Buka, et al[47] found strong agreement between self-reported smoking and serum cotinine levels.
Some medications, such as antihistamines and promethazine, may have been prescribed
to suppress allergy symptoms and information was not available regarding their intended
use. Moreover, dosage data on prescribed medications and other factors (e.g., environmental
toxins) were not available in the electronic medical records. To better understand
the potential impact these and other unmeasured confounders may have, we calculated
the E-value for the main result ([Supplementary Appendix C] [available in the online version]). Our finding suggests that the observed result
would be explained by confounding bias only if the independent associations of the
potential confounders with both the exposure and outcome of interest was of unrealistic
magnitude (up to a 2.43-fold difference). Although causality cannot be proven based
on this observational study, HG fulfills the Bradford–Hill criteria as a risk factor
in terms of biological plausibility and timing.
Conclusion
Our findings provide additional evidence that the intrauterine environment has a significant
impact on ASD risk. However, HG may stand out as a risk factor because it is clinically
obvious and may be useful for identifying at-risk children that may benefit from closer
surveillance and earlier diagnosis. Early intervention with behavioral and developmental
therapy can result in better long-term cognitive and behavioral function, and improve
symptoms.[48] More aggressive diagnosis and treatment of HG may provide new opportunities to prevent
this life-long disability and its impact on the families and society.