Search strategy

A medical information specialist (J.L.) performed a search in MEDLINE and Embase from inception to September 6, 2021. We used respectively Medical Subject Headings (MeSH) or Emtree terms and text words for the concepts: (1) hyperemesis gravidarum and (2) child, offspring, pregnancy outcomes, or long-term effects. Animal studies, reviews, case reports, editorials, and conference abstracts (Embase) were excluded. No other limitations, including date and language, were applied. Complete search strategies are detailed in Appendix A1 . The search also included perinatal outcomes, which will be discussed in a separate systematic review that is currently in progress. All references were imported in EndNote (version X9.3.3; Clarivate, London, United Kingdom) and duplicates were removed. Reference lists and citing articles of identified relevant papers were checked for additional relevant studies using Web of Science.

Protocol deviations

For practical reasons, a number of deviations from the published PROSPERO protocol were necessary. We did not search the Cochrane Central Register of Controlled Trials (CENTRAL), which only contains controlled clinical trials and mainly has added value as a source for gray literature. Because we expected to find evidence from observational studies only and we excluded conference abstracts and other gray literature, we limited our search to MEDLINE and Embase. In addition, none of the relevant studies that were identified by cited reference searching before constructing the actual search were found in CENTRAL. We aimed to ensure search comprehensiveness by repeating cited reference searching after search completion. Furthermore, we decided not to use the Risk of Bias in Non-randomized Studies of Exposure (ROBINS-E) tool because it had unfortunately remained “under development” during the entire review process.

Study selection

Two reviewers (K.N. and L.J.) independently screened titles and abstracts using Rayyan (Rayyan System Inc., Cambridge, MA), after which potentially eligible studies were obtained in full text. Two reviewers (K.N. and L.J.) independently performed a second eligibility check for studies in the full text. Any disagreements were discussed until consensus was reached or, if necessary, a third reviewer was consulted (R.P.).

Eligibility criteria were:

• •
  

  Studies reporting on long-term health outcomes of offspring born to mothers with severe nausea and vomiting in pregnancy (NVP) or HG, as reported by the authors.

  
  
• •
  

  Long-term health outcomes included: general health, growth development, cardiometabolic outcomes, cognitive development, behavioral development, neurodevelopment, mental health, and cancer.

  
  
• 
  

  Exclusion criteria were:

  
  
• •
  

  Case reports, case series, letters to the editor, conference abstracts, and reviews.

  
  
• •
  

  Studies not reporting a control group, unless offspring’s long-term health across the disease spectrum of severity of HG was assessed.

Data extraction

A piloted data extraction form was used to extract data by 1 reviewer (K.N.), which was critically appraised by a second reviewer (L.J.). Any disagreements were solved by consensus, and in case of persistent disagreement, a third reviewer was consulted (R.P.). Authors were contacted by email if data were unclear or missing.

Assessment of risk of bias

A quality assessment was performed independently by 2 reviewers (K.N. and L.J.). Any disagreements were discussed until consensus was reached. The Newcastle-Ottawa Scale was used, which consists of 8 questions with a maximum score of 9. Studies scoring ≥7 were considered as good-quality, ≥5 as fair-quality, and ≤4 as poor-quality. Low quality on assessment was not a reason for exclusion.

Data synthesis

Findings were described by meta-analysis, where sufficient data were available, or otherwise described narratively. In the meta-analyses, data were presented as odds ratios (ORs) with corresponding 95% confidence intervals (CIs). Random-effects models according to the Mantel–Haenszel method were used on the basis of anticipated heterogeneity. We used I ² statistics to assess heterogeneity, with I ² values >75% considered as high heterogeneity. In those cases, we did not provide a pooled OR and we performed a sensitivity analysis, if possible. We also assessed heterogeneity by calculating 95% prediction intervals (PIs) of pooled ORs of meta-analyses that included at least 3 studies, to give an estimate of an interval in which 95% of effects might be found in future, comparable studies. We assessed publication bias by analyzing funnel plots if at least 10 studies reported the same outcome, according to the Cochrane Handbook for Systematic Reviews of Interventions. P values below.05 were considered statistically significant. Review Manager (RevMan, version 5.4; Cochrane, London, United Kingdom) was used to conduct meta-analyses.

Strength of the evidence

Strength of the evidence of all meta-analyses was assessed by the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) method according to the GRADE handbook. Two reviewers (K.N. and L.J.) independently graded evidence using GRADEpro (Evidence Prime Inc., Ontario, Canada). Because only observational studies were included, the initial quality of evidence started at very low and was upgraded if there were: (1) a large magnitude of effect (1 level up in case of a relative risk [RR] <0.5 or >2; 2 levels up in case of a RR <0.2 or >5), (2) signs of a dose–response relationship, or (3) plausible residual confounding. Evidence was rated as very low, low, moderate, or high quality.

Study selection

Our search identified 1360 unique studies. Nineteen studies were considered eligible and included in this systematic review, as shown in Figure 1 . ^,

PRISMA flow diagram: selection process of articles

HG , hyperemesis gravidarum; NVP , nausea and vomiting in pregnancy.

Nijsten. Hyperemesis gravidarum and children’s health. Am J Obstet Gynecol 2022 .

Study characteristics

Characteristics of included studies are summarized in Table 1 . Two studies of Fejzo et al ^, were follow-up survey studies of children of the same population at the ages of 8 and 12, respectively. Both studies were included in this systematic review, but only the latter was included in meta-analysis on the basis of our ability to produce 2×2 tables and because of a longer follow-up period. Furthermore, Wang et al validated their results from an American cohort in a different, Danish cohort, thus these 2 study populations were described separately in this systematic review.

Of the studies included, 12 were cohort studies and 8 case–control studies. HG diagnosis was based on self-reports in 7 studies, ^{, , , , , ,} interviews in 7 studies, ^{, , , , , ,} International Classification of Diseases (ICD) codes in 5 studies, ^{, ,} and medical records in 1 study. In total, 1,814,785 children were included in this systematic review, of whom 36,546 were born to women who experienced HG.

Risk of bias of included studies

Four studies (3 cohort and 1 case–control study) were rated as poor-quality, as shown in Table 2 . ^{, , ,} These studies scored low on either the selection domain, because of being self-selected survey studies, and/or on the comparability domain, because of not adjusting for confounders in statistical analyses. Three studies were rated as fair-quality (2 cohort and 1 case–control study). ^{, ,} Twelve studies, including 6 cohort and 6 case control studies, were rated as good-quality. ^{, , , , ,} We were not able to conduct funnel plots to assess publication bias because of the low number of studies included in meta-analyses.

Anthropometry

Five studies including 14,423 children reported on anthropometry measures, as shown in Supplemental Table 1 . ^{, , , ,} Four studies reported on height. ^{, , ,} Two studies reported on boys and girls separately (n=3114 children) and found contradictory findings: the first study found that girls but not boys from mothers with HG were taller at ages of 12, 18, and 24 months, whereas the second study found that boys but not girls of mothers with HG were taller at 72 months (adjusted β, 0.64 SD; 95% CI, 0.23–1.04). Two studies (n=6540 children) that analyzed boys and girls together did not find any differences in height at ages 4 to 11 and 16. ^,

Three studies reported on weight. ^{, ,} The 2 studies that evaluated boys and girls separately (n=3114 children) again showed conflicting findings: the first study showed that girls but not boys exposed to HG were heavier at 12, 18, and 24 months, whereas the other study found that exposed girls were lighter (adjusted β, −0.53 SD; 95% CI, −1.03 to −0.03) and boys heavier at 5 years (adjusted β, 0.57 SD; 95% CI, 0.05–1.08). The third study that evaluated weight in boys and girls (n=4760 children) found that HG-exposed children weighted more than nonexposed children at age 2 (9838±1712 vs 9581±1440 g).

Four studies reported on children’s body mass index (BMI). One study (n=4760 children) found higher BMI among 6-year-old HG-exposed children than among nonexposed children (male and female: adjusted β, 0.08; 95% CI, 0.00–0.17), whereas another study (n=1172 children) found lower BMI among 66-month-old female offspring exposed to HG (adjusted β, −0.57 SD; 95% CI, −1.09 to −0.05) but no differences among male offspring. The 2 other studies did not find differences in BMI among HG-exposed and nonexposed children at 4 to 11 and 16 years (n=6540 children). ^,

Lastly, 1 study including 4760 children found that, at age 6, those born to women with HG had higher total body fat mass (adjusted β, 0.12; 95% CI, 0.03–0.20), higher android/gynoid fat mass ratio (adjusted β , 0.11; 95% CI, 0.02–0.21), and higher abdominal preperitoneal fat mass area (adjusted β, 0.10; 95% CI, 0.00–0.20) than those born to women without HG, whereas 2 other studies (n=6540 children) did not find any differences in waist to hip ratio, android to gynoid fat mass ratio, or total body fat percentage at ages 4 to 11 and 16. ^,

Cardiometabolic health

Neurodevelopment and mental health

Mental health

All 8 included studies reporting on mental health found an increased risk in HG-exposed children compared with nonexposed children, as narratively summarized in Supplemental Table 3 . A meta-analysis including 2 studies (n=1,109,629 offspring, including children aged up to 18 and adults) that reported on emotional problems found no significant association (OR, 1.19; 95% CI, 0.89–1.61; I ² , 0%), as shown in Figure 2 (GRADE level of evidence: very low quality; Supplemental Table 4 ). ^, One study assessed affective problems as a subdomain of the Child Behavior Checklist (CBCL) and found higher scores among HG-exposed than nonexposed children at 2 and 4 years after adjustments for confounding factors.

• 
  

  Four studies (n=10,490 offspring varying in age from 4 years to adulthood) found that HG-exposed offspring were at higher risk of anxiety than nonexposed offspring on the basis of CBCL subdomains ^, and self-reported symptoms. ^, When pooling these last 2 studies in meta-analysis, we found that HG exposure was significantly associated with anxiety disorder in offspring (OR, 1.74; 95% CI, 1.04–2.91; I ² , 0%) ( Figure 2 ) (GRADE level of evidence: very low quality; Supplemental Table 4 ). Two studies (n=9473 offspring) found a significant association between HG exposure and CBCL depression scores at ages 9 to 11, ³⁴ and depression rates in adulthood (unadjusted OR, 6.35). The latter of the 2 also found that bipolar disorder occurred more often in HG-exposed offspring (unadjusted OR, 4.90).

  
  
• 
  

  Two studies (n=778 children, aged 8–9 and 11–13), which were likely to partially include the same children because of including follow-ups of the same original study population, found higher rates of social developmental delay or social anxiety in HG-exposed children than in nonexposed children (unadjusted OR, 5.02 and 3.58). ^,

Meta-analyses of neurodevelopment and mental health outcomes

Nijsten. Hyperemesis gravidarum and children’s health. Am J Obstet Gynecol 2022 .

The effect of hyperemesis gravidarum severity and treatment on neurodevelopment and mental health in offspring

Few studies also assessed differences in children’s neurobehavioral outcomes across the disease spectrum of HG. One study (n=819 offspring with a mean age of 32) found no significant differences in autism, behavioral, emotional, sensory, and learning disorders between offspring of HG patients with and without severe weight loss (>15% of prepregnancy weight). The second study (n=418 children, aged 8–9) found that the presence of early HG symptoms (below 5 weeks’ gestation) was significantly associated with neurodevelopmental delay in children, but found no differences between in- and outpatient care or between different medications and treatments. Another study (n=1172 children, aged 1–4.5) found that severe NVP without admission was more often associated with reduced neurobehavioral development in children than severe NVP with admission.

Two studies assessed associations between ASD rates and HG severity and treatments. The first study (n=360 children, aged 11–13) did not find any differences related to different treatments (antiemetics or tube feeding), in- or outpatient care, and early onset of symptoms (below 5 weeks’ gestation). The second study (n=469,789 children, aged 2–17) found a higher ASD incidence rate among children of women diagnosed with HG in the first (adjusted HR, 1.58; 95% CI, 1.40–1.79; P <.001) or second trimester (adjusted HR, 1.36; 95% CI, 1.05–1.75; P= .02) and children of HG patients with metabolic disturbances that required rehydration or tube feeding (adjusted HR, 1.41; 95% CI, 1.05–1.90; P =.02). No differences were found in ASD rates between early and late maternal hospitalization for HG.

Brain morphology and its association with cognitive and psychological symptoms in children

One study (n=10,710 children, aged 9–11) assessed brain morphology in children and measured the cortical area, volume, and thickness of each brain region. They found that the total cortical volume and area were significantly smaller in children exposed to severe NVP, and that these smaller volumes and areas mediated associations between severe NVP exposure and cognitive and psychological symptoms.

Cancer risk

Data on offspring’s cancer risk was available in 5 studies (n=173,502 offspring, aged up to 40). ^{, , , ,} Four studies reported solely on testicular cancer and assessed whether HG during pregnancy was a risk factor for the disease. ^{, , ,} The fifth study reported whether HG was a risk factor for developing multiple types of childhood cancer, including testicular cancer. Meta-analysis including these 5 studies showed that HG was significantly associated with testicular cancer in male offspring (OR, 1.60; 95% CI, 1.07–2.39; I ² , 0%; 95% PI, 0.83–3.08), as shown in Figure 3 (GRADE level of evidence: very low quality) ( Supplemental Table 4 ). Sensitivity analyses showed that a significant association between HG exposure and testicular cancer in male offspring only remained after omitting the study of Vandraas et al ( Supplemental Figures 3–7 ), which means that this study had a large impact on the results. The study of Vandraas et al is a large case–control study (n=162,514 offspring) that did not find a significant association between HG exposure and testicular cancer in offspring.

Meta-analysis of testicular cancer

Nijsten. Hyperemesis gravidarum and children’s health. Am J Obstet Gynecol 2022 .

No association was found between HG and multiple childhood cancers in 1 study including 162,514 offspring. However, when dividing their study population into offspring aged 1 to 10 years and 10 to 20 years, HG was significantly associated with lymphoma in offspring aged 10 to 20 years (adjusted RR, 2.08; 95% CI, 1.11–3.90).

Main findings

Our systematic review identified 19 studies reporting on children’s long-term health after HG exposure in utero. We were able to conduct meta-analyses for 7 outcomes and found that maternal HG was significantly associated with anxiety disorder and sleep problems in offspring, whereas no associations were found with emotional disorders and autism. Maternal HG was also significantly associated with testicular cancer in male offspring in meta-analyses, although there were signs of heterogeneity based on the 95% PI, which is considerably wider than the CI and which crosses 1. All studies reporting on ADD/ADHD and ASD found an increased risk among children of mothers with HG, but showed high heterogeneity when pooling these results in meta-analysis. Narrative synthesis showed that most studies reported an increase in cognitive and motor problems in children of mothers with HG. One study showed that children’s brain morphology was affected, with children exposed to severe NVP having smaller total cortical volume and area than nonexposed children. Inconsistent associations were found between HG exposure and anthropometry and cardiometabolic disease risk markers. Studies reporting on the effect of HG disease severity or treatment on children’s long-term health had inconsistent findings. All evidence included in this systematic review is based on a small number of studies with evidence of meta-analyses graded to be of very low to low quality.

Strengths and limitations

This systematic review has several strengths. A research protocol was published online before conducting this review. We were able to include studies on multiple long-term health domains, including cardiometabolic health, neurodevelopment, mental health, and cancer risk. We also managed to collect additional data from corresponding authors of included studies to extend our meta-analysis. Moreover, two-thirds of included studies were rated as fair- to good-quality, and we included a relatively large group of offspring that was exposed to HG.

Limitations of this review include heterogeneity in reported outcomes and HG diagnosis. Heterogeneity in HG diagnoses is a problem that has been previously identified in a systematic review that concluded that only 2 criteria (nausea and vomiting) were commonly used in diagnosing HG. Heterogeneity led to variances in HG rates among the included studies in our review: cohort studies diagnosing HG on the basis of self-reports or interviews reported rates between 9.7% and 21.9%, whereas cohort studies including HG patients on the basis of ICD codes reported much lower rates varying from 0.6% to 3.1%. Although ICD codes have been shown to be only valid for diagnosing women with mild HG, they were used to diagnose HG patients in the 4 largest studies included in this review. ^, Our results could have been influenced by the inclusion of mainly case-control studies, which are prone to recall bias, and by the use of different methods of disease ascertainment, which may have led to over- and underestimation of HG or disease severity.

In addition, studies reported on many different long-term health outcomes and used different methods to ascertain these. Therefore, we were not able to aggregate results in meta-analysis for most outcomes, and our results were mostly presented using narrative synthesis. Moreover, evidence of outcomes that were assessed by meta-analyses was graded as very low or low quality. Furthermore, the Diagnostic and Statistical Manual of Mental Disorders was updated in 2013, combining several disorders, including autism disorder, into 1 diagnosis: ASD. This hampered us in conducting meta-analysis for ASD and autism combined because both outcomes are used interchangeably in current literature. Lastly, some of the studies included self-selected participants that self-reported offspring outcomes, plausibly leading to higher rates of adverse health outcomes and possible recall bias. ^{, ,}

Comparison with existing literature

Our systematic review and meta-analysis showed that HG exposure is associated with neurodevelopmental and mental health problems and a decrease in brain volume in offspring. Thus far, only few studies have evaluated the effect of maternal undernutrition during pregnancy on children’s neurodevelopment and mental health. Research from the Dutch famine found that in-utero exposure to undernutrition led to an increased risk for developing schizoid or antisocial personality disorders, ^, whereas animal studies found that maternal protein restriction during pregnancy had a negative effect on anxiety and cognitive behavior in offspring. Different mechanisms can be hypothesized to be the underlying cause, for example when interpreting our results in the context of ASD research. A recent systematic review including 36 studies found that an appropriate intake of folic acid and vitamin D could protect against ASD in offspring. Conversely, this could mean that a reduced availability of these nutrients, potentially caused by HG, may lead to an increased ASD risk in children. Another recent systematic review and meta-analysis demonstrated that small-for-gestational-age (SGA) neonates are at increased risk of developing ASD. Given that previous research showed that HG children are more often born SGA, an alternative explanation could be that the risk of HG-exposed children developing ASD originates in fetal growth restriction. Genetics also play an important role—a large multicountry cohort study found a high heritability rate for ASD of 80%. However, no evidence of genetic factors being involved was found in a self-reported survey study comparing HG-exposed offspring with unexposed siblings and offspring without family history of HG.

We did not find evidence of a consistent effect of HG on offspring anthropometry and cardiometabolic disease risk markers. These findings are at odds with the large body of evidence on effects of maternal undernutrition in pregnancy. Animal experiments have linked intrauterine undernutrition to adverse health outcomes in offspring in later life. ^, In addition, studies about the Dutch, Chinese, and Nigerian famines showed that adults exposed to maternal undernutrition in utero more often had type II diabetes mellitus, hypertension, overweight, and coronary heart diseases in later life. ^{, , ,} This may be explained by the fact that the studies in our systematic review included children of young ages, which precludes any definitive statement about the possible effects HG may have on adult offspring. Alternatively, it could be that women included in the studies had relatively mild HG and that only women with severe HG are comparable with pregnant women during famines in terms of maternal undernutrition. Finally, it is possible that HG, despite incurring maternal undernutrition, simply does not have effects on cardiometabolic health in the next generation. Moreover, because of the presence of age-specific reference intervals for anthropometry and cardiometabolic laboratory measures, we were not able to pool results in meta-analysis. More research is warranted to draw firm conclusions about possible adverse cardiometabolic disease risks for HG-exposed children.

HG was associated with testicular cancer in male offspring, with an OR of 1.60, which has been previously hypothesized to be caused by increased estrogen levels during pregnancies complicated by HG. ^, Nonetheless, these results should be interpreted with caution. Although there were no signs of heterogeneity on the basis of the I ² of 0, the 95% PI suggested otherwise, estimating that in 95% of subsequent studies the true effect estimate could range from 0.83 to 3.08. In addition, only a small number of HG cases were included in most studies, and in 4 out of 5 studies HG was assessed by questionnaires or interviews long after the pregnancy occurred, which could have led to recall bias. ^{, , ,} Importantly, the fifth and most recent large study diagnosed HG on the basis of ICD codes and did not find an association between HG and testicular cancer in offspring.

We were unable to assess whether maternal treatment interventions can have a preventive role in avoiding health sequelae among the offspring. Only 3 studies that assessed neurodevelopment and mental health, but none of the studies on other long-term health conditions, evaluated the association with treatment interventions and showed variations in results. Sensitivity analysis showed that HG was associated with more offspring ASD only in studies from the United States, but not in Denmark. This might indicate that differences in treatment or accessibility to health care between the United States and Denmark alter offspring’s ASD disease risk, which is an important matter to be addressed in future research because it has direct implications for healthcare policy choices. Lastly, yet importantly, HG was not associated with more offspring ASD when excluding studies with a poorer study design.

Conclusions and implications

This systematic review and meta-analysis showed that there is an increased adverse long-term health risk for children exposed to HG during gestation in terms of neurodevelopmental problems and mental well-being, whereas the impact on cardiometabolic disease risk and testicular cancer remains unclear. Conclusions are based, however, on a small number of studies, with evidence from meta-analyses being graded as very low to low quality. Altogether, long-term health research in HG is still in its infancy, and more research with long-term follow-up is needed to determine the pathophysiology of these associations and to further explore the role of (early) treatment to reduce adverse long-term health effects in offspring.