Maternal prenatal infection and anxiety predict neurodevelopmental outcomes in middle childhood.

Child Neurodevelopment and Psychiatric Symptoms

Several measures were collected to assess neurodevelopment and psychiatric symptoms when the children were approximately 8 years of age. The full-scale intelligence quotient (IQ) from the Wechsler Intelligence Scale for Children (WISC III, 3rd U.K. edition; Weschler et al., 1992 ) was administered at in-person assessment that took place at a half-day clinic. Except for the coding subtest, which was administered in the standard form, administration of the WISC involved alternate questions for every subtest, starting with the first item, to shorten the test. Social and communication problems associated with autism were assessed using the Social and Communication Disorders Checklist (SCDC; Skuse et al., 2005 ), a 12-item maternal report screening measures that assesses autistic-like traits, including problems in social reciprocity and verbal/nonverbal communication (sample items include, “Not aware of other people’s feelings”; “Does not pick up on body language”); the measure has been used widely in other research and has demonstrated high reliability and validity ( St Pourcain et al., 2018 ); internal consistency (alpha) was .88. Higher SCDC scores reflect more social-communication deficits.

Symptoms of ADHD, anxiety, depression, and disruptive behavior were derived from the Development and Well-Being Assessment (DAWBA; Goodman et al., 2000 ), a validated psychiatric assessment tool widely used in epidemiological studies ( Ford et al., 2003 ). The parent-completed DAWBA questionnaire constituted the primary source of information for the diagnosis of psychiatric disorders. In addition, where available, complementary sources of information on a child's emotional, behavioral, and academic characteristics were consulted, which included teacher version of the DAWBA that addressed potential hyperactivity and conduct disorder, teacher-reported Strengths and Difficulties Questionnaire (SDQ), and test results from reading and spelling competency. Number of symptoms for each disorder was recorded and diagnostic assessments were coded according to DSM–IV criteria by two experienced psychiatrists taking into account symptoms and burden or impact.

Covariates and Potential Effect Modifiers

Maternal education and household crowding at time of pregnancy were used as measures of socioeconomic status. Women endorsed their highest educational achievement within the U.K. educational system according to four categories (1 = Certificate of Secondary Education/vocational training; 2 = O-levels, equivalent to modern day General Certificate of Secondary Education ; 3 = A-levels, comparable to college entrance examinations ; 4 = university degree or higher degree ). Household crowding was calculated by dividing the number of people in the household by the number of rooms; four categories were created ranging from low to high (1 = 0–.50 ; 2 = .50–.75 ; 3 = .75–1.00 ; 4 = > 1.00 ). Mothers reported whether or not they were ever diagnosed with a severe mental illness (yes/no). Respondents also provided information on medication for prenatal depression and anxiety. Prenatal and perinatal covariates include parity, prenatal smoking (defined as the number of cigarettes smoked in the 2 weeks prior to assessment) and prenatal alcohol use (defined as the number of alcohol units in the week prior to assessment); prepregnancy BMI based on self-reported weight and height; gestational age (in weeks) and birth weight (in grams) extracted from medical records; maternal age in years was recorded at the completion of the initial prenatal questionnaire. Gestational age at the completion of the prenatal questionnaire, child sex, and child age (in months) at the time of testing were also considered as possible covariates.

Data Analysis

The main analyses, which appear after descriptive data, were based on a definition of prenatal infection from the composite measure of symptoms indicating active systemic infection (diarrhea, urinary infection, influenza/flu, thrush/candida, and herpes); number of infections were categorized as 0, 1, 2, and 3 or more. Child outcomes were cognitive ability based on in-person IQ testing; social and communication problems associated with autism from parent report; and number of psychiatric symptoms from the diagnostic interview. Hypothesis testing was based on generalized linear models ( McCullagh & Nelder, 1998 ), which have several advantages over ordinary linear regression; for example, error distributions are not considered normal, response variables are not required to have normal distributions, parameters are assessed according to maximum likelihood estimation. B coefficients using the existing scaling of the predictor variables were first estimated; in order to obtain beta coefficients, the models were reanalyzed using standardized variables. Generalized linear models for hypotheses were conducted in SPSS (Version 26). Based on the existing literature and available data, we included the following covariates on an a priori basis in all prediction models: maternal age; prepregnancy BMI because it has been considered as an index of maternal prenatal inflammation; birth weight and gestational age because they are measures of perinatal risk that may be associated with prenatal exposures; prenatal smoking because it is an additional index of prenatal risk exposure that may be confounded with target prenatal exposures and child outcomes; maternal education and household crowding because they provide a measure of socioeconomic status (and household crowding may also index prenatal illness exposure); postnatal maternal anxiety was included to test the specific prediction from prenatal anxiety; child age at assessment was included to adjust for variation in age at assessment; child sex was included because it is associated with many neurodevelopmental outcomes. None of the other covariates listed above were retained in prediction models because none was reliably associated with child outcomes. Sensitivity analyses considered alternative scaling and measurement of prenatal infection. A series of exploratory analyses examined moderation of effects by child sex and moderation of effects between the two target prenatal exposures; additionally, although our focus is on prenatal anxiety, we also examined prenatal depression in supplementary analyses. Given the large sample size, we attended to effect sizes for interpretation.

Results

Descriptive and demographic data on the sample are provided in Table 1 ( n = 7,042 pregnancies). Comparisons between those with no versus those with any infection identify a number of differences. Using Cohen’s d as a measure of effect size ([ M 2 - M 1 ]/ SD pooled ), the most notable difference, .37, was for prenatal anxiety; smaller (<.15) but notable differences were detected for smoking and prepregnancy BMI. Of the 7,042 women in this study, 32.7% reported one infection in pregnancy, 8.2% reported two infections, and 1.4% reported three or more infections; that is, 42.3% reported any infection in pregnancy. Self-reported rates of all types of infection varied: 6% for influenza, 5% for urinary infection, 29% for diarrhea, 13% for thrush/candida, and .3% for herpes. The broader category of infection, which also included vaginal bleeding and nonspecified “other infection” identified 46%. Rates of child psychiatrist-confirmed disorders were low (2% for any ADHD disorder; 3.1% for any oppositional-conduct disorder; 2.9% with any anxiety disorder; .4% for any depressive disorder, and .4% for any pervasive developmental disorder). Accordingly, prediction analyses focus on number of psychiatric symptoms rather than the presence/absence of disorder.

Of the 7,042 participants identified for the analyses, the rate of missing data was <5% on covariates and predictors, with two exceptions: prepregnancy BMI had a missing data rate greater than 5% (6,506 cases were available from the 7,042) and in-person IQ data were available on 5,213/7,042. There were 6,800 children with data on the SCDC and 6,866 with data on the DAWBA.

Bivariate correlations, displayed in Table 2 , indicate modest overlap between prenatal infection, prenatal anxiety, smoking, and sociodemographic risk. There was a modest positive association between prenatal infection and prenatal anxiety, as shown in the correlation table and according to the means when assessed for 0–4 infections: means ( SD ) of prenatal anxiety according to number of infections were 4.32 (3.23) for zero infections, 5.41 (3.53) for one infection, 5.99 (3.70) for two infections, and 7.73 (4.20) for three or more infections; F (3, 6833) = 96.57, p < .001). A similar pattern was obtained for prenatal depression: the means ( SD ) of EPDS according to number of infections were 5.98 (4.65) for zero infections, 7.25 (4.88) for one infection, 8.08 (5.11) for two infections, and 9.98 (6.07) for three or more infections; F (3, 6980) = 71.01, p < .001). In addition, prenatal infection and prenatal anxiety were also modestly associated with prenatal smoking.

Prenatal Prediction of Child IQ

Bivariate analyses indicated that child IQ, based on the full-scale WISC, showed a dose–response pattern of association with number of maternal prenatal infections, that is, each increase in infection was associated with lower IQ (means [ SD ] were 106.21 [16.21] for zero infections, 105.00 [16.52] for one infection, 104.40 [15.51] for two infections, and 100.11 [15.12] for three or more infections; F (3, 5209) = 5.46, p < .001); similar dose–response patterns were obtained for the Verbal IQ (means [ SD ] were 109.08 [16.73] for zero infections, 107.76 [16.64] for one infection, 106.96 [16.10] for two infections, and 102.60 [14.01] for three or more infections; F (3, 5239) = 6.24, p < .001) and for Performance IQ (means [ SD ] were 101.28 [16.76] for zero infections, 100.48 [17.24] for one infection, 100.22 [16.43] for two infections, and 97.06 [16.15] for three or more infections; F (3, 5228) = 2.23, p = .08). The bivariate association between full-scale IQ and prenatal maternal anxiety was small in magnitude [ r (5213) = −.07, p < .001]. However, in the regression model that adjusted for covariates, neither prenatal infection nor prenatal anxiety was significantly associated with child IQ ( Table S1 in the online supplementary materials).

Prenatal Prediction of Social and Communication Deficits

As shown in Figure 1 , there was evidence of a dose–response pattern between prenatal infection and problem severity on the SCDC: mean problem severity on the SCDC increased with each prenatal infection reported (means [ SD ] were 2.53 [3.39] for zero infections; 3.01 [3.95] for one infection; 3.10 [3.98] for two infections; 3.41 [4.29] for three or more infections; F (3, 6796) = 11.37, p < .001). Bivariate correlation analysis indicated that problems on the SCDC were significantly and modestly associated with prenatal anxiety [ r (6769) = .15, p < .001].

Note . See the online article for the color version of this figure.

Regression analyses predicting continuous scores on the SCDC from prenatal exposures and covariates are presented in Table 3 . Results indicate that both prenatal infection (scored 0 to 3 or more) and prenatal anxiety were both significantly associated with greater disturbance in social and communication problems after adjusting for covariates. The (adjusted) increase in score between none and 3 or more infections (i.e., .4) is small, but notable in this nonselected community sample.

Sensitivity analyses indicated that the prediction of SCDC was robust across alternative measures of prenatal infection. For example, the regression coefficient for none versus any infections was .30 ( SE .10; p < .01). Similar effects, in terms of parameter estimates, were also found when specific infections were assessed separately, such as urinary tract infection (B .31 SE .21, p = .13) and influenza (B .42 SE .20, p = .04; the SE for these analyses were correspondingly higher given the smaller number of cases). Further sensitivity analyses indicated that including child full-scale IQ reduced, but only modestly, the effects of prenatal infection or prenatal anxiety on SCDC (for prenatal anxiety, the likelihood ratio chi-square was 16.99, p < .001 vs. 11.58, p < .001 for the model that included IQ as a covariate; for prenatal infection, the likelihood ratio chi-square was 9.89, p = .019 vs. 7.51, p = .057 for the model that included IQ as a covariate). Additionally, although there was a sizable main effect of sex on SCDC, there was not significant evidence that the prediction of SCDC differed significantly by sex for either prenatal exposure ( p = .278 for the interaction between child sex and prenatal infection; p = .182 for the interaction between child sex and prenatal anxiety). There was also no evidence of a significant interaction between prenatal infection and prenatal anxiety in predicting SCDC: the interaction term was nonsignificant in the model ( p = .537); Figure S1 in the online supplementary material displays the additive effects of prenatal infection and prenatal anxiety on SCDC. Lastly, analyses that included prenatal depression rather than anxiety indicated substantively similar effects (see Table S2 in the online supplementary materials).

Prenatal Prediction of Psychiatric Symptoms

Regression analyses for psychiatric symptoms based on the DAWBA are reported in Table 4 . Results for symptoms of ADHD was strikingly similar to those for SCDC. That is, both prenatal infection and prenatal anxiety independently predicted ADHD symptoms after adjusting for covariates. Effect size estimates (adjusting for all covariates) indicated a difference of 1.35 symptoms of ADHD between those exposed to none versus three or more infections. A series of supplementary analyses, parallel to those for SCDC (above), confirmed the robustness of these results. For example, supplementary analyses indicated that the prediction from prenatal infection to ADHD symptoms was robust to alternative scaling of prenatal infection (e.g., the prediction from any infection was B = .73 [ SE .20], p < .001; for influenza only, B = 1.29 [ SE .37], p = .001). Furthermore, there was no evidence of a significant difference in prediction from either prenatal exposure by child sex ( p = .275 for the interaction between prenatal infection and child sex; p = .624 for the interaction between prenatal anxiety and child sex); neither was there evidence of a significant interaction between prenatal anxiety and prenatal infection in predicting ADHD symptoms ( p = .856 for the interaction).

Results from the same prediction model (see Table 4 ) indicate that prenatal maternal anxiety was a significant predictor of child anxiety and depressive symptoms, adjusting for covariates; the prediction of disruptive behavior was weaker and not significant at p < .05. In contrast, prenatal maternal infection was reliably ( p < .05) associated with child anxiety symptoms but only marginally with depressive and disruptive behavior symptoms. The effect size of prenatal prediction, in terms of the impact on numbers of symptoms, varied considerably. In the case of prenatal infections, the difference in child symptoms between those whose mothers reported no infections and those whose mothers reported three or more was 1.35 for ADHD but .29 for anxiety, and −.06 for disruptive behavior.

Discussion

Analyses of this prospective longitudinal study of a large community sample indicated that prenatal anxiety and infection were reliably, independently, and additively associated with ADHD symptoms and social and communication difficulties, two key markers of neurodevelopment in middle childhood. The robustness of the prediction is implied from a dose–response pattern across an 8-year period from exposure to outcome, detailed clinical assessment, consistency across multiple definitions of prenatal infection, large sample size, and adjustment for perinatal and postnatal confounds. An additional novel and notable finding is that the prediction from prenatal infection extended to child anxiety (and was marginal for depressive and disruptive behavior symptoms). The findings further substantiate a developmental and conceptual model of psychopathology in which clinical disturbances can be traced to prenatal exposures, and raise novel questions about mechanisms of action, and their timing of influence.

Research findings suggest that multiple types of prenatal exposure, from chemical and pollution agents and specific micronutrients to, as in the current study, prenatal infection and mood disturbance, are reliably associated with child health and development. A key conceptual point in these studies, and the DOHaD and MIA models with which they are affiliated, is that brain and behavioral development begin before birth and are susceptible to prenatal exposures. However, a key limitation of these studies is the focus on a singular prenatal exposure, which may mis-specify effects and implied mechanisms to the extent that different types of risk exposures are confounded—as is often found. Furthermore, many specific types of risks are inherently variegated. For example, studies of the Dutch Hunger Winter ( Schulz, 2010 ) focus on nutritional deprivation, but the impact of the exposure likely also derives from prenatal distress and changes in many kinds of health behaviors, exposures, and health care. Similarly, studies of pregnant women during the period affected by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic must account for increases in stress as well as the multiple kinds of changes in economic, nutritional, health care, and health behaviors that may affect the health of the mother, pregnancy, and child—rather than just one or other of these. To that end, we sought in the current article to examine two of the more prevalent and robust risk factors for child neurodevelopment that derive from separate research traditions, with disjunctive conceptual models and wholly separate empirical bases.

Much is known about how the health of the fetus may be disrupted by prenatal infection by viruses, bacteria, and parasites ( Adams Waldorf & McAdams, 2013 ). Importantly, there is growing recognition that prenatal infections that may shape fetal brain development extend beyond the classic TORCH agents ( Toxoplasma gondii , rubella, cytomegalovirus, herpes simplex virus) to a much broader array of agents associated with prenatal inflammation, which may be driven by psychological symptoms, stress, or health behaviors such as obesity. Links between inflammation and developmental and physical disability ( Grether & Nelson, 1997 ) and brain disorders ( Brown et al., 2005 ; Croen et al., 2019 ) have been noted for some time, but there has been limited attention to competing exposures and very few studies consider the broader phenotypes of neurodevelopment and mental health. In that context, the current study is unique in assessing a broad spectrum of neurodevelopmental and psychiatric conditions in young children from parent and clinician evaluation, variability in symptom expression (rather than disorder presence/absence), and consideration of covariation between prenatal infection and prenatal anxiety.

The current findings extend the MIA model by demonstrating dose–response patterns of outcomes across a continuous range of symptomatology. Specifically, maternal prenatal infection was associated with a continuous trait of social and communication problems assessed in a large community sample. The implication is that prenatal infection may have an even larger public and clinical health impact because of its influence on the broader (endo)phenotype underlying a clinical diagnosis of autism. A more novel prediction was to child anxiety symptoms. That is significant insofar as it implies that the psychiatric and behavioral phenotypes—and their underlying neurodevelopmental mechanisms—that may originate with prenatal infection may be broader than autism and schizophrenia, which have dominated this line of clinical research.

The findings suggest some specificity in effects. So, for example, the largely independent predictions from prenatal infection and anxiety imply distinct and specific underlying mechanisms: additive, independent prediction would not result if these exposures operate via wholly shared mechanisms. Our findings do not directly implicate inflammatory or stress physiology mechanisms, but they do indicate that prenatal infection and prenatal anxiety operate separately. That is a novel observation that requires detailed studies of maternal prenatal biology. Second, the predictions from prenatal infection were slightly less robust than those from prenatal anxiety, after adjusting for covariates (see Table 4 ). We also found a lack of significant prediction to disruptive behavior from prenatal infection, after adjusting for covariates; notably, there was also a weaker and nonsignificant prediction of disruptive behavior from prenatal anxiety (e.g., compared to other symptom clusters). Individual differences in disruptive behavior may simply be less reliably predicted from these prenatal exposures, or it may be that the prenatal prediction from these factors is confounded by covariates and postnatal exposures. Understanding why there may be differential prediction from prenatal e