Lower cognitive scores among toddlers in birth cohorts with acute respiratory illnesses, fevers, and laboratory‐confirmed influenza

Abstract Background We established cohorts to assess associations between viral influenza and cognitive development to inform the value proposition of vaccination. Methods From 2014 through 2017, we called women seeking care at four prenatal clinics in Panama and El Salvador to identify acute respiratory illnesses (ARIs). Within 2 weeks of childbirth, mothers were asked to enroll their neonates in the cognitive development study. Staff obtained nasopharyngeal swabs from children with febrile ARIs for real‐time reverse transcription polymerase chain reaction (rtPCR) detection of viral RNA. Toddlers were administered Bayley developmental tests at ages 12 and 18–24 months. We used multilevel linear regression to explore associations between Bayley scores, ARIs, fever, and laboratory‐confirmed influenza, controlling for maternal respiratory or Zika illnesses, infant influenza vaccination, birth during influenza epidemics, and the number of children in households. Results We enrolled 1567 neonates of which 68% (n = 1062) underwent developmental testing once and 40% (n = 623) twice. Children with previous ARIs scored an average of 3 points lower on their cognitive scores than children without ARIs (p = 0.001). Children with previous fevers scored an average of 2.1 points lower on their cognitive scores than afebrile children (p = 0.02). In the second year, children with previous laboratory‐confirmed influenza scored 4 points lower on their cognitive scores than children without influenza (p = 0.04, after controlling for first Bayley cognitive scores). Conclusions ARIs and fever during infancy were associated with lower Bayley scores at 12 months, and laboratory‐confirmed influenza was associated with lower cognitive scores at 24 months suggesting the potential value of vaccination to prevent non‐respiratory complications of influenza.

Conclusions: ARIs and fever during infancy were associated with lower Bayley scores at 12 months, and laboratory-confirmed influenza was associated with lower cognitive scores at 24 months suggesting the potential value of vaccination to prevent non-respiratory complications of influenza.  3 Neurologic complications typically occurred within 5 days from symptom onset and disproportionally among racial and ethnic minorities and children. 4,5 More than 20% of hospitalized young children with influenza had febrile seizures, compared with 2% to 5% of other children, double that of parainfluenza or adenovirus. 6 Pathogenesis of acute influenza-associated neurological complications is not well understood and may include temporary and/or subacute neurological changes precipitated by cytokine and immune dysregulation. 2,7,8 Mouse models suggest that influenza A virus illnesses may also be associated with long-term neuroinflammation and cognitive deficits. 9 Influenza infections in mice have been associated with recruitment and upregulation of astrocytes; dendritic blunting; remodeling of the hypothalamus responsible for motivation, emotion, learning, and memory; and poor performance on maze challenges. In other mouse models, neonatal influenza infection led to a proinflammatory cytokine and neural dysfunction. 10 Although such mouse models suggest proinflammatory changes were associated with acute, subacute, and long-term changes in rodent brains, the implications to humans remain unclear.
The few existing human birth cohort studies suggest febrile respiratory illnesses are associated with subacute changes in brain function. 11 A birth cohort in Bangladesh, for example, found an association between days of febrile illness, proinflammatory cytokines, and lower language and motor development scores during infancy, 12 but findings are difficult to interpret because of limited study duration and insufficient adjustment for factors associated with cognitive development. 13 It is unclear if studies that could better account for maternal and infancy factors would still find associations between febrile respiratory illnesses and delayed cognitive development. Even if such findings were replicable, it is unclear whether cognitive development would be specifically associated with vaccine preventable influenza. We believe it is important to identify potential associations between influenza and cognitive development because; if these exist, this might strengthen the value proposition of influenza vaccination among infants aged ≥6 months to prevent such cognitive delays in settings where respiratory illnesses alone seem insufficiently compelling to providers and parents to achieve high vaccination coverage. 14 We therefore established a cohort in Panama and El Salvador, at sites where we had completed an oseltamivir clinical trial among young children, 15  We used univariate (for the first Bayley score) and bivariate (for the second Bayley score, controlling for the first score) multilevel mixed effects regression models with individuals at the first level, clinics at the second level, and countries at the third level, to examine associations between cognitive Bayley test scores and maternal and infancy exposures frequently associated with development. Multilevel models were used to account for possible clustering by clinic and country. Maternal exposures included age, pre-existing conditions (Table 2), substance abuse, education, monthly household income (as a binary variable of being the lowest income quintile-<$US400 per month income-or not), 13 presence of fever or acute respiratory symptoms, laboratory-confirmed influenza, respiratory syncytial virus, parainfluenza viruses 1-3, adenoviruses, human metapneumovirus, rhinovirus and Zika. Infant exposures included race/ethnicity, premature birth, born during influenza season (defined as birth during epidemic weeks when the proportion of samples testing positive for influenza at the national reference laboratories where above the annual mean for >2 consecutive epidemic weeks 20 ), female sex, twin, weight, length and height for age, acute respiratory infection (as yes/no), total number of acute respiratory infections, hospitalization with bronchopneumonia, fever, influenza vaccination (none, partial, and full), and laboratory-confirmed influenza. We assumed mothers who did not report fever or respiratory symptoms did not have respiratory illnesses. We also assumed children with ARIs who were untested for respiratory viruses would have tested negative for influenza if these illnesses occurred during epidemiologic weeks without any influenza detections through national surveillance at reference laboratories in each country.
We used stepwise selection to build multilevel mixed effects linear regression models with individuals at the first level, clinics at the second level, and countries at the third level, with Bayley cognitive scores at ages 1 or 2 as dependent variables; variables that had univariate or bivariate associations of p ≤ 0.1 were considered for the model. We also tested variables that have been previously reported to influence cognitive development or been associated with risk of respiratory illnesses: maternal Zika illness and respiratory virus illnesses (tested as individual and combined variables), infant influenza vaccination, number of household members, children, and children aged <5 years, regardless of their significance in univariate or bivariate regressions. We excluded variables associated with study retention, and we repeated the analysis excluding twins. All analyses were done in STATA 15 (College Station, TX: To assess potential bias from missing data because of loss to follow-up, we used t tests and chi-squared tests to examine associations between baseline variables and participants who dropped out of the study. Variables associated with dropout and their potential to introduce bias are discussed in limitations. The study was funded through CDC cooperative agreement 5U01IP000791-05. The protocol was reviewed and approved by the  Children from Indigenous families were more likely to drop out before study completion (Table 1, p < 0.0001).

| Maternal characteristics
Most mothers in El Salvador described themselves as

| DISCUSSION
ARI and fever were associated with lower cognitive scores among children aged 12 months. The clinical significance of these lower cognitive scores is unclear and will be studied in a subsequent phase of the cohort. It is also unclear why ARI and fever were associated with lower cognitive scores, but it is possible that this finding might have been associated with increased proinflammatory interleukins levels during illness. 12 For example, it is possible that cognitive function among children with ARIs might have been affected by proinflammatory interleukins precipitating subsequent brain remodeling, as suggested by animal models infected with influenza viruses. 9,10 One might also speculate that ARIs led to Eustachian tube dysfunction, serous otitis, or other upper respiratory tract sequela that affected audition; while we did not perform audiometry, children with ARIs that occurred days before Bayley testing had similar cognitive scores to those with ARIs that occurred months before, when effects of upper respiratory tract dysfunction might have dissipated among most children.
Children with laboratory-confirmed influenza illnesses, at any time before their second Bayley, also scored lower in their second- year cognitive tests than children without influenza; these differences were statistically significant when controlling for the first cognitive test score and ARIs. Influenza triggers a unique cytokine response among respiratory viruses, 21 often manifests with fever among children, 22 and sporadically leads to encephalitis. 23  We identified maternal and childhood risk factors commonly associated with lower cognition, 13 but after accounting for country of origin, maternal and childhood risk factors, and/or baseline Bayley scores, only ARIs, fevers, and/or influenza illnesses were significantly associated with Bayley scores. Other household factors such as having mothers with elementary school education or less, negatively affected scores, but these differences did not remain significant in multivariate analyses. It is therefore possible that some associations were masked by ARI case status and difficult to independently identify because of limited sample size.
Our study had strengths and limitations. Unlike previous cohorts, we carefully followed mothers during pregnancy, recorded gestational age using ultrasound, and swabbed ARIs. We then followed children for 2 rather than 1 year and documented findings following STROBE guidelines. 12 However, our study had important limitations. First, we had unexpectedly high losses to follow-up from enrollment of a very mobile population, missed swabbing opportunities, and potentially misclassified influenza case status because of passive follow-up and an insensitive ARI case definition for testing. 22 Several baseline variables were associated with loss to follow-up, notably Mestizo race/ ethnicity and poor economic status were associated with study retention; these variables were also associated with first or second cognitive scores in univariate/bivariate models and excluded from multivariate models. These variables will be important to test in future studies. While Bayley III scores were internally consistent (coefficients ranged from 0.86 to 0.91), it is possible that the instrument imperfectly assessed cognitive development among some racial and/or ethnic minority 29 children or whose households were bilingual. 30 Although we initially planned to obtain blood work to document proinflammatory marker levels, 12 we did not have the resources to do so or to systematically seek to identify encephalitis, electroencephalograms, and audiology assessments from ill children to explore mechanisms of action and identify ARI sequelae. 23 We also did not have the resources to systematically document neurodevelopmental diagnoses, maternal depression, postnatal exposure to violence, home stimulation, micromalnutrition, toxicants, concomitant illnesses, or visual and/or auditory impairment 31 that might have been attributed to lower cognitive scores but hope to do so in our future studies. Last, while our findings seem consistent with those of our cohorts in Asia, our results may not be generalizable to other populations.

| CONCLUSION
ARIs and fevers during infancy were associated with lower Bayley cognitive scores. Children with laboratory-confirmed influenza illnesses during the first year of life scored lower in their second cognitive tests than those without influenza, but this difference was not statistically significant. Children who had laboratory-confirmed influenza before their second year of life, however, had significantly lower cognitive scores when controlling for ARIs and country. It is unclear if children with lower cognitive scores catch up to their peers after 24 months of age. Larger and longer duration cohorts might be warranted to further explore associations between influenza and cognitive development, identify lasting impact beyond the age of 2 years, explore mechanisms of action, and determine the potential value of vaccines, antipyretics, and antiviral treatments in mitigation. Panama and El Salvador might benefit from program evaluations to optimize vaccination coverage and empiric ARI treatment with antivirals among very young children.

ACKNOWLEDGMENT
The study was funded through Centers for Disease Control and Prevention cooperative agreement 5U01IP000791-05.