Autism Spectrum Disorder Diagnoses and Congenital Cytomegalovirus Free

We used 2014 to 2020 administrative data from the Transformed Medicaid Statistical Information System Analytic Files database from the Centers for Medicare and Medicaid Services, which includes information on enrollment, inpatient and outpatient services, and outpatient drug claims. Deidentified data on >100 million Americans, including children and pregnant women, enrolled in state Medicaid or Children’s Health Insurance Program (CHIP) plans are included in the database. The use of this database was determined not to require institutional review board approval.

Our study population consisted of children who were continuously enrolled in Medicaid or CHIP from birth through ≥4 to <7 years. Children with cCMV were identified by the presence of an International Classification of Diseases, Ninth or Tenth Revisions, Clinical Modification diagnostic code for cCMV infection (771.1, P35.1) or cytomegaloviral disease (078.5, B25.×) recorded within 45 days of birth, as in our previous studies.^28 –30  ASD was defined as 2 or more inpatient or outpatient visits with diagnostic codes for ASD (299.0×, 299.8×, 299.9×, F84.0, F84.5, F84.8, F84.9) ≥6 days apart at a minimum age of 1 year (Supplemental Table 4).³¹  The date of the first ASD diagnostic code was used at the time of initial ASD diagnosis.

We assessed the presence of cooccurring diagnoses that have been reported to be associated with cCMV and ASD, including preterm birth, LBW, hearing loss, and CNS anomaly or injury defined as diagnoses of brain anomaly, microcephaly within 45 days of birth, cerebral palsy, epilepsy, or chorioretinitis (Supplemental Table 4).^32 –35 

We calculated the prevalence and 95% confidence intervals (CIs) of cCMV and ASD by sex, region, birth outcomes (preterm birth and LBW, preterm birth only, LBW only, or neither), the presence of individual or collective CNS anomaly or injury codes, and hearing loss.

We used Cox proportional hazards regression models to estimate hazard ratios (HRs) and 95% CIs for ASD in children with and without cCMV, overall, adjusted for birth year, region, and sex. We also conducted stratified analyses by sex, combinations of preterm birth and LBW, and presence of any CNS anomaly or injury to assess potential differences by these characteristics. We elected not to adjust for race and ethnicity in our primary analysis because of missingness of this variable (24%) but included it in a complete-case sensitivity analysis with the 76% of the sample with non-missing values (Supplemental Table 5).

We generated a Kaplan–Meier curve to compare the risk of having a diagnosis of ASD among children with and without cCMV over time. Observation time started at birth. Children who met the ASD case definition were censored when the first ASD diagnostic code was recorded, and others who did not experience the event, ASD diagnosis, were censored at either death, disenrollment, or the end of the study period follow-up time. For all analyses, we used SAS version 9.3 (SAS Institute Inc, Cary, NC).

A total of 2 989 659 children who were continuously enrolled in Medicaid or CHIP from birth through ≥4 to <7 years were included in this analysis. Overall, we identified 1044 (3.5 per 10 000) children with cCMV and 74 872 (25.0 per 1000) children with ASD (Table 1; Supplemental Fig 2). The administrative prevalence (prevalence of diagnoses) of cCMV was similar by sex but varied by region, with lower prevalence in the West as compared with other regions. The prevalence of diagnosed ASD was higher in males than in females (37.2 vs 12.3 per 1000 children) and in the Northeast as compared with other regions.

The prevalences of both cCMV and ASD diagnoses were higher among children with diagnoses of preterm birth and LBW, LBW only, or preterm birth only, compared with children who were not coded as either preterm birth or LBW (Table 1). The prevalences of CNS anomaly or injury codes and hearing loss were also higher among children diagnosed with cCMV or ASD. For example, hearing loss was 45 times and 5 times as prevalent among children with cCMV and ASD than among those without those diagnoses, respectively.

The Kaplan–Meier curve revealed that children with a cCMV diagnosis were more likely to subsequently have a diagnosis of ASD (Fig 1). Among those who had an ASD diagnosis, the median age at the first claim with an ASD diagnosis was 39 (interquartile range: 31–50) months for children with a cCMV diagnosis and 37 (interquartile range: 30–45) months for children without a cCMV diagnosis. The overall prevalence of ASD diagnosis at the end of the study period was 64.2 (95% CI: 49.3–79.0) per 1000 for children with a cCMV diagnosis and 25.0 (95% CI: 24.9–25.2) per 1000 for children without a cCMV diagnosis (Table 1).

The unadjusted HR for having a diagnosis of ASD was 2.61 (95% CI: 2.03–3.28) comparing children with and without cCMV. The HR adjusted for birth year, region, and sex was 2.54 (95% CI: 1.98–3.19; Table 2). Stratification by sex revealed a much stronger association between cCMV and ASD was observed for females (adjusted HR: 4.65; 95% CI: 3.09–6.59) than for males (adjusted HR: 1.95; 95% CI: 1.40–2.61). Stratification by preterm birth and LBW resulted in slightly lower adjusted HRs (1.79 to 2.15). Stratification by CNS anomaly or injury revealed larger (but not statistically significant) differences with an adjusted HR of an ASD diagnosis among children diagnosed with cCMV of 1.00 (95% CI: 0.74–1.31) in the stratum with CNS anomaly or injury and of 1.60 (95% CI: 0.99–2.40) than in the stratum without any CNS anomaly or injury.

In a sensitivity analysis in which race and ethnicity was included for the subset of cases with non-missing race and ethnicity records, the overall association changed from HR 2.45 to adjusted hazard ratio 2.65 (95% CI: 1.94–3.51; Supplemental Table 5).

In this study of administrative claims data for nearly 3 million children enrolled in Medicaid or CHIP, we found that children with a cCMV diagnosis were ∼2.5 times as likely to have an ASD diagnosis compared with children without cCMV. Although there was no association between cCMV and ASD diagnoses among children who had CNS anomaly or injury, the likelihood of an ASD diagnosis among children with cCMV diagnoses was 1.6 times greater for children without cCMV among children without CNS anomaly or injury codes. Although the association between cCMV and ASD diagnosis was not statistically significant for those with or without CNS injury, the HR reduced substantially when stratifying by CNS involvement. These results are consistent with some, but not all, of the association between cCMV and ASD potentially being attributable to brain anomalies caused by cCMV. In addition, when we stratified by preterm birth or LBW, the likelihood of an ASD diagnosis among children with cCMV was approximately twice that of children without cCMV in all strata. Our findings suggest that even among children with cCMV who are not born with brain anomalies or preterm +/− LBW, an increased risk for ASD is still observed.

When our models were stratified by sex, we found the risk of having an ASD diagnosis in females with a cCMV diagnosis was 4.6 times greater than of those without a cCMV diagnosis, and nearly 2 times higher in boys with a cCMV diagnosis. In general, neurodevelopmental disorders, such as ASD, disproportionately affect males, which has been hypothesized to be due to genetic factors (eg, X-linked genetic conditions) and sex hormone involvement.^36 ,37  It may be that children with cCMV receive more neurodevelopmental surveillance and services at baseline and may be more likely to receive a diagnosis of ASD, whereas ASD is thought to be underdiagnosed in the general population of girls because of more subtle and nonstereotypical presentations.^38 ,39 

Maternal CMV infection can activate an inflammatory state which may, in turn, impact fetal brain development, thereby increasing the risk of ASD.¹³  In the most severely affected pregnancies, the virus may enter and replicate in fetal CNS cells, leading to alterations in embryogenesis.^40 ,41  These differences in CNS development in utero may, in some cases, be subtle and not manifest as visible pathology on neuroimaging or as medical conditions such as cerebral palsy or epilepsy. The authors of previous studies have suggested that children with symptomatic cCMV may be at particularly elevated risk of ASD.^19 ,20  We found that the overall prevalence of ASD diagnosis was higher among children with a cCMV diagnosis versus those without, even when adjusting for potential confounders (birth year, sex, and region) in the Cox proportional hazard model and when stratifying by preterm birth or LBW. Our study does not address the question of whether children with asymptomatic cCMV infection have a higher risk of ASD because it is not possible to reliably identify such children on the basis of administrative data.⁴²  Our sample of children with cCMV appears heavily weighted toward children with diagnoses for CNS anomaly or injury. Estimates of the overall proportion of CNS diagnoses in a representative cohort of children with cCMV are not available, but the 21% prevalence of cerebral palsy among children with cCMV in our sample is roughly 7 times greater than the prevalence of cerebral palsy in representative samples of children with cCMV¹⁸  and 70 times greater than the population prevalence of cerebral palsy.⁴³ 

Our study is not without limitations, including inherent limitations of studies using administrative databases, notably the misclassification of conditions resulting from clinical underdiagnosis and inconsistent coding practices.^44 ,45  Cases identified using these Medicaid administrative data are not generalizable to all infants with cCMV infection. cCMV screening in newborns in the United States is primarily conducted for infants with symptoms suggestive of cCMV, and in places without routine screening for cCMV, most infants with cCMV go undiagnosed.^42 ,46  Consequently, the administrative prevalence of cCMV in our sample (3.5 per 10 000) was substantially lower than the population prevalence of cCMV infection in large newborn screening studies in the United States (4.5 per 1000).⁴⁷  Nondifferential misclassification of both cCMV and ASD status as a result of coding or diagnostic errors could lead to underestimation of HRs toward the null.^2 ,45  Validation studies for International Classification of Diseases, Ninth or Tenth Revisions, Clinical Modification codes for cCMV in administrative data are lacking, whereas administrative ASD diagnoses in North American children have a high positive predictive value for medically diagnosed ASD.^31 ,48  Differential misclassification, on the other hand, could lead to an overestimation of associations.⁴⁵  For example, if children with recognized cCMV are more closely monitored for developmental milestones, they may be more likely to be clinically diagnosed with ASD, as compared with children without cCMV. We had limited ability to adjust for potential confounders, given limitations of Medicaid claims data. The only potential indicator of socioeconomic status of children in the data is eligibility for CHIP versus Medicaid, but we found low continuity of enrollment in CHIP. Finally, although it is possible that requiring at least 4 years of enrollment could introduce selection bias based on factors that influence long-term participation in Medicaid, results were similar using a 1-year enrollment requirement (data not shown). It should also be noted that maternal–infant records were not linked; therefore, this analysis could not include potentially confounding pregnancy factors or maternal characteristics that may be associated with ASD risk.

Given the high costs associated with management of ASD in children and adults,⁴⁹  the demonstration of a potentially causal impact of cCMV on risk of ASD could be an important component of the economic impact of cCMV.^50 ,51  However, the confirmation of a heightened risk of ASD among children with cCMV infection may require data on prospectively monitored cohorts of children in which cCMV infections are identified through universal screening programs, which have recently been initiated in a few jurisdictions.⁵²  The collection of long-term follow-up data for screened cohorts, including information on ASD and other neurodevelopmental diagnoses that are linked to cCMV screening and diagnostic results, could yield more accurate estimates of the association of cCMV infection with ASD risk, both overall and stratified by the presence of cooccurring conditions. Testing for CMV DNA in stored newborn dried blood spots for samples of children with and without ASD could potentially be informative for research purposes.⁵³ 

Our findings also have clinical implications. Clinicians may want to be proactive in monitoring for early signs of ASD in children with a diagnosis of cCMV, especially those with hearing loss, the risk of which is elevated among both children with cCMV and those with ASD.^54 ,55  Making a diagnosis of ASD can be challenging in children who are deaf or hard of hearing.^56 ,57  Although cCMV can only be diagnosed through testing of biological specimens collected in the first few weeks of life,⁵⁸  in a few states, clinicians with parental consent might be able to test stored residual dried blood spot specimens for viral DNA for children suspected of having cCMV.⁵⁹  As noted above, a finding of cCMV may prompt additional monitoring for developmental outcomes and need for developmental supports.

In this cohort study, we found that children with Medicaid or CHIP coverage who had a diagnosis of cCMV were more likely to have a diagnosis of ASD than children without a diagnosis of cCMV. Results of the stratified analysis suggest that a diagnosis of ASD is more common among children with cCMV even in the absence of CNS involvement. Future work examining the prevalence of and risk factors for ASD among children with cCMV identified through a universal screening, as well as studies examining the prevalence of CMV identified on newborn dried blood spot specimens in matched cohorts of children with ASD, could better inform our understanding of this association.

ASD

autism spectrum disorder

cCMV

congenital cytomegalovirus

CHIP

Children’s Health Insurance Program

CI

confidence interval

CMV

cytomegalovirus

CNS

central nervous system

HR

hazard ratio

LBW

low birth weight