Sensitivity of Autism Spectrum Disorder Prevalence Estimates to Private and Home School Enrollment and Denominator Choice in North Carolina

Background:   Accurate estimates of the prevalence of autism spectrum disorders (ASDs) over time are important for planning services and generating hypotheses about causality.  The Autism and Developmental Disabilities Monitoring (ADDM) Network has generated population-based prevalence estimates bi-annually since 2000.  ADDM methodology is considered to be more complete than individual administrative data sources, in that standardized criteria are applied to existing developmental evaluation records of children from both health and public educational settings without requiring a previous diagnosis.  Still, limitations arise from incomplete enumeration of children with an ASD because not all agencies serving such children can be included.  Limitations also arise from misspecification of the underlying population denominator; ADDM uses the number of 8-year old children projected by the US census.

Objectives:   We evaluated the degree to which ASD prevalence estimates from North Carolina (NC) ADDM were influenced by 1) under-counting of children with an ASD in private and home-school settings, and by 2) using previously-considered alternate denominators. 

Methods:   We estimated ASD prevalence in 8 counties in central NC for 8 year olds in 2002, 2004, 2006, and 2008.  Children meeting the NC ADDM case definition for ASDs were classified based on whether their composite developmental record included a public school contribution, indicating public school enrollment.  Denominators were derived using 1) birth counts based on NC birth certificates and 2) total school enrollment counts available from NC administrative data summing 8-year old children in public, private, and home school settings. 

Results: NC ADDM measured an ASD prevalence of 9.8/1000 children during this period, including 738 children with an ASD with a public school record and 82 without.  The number of 8-years olds in a private or home-school (non-public) setting was estimated to be 10,097 (12.4% of all school enrollment).  If the true proportion of ASDs in the non-public setting equaled that in the public setting, 104 children with an ASD would be expected (leading to a 3% increase in overall prevalence).  To increase the overall ASD prevalence by 10%, the proportion of children with an ASD coming from the non-public sector would need to be 16.2/1000.   Using alternate denominators resulted in small increases in ASD prevalence estimates, which were 5% higher using birth cohorts and 3% higher using school enrollment data.

Conclusions:   In NC, ADDM prevalence estimates of ASDs may underestimate true prevalence.  Prevalence estimates would be 3-5% higher if birth cohorts or school enrollment counts more accurately enumerated the underlying population.  Assuming that children with an ASD are just as likely to be in private or home school settings as in public schools, prevalence may be 3% higher if developmental evaluation records from these settings could be included.  To have a substantial impact on overall prevalence (> 10%), the proportions of children with an ASD in the non-public setting would need to be > 1.6 times as high as what we observed among children with public educational contact.  Studies that directly ascertain the extent of ASDs by educational setting would be needed to inform whether this increase is likely.