Background: Large-scale autism screening studies have yielded empirical support for early screening for ASD. However, they have not accurately estimated sensitivity, specificity, and negative predictive value (NPV), because children who screen negative have not been systematically followed through the risk period. The Children’s Hospital of Philadelphia (CHOP) implemented universal screening for ASD documented in the Electronic Health Record (EHR) in 2009. As these children age and continue to be served within our healthcare system, we have the opportunity to follow children screened for ASD and fill important gaps in the screening literature.
Estimate sensitivity, specificity, positive predictive value (PPV), and NPV in a large sample of children screened for ASD in primary care and followed until age 4-6 years using data from the EHR.
Universal screening of children age 16-30 months was conducted using the Modified Checklist for Autism in Toddlers (M-CHAT; Robins et al., 2001) at CHOP primary care clinics. Parents completed the M-CHAT and pediatricians were asked to complete the Follow-Up Interview when appropriate and document results in the EHR. Children were included if (1) the M-CHAT was completed between 16-30 months, and (2) the child presented to primary care for follow-up at ≥4 years. Diagnostic outcomes were abstracted from EHR records from pediatricians, developmental behavioral pediatricians, psychologists, psychiatrists, and neurologists within the CHOP system.
21,543 children were screened using the M-CHAT during routine clinical between 2010-2012. 77.96% had complete data recorded in the EHR and 59.44% returned for a subsequent CHOP primary care visit at ≥ 4 years. Of the 7,602 children with follow up information available, 6.77% screened positive for ASD on the M-CHAT and 2.10% had a documented diagnosis of ASD by age 4. Sensitivity was estimated at 38.12% and specificity was 93.90%. PPV was 11.84%, while NPV was 98.60%. When outcome was broadened to include any neurodevelopmental or psychiatric diagnosis (i.e., developmental/language delay, intellectual disability, ADHD, other behavior disorder, anxiety), PPV was 58.83% and NPV was 76.18%. The effect of child and family characteristics (i.e., age at screening, race, ethnicity, prematurity, SES, primary language spoken) on screening accuracy will be discussed.
This study is among the first to examine universal autism screening and subsequent outcomes within a large pediatric healthcare system. Important differences from previous large-scale screening studies include follow-up data after the primary risk phase on large numbers of children, and “real world” use of the M-CHAT with diagnoses made by CHOP providers (or community diagnoses documented in the EHR by pediatricians) rather than assessments in the research lab. Our data suggest a substantial proportion of children with ASD were missed by universal screening at 16-30 months. Just as there is no single cause or effective treatment across the entire autism spectrum, there is likely not a single screening measure or screening age that will identify all individuals with ASD. Future research should determine the type of child best identified through traditional early screening, and what additional methods would enhance classification accuracy.