Does a Claims Diagnosis of Autism Mean a True Case?

Background: Although Autism spectrum disorders (ASD) affect a large and heterogeneous group of children
and adolescents, most studies of health outcomes in ASD to-date have included relatively small clinical and
research samples, limiting their generalizability. Thus secondary analysis of a large administrative dataset
containing medical, pharmacy and behavioral health claims for children with ASDs may be potentially useful
for carrying out observational research studies. The extent to which claims-based approaches for identifying
ASD cases in administrative data are accurately identifying true ASD cases has not been well-studied.

Objectives: To conduct a validation study of claims-based ASD case identification against medical charts.

Methods: From a large cohort of children enrolled in a large private health plan for 6 months between 2001-9
we sampled 1) children with 2+ claims with an ASD ICD-9 (n=180); 2) children with 1 claim with an ASD
ICD-9 (n=180); and 3) children without claims with ASD ICD-9 codes who did have at least one claim for a
neurodevelopmental condition (e.g., developmental delays, intellectual disabilities, language disorders)
(n=58). Within each group the sample was further stratified for age (<8 years and >8 years), length of
enrollment (<18 months and >18 months), and provider type (ASD specialist vs. other specialist vs. primary
care). With provider consent, a single chart from a single provider was available for analysis limited to the
time period of enrollment in the database/health plan. Following a protocol modeled after that used by the
CDC ADDM Network, charts were examined and abstracted for information related to ASD diagnosis
(including behavioral descriptions, history of developmental delay, referral for ASD assessment, evidence of
developmental plateau or regression, tests or assessments for ASD, and other health conditions or concerns).
Review of abstracted information was conducted by an experienced clinician also following the CDC ADDM
protocol classifying children as confirmed ASD cases, suspected ASD cases or not meeting criteria. Results:
Preliminary results are available on the first 111 (out of 418) abstracted and reviewed charts. Positive
predictive value having 2+ claims with an ASD ICD-9 codes was 88.9% [95% CI 79.6, 98.2] while the
positive predictive value of having just one claim with an ASD ICD-9 was 68.3% [95% 53.4,83.4]. Predictive
value was negative in the sample for children without ASD ICD-9 codes but with other ICD-9 codes for
neurodevelopmental conditions. Only 4.5% of the subjects found not to meet chart review criteria for either
confirmed or suspect ASD had clear ASD rule-out information in the medical chart.

Conclusions: An ASD case finding approach relying on the presence of 2 or more claims with ASD ICD-9
shows promise as a valid means to identify true ASD cases without bringing in substantive numbers of false
positives.