22612
Development and Validation of a Streamlined Autism Case Confirmation Approach for Use in Epidemiologic Risk Factor Research
Objectives: To evaluate the performance of three streamlined ASD case confirmation approaches, alone and in combination, for potential application in large epidemiologic studies.
Methods: We conducted a multi-site, cross-sectional criterion validity study in a convenience sample of 388 three-year olds scheduled for neurodevelopmental evaluation that broadly represent children who would fail a stage 1 screen like the M-CHAT. ASD case classification as determined by a battery of three novel assessment instruments (the Early Video-guided Autism Screener E-VAS; the Autism Screening Interview, ASI; the Screening Tool for Autism in Toddlers Extended, STAT-E), each of which designed to be administered in less than 30 minutes by lay staff, was compared to case-classification based on the DSM-based diagnostic assessment from a qualified clinician. Sensitivity and specificity of each instrument alone and in combination were estimated. Order of instrument administration was randomly assigned. Alternative cutpoints were identified under different criteria and two-stage cross validation was used to avoid overfitting. Findings were interpreted in the context of a simulated prospective pregnancy cohort study seeking to estimate relative risks (RRs) of varying strengths for dichotomous exposures of varying prevalences using a two-stage case finding approach where the first stage would be a screener like the M-CHAT and the second stage would be case confirmation based on the instruments tested here. The bias introduced in RR estimation under different case-confirmation approaches was calculated, informed by the sensitivity and specificity estimates for each of the alternative approaches developed here.
Results: Under the instruments’ initial recommended cutpoints, sensitivity ranged from 0.63 to 0.94 and specificity from 0.34 to 0.70. Alternative cutpoints that gave equal weight to sensitivity and specificity resulted in sensitivity estimates ranging from 0.56 to 0.84 and specificity ranging from 0.45 to 0.77. Additional sets of alternative cutpoints holding sensitivities at 50% resulted in specificities near 80%. And, while these were lower than the >80% for both sensitivity and specificity that comes with using ADOS and ADI-R in combination (Kim and Lord, 2012), this lower sensitivity and specificity had only modest effect on the extent of bias in dichotomous exposure RR estimates. Several strategies (including use of the direct observation STAT-E alone and the combined administration of the parent interview ASI and parent self-report E-VAS) resulted in performance suitable for application as a second stage case-confirmation tool in an epidemiologic study like that considered here.
Conclusions: Use of more streamlined methods of case-confirmation in large-scale prospective cohort epidemiologic investigations of ASD risk factors appears feasible.