31497
A Computer Administered Autism and Developmental Disorder Screen for Toddlers (CAADST) Efficiently Improves Identification at the 18-Month Pediatric Visit
To achieve early identification, “autism specific” screening is recommended by the American Academy of Pediatrics (AAP) at both 18 and 24-month visits. The Modified Checklist for Autism in Toddlers- Revised (M-CHAT-R) screen has been shown to require a follow-up interview (M-CHAT-R/F) (Kleinman, 2008) and may be less accurate at the 18-month visit than 24-months (Pandey, 2008) (Sturner, 2017). Preliminary data suggested that algorithms of individualized item presentation with computer assisted administration and scoring may improve accuracy for younger toddlers. The AAP also recommends screening for other developmental disabilities at the 18-month visit using a “broadband developmental screen”. It has been suggested that “autism specific” and “broad band” items may have overlapping clinical value when considered together. Yet the predictive performance of combined screening has not been examined when screen positive cases as well as screen negative children are considered.
Objectives:
To develop optimal decision trees of parent administered items to improve the accuracy of detecting autism and other developmental disorders at the pediatric 18-month visit.
Methods:
Parents of 11,878 children 16-20 months old completed the M-CHAT-R before 18-month pediatric visits via an online system (CHADIS). Positive screens (96) and practice matched controls (314) were recruited and parents completed additional items derived from prospective studies of autism outcomes that showed promise for identification of autism (First Years Inventory (FYI); Parents Observation Checklist [POC]; MacArthur-Bates Communicative Development Inventory; Parent Observation of Social Interaction (POSI)). The children had diagnostic autism evaluations using the Autism Diagnostic Observation Schedule- Toddler version and Mullen Scale of Early Development with clinical impression of autism considered to be a positive for autism. Developmental disorder (DD) was defined as the typical entry criteria for early intervention services (score >1½ SD below the mean on >=2 subscales or >2 SD on 1 subscale). Optimized decision trees for predicting autism and DD diagnoses were created through application of recursive partition models (CART) of the potential predictor variables using a random selected subset of the dataset and tested with the holdback sample resulting in decision rules (Computer Assisted Autism and Developmental Screen for Toddlers or CAADST) for presenting varying numbers (7 to 21) of items to subgroups of parents.
Results:
ROC analyses of CAADST showed improved sensitivity (0.92) compared to both M-CHAT-R/F (0.27); M-CHAT-R (0.62) and M-CHAT-R + ASQ (0.75) and higher Negative Predictive Values (NPV; 0.92 compared to M-CHAT-R (0.76); M-CHAT-R/F (0.68); or M-CHAT-R + ASQ (0.81); with lower rates of specificity (0.60 vs 0.93) and PPV (0.56 vs 0.70) compared to M-CHAT-R/F but similar specificity and Positive Predictive Value (PPV) to both M-CHAT-R and M-CHAT-R + ASQ.
Conclusions:
An algorithm for computer administered parent report items correctly identified more toddlers with autism and/or developmental disorder with greater certainty that a child with a negative screen does not have a problem than the standard combination of broadband and autism specific tools in a community sample using fewer items and no follow up interview.