Category: Autism Spectrum and Developmental Disorders

PS6- #B51 - Predicting Responsiveness to a Community-Based CBT Intervention for Adults With Autism Using Innovative Artificial Intelligence

Friday, Nov 17
2:45 PM – 3:45 PM
Location: Indigo Ballroom CDGH, Level 2, Indigo Level

Keywords: Autism Spectrum Disorders | Community-Based Assessment / Intervention | Statistics

Moving evidence-based interventions into community practice is an important current movement within services research due to the well-documented gap between research-based interventions and routine care. While community programs often welcome the use of evidence-based interventions (EBIs)  there is also a concern about costs related to such implementation (staff training, dosage of sessions, material fees, fidelity measurement) that may influence community agency decision making in using EBIs. The use of a precision medicine approach of identifying what intervention is indicated for whom could facilitate the use of EBIs in community care by making them more cost effective and delivered to those indicated.  This exploratory study used an innovative statistical approach, artificial intelligence methods, to predict which participants had positive outcomes after completing a CBT intervention to develop cognitive and social skills called SUCCESS (Supported employment, Comprehensive Cognitive Enhancement and Social Skills).  The aim of the study was to predict participant responsiveness to the SUCCESS intervention using artificial intelligence statistical methods utilizing variables from each study measure. Sample included 19 adults with autism (a total of 24 will be included for final analyses). The model included 13 measures of the following constructs: executive functioning skills (DKEFS, BRIEF-A, Cognitive Problems and Strategies Questionnaires), social skills (SSPA, SRS-2, Social Problems and Strategies Questionnaires), mental health problems (Achenbach Adult Self Report), functional impairment (QLESQ), self-efficacy rating(General Self Efficacy Scale), and daily living skills (Adaptive Functioning Report, Insomnia Sleep Inventory) which equated to 350 individual variables that were included in the model. Using caret package in R language and artificial intelligence algorithm (non-parametric), hierarchical cluster analysis showed that participants with successful outcomes were relatively well-discriminated from those with non-successful outcomes to the intervention. Based on the support vector machine analysis, the best model resulted in using 30 variables of the original 350 which resulted in an overall class prediction accuracy of 80% as well as a sensitivity and specificity of 80% and 85%, respectively.  Interestingly all 30 variables included in final model were either from executive functioning or social skill constructs. Artificial intelligence may be a new fruitful methodology to predict which adults with autism would be the most responsive to a CBT intervention to enhance executive functioning and social skills in community settings allowing for targeted community services and ultimately reducing overall costs.

Enrique I. Velazquez

Post-Doctoral Fellow
Rady Children's Hospital San Diego (RCHSD), California

Mary Baker-Ericzen

Associate Research Scientist
Rady Children's Hospital-San Diego
San Diego, California