Estimating Social Communication Functioning (ACSF:SC) from ADOS-2 Data: Development of an Algorithm

Background: Innovative algorithmic tools have exciting potential for extracting novel information from commonly used measures used in ASD, such as the item-level ADOS-2 data that are abundantly available in existing datasets, vastly increasing the utility of such datasets for research purposes. The Autism Classification System of Functioning: Social Communication (ACSF:SC) allows parents and professionals to categorize the social communication functioning of children aged 3-5 years into one of five meaningfully distinct levels. Grounded on WHO’s International Classification of Functioning, Disability, and Health (ICF) framework, the ACSF:SC differs from traditional measures, including severity metrics, by assessing abilities rather than deficits. At IMFAR 2016, we presented pilot results regarding the feasibility of developing and validating an algorithm to generate estimates of social communication functioning (descriptive ACSF:SC levels) from existing ADOS data, including (a) results of a modified Delphi process to identify ADOS items most relevant to the social communication construct used in the ACSF:SC, and (b) a preliminary decision tree algorithm predicting ACSF:SC from a limited data sample comprising paired SRS-2 social communication subscale and ACSF:SC assessments.

Objectives: We now present the initial decision tree algorithm developed from analysis of prospectively collected paired ADOS-2 and ACSF:SC data from multiple diagnostic clinics in Ontario, Canada.

Methods: Classification and Regression Tree (CART) analysis is being used to develop and validate the ADOS-to-ACSF:SC algorithm. CART requires paired data (i.e., contemporaneous ADOS-2 assessment and ACSF:SC level in the same child). A required sample size of n=300 was estimated from pilot work. Data will be split on response items within A and B groupings of the ADOS-2 when developing the algorithm. CART interim and final analyses are being run using IBM SPSS Statistics per the pilot work.

Results: In work to date, we have gained the necessary ethics approvals and site agreements, and begun recruitment and data collection from partners at 6 diagnostic clinics from varying regions across Ontario. Generating ACSF:SC tool adoption and enthusiasm was important for this recruitment. At the time of writing (approximately 3 months into data collection), paired data from 29 cases have been collected. This number of cases was insufficient to conduct interim analysis to reliably identify ADOS-2 items that are predictive of ACSF:SC level. Data collection will continue, and a CART analysis and decision tree based on the latest data will be presented in the poster.

Conclusions: An ADOS-to-ACSF:SC algorithm would increase the utility of available ADOS-2 data by providing a means to conduct secondary analysis to characterize the social communication functioning of children within historic datasets. For example, such estimates derived from previously collected data could facilitate examination of longitudinal trajectories of social communication functioning in ASD; work is currently ongoing to extend the age range of the ACSF:SC beyond age 3-5 years. This methodology has value as a template for developing additional algorithmic means for deriving ICF-based ratings of functioning in ASD (not limited to social communication) and creating longitudinal perspectives from previously collected, routinely available data generated by commonly used instruments—extending opportunities for secondary data analysis.