Online concerns of parents suspecting autism spectrum disorder in their child: Content analysis of signs and automated prediction of risk

By: Ben-Sasson A., Yom-Tov E.
Published in: Journal of Medical Internet Research
SDGs : SDG 03  |  Units: Social Welfare & Health Sciences  | Time: 2016 |  Link
Description: Background: Online communities are used as platforms by parents to verify developmental and health concerns related to t heir child. The increasing public awareness of autism spectrum disorders (ASD) leads more parents to suspect ASD in their child. Early identification of ASD is important for early intervention. Objective: To characterize the symptoms mentioned in online queries posed by parents who suspect that their child might have ASD and determine whether they are age-specific. To test the efficacy of machine learning tools in classifying the child’s risk of ASD based on the parent’s narrative. Methods: To this end, we analyzed online queries posed by parents who were concerned that their child might have ASD and categorized the warning signs they mentioned according to ASD-specific and non-ASD-specific domains. We then used the data to test the efficacy with which a trained machine learning tool classified the degree of ASD risk. Yahoo Answers, a social site for posting queries and finding answers, was mined for queries of parents asking the community whether their child has ASD. A total of 195 queries were sampled for this study (mean child age=38.0 months; 84.7% [160/189] boys). Content text analysis of the queries aimed to categorize the types of symptoms described and obtain clinical judgment of the child’s ASD-risk level. Results: Concerns related to repetitive and restricted behaviors and interests (RRBI) were the most prevalent (75.4%, 147/195), followed by concerns related to language (61.5%, 120/195) and emotional markers (50.3%, 98/195). Of the 195 queries, 18.5% (36/195) were rated by clinical experts as low-risk, 30.8% (60/195) as medium-risk, and 50.8% (99/195) as high-risk. Risk groups differed significantly (P<.001) in the rate of concerns in the language, social, communication, and RRBI domains. When testing whether an automatic classifier (decision tree) could predict if a query was medium- or high-risk based on the text of the query and the coded symptoms, performance reached an area under the receiver operating curve (ROC) curve of 0.67 (CI 95% 0.50-0.78), whereas predicting from the text and the coded signs resulted in an area under the curve of 0.82 (0.80-0.86). Conclusions: Findings call for health care providers to closely listen to parental ASD-related concerns, as recommended by screening guidelines. They also demonstrate the need for Internet-based screening systems that utilize parents' narratives using a decision tree questioning method. © Ayelet Ben-Sasson, Elad Yom-Tov.