Risk Assessment for Parents Who Suspect Their Child Has Autism Spectrum Disorder: Machine Learning Approach

Ayelet Ben-Sasson, Diana L Robins, Elad Yom-TovJournal of Medical Internet Research 2018 · OpenReview

Abstract

Background: Parents are likely to seek Web-based communities to verify their suspicions of autism spectrum disorder (ASD) markers in their child. Automated tools support human decisions in many domains and could therefore potentially support concerned parents. Objective: The objective of this study was to test the feasibility of assessing ASD risk in parental concerns from Web-based sources, using automated text analysis tools and minimal standard questioning. Methods: Participants were 115 parents with concerns regarding their child’s social-communication development. Children were 16- to 30-months old, and 57.4% (66/115) had a family history of ASD. Parents reported their concerns online and completed standard screeners (M-CHAT-R/F and ASQ). An algorithm predicted ASD risk using a combination of the parent's text and a single screening question, selected by the algorithm to enhance prediction accuracy. Results: Screening measures identified 58% to 88% of children at risk. Prediction accuracy was significantly higher when text was combined with an automatically selected M-CHAT-R question (AUC 0.88) than from text alone. Conclusions: The internet can be harnessed to prescreen for ASD using parental concerns by administering a few standardized screening questions to augment the process.

Introduction

There is a significant "detection gap" in autism diagnosis. While many parents notice developmental red flags when their child is between 12 and 19 months old, the average age of diagnosis remains over three years. This delay is critical because early intensive intervention is most effective during the toddler years. Currently, parents often turn to the internet to "vibe check" their concerns in web communities. However, they frequently encounter unreliable advice or fail to share these digital interactions with their pediatricians. This study explores an automated, low-friction screening tool that accurately assesses risk by combining the natural language of a parent’s concern with structured clinical screening data.

Methods

The study utilized a multi-stage machine learning pipeline to predict ASD risk among 115 parents of children aged 16 to 30 months. The core innovation of this approach is the Active Question Selection mechanism.

The automated screening process combining parental text with targeted questions.
  1. Text Analysis: Parents provided open-ended descriptions of their concerns. Natural Language Processing (NLP) was used to represent this text using single words, bigrams, trigrams, and lexical affinities.
  2. Active Question Selection: Rather than requiring a full 20-question survey, the algorithm analyzed the parent's text and then "decided" which single question from the Modified Checklist for Autism in Toddlers-Revised (M-CHAT-R) would most improve the prediction accuracy for that specific child.
  3. Risk Prediction: A Regression Tree algorithm was trained to predict risk based on the text alone, or the text plus the answer to the selected question.
  4. Validation: The model was evaluated using leave-one-out cross-validation, comparing its predictions against established "gold standard" screeners: the M-CHAT-R/F and the Ages and Stages Questionnaire (ASQ).

Results and Key Insights

The study found that text alone is insufficient for accurate screening. Using only the parent's open-ended text resulted in mediocre predictive power (AUC of 0.54 for M-CHAT-R risk). Parents often describe symptoms vaguely or focus on only one area, missing other clinical markers.

However, adding just one targeted question jumped the predictive accuracy significantly to an AUC of 0.88. This suggests that machine learning can effectively "fill in the blanks" of a parent's narrative if it knows exactly what piece of missing information is most critical. The algorithm most frequently selected three specific M-CHAT-R questions to clarify risk:

  1. Does the child follow a point?
  2. Do they engage in make-believe play?
  3. Is there a concern about deafness (often a proxy for not responding to their name)?

For 17% of the ASQ risk predictions, the algorithm determined that no additional questions were needed beyond the initial text to reach a confident estimate. Furthermore, children with a family history of ASD were three times more likely to show risk on screening measures, and their parents used more specific terminology in their descriptions.

Discussion and Future Work

While promising, the study has limitations, including a relatively high-risk sample (many with family history) and the use of other screening tools rather than full clinical diagnoses as the "ground truth." Future work aims to integrate this into a real-time, web-based support system. Such a tool could act as a digital "triage," prompting parents to seek professional advice earlier and effectively shortening the years-long gap between a parent's first concern and a formal diagnosis.

Summary updated

Abstract

Background: Parents are likely to seek Web-based communities to verify their suspicions of autism spectrum disorder markers in their child. Automated tools support human decisions in many domains and could therefore potentially support concerned parents.Objective: The objective of this study was to test the feasibility of assessing autism spectrum disorder risk in parental concerns from Web-based sources, using automated text analysis tools and minimal standard questioning.Methods: Participants were 115 parents with concerns regarding their child’s social-communication development. Children were 16- to 30-months old, and 57.4% (66/115) had a family history of autism spectrum disorder. Parents reported their concerns online, and completed an autism spectrum disorder-specific screener, the Modified Checklist for Autism in Toddlers-Revised, with Follow-up (M-CHAT-R/F), and a broad developmental screener, the Ages and Stages Questionnaire (ASQ). An algorithm predicted autism spectrum disorder risk using a combination of the parent's text and a single screening question, selected by the algorithm to enhance prediction accuracy.Results: Screening measures identified 58% (67/115) to 88% (101/115) of children at risk for autism spectrum disorder. Children with a family history of autism spectrum disorder were 3 times more likely to show autism spectrum disorder risk on screening measures. The prediction of a child’s risk on the ASQ or M-CHAT-R was significantly more accurate when predicted from text combined with an M-CHAT-R question selected (automatically) than from the text alone. The frequently automatically selected M-CHAT-R questions that predicted risk were: following a point, make-believe play, and concern about deafness.Conclusions: The internet can be harnessed to prescreen for autism spectrum disorder using parental concerns by administering a few standardized screening questions to augment this process.J Med Internet Res 2018;20(4):e134doi:10.2196/jmir.9496
BibTeX
@article{bensasson2018risk,
  author = {Ayelet Ben-Sasson and Diana L Robins and Elad Yom-Tov},
  title = {{Risk Assessment for Parents Who Suspect Their Child Has Autism Spectrum Disorder: Machine Learning Approach}},
  journal = {Journal of Medical Internet Research},
  year = {2018},
  url = {https://openreview.net/forum?id=r8jJZHx8R9},
}
Figure