The Development and Validation of Models of Risk for Behaviours That Challenge in Children With Developmental...

Area of Science:

• Developmental Psychology
• Machine Learning in Healthcare
• Behavioral Science

Background:

• Children with developmental disabilities frequently exhibit behaviours that challenge (BtC).
• Identifying risk markers is crucial for early intervention strategies.
• Predicting BtC can improve support for affected children and families.

Purpose of the Study:

• To develop and validate machine learning models for predicting risk of BtC in children with developmental disabilities.
• To assess the models' ability to predict specific types of BtC, including self-injurious behavior, aggression, and property destruction.
• To evaluate the external validity of these prediction models for future behavioral outcomes.

Main Methods:

• Caregiver-reported data on behaviours that challenge were collected using the Self-injury, Aggression and Destruction Screening Questionnaire (SAD-SQ).
• Machine learning algorithms, including random forest classifiers and Gaussian mixture models (GMM), were trained and tested on a dataset of 778 individuals.
• External validation was performed on a separate dataset of 121 individuals, assessing predictions over a 12-month period.

Main Results:

• Random forest classifiers and GMM algorithms demonstrated high accuracy in predicting 'any BtC', correctly classifying 83.5% of individuals at risk.
• The models showed good performance in predicting the persistence and incidence of BtC over 12 months, with accuracy rates of 83.5% and 83.3%, respectively.
• The study achieved fair to good recall and precision for predicting various BtC categories.

Conclusions:

• Novel machine learning models effectively predict behaviours that challenge in children with developmental disabilities.
• These predictive models hold significant potential for clinical application in guiding early preventative interventions.
• The findings support the use of AI-driven tools to enhance support and reduce the impact of BtC in this population.