New research reveals that employing a machine-learning approach to analyze physiological changes recorded by a wearable biosensor can aid in predicting aggressive behavior in young patients with autism within psychiatric facilities before such behavior transpires. Autism, affecting 1 in 36 children, is often accompanied by challenging behaviors such as aggression, self-injury, and emotion dysregulation in up to 80% of cases. Traditional diagnostic approaches face difficult challenges due to the inherent difficulty individuals with autism encounter in regulating their emotions and expressing internal states. This predicament not only presents risks to the affected individuals but also creates considerable challenges for families, support professionals, and healthcare systems. Addressing this, a recent prognostic study, titled “Wearable Biosensing to Predict Imminent Aggressive Behavior in Psychiatric Inpatient Youths with Autism,” explores the potential of utilizing wearable biosensors and machine learning to forecast aggressive behaviors in psychiatric inpatient youths with autism.
The study, conducted across four psychiatric inpatient hospitals from March 2019 to March 2020, enrolled 70 psychiatric inpatients diagnosed with autism, exhibiting self-injurious behavior, emotion dysregulation, or aggression toward others. The wearable biosensor recorded cardiovascular activity, electrodermal activity, and motion, while live behavioral coding of aggressive behavior was performed concurrently. The results demonstrated that logistic regression, among various classifiers, provided the most consistent and accurate predictions, anticipating aggressive behavior 3 minutes before onset with a mean area under the receiver operating characteristic curve (AUROC) of 0.80.
Despite the notable advancements made in the study, it is important to acknowledge specific limitations that warrant consideration. One such limitation revolves around the participant demographic, which was relatively restricted, and the occurrences of aggressive behavior exhibited great variation among individuals. To improve the applicability and relevance of these findings, it is necessary for future trials to include a more diverse population, ensuring comprehensive data that enhances the generalizability of the study’s conclusions. The study highlighted the potential for leveraging advanced machine learning methods to further refine predictive capabilities. Specifically, the exploration of domains like ground truth behavior labels and reinforcement learning was highlighted as avenues for future investigation. This recognition of the need for ongoing refinement and exploration emphasizes the study’s commitment to pushing the boundaries of current knowledge and methodologies.
The researchers emphasized the promising applications of biosensor data and machine learning in objectively identifying impending aggressive behaviors. This represents a promising breakthrough, particularly for the often understudied and underserved segment of the autism population. The study’s results also provide a strong basis for the creation of adaptive intervention mobile health systems. Envisioning opportunities for preemptive intervention and a reduction in the unpredictability of aggressive behavior, this research program holds the potential to transform care for individuals with autism. As the research program continues, the objective is to empower inpatient youths with autism, promoting active participation in their homes, schools, and communities. This empowerment is envisioned through the innovative integration of technologies like biosensors and the sophisticated analytics of machine learning. By addressing a key aspect of their condition, the study contributes to a comprehensive approach aimed at improving the quality of life for individuals with autism through cutting-edge technologies and predictive analytics.
