Predicting Aggression Onset in Minimally-Verbal Youth with Autism Spectrum Disorder Using Preceding Physiological Signals

Background: Unpredictable and potentially dangerous aggressive behavior by youth with autism spectrum disorder (ASD) isolates them from foundational educational, social, and familial activities, thereby markedly exacerbating the morbidity and costs associated with the condition. As many as 2/3 of youth with ASD display aggression, which is one of the primary reasons they use behavioral healthcare services. Aggression presents imminent safety risks for the individual and others in the environment. Families report that aggression increases their stress, isolation, and financial burden, and decreases available support options. Aggression to others is particularly impairing and treatment refractory in the 30-40% of youth with ASD who are minimally verbal (MV-ASD). Their inability to self-report distress can lead to behaviors that seem to occur without warning, sometimes long after any observable trigger.

Objectives: Aggression to others may represent a maladaptive attempt to express or modulate physiological arousal arising from distress. Thus, we hypothesize that physiological arousal precedes aggressive behavior. Our objective is to test whether the proximal onset of aggression can be predicted from preceding physiological signals.

Methods: In this IRB approved study, we collected physiological data from the wrist using the commercially available E4 by Empatica, Inc. that wirelessly measures heart rate, heart rate variability, electrodermal activity, skin temperature, and physical motion from 20 MV-ASD inpatient youth. E4 signal parameters were derived using time-series statistics in a past interval of time (t_p). Ridge-regularized logistic regression models were used for binary decision making for aggression onset in an upcoming time interval (t_f) using temporal (time elapsed since last observed aggression) and E4 signals. Model prediction performance was calculated using 5-fold cross-validation. For all binary prediction models, receiver operator characteristic (ROC) curves and their corresponding area under the curve (AUC) values were calculated to represent accuracy based on true (sensitivity)/false (1-specificity) positive rates. Classification of varying values of t_p and t_f was assessed using both global (a single classifier consisting of concatenated time-series data across all sessions and participants) and person-dependent (data only pooled across sessions within person) models.

Results: All participants tolerated the E4 after desensitization and usable data was obtained in all cases. Sixty-nine independent naturalistic observational sessions were collected, totaling 87hrs. Out of 548 total aggressions observed with concurrent E4 data, our results [Fig 1] demonstrate that, on average, models with all signals included (i.e., time since last aggression and E4 signals) predict the onset of aggression 1min before it occurs using 3min of prior data with 0.71 AUC for global and 0.84 AUC for person-dependent models.

Conclusions: By linking observable aggressive behavior to preceding physiological signals in MV-ASD, we move the field of problem behavior assessment towards a new biologically-based, data-informed approach that is focused on prospective monitoring, prevention, and eventually real-time intervention, addressing a historically intractable problem for a segment of the ASD population who is arguably the most in need of innovative approaches.