Goodwin 2019 — Predicting Aggression to Others in Youth with ASD Using a Wearable Biosensor
Citation: Goodwin, M.S., Mazefsky, C.A., Ioannidis, S., et al. (2019). Autism Research, 12(8), 1286-1296.
DOI: 10.1002/aur.2151
Type: Peer-reviewed study
Key Contribution
First study to demonstrate that wearable biosensor data can predict aggression in youth with ASD minutes before it occurs. Person-dependent AUC 0.84.
Design
- 20 youth with ASD (6-17 yrs, 75% male, 85% minimally verbal)
- 69 naturalistic observation sessions in inpatient psychiatry unit
- Wrist-worn biosensor (cardiovascular + EDA + accelerometry)
- Ridge-regularized logistic regression
- 3-min context window → predict 1 min before aggression
Key Results
- Global model AUC: 0.71
- Person-dependent model AUC: 0.84
- Biosensor data contributed unique predictive value beyond temporal features
- Model performance improved with more training data, not longer observation
Impact
Seminal paper spawning a line of research. Directly replicated and extended by Imbiriba 2023 (70 participants, 4 sites). Data available via SFARI Base.
Linked To
- Dataset: goodwin-aic
- Dataset: imbiriba-jama
- Paper: imbiriba-2023