Goodwin — Autism Inpatient Collection (AIC)
Citation: Goodwin, M.S., Mazefsky, C.A., Ioannidis, S., et al. (2019). Autism Research, 12(8), 1286-1296.
DOI: 10.1002/aur.2151
Access: SFARI Base — data access request required, may need IRB/data use agreement
Tier: 1
Overview
The foundational study for wearable aggression prediction in ASD. 20 youth with ASD (ages 6-17, 75% male, 85% minimally verbal) in a specialized inpatient psychiatry unit. 69 naturalistic observation sessions.
Signals
- Cardiovascular (HR)
- Electrodermal activity (EDA)
- Accelerometry (motion)
Device: wrist-worn biosensor (Empatica E4 or equivalent).
Performance
- Global model AUC: 0.71
- Person-dependent model AUC: 0.84
- Prediction window: 1 minute before aggression
- Context window: 3 minutes of prior biosensor data
- Model: ridge-regularized logistic regression
Key Findings
- Biosensor data contributed unique predictive information beyond temporal features
- Person-dependent models significantly outperformed global models
- Larger training datasets improved prediction more than longer observation of aggression episodes
- AIC phenotypic data available through SFARI Base
Linked To
- Paper: goodwin-2019
- Dataset: imbiriba-jama (follow-on study, same AIC source)