Imbiriba 2023 — Wearable Biosensing to Predict Imminent Aggressive Behavior in Psychiatric Inpatient Youths With Autism
Citation: Imbiriba, T., Demirkaya, A., Singh, A., et al. (2023). JAMA Network Open, 6(12), e2348898.
DOI: 10.1001/jamanetworkopen.2023.48898
Type: Peer-reviewed study (prognostic)
Key Contribution
Largest replication and extension of wearable aggression prediction in ASD. 70 participants across 4 inpatient sites. First to separately model SIB, emotion dysregulation, and ATO.
Design
- 70 participants (5-19 yrs, 88.6% male, 45.7% minimally verbal)
- 429 sessions, 497 hours, 6665 aggressive behaviors
- Empatica E4 wristband on non-dominant wrist
- Models: logistic regression, SVM, neural networks, domain adaptation
Key Results
- Logistic regression best overall
- AUROC 0.80 predicting 3 min before onset (CI: 0.79-0.81)
- Person-dependent outperformed population models
Impact
Established multi-site generalizability of the Goodwin approach. Domain adaptation promising for cross-subject generalization. Published in high-impact general medical journal (JAMA Network Open).
Linked To
- Dataset: imbiriba-jama
- Dataset: goodwin-aic
- Paper: goodwin-2019