Prediction Model for Favorable COVID-19 Patient Outcomes
summary
Retrospective model trained on 3,345 Covid-positive adults and prospectively validated on 474 predicts a favorable 96-hour outcome (no ICU transfer, intubation, ED return, hospice, death, or >6 L O2). Ensemble + 13-feature parsimonious logistic runs every 30 minutes in the EHR; average precision ~89–91% and discrimination ~95% (retrospective) / ~87% (prospective). Live deployment reports PPV 93.3% with 41% sensitivity, a 50% digital RCT concealment, and measurable reduction in provider uncertainty.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:42New York City
- 1:43About Yin
- 2:42Why am I here
- 3:30Challenges
- 6:12Big Picture
- 7:39Paper
- 10:15Use Cases
- 10:36Development
- 12:07Logistic Regression
- 12:42Clinical Workflow
- 13:40Monitoring
- 14:30Evaluation
- 15:17Observations
- 17:42Success Principles
- 19:20Communicate the Surprise
related talks
23mTransferring clinical prediction models across hospitals...
32mSHAP and Beyond
29mBart Steverink & Laurens Schinkelshoek
20mML applications in healthcare - Saudi Arabian hospital case study
54mCrossing the Last Mile with Predictable Hospital Readmissions
31mBayesian Risk Analysis For Large Multi-Modal Data
39mBest Practices for Prototyping ML Models for Healthcare
23mEngineering AI-Powered Predictive Healthcare Systems