Anoop M Mayampurath: Professor, Department of Pediatrics, University of Chicago at Cypher 2017
summary
Hospital team develops and deploys an ECOT early warning score using vital signs in six-hour blocks with discrete-time survival analysis on ~850,000 patients / ~1M admissions, producing a color-coded risk display nurses use to trigger rapid response. Moving models from logistic regression to random forest raises AUC from 0.75 to 0.8. Neighbourhood effects appear: one cardiac arrest on a ward raises a neighbour’s six-hour risk by 18% and two or more by 53%, and each ward crash adds ~3 days to others’ length of stay.
AI-written summary · about
chapters (12)
- 0:00Start
- 2:40Use of Machine Learning To Predict Clinical Deterioration
- 3:19Research Interests
- 6:14Causes of Mortality
- 6:29Medical Error
- 8:04Hospital-Acquired Infection
- 12:58Patients Predict
- 15:10Ecot Score
- 16:38Discrete Time Survival Analysis
- 19:56Use of Domain Knowledge
- 21:56Patient Privacy
- 26:04Does the Location of Patient Admission Matter
related talks
08min conversation with AIM
27mCitizen Data Science campaign
16mSee if talking to the doctor is right for you
08mA Healthcare Analyst Responds to Strata Rx
19mLearning from eight years of data science mistakes
54mR Not Only In Production
18mData Science Careers via A/B Testing
21mApplied AI & Data Analytics