Data Mining in the EMR Gold Rush
summary
Federal incentives produce roughly a thousandfold increase in EMR data, driving claims of 250 MB per patient per year and hundreds of terabytes to analyze. The stack uses visualization (a deployed horizon graph, Protovis/InfoVis/Gephi-style clouds) plus machine learning (an open-source neural net for PubMed matching and genetic algorithms to surface unexpected clusters). Prototypes predict no-shows, estimate payer mixes, and highlight lab-patterns; deployments with clinics report workable, iteratively refined tools and serious HIPAA privacy tradeoffs.
AI-written summary · about
chapters (10)
- 0:00Intro
- 1:37n. The automatic extraction of useful, often previously unknown information from large databases or data sets.
- 8:45HOW TO GLANCE AT A PATIENTS OVERALL HEALTH?
- 11:39HOW TO CONNECT RELEVANT PUBMED DATA TO AN INDIVIDUAL PATIENT?
- 17:22NEURAL NET STACKED BAR
- 18:11HOW TO ANALYZE LAB RESULTS SIMULTANEOUSLY FOR A SINGLE PATIENT?
- 19:18HOW TO ANALYZE UDS DATA?
- 21:02HOW DOES GEOGRAPHY IMPACT PATIENT WELLNESS?
- 22:11HOW TO FIND ARBITRARY PATTERNS?
- 24:48HOW TO ANALYZE DIAGNOSES IN A PATIENT POPULATION?
related talks
05mHadoop and Healthcare
31mAI in healthcare: Data driven transformation
33mHealthcare "big" Data from Data Management to Web Apps
07mData Collection and Reorganization
48mAnalytics and AI on FHIR
10mA Story about Big Data in Healthcare
08mBig data answering healthcare's biggest inquiries
01h 03mMerck drives innovation with a “real-world data exchange”