gistfile1.md

Shahid Shah

CEO, Netspective

Slides

Health Data Integration Complexity

• The kinds of data that are most interesting.
• Pitfalls
• Data changes the questions we ask

Scientific Discovery

• Old way:

  1. Identify problem
  2. Ask questions
  3. Collect data
  4. Answer questions

• New way:

  1. Identify data
  2. Generate questions
  3. Mine data
  4. Answer questions

• Always store data!

  • Just because you don't know how to synthesize it, doesn't mean you shouldn't store it.
  • Keep all your data around indefinately.

Modernization

• Permissions-oriented systems discourage data exchange/discovery/hacking.
• Allow clinicians to access data by default
• Getting data pushed to you is cheaper than pulling from external sources
• Industry-specific (legacy) formats aren't always necessary
  • Aka "fuck the police".
• Emit data using semantic markup so it is search/mine-able

gistfile2.md

Michael Yuan

CEO, Ringful

You have data now what?

Multiple regression-based analysis

• Generates lots of hypotheses
• Hard to discern causal links
• Requires little domain knowledge

Principal Component Analysis

• Identify related cluster of factors.
• Reduce over-fitting

Probability-Base analysis

• neural networks, bayesian
• Requires explicit domain knowledge

gistfile3.md

Rachel Kalmar

Data Scientist, Misfit Wearables

Different flavors of open

• Pay per API call
  • Not well suited for small business/adoption
• Reciprocal data sharing
• Pay for data download
• Free API access to trusted parties
• Build your own playform
• Open Data Share and Share-alike
• Completely closed
• Completely open
  • Twitter, genome project, etc
  • Hard reveneue model