Decision Support for Patients

• 

  explanation of NN

• 

  Screenshots from the front end

• 

  The LTSM vs RNN implementation error

Inspiration

Patients should have visibility into downstream implications of their decisions, but “a little knowledge can be a dangerous thing*”. Growing transparency around procedure cost (Castlight, CMS Open Data, etc) gives patients short term cost implications, but patients lack information on the long term implications of care delivery decisions.

What it does

Uses a neural network implementation to predict claims cost after a particular sequence of events so that patients and physicians can manage healthcare costs

How I built it

Using a theanets python package for neural networks and javascript for the front end.

Challenges I ran into

Performance issues when including additional features from a claim in the neural network sequence

Accomplishments that I'm proud of

Dramatically reduced the error from a random prediction of cost based on cohort of patients

What I learned

Will need additional data sources to improve accuracy on claims

What's next for Decision Support for Patients

Improving features from claims and expanding to clinical datasets using FHIR

Built With

• javascript
• python