locked
Drug Safety Surveillance (Migrated from community.research.microsoft.com) RRS feed

  • Question

  • frank.defalco posted on 05-19-2010 3:21 PM

    Hi.  I'm working on a project related to drug safety surveillance and came across Infer.Net.  I am hoping someone might be able to point me in a direction that would highlight the capabilities of Infer.Net as it relates to this type of a problem.

    A little background; I'm a developer / architect and don't necessarily have the background in computer learning / statistics that I think I would need to be able to arrive at an interesting use case on my own.  I work in health informatics supporting epidemiology and outcomes research groups.  Most of my time is spent managing medical records data.  I also get to spend some of my time exploring novel ways of exploring / analyzing the data.

    The data we have in our systems includes diagnoses, procedures, prescriptions, and demographics.  A relatively simple example of an analysis we might do would be to try and determine the effect of a particular drug on a specified outcome.  I understand that a logistic regression is typically used in this type of scenario, where the drug prescribed, age, gender, race, and outcome can be fed into a logistic regresssion to determine the effect of the drug on the occurrence of the outcome.

    Could someone describe how I might leverage Infer.Net in this type of work?

    Thanks,

    Frank

    Friday, June 3, 2011 5:42 PM

Answers

  • jwinn replied on 05-26-2010 3:19 AM

    Hi Frank,

    We have only limited resources to help people design models for new problems- we have to focus on supporting  Infer.NET-specific questions.  That said, the following examples may be of interest:

    • clinical trial- a simple model for determining if a drug is effective in a controlled trial, which does not depend on patient-specific information
    • Bayes Point Machine - a classifier which can use age, gender etc. features to predict a binary outcome

    By combining these two models, you should be able to achieve what you are aiming for.

    Best

    John W.

    Friday, June 3, 2011 5:42 PM