Statistical methods competition

Posted on by

From David Madigan:

The Observational Medical Outcomes Partnership (OMOP) seeks new statistical and data mining methods for detecting drug safety issues through the OMOP Cup Methods Competition.

Observational data analyses algorithms are routinely used throughout the world to monitor the safety of prescription drugs. With the increasing attention and implementation of electronic healthcare information, new and improved analysis approaches are urgently needed. The OMOP Cup is your opportunity to contribute new methods to monitor drug safety.

The OMOP Cup provides a simulated dataset of 10 million hypothetical patients with drug and condition information. Known drug safety issues are present in the simulated dataset and your charge is to effectively and efficiently identify these issues. The OMOP Cup has two challenges:

Challenge 1: Explores how well methods work when against an entire dataset, targeting the accurate classification of which drugs are associated with which outcomes.

Challenge 2: Evaluates the timeliness of detection of drug-event associations by having methods run against data sequentially as it accumulates over time.

The total prize money is $20,000. Winners will be required to place their algorithms in the public domain. Details about the OMOP Cup Methods Competition can be found at: http://omopcup.orwik.com. The competition is being conducted by the Observational Medical Outcomes Partnership (http://omop.fnih.org), a public-private research partnership of the Foundation for the National Institutes of Health.

$20,000 doesn’t seem like a lot, on the other hand if it’s interesting it could be fun to do for free. Maybe they could also give the winner a year’s supply of Vicodin or something like that?

2 thoughts on “Statistical methods competition”

  1. Here's the competition home page:

    http://omopcup.orwik.com/

    And here are the official rules [pdf]:

    http://competition-files.s3.amazonaws.com/OMOP_Cu…

    And most interestingly, the call for proposals from OMOP to generate the simulated data:

    http://omop.fnih.org/sites/default/files/OMOP%20R…

    That should help with predictions, as it goes over how they were planning to simulate the data, right down to numerical constants and classes of drug/disease relationships to generate.

    Just to clarify, winners are not "required to place their algorithms in the public domain" as David's e-mail said, but in the words (and capitalization) of the official rules:

    ENTERING A ENTRY IN THIS COMPETITION CONSTITUTES ENTRANT'S IRREVOCABLE ASSIGNMENT, CONVEYANCE AND TRANSFERENCE TO SPONSOR OF ANY AND ALL RIGHT, TITLE AND INTEREST IN THE ENTRY INCLUDING, WITHOUT LIMITATION, ALL COPYRIGHTS.

    If you're used to dealing with software licenses, the real shocker is this, also from the official rules:

    e) the Participant will indemnify and hold harmless the Released Parties (as defined in Rule XIII below) from any claims and damages (including reasonable attorneys' fees) arising from or relating to the breach or alleged breach of your representations and warranties herein, your Entry, in connection with this Competition.

    Looks like I'll be looking for another contest.

  2. Hi Bob, good catch! Apparently they posted the wrong version of the legalese. A corrected version will be posted this morning. OMOP has an extensive network of methods developers and all algorithms and methods will be in the public domain including competition entries.

    As regards the simulator, yes, the specifications and shortly the code will be in the public domain. This might help a competitor who really digs into it, and good luck to anyone who attempts to do so!

    Regards, David

Comments are closed.