Methods Discussion
A discussion on the methods useful for the competition
-
I’ve posted a set of benchmarks based on the Disproportionality Analysis program produced by the OMOP team. You can find it at http://omop.fnih.org/MethodsLibrary as a zip file, with the program implemented in SAS. Here are the parameters used in each benchmark. More details about the methods are in the pdf contained in the zip file.
The beginning of each benchmark is the disproportionality metric
- PRR – Proportional Reporting Ratio
- BCPNN – Bayesian Confidence Propogation Neural Network
- EB05 – Empirical Bayes geometric mean, 5th percentile of posterior distribution
All of the DPA benchmarks have the following parameters:
- CONDITION_TYPE: 2 (incident cases)
- SURVEILLENCE_WINDOW: 0
- DRUG_PERSISTENCE_WINDOW: 30
- CONDITION_PERSISTENCE_WINDOW: 30
After the name of metric, the remaining letters signify additional options:
- P – ‘Distinct Patients’ counting scenario
- M – ‘Modified SRS’ counting scenario
- S – stratification by age, sex, and year (If this is absent, no stratification)
-
I’ve posted a few more challenge 1 benchmarks based on a method called Bayesian Logistic Regression (BLR), available at http://omop.fnih.org/MethodsLibrary. Here’s a summary of the parameters for the three variants:
BLR – Normal prior, variance 1, not stratified
BLR-S – Normal prior, variance 1, stratified on age and sex
BLR-L – Laplace prior, variance 1, not stratified -
Hi Eric,
Will you have a chance to post either BCPNN or BLR benchmark for Challenge 2? No pressure :)
Harris.
-
I would love to run the DP and BLR code posted at the Methods Library, but I don't have access to SAS. Any suggestions on workarounds, e.g. open-source equivalents? SAS Learning Edition appears to be limited to 1500 observations in the input. Thanks.
Jeff Howbert
-
Since we took the OMOP competition and we did not have the chance to talk to each other, we would like invite OMOP competition participants to present your current research on INFORMS meeting which is the largest conferece on operations research and management sciences area. This year, we host the session "Efficient Correlation Search on Complex Data" and we hope you can present a topic in this session. Since the solutions to the OMOP Cup are more like the side-product of your research, you can also choose any topic related to correlation search or pattern searching on complex data instead of the solutions to the OMOP Cup. If you are interested, please contact Lian Duan at lian-duan@uiowa.edu. The INFORMS meeting will be held in Austin, Texas on November 7-10. There are always 75 parrelled sessions being held during the conference period. The three major topics in this conference are optimazation, statistics, and data mining. We believe there are always some interesting topics to you. Thank you very much for your time!
5 comments