I want to draw your attention to two papers that were recently published in JAMIA by members of the OHDSI community.
@dsontag and his team published their 'Anchor and learn' method: http://www.ncbi.nlm.nih.gov/pubmed/27107443
@nigam, @Juan_Banda and others in the Stanford team published their phenotype process using noisy labeling: http://www.ncbi.nlm.nih.gov/pubmed/27174893
We all know that one of the biggest challenges facing our community is developing phenotype definitions that can accurately classify which patients have a given clinical disease. With thousands of clinical disease states of relavance to patients and providers, it is infeasible for us to achieve our community objective of generating reliable evidence at scale if we have to manually curate rule-based cohort definitions one-at-a-time. A novel solution is to consider developing a (semi-)automated process that treats the problem as a predictive modeling problem. This has the potential to simultaneously increase efficiency, accuracy, and generalizability. I firmly believe this is an important direction our community must pursue.
@Juan_Banda has now implemented this approach within the OHDSI community as the APHRODITE package (https://github.com/ohdsi/aphrodite). @nigam has been looking for volunteers within the community to use APHRODITE to both learn and apply models across our data network. I hope reading these two publications will provide everyone additional motivation to join this journey with them.
Great work @dsontag, @nigam, @Juan_Banda!