Plenge et al,
Discussed a vexing problem where our gold standard subsample of the data mart appears to have sicker (or more annotated) patients than the overall RA data mart. That is the random sample appears biased. Not clear why and Kat is working with Vivian on figuring out why.
Gainer, Smoller, Perlis, Kohane, Churchill
Reviewed the gold-standard efforts for the NLP of notes. Described how we are going to go through 120 patient cases (with dozens to hundreds of notes apiece). Also reviewed the finding of substance abuse as a confounder.
Keogh, Churchill, Kohane, Murphy, Gainer, Weber
Discussed augmentation of NLP efforts
Discussed interplay with CTSA.
Discussed upcoming national meetings.
Discussed March 18th milestone in collaboration with UW/UCSF/UCD
This announcement by NIH of several new collaborative projects across CTSA includes the highlight of one relevant to i2b2:
"The University of Washington project, led by Dr. Nicholas Anderson, will develop a mechanism allowing researchers at three large, geographically distributed medical centres to easily access large shared data sets to assist in designing research studies and generating hypotheses. This team, which includes investigators from the University of California, Davis, and the University of California, San Francisco, will extend Harvard University's i2b2 software architecture to support cross-institution searches."
This is a gratifying large scale adoption of i2b2 software and the leadership of our colleagues in the Northwest is much appreciated.
Visitors from South Carolina (led Iain Sanderson), sat in on our Rheumatoid Arthritis DBP led by Robert Plenge.
Discussed the latest analyses of the increment in accuracy derived from NLP and the differences in cohort sizes and misclassification that would arise with and without various data types.
Reviewed national CTSA efforts in NLP.
Reviewed the 1.3 i2b2 release just made available.
Jong Min from Marcy MacDonald's lab (who ran one of the first i2b2 DBP's) presented follow-up on his studies of gene expression and variation in length in the huntingtin gene.
Qing, Zak, Susanne, Shawn, Jordan, Roy, Griffin, Dan, Vivian
Discussed the definition of normality for the imaging controls and how far back in time we could pull normal MRI's.
Reviewed the validation methodology for cohort ascertainment. This include the review of the flow from the original RPDR database all the way through the various NLP filters and human validation steps. Clear consensus that we need to focus on Positive Predictive Value and Negative Predictive Value. Significant discussion of what constitutes a positive case.
