Death and its predictors

Professor Tianxi Cai ran a very interesting discussion about generating risk profiles for biomarkers and the problem in adjusting out the diseases that are along the causal pathway to mortality of those biomarkers. As a result of such an over adjustment, most biomarkers will give counter-conventional-wisdom associations.

NLP

Guergana's team: discussed speed issues.

cTAKES with full UMLS module takes about 150 notes/hour/core. Discussed various architectural tradeoffs.

Accelerating NLP

Tianxi Cai led a discussion of "Semi Automated Model Building Using Knowledge Bases" using bibliographic external sources of medical knowledge to generate the candidate terms for a NLP filter to phenotype a particular characteristic.

NLP games

Cai, Savova et al.

Discussed tradeoffs between performing string matching (hard to scale to the general case, but quite adequate accuracy for well specified and specific narrow cases) and general NLP. Also discussed how to enrich prior probability for a disease of interest to increase performance of high specificity NLP.

Identity Management

Discussed a variety of techniques to include different quality master patient indices unified with the identity system within i2b2.

Security Policy and Identity Management

Churchill, Kohane, Murphy, Wattanasin, Bickel, Simons Discussed different configurations of identity management to support a range of solutions of i2b2 database that are along the spectrum of identified to de-identified.

Review of methodology

cTAKES discussion

Savova et al.,

Discussed tighter i2b2 integration with cTAKES. Discussed representation of modifiers and how to communicate to the community which subset of which standardized ontology was used for each term. Pei demonstrated a very impressive preliminary integration.

AUG meeting and other planning

Kohane, Churchill, Murphy, Weber, Mendis, Bicket et al.,

Discussed the agenda for this Summer's AUG and at next week's TBI Summit at AMIA.

Discussed impending very LARGE implementations of SHRINE (independent of this core group).

NLP and i2b2 Status Report

Guergana led a discussion on the current state. Total team is currently 7-9 individuals (depending on how directly they are involved for a given project).

Currently, in i2b2, NLP is cast as a supervised classification task that is applied to patients that are filtered according to various criteria (e.g. lab values, ICD-9 codes). Domain experts typically annotate (extensively) 100+ charts as part of the gold-standard (used for supervised learning). A subset of these annotations are done twice (at least) to assess inter-annotator agreement.

The cTAKES has processed 28M documents at Partners Healthcare System to date.

Reviewed recent results with IBD (NLP alone almost as good as NLP+Codified data), Multiple Sclerosis (NLP helped but the combination of NLP + codified is significantly better).

Finding the timing of the myocardial infarction

Liao, Shaw, Tsai, Kohane, Churchill, Savova (Murphy absent)

Discussed various ways to annotate the timing of heart attacks of heart attacks. Did a little review of the temporal utility package.

Governance models

We met this morning to discuss several alternative governance models. Given the spreading use and an increasingly vibrant community, we will be developing an even more pro-active and long-term plan to support i2b2's continued growth. Details to be presenting at the Summer i2b2 AUG meeting.

Topic models and unsupervised NLP

Very nice presentation from Tim Miller today about work at i2b2 on this.

Review of Adaptive Lasso

Professor Cai reviewed for us the Adaptive Lasso function. She also provided the framework to understand the similarities between various machine learning formalisms (e.g. Hinge Loss function and SVM)

Also discussed the fundamental bias vs variance tradeoff. Approximation error (i.e. from selection from the feature space). Variance (i.e. sample error). If you have lower approximation error, for example, you can afford a high sample error (and vice versa).

SHRINE & i2b2

Murphy, Churchil, Kohane et al

Reviewed the next steps in disseminating SHRINE and necessary architecture additions.

Technical Futures

Shawn, Kohane, NIch, Mike, Zak, Susanne, Bickel Discussed several important minimal commitment abstractions useful for temporal reasoning and for suggesting related queries.

Distributed queries

Griffin Weber, Shawn Murphy, Bill Simons, Susanne Churchill, Zak Kohane Discussed latest national developments for SHRINE queries. Very significant support and enthusiasm from the AUG. Reviewed temporal query capabilities and roadmap.

i2b2 Version 1.6 release preparations

Kohane, Churchill, Murphy, Mendis, Wattansin Last minute preparation for documentation and squishing any remaining bugs.

SMART integration with i2b2 reviewed

Discussed how additional SHRINE influences future directions of i2b2. Then demonstrated the latest i2b2/SMART integration (quite impressive!).

Discussion of which pieces have to be locally vs centrally hosted

For the next generation of i2b2, we are trying to minimize how many cells have to be locally hosted. Even more crucially, how many cells have to be locally customized and supported and how many can be generic. In other words, can we or should we reposition any abstraction barriers?