Clinical computing and the repository George Hripcsak Jim

Clinical computing and the repository George Hripcsak Jim Cimino Pete Stetson

Web. CIS Data sources MED Info resources Repository CDW Data mart

George Hripcsak • MD – internal medicine • MS – biostatistics • Professor and Vice Chair

Repository • 13 years • > 2, 000 patients • > 5, 000 narrative reports • Data are useful only for the purpose for which they were collected • Data collected without a purpose are useless

• We are overflowing with information • We just need to know how to tap it

Data mining • We can use the information in the repository to discover knowledge about medicine and biology • So close and yet so far

Range of projects Theoretical Practical

(Use of natural language processing) • Carol Friedman, Steve Johnson

Temporal representation and reasoning 2003 -09 -06 -08: 11: 34. 243563 My pneumonia was a while before the stroke

Case-based reasoning • Match on a case level (gestalt) to overcome inaccuracies and missing data

Evaluation • Use computer intensive techniques to figure out whether you have a significant data mining result

Summarization • Give providers a quick summary of the important facts about a patient

Patient safety • Detect (and someday prevent) medical errors using the repository

Jim Cimino • MD – internal medicine • Postdoctoral fellowship (MGH) • Professor

Data Re-View and Re-Use • View: look at data in their original form • Re-View: alternate views of clinical info • Use: original intent for collecting data • Re-Use: any other use

Data Re-View • For other users: Pat. CIS

Data Re-View • For other users: Pat. CIS • For other platforms: Palm. CIS

Data Re-View • For other users: Pat. CIS • For other platforms: Palm. CIS • Problem-oriented views: Qing. CIS

Data Re-View • For other users: Pat. CIS • For other platforms: Palm. CIS • Problem-oriented views: Qing. CIS • Using patient data to improve information retrieval: Mendon. CIS

Data Re-View • For other users: Pat. CIS • For other platforms: Palm. CIS • Problem-oriented views: Qing. CIS • Using patient data to improve information retrieval: Mendon. CIS

Data Re-Use • Examples of reuse: – Summary reports – Automated decision support – Context-specific information retrieval (infobuttons) • Challenge: – Mapping concepts of raw data to concepts of reuse • Solution: – Medical Entities Dictionary

MED Structure Medical Entity Substance Chemical Laboratory Specimen Anatomic Substance M Substance Laboratory Test en Glucose Diagnostic Procedure Laboratory Procedure im Bioactive Substance c pe Carbohydrate nce a t s d Subample S s. S Ha Plasma Specimen Event easured Plasma Glucose Test Part of CHEM-7

Translations with the MED Intravascular Gentamicin Tests Serum Gentamicin Level Summary Reports is-a Sub Injectable Gentamicin stan t re ce M easu r ed Gentamicn Sensitivity Test n die Gentamicin resy u s a it Me nsitiv Se g s in Ha Decision Rule Eti olo gy Gentamicin Toxicity Drug Information Expert System

Research Issues • Identifying information needs • Developing automated solutions • Conceptual mapping • Terminology management

Pete Stetson • Training: – MD – Internal Medicine – MA – Informatics (Columbia’s RMA) • Positions: – Assistant Professor – Associate Program Director, IM Residency – Chair, Institutional Decision Support and Clinical Alerts Committee

Projects “Real solutions for real users. ” Patient Safety: – Detection: • Computerized algorithms to screen the electronic medical record for adverse events (CLIPS) – Prevention: • Clinical Alerts • Information Access – Palm. CIS • Clinical Communication – – Understanding how physicians talk to eachother (Sign-outs) Increasing the portability of patient information (Palm. CIS) Matching patients to providers (Patient-Provider Index) Improving capture of narrative data (Hospitalist e. Note)

Potential Student Projects • Detection: – Using NLP to screen narrative data for evidence of adverse events • Prevention: – Development and ownership of alerts (eg: K+ > 6. 0) – User interface design for new Palm. CIS platforms – Development of an algorithm to match patients to their primary providers (PPI) – User interface design, evaluation, real-time NLP, summarization views of clinical notes (Hospitalist e. Note)

Hospitalist e. Note Architecture Web. CIS User MD Input CDR CDW Billing Office MD Review Billing Compliance Officer To IDX