NLP Working Group Chunnan Hsu Biomedical Informatics UC

  • Slides: 9
Download presentation
NLP Working Group Chunnan Hsu Biomedical Informatics UC San Diego October 13, 2016 Supported

NLP Working Group Chunnan Hsu Biomedical Informatics UC San Diego October 13, 2016 Supported by the Patient-Centered Outcomes Research Institute (PCORI) Contract CDRN-1306 -04819 1

Current members § § § § Chunnan Hsu Ramana Seerapu Scott Duvall Olga Patterson

Current members § § § § Chunnan Hsu Ramana Seerapu Scott Duvall Olga Patterson Hua Xu Michael Matheny Glenn Gobbel Tsung-Ting Kuo Supported by the Patient-Centered Outcomes Research Institute (PCORI) Contract CDRN-1306 -04819 2

Scope § The NLP working group is tasked to accurately extract phenotypes for three

Scope § The NLP working group is tasked to accurately extract phenotypes for three clinical conditions: Kawasaki Disease (KD), Weight Management / Obesity (WM/O), and Congestive Heart Failure (CHF), from tens of millions of clinical notes shared by participating institutes in p. SCANNER, and seamlessly integrate with the shared structured data. Supported by the Patient-Centered Outcomes Research Institute (PCORI) Contract CDRN-1306 -04819 3

Milestones § Refine the problem definitions of the extraction of common data elements to

Milestones § Refine the problem definitions of the extraction of common data elements to guide the finalization of a common evaluation guideline, mapping to a common output data model, and design, reuse, and sharing of NLP tools § Finalize output format definitions with OHDSI § Create secure, privacy-preserving, cross-institution clinical NLP infrastructure where tens of millions of clinical notes can be processed and the quality of processing can be assessed semi-automatically. § With the infrastructure, create large data warehouse of NLP extracted data from clinical notes to support phenotyping of the three p. SCANNER use case conditions. Supported by the Patient-Centered Outcomes Research Institute (PCORI) Contract CDRN-1306 -04819 4

Workflow UCSD AD/VPN i. DASH VPN Pool i. DASH 2 -Factor Authentication i. DASH

Workflow UCSD AD/VPN i. DASH VPN Pool i. DASH 2 -Factor Authentication i. DASH VDI Remote Desktop EHR Data Warehouse IRB Approval . . … …. . . Common Data Elements i. DASH Midas Interface i. DASH Midas Annotated Clinical Notes NLP Ensemble Pipeline . . . … …. . . . . Clinical Notes Cohort Identification Interface Target Phenotype N 112 N 431 N 886 P 002 P 293 P 534 Retrieved Patient ID List . . . . … …. . . . . Phenotype Database OMOP Common Data Model Annotation Review Interface (Remote Desktop) Reviewed Annotations Cohort Identification from Clinical Texts (CICT) Retrieved Note ID List Supported by the Patient-Centered Outcomes Research Institute (PCORI) Contract CDRN-1306 -04819 5

Current Progress UCSD AD/VPN i. DASH VPN Pool i. DASH 2 -Factor Authentication i.

Current Progress UCSD AD/VPN i. DASH VPN Pool i. DASH 2 -Factor Authentication i. DASH VDI Remote Desktop EHR Data Warehouse IRB Approval . . … …. . . Common Data Elements i. DASH Midas Interface i. DASH Midas Annotated Clinical Notes NLP Ensemble Pipeline . . . … …. . . . . Clinical Notes . . . . … …. . . . . Annotation Review Interface (Remote Desktop) Reviewed Annotations Cohort Identification from Clinical Texts Supported by the Patient-Centered Outcomes Research Institute (PCORI) Contract CDRN-1306 -04819 6

CICT: NLP Ensemble Pipeline De-identified Clinical Notes. . … …. . . Common Data

CICT: NLP Ensemble Pipeline De-identified Clinical Notes. . … …. . . Common Data Elements NLP Preprocessor Deidentification NLP Toolkit NLP Ensemble Sentence Splitter Annotated Clinical Notes c. TAKES Intersection (Good) Annotation Tags Union (Iffy) Extracted Data Elements Meta. Map Encoding Converter NLP Postprocessor CLAMP EFex Supported by the Patient-Centered Outcomes Research Institute (PCORI) Contract CDRN-1306 -04819 . . . … …. . . . . Common Data Model 7

CICT: Secured Environment User Institution Data Center UCSD Computer Center HIPPA Cloud User Institution

CICT: Secured Environment User Institution Data Center UCSD Computer Center HIPPA Cloud User Institution VPN User i. DASH Institution Firewall i. DASH Midas EHR Data Warehouse User Institution Firewall User Institution i. DASH Firewall SSH / Remote Desktop i. DASH VM Terminal 2 -Factor Authentication UCSD 2 -Factor Authentication User Institution Supported by the Patient-Centered Outcomes Research Institute (PCORI) Contract CDRN-1306 -04819 8

Current Progress UCSD AD/VPN i. DASH VPN Pool i. DASH 2 -Factor Authentication i.

Current Progress UCSD AD/VPN i. DASH VPN Pool i. DASH 2 -Factor Authentication i. DASH VDI Remote Desktop EHR Data Warehouse IRB Approval . . … …. . . Common Data Elements i. DASH Midas Interface i. DASH Midas Annotated Clinical Notes NLP Ensemble Pipeline . . . … …. . . . . Clinical Notes . . . . … …. . . . . Annotation Review Interface (Remote Desktop) Reviewed Annotations Cohort Identification from Clinical Texts http: //textmining. ucsd. edu: 5005 Supported by the Patient-Centered Outcomes Research Institute (PCORI) Contract CDRN-1306 -04819 9