David Hanauer , MD, MS University of Michigan - PowerPoint Presentation

1. David Hanauer, MD, MSUniversity of Michiganhanauer@umich.eduTwitter: @DavidHanauer2MARCH Informatics Committee MtgJuly 7, 2020http://project-emerse.org/presentations.html

2. these slides can be found at:If you’re thinking of taking noteslink will be onmost slideshttp://project-emerse.org/presentations.html

3. 3slides2live demo1Q&Aroadmap fortoday’spresentationhttp://project-emerse.org/presentations.html

4. What is EMERSE?A search engine, kind of like Google, for free text medical record documents(technically, it’s an information retrieval system) Simple to use Designed to be self-serve* http://project-emerse.org/presentations.html*Once access is approved

5. What is EMERSE?A research tool, available at no costOpen source, Apache 2.0 licensingPicking the right tool for your research is importantmultiple tools are often neededhttp://project-emerse.org/presentations.html

6. Unstructured vs Structured DataEMERSE is for this…i2b2 is for this…Unstructured Data (free text)Structured DataMrs. Jones is a 56 year old female with a h/o HTN, T2DM, and, and high cholesterol who comes to the clinic today with a 3 day h/o coughing, fever, and severe headache.WBC: 5.6Total cholesterol: 182Weight: 67.4AST: 30ALT: 52http://project-emerse.org/presentations.html

7. 80% of EHR data are in unstructured free textstructured dataunstructured data/free texthttp://project-emerse.org/presentations.html

8. 80% of EHR data are in unstructured free text"potential cases may be missed or wrongly dated when coded data alone are used..” (PMID: 23964710)"structured data is best used for first pass filtering of EHR data…prior to a more nuanced second pass using unstructured data" (PCMID: PMC4333685)"...urinary incontinence, an important patient-centered outcome following prostate cancer treatment, was reported almost exclusively in the free text of EHRs and was rarely coded as an ICD 9 diagnosis code.” (PMCID: PMC4899050) "Claims and structured EHR data give an incomplete picture of burden related to geriatric syndromes. Geriatric syndromes are likely to be missed if unstructured data are not analyzed." (PMID: 29972595)http://project-emerse.org/presentations.html

9. What isn’t EMERSE?It is not: natural language processing (NLP)but full NLP may not always be neededand NLP can be hard to use or customizehttp://project-emerse.org/presentations.html

10. EMERSE vs NLPEMERSE: augmented intelligenceNLP: artificial intelligencehttp://project-emerse.org/presentations.html

11. EMERSE vs NLPEMERSENLPhttp://project-emerse.org/presentations.html

12. With EMERSE: find cohortsEMERSE allows you to find cohorts based on things mentioned in the notesdiseasesdrugssymptomsmore!http://project-emerse.org/presentations.html

13. Find cohortsIt’s perfect for finding diagnoses that don’t have a specific ICD-10 code… like leiomyosarcoma…https://www.icd10data.com/search?s=leiomyosarcomahttp://project-emerse.org/presentations.html

14. Highlight documentsfor chart reviewhttp://project-emerse.org/presentations.htmlhttps://www.ncbi.nlm.nih.gov/pmc/articles/PMC7256550/A chest x-ray was obtained and revealed the presence of patchy consolidations in bilateral lower lobes. CT chest/abdomen/pelvis (Fig. 1 ) was notable for diffuse ground-glass opacities in the periphery of bilateral lungs as well as splenomegaly with a splenic hypodensity measuring 8.6 × 0.7 cm and surrounding peri-splenic edema, consistent with a splenic infarct. Hydroxychloroquine and a one-time dose of Tocilizumab were administered along with supportive measures. A continuous heparin infusion was also initiated and patient was admitted to the general medical ward for further workup.

15. Chart reviewhttp://project-emerse.org/presentations.htmlResearch team at Michigan Medicine usedEMERSE to identify presenting symptoms forCOVID-19, including:dyspnearhinorrhea or congestionvomitingwheezingmyalgia or arthralgiaabdominal paincoughfatigue or malaisechest pain or tightnesssputum productionheadacheloss of smell or tasteblood in sputumloss of appetitealtered mental statussore throatdiarrheaweaknessfever or chillsnausealightheadedness or syncope

16. EMERSE query expansionhttp://project-emerse.org/presentations.html

17. EMERSE query expansionhttp://project-emerse.org/presentations.htmlCan use “EMERSE Synonyms” or upload your own dataset (e.g., ICD-10, human phenotype ontology, MeSH terms)Simple file formathttp://project-emerse.org/documentation/administrator_guide.html#truepreparing-a-synonyms-file

18. EMERSE Synonymshttp://project-emerse.org/presentations.htmlManually curated over past 12+ yearsTerms/phrases not found in standard vocabularies, taxonomies, ontologies (but found in clinical notes)Academic and non-academic licensing available*> 1.1 million unique rowsDetails are at: http://project-emerse.org/synonyms.htmlTo explore a subset, visit: http://www.davidhanauer.com/EMERSE_synonyms_T_Terms/*David Hanauer is entitled to a portion of the license fees

19. EMERSE Synonymshttp://project-emerse.org/presentations.htmlCategoryExampleAcronymsCXRchest x-rayAbbreviationschemotherapychemoProfessional and consumer termsemesispukingMisspellingsdiarrieadiahhreaTrade and generic drug namesTylenolacetaminophenChemotherapy regimensFOLFOX, folinic acid, fluorouracil, oxaliplatinWord Stemssleep, sleeping, sleepyCategoryExampleTrue synonymsfrigidcoldPhrasing variationslido w/ epiepi-lidocaineMalapropisms/usage errorsbeast massprostrate cancerRoot word variationsadenotonsillectomytonsilloadenoidectomyHyphenationspseudosarcomatouspseudo-sarcomatousIdiomsunder the knifeunder the weatherNeologismstrabecularity(vs trabeculation)

20. EMERSE Bundleshttp://project-emerse.org/presentations.htmlSaved sets of search termsShared among usersHelps standardize searchesSupports reproducibility - Some teams publish Bundles in their publication’s appendix (example: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3860174/)

21. What’s in EMERSE?Free text notesadmission noteshistory and physicalprogress notespathology reportsradiology reportsmore!@ Michigan: 2.5 million patients 170 million notes 1995-presenthttp://project-emerse.org/presentations.html

22. What’s not in EMERSE?Structured datadiscrete lab valuesvital signsbilling codesOther tools such as i2b2 exist for structured datahttp://project-emerse.org/presentations.html

23. Typical workflowinitial cohort identifiedwith a cohort discoverytool (i2b2) using structured data (or via EMERSE‘Find Patients’)http://project-emerse.org/presentations.html

24. Typical workflowadditional details abstracted from the free text using EMERSEhttp://project-emerse.org/presentations.html

25. Typical workflowData recorded in an electronic data capture system (REDCap)http://project-emerse.org/presentations.html

26. Publications using EMERSE354 papers54 abstractsFull list can be found at:http://project-emerse.org/publications.htmlhttp://project-emerse.org/presentations.html

27. Where is EMERSE?installedinstallinghttp://project-emerse.org/presentations.htmlU of CincinnatiUNC-Chapel HillUCSFU of MichiganU of KentuckyU of VirginiaCase Western ReserveColumbia

28. Getting EMERSElocal installation requiredneed support of IT teamsimilar to i2b2EHR-agnosticruns independent of EHRneed a way to extract documents to send to EMERSEhttp://project-emerse.org/presentations.html

29. Test driving EMERSEonline demovideos on websitevirtual machine - could be used for individual projectshttp://project-emerse.org/presentations.html

30. project-emerse.orghttp://project-emerse.org/presentations.html

31. hanauer@umich.eduhttp://project-emerse.org/presentations.html

32. AcknowledgementsU of MichiganRogel Cancer Center[P30CA046592]Michigan Institutefor Clinical andHealth Research(MICHR)[UL1TR000433]U of MichiganHealth Informationand TechnologyServices (HITS)NCI InformationTechnology forCancer Research(ITCR)[U24CA204863]http://project-emerse.org/presentations.html