Clinical NLP in North Germanic Languages Clinical NLP

  • Slides: 17
Download presentation
Clinical NLP in North Germanic Languages Clinical NLP in Languages Other Than English S

Clinical NLP in North Germanic Languages Clinical NLP in Languages Other Than English S 07 Sumithra Velupillai KTH, Sweden & King’s College, London Twitter: #AMIA 2017

Disclosure I and my spouse/partner have no relevant relationships with commercial interests to disclose.

Disclosure I and my spouse/partner have no relevant relationships with commercial interests to disclose. AMIA 2017 | amia. org 2

Clinical NLP in North Germanic Languages • This part of the panel will focus

Clinical NLP in North Germanic Languages • This part of the panel will focus on: • Swedish, Danish, Norwegian • Past and ongoing projects • Examples 3

HEALTH BANK –The Swedish Health Research Bank (Stockholm EPR Corpus) • Karolinska University Hospital

HEALTH BANK –The Swedish Health Research Bank (Stockholm EPR Corpus) • Karolinska University Hospital • Take. Care Intelligence • First ethical permission 2008, now on 7 th • First database 2006 -2008, 4 th to 2014 • ~ 2 million patient records • Now also in the process of being linked to primary care data • Prof. Hercules Dalianis, Dept. Computer and Systems Sciences: hercules@dsv. su. se 4

Health record content • Randomized identification number, gender, age • Documentation date stamps •

Health record content • Randomized identification number, gender, age • Documentation date stamps • Blood, laboratory values, ICD-10 codes • Drugs – ATC-codes • Free-text • Physician notes, nursing notes, radiology reports, etc. 5

Projects and Applications • Automatic surveillance of healthcare-associated infections • Detection and exploration of

Projects and Applications • Automatic surveillance of healthcare-associated infections • Detection and exploration of adverse drug events • Diagnosis code assignment (ICD-10) • Text mining in the cancer domain (Cervical cancer, pathology) • Text simplification of clinical narratives • Clinical entity recognition, semantic modifiers (negation, uncertainty, time expressions) • Comorbidity analysis • Publications: http: //dsv. su. se/en/research-areas/health/results 6

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76 -årig kvinna med hypertoni och angina pectoris. Trolig hjärtinfarkt 2 år sedan. Inkommer med centrala bröstsmärtor utan utstrålning. 76 -year old woman with hypertension and angina pectoris. Possible heart attack 2 years ago. Admitted to hospital with central chest pain without radiation. 7

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76 -årig kvinna med hypertoni och angina pectoris. Trolig hjärtinfarkt 2 år sedan. Inkommer med centrala bröstsmärtor utan utstrålning. 76 -year old woman with hypertension and angina pectoris. Possible heart attack 2 years ago. Admitted to hospital with central chest pain without radiation. 8

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76 -årig kvinna med hypertoni och angina pectoris. Trolig hjärtinfarkt 2 år sedan. Inkommer med centrala bröstsmärtor utan utstrålning. 76 -year old woman with hypertension and angina pectoris. Possible heart attack 2 years ago. Admitted to hospital with central chest pain without radiation. 9

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76 -årig kvinna med hypertoni och angina pectoris. Trolig hjärtinfarkt 2 år sedan. Inkommer med centrala bröstsmärtor utan utstrålning. 76 -year old woman with hypertension and angina pectoris. Possible heart attack 2 years ago. Admitted to hospital with central chest pain without radiation. 10

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76 -årig kvinna med hypertoni och angina pectoris. Trolig hjärtinfarkt 2 år sedan. Inkommer med centrala bröstsmärtor utan utstrålning. 76 -year old woman with hypertension and angina pectoris. Possible heart attack 2 years ago. Admitted to hospital with central chest pain without radiation. 11

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76

Example modules • Symptom and diagnosis • Negation • Uncertainty • Time information 76 -årig kvinna med hypertoni och angina pectoris. Trolig hjärtinfarkt 2 år sedan. Inkommer med centrala bröstsmärtor utan utstrålning. 76 -year old woman with hypertension and angina pectoris. Possible heart attack 2 years ago. Admitted to hospital with central chest pain without radiation. 12

Danish – Dictionary construction for ADE • Danish summaries of product characteristics – undesirable

Danish – Dictionary construction for ADE • Danish summaries of product characteristics – undesirable effects • Group-based lexicon • 7 groups, e. g. independent event, location, preposition • 4 groups for filtering, e. g. negation, temporal triggers • Post-coordination rules • Tested on Danish psychiatric records • Eriksson R 1, Jensen PB, Frankild S, Jensen LJ, Brunak S. Dictionary construction and identification of possible adverse drug events in Danish clinical narrative text. J Am Med Inform Assoc. 2013 Sep-Oct; 20(5): 947 -53. doi: 10. 1136/amiajnl-2013 -001708. Epub 2013 May 23. 13

Big. Med - Norway • Lighthouse project funded by Norwegian Resource Council 2017 -2020.

Big. Med - Norway • Lighthouse project funded by Norwegian Resource Council 2017 -2020. • Goal: novel big data solution that integrates patients’ records information, genomics data and lab data, as well as all scientific publications. • Multidisciplinary team - ICT, biomedical, law and health economy researchers and industrial partners. • Contact: Lilja Øvrelid: liljao@ifi. uio. no 14

Big. Med - Norway 15

Big. Med - Norway 15

Opportunities and Challenges + Resources in non-English languages §Annotated corpora, lexicons, rule-based modules +

Opportunities and Challenges + Resources in non-English languages §Annotated corpora, lexicons, rule-based modules + Aggregated data analysis and visualization (no PHI) - Still limited formal resources (e. g. UMLS) - Existing resources specific to dataset - Language-specific challenges: - compounds - word order - inflections 16

Thank you! Email me at: sumithra@kth. se

Thank you! Email me at: sumithra@kth. se