Effect of Out Of Vocabulary Terms on Inferring Eligibility Criteria for a Retrospective Study in Hebrew EHR

What it is about

The authors classify Hebrew clinical notes from a paediatric neurology unit (894 visits by 516 patients in 2009, about 226K tokens) as eligible or not for a retrospective epilepsy study. The pipeline runs morphological analysis and segmentation on Hebrew words, aligns the useful words to UMLS concepts and feeds the most frequent noun lemmas to a classifier. The main obstacle is out-of-vocabulary terms: 30% of the token types were missing from the morphological lexicon, against 7.5% in newswire, and 71.5% of a sample of unknown tokens were transliterated medical English. The authors build a transliteration lexicon of 2,507 lemmas (75% precision) by transliterating UMLS English terms into Hebrew spellings and matching them to unknown corpus forms. That lexicon halves segmentation errors and raises the average F-score of an SVM from 83.6% to 87%; with decision trees the score rises from 76.8% to 82.6%.

Ideas

• I1. Classify unknown tokens before segmenting them: transliteration, misspelling or missing lexicon word (the paper's split is 71.5%, 16% and 13.5%). Transliterations go to a loan-word path; the other two go to the ordinary prefix-suffix search.
• I2. A transliteration lexicon built from an English term list: generate the Hebrew spellings an English term may take, match them to unknown forms in a corpus, and keep the pairs that pass a heuristic filter. Check a random sample by hand to estimate precision, as the paper does.
• I3. A segmentation test with variant forms: the paper's drug name appears as three segmentations (plain, with the preposition b+, with h+ for "the"). A gold list of such variant groups tests whether the segmenter reaches one lemma for all of them.

For Mova

I1 and I3 go into Prepositions, prefixes, conjunctions and particles, where the prefix splits are described. I2 goes into 8. Lexicon and register as the loan-word section, and into tools/langseg if its rules need a transliteration class.

Cite

The paper itself, not this card. BibTeX from the library's record of the work.

@inproceedings{cohen2013effect,
  title = {{Effect of Out Of Vocabulary Terms on Inferring Eligibility Criteria for a Retrospective Study in Hebrew EHR}},
  author = {Cohen, Raphael and Elhadad, Michael},
  booktitle = {Proceedings of the 2013 Workshop on Biomedical Natural Language Processing},
  year = {2013},
  url = {https://aclanthology.org/W13-1915/}
}

References

5 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. The Electronic Health Record (EHR) contains information useful for clinical, epidemiological and genetic studies. This information of patient symptoms, history, medication and treatment is not completely captured in the structured part of the EHR but is often found in the form of free-text narrative. A major obstacle for clinical studies is finding patients that fit the eligibility criteria of the study. Using EHR in order to automatically identify relevant cohorts can help speed up both clinical trials and retrospective studies (Restificar, Korkontzelos et al. 2013). While the clinical criteria for inclusion and exclusion from the study are explicitly stated in most studies, automating the process using the EHR database of the hospital is often impossible as the structured part of the database (age, gender, ICD9/10 medical codes, etc.’) rarely covers all of the criteria. Many resources such as UMLS (Bodenreider 2004), cTakes (Savova, Masanz et al. 2010), MetaMap (Aronson and Lang 2010) and recently richly annotated corpora and treebanks (Albright, Lanfranchi et al. 2013) are available for processing and representing medical texts in English. Resource poor languages, however, suffer from lack in NLP tools and medical resources. find
2. Dictionaries exhaustively mapping medical terms to the UMLS medical meta-thesaurus are only available in a limited number of languages besides English. NLP annotation tools, when they ∗ Supported by the Lynn and William Frankel Center for Computer Sciences, Ben Gurion University Proceedings of the 2013 Workshop on Biomedical Natural Language Processing (BioNLP 2013), pages 116–119, Sofia, Bulgaria, August 4-9 2013. c 2013 Association for Computational Linguistics Michael Elhadad Computer Science Dept. find
3. Ben-Gurion University in the Negev elhadad@cs.bgu.ac.il exist for resource poor languages, suffer from heavy loss of accuracy when used outside the domain on which they were trained, as is well documented for English (Tsuruoka, Tateishi et al. 2005; Tateisi, Tsuruoka et al. 2006). In this work we focus on the problem of classifying patient eligibility for inclusion in retrospective study of the epidemiology of epilepsy in Southern Israel. Israel has a centralized structure of medical services which include advanced EHR systems. However, the free text sections of these EHR are written in Hebrew, a resource poor language in both NLP tools and hand-crafted medical vocabularies. Epilepsy is a common chronic neurologic disorder characterized by seizures. These seizures are transient signs and/or symptoms of abnormal, excessive, or hyper synchronous neuronal activity in the brain. Epilepsy is one of the most common of the serious neurological disorders (Hirtz, Thurman et al. 2007). find
4. Corpus We collected a corpus of patient notes from the Pediatric Epilepsy Unit, an outpatient clinic for neurology problems, not limited to epilepsy, in Soroka Hospital. This clinic is the only available pediatric neurology clinic in southern Israel and at the time of the study was staffed by a single expert serving approximately 225,000 children. The clinical corpus spans 894 visits to the Children Epilepsy Unit which occurred in 2009 by 516 unique patients. The corpus contains 226K tokens / 12K unique tokens. find
5. Figure 1 – Decision Tree for inclusion/exclusion. Sodium Valproate (dplpt) is a key term which is often segmented incorrectly. The patients were marked by the attending physician as positive or negative for epilepsy. In the study year, 2009, 208 patients were marked as positive examples and 292 as negative. The inclusion criteria were defined as history of more than one convulsive episode excluding febrile seizures. In practice, the decision for inclusion was more complex as some types of febrile seizure syndromes are considered a type of epilepsy while some patients with convulsion were excluded from the study for various reasons. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2013-cohen-effect-out-vocabulary-terms-inferring/