Consumer Health Terminology Acquisition
_{from UMLS Metathesaurs and MEDLINE N-gram Set}

I. Introduction

In addition to terms in the Lexicon, terms from UMLS are retrieved as medical terms and for a better coverage. Lexicon and this retrieved medical terms are used together as dictionary in cSpell.

II. Algorithm for Medical Terms

Medical terms are generated by the following steps:

• Terms from the latest UMLS:
  • Go through MRCONSO.RRF
  • English String (LAT = ENG)
  • Preferred Term (TS = P, SST = PF, ISPREF = Y)
  • Not Obsolete (not used)
  • Not Abb/Acr (Not used)
  • Semantic type match (from STDEF: St_abb -> STI, MRSTY.RRF: STI -> CUI)

    Category	Semantic Types
    Problem	acab, anab, bact, cgab, dsyn, inpo, mobd, neop, patf, sosy
    Interventions	diap, lbpr, topp
    Drugs	aapp, antb, clnd, drdd, phsu, vita nsba, strd (removed after 2014AA-)
    Anatomy	bdsy, blor, bpoc, bsoj, gngm, orga, orgf, phsf, tisu
    population	aggp famg, podg, popg

  • Only lower case term (mixed case and upper case contains lots of Abb/Acr, also mostly overlap with lowercase only terms)

  • Program:

    shell>cd ${PRE_PROCESS}/bin
2017AB
3
35

  • output:
    ${PRE_PROCESS}/data/Umls/${RELEASE_AX}/outData/umlsDicBySt.data.ewLc
    (ew: element word, Lc: lowercase)

• Terms form the useful legacy data (from Gopher and problem list). These terms are static and are retrieved from baseline dictionary (4 files):
  • Input:
    • noCui.data.expo
    • noCui.data.prob
  • Program:

    shell>cd ${PRE_PROCESS}/bin
2017
4
45
46

  • output:
    ${PRE_PROCESS}/data/Baseline/outData/noCui.ewLc.data

• Retrieved unigrams from UMLS resources and combine with legacy data:
  • Algorithm:
    • Tokenized unigram
    • coreTerm - remove punctuation at the leading/ending position
    • Filter out digit/punctuation, numbers, unit/measurement
    • Filter out terms/possessive already in Lexicon (or general English dictionary)
    • Customized dictionary (add and remove terms)
  • Program:

    shell>cd ${PRE_PROCESS}/bin
2017
5
53
54

  • input (${PRE_PROCESS}/data/cSpellDic/${YEAR}/inData/):
    • ewToBeAdded.data
    • ewToBeRemoved.data
    • umlsDicBySt.data.ewLc
    • noCui.ewLc.datao
    • lexicon.enEwLc.dic.addRm
  • output (${PRE_PROCESS}/data/cSpellDic/${YEAR}/outData/):
    • Med.l.dic (l: for using lexicon as English dictionary)
    • EngMed.l.dic

• Finalized words from above derived Med.l.dic:
  • Get words that are in both Med.l.dic and medline.dic
    56
  • Exclude words that are in Lexicon to gen Dic
    55
57

  • output (${PRE_PROCESS}/data/cSpellDic/${YEAR}/outData/):
    • Med.cm-l.dic (cm: consumer and medline)
    • EngMed.cm-l.dic (-l: exclude lexicon)

• Generate LexBuild Candidate list from Med.cm-l.dic:
  • Generate candidate list from Med.cm-l.dic
    60
    Output: ${PreProcess}/data/LexBuild/${YEAR}/outData/cCandidates.data
  • Sort by groouping base and plural forms of element words
    61
    Output: ${PreProcess}/data/LexBuild/${YEAR}/outData/cCandidates.data.gbp