Does SNOMED CT post-coordination scale? Daniel KARLSSON a , Mikael NYSTRÖM a , and Ronald CORNET a, b a Department of Biomedical Engineering, Linköping University, Sweden b Department of Medical Informatics, Academic Medical Center – University of Amsterdam, The Netherlands
Post-coordination • Definition after release • A limited set of constructs • Equivalence axioms • Conjunction • Existential restriction • Role-grouping • Example 12646000 | repair of incarcerated inguinal hernia (procedure) | : { 425391005 | using access device (attribute) | = 86174004 | laparoscope, device (physical object) | }
Post-coordination • Why? • Combinatorial explosion of pre- coordinated content • Do people post-coordinate? • Yes, in information models... • and in extensions, maps, expression libraries, etc. • Is post-coordination feasible? • Needs constraints • Needs consumable technical solutions
Assumption and Research question • The “practical” size of SNOMED CT >> size of international release [Pauker, 1976; Wyatt, 1991] • Does it scale (in a technical sense)? Experiment: • Add large amounts of post-coordinated expressions in batches • Measure classification time
Material • Post-coordinated expressions • SNOMED CT release July 2013 • all possible refinements of existing clinical findings using 2 attributes (n=7 million, distinct) 404684003 | clinical finding (finding) | : { 363698007 | finding site | = X, 116676008 | associated morphology | = Y }
Method • Two tests • Add 10 000 expressions at a time until 1 million added or 4 hours passed • Add 1 million at a time until system fails • Three reasoners • ELK, Snorocket, HermIT • No incremental classification • Reasonable hardware
Results
Log-log plot of classification time
Classification time Reasoner SNOMED CT SNOMED CT Number of Order of only (s) and 100 k iterations / 4h growth (S.E.) expressions (s) ELK 0.4.1 3.6 6.0 101 1.6 (0.019) Snorocket 2.1.1 14 39 36 3.3 (0.026) HermIT 1.3.8 1900 - 6 4.0 (0.19)
Linear plot of classification time
Discussion • SNOMED CT post-coordination does scale ! • Real-life post-coordinated expressions - how different?
Discussion • EL-specific reasoners perform well, > EL reasoners not so well • Implications for more expressive languages for SNOMED CT • Wish list • Standardised test procedures and test sets
https://github.com/LiU-IMT/post_coord_test.git www.liu.se
Recommend
More recommend