How can ontologies help with electronic health records? Kent Spackman, MD PhD
Bottom line: ICBO, 30th July 2011 • The world is not made up of questions and answers • And a question-answer view impedes data re-use and standardization • But in health care (at least), nearly everyone thinks of data that way • Ontologies hold (one) key to solving that problem
Decision support ICBO, 30th July 2011 • Central to the Value of Semantic Interoperability • Numerous studies document the ability of computerized decision support to decrease costs and improve quality • But use is limited • One major barrier is lack of standardization • Clinical terminology/ontology standards help fill this need • but we aren’t there yet
Ontology enables decision support: influenza vaccination • decision support program criterion: • chronic cardiorespiratory disorders • patient record: • mild persistent asthma ICBO, 30th July 2011
Ontology enables decision support: hemoglobin A1C interpretation • decision support program asks for: • hereditary anemia due to disturbance of hemoglobin synthesis • patient record says: • Aγ β+HPFH and β0thalassemia in cis ICBO, 30th July 2011
Ontology enables decision support: antibiotic therapy • decision support program asks for: • bacterial effusions • patient record says: • tuberculous ascites ICBO, 30th July 2011
What’s the problem? • No single barrier • Inertia of existing systems • Cost of change & lack of clear return for investments in change • Barriers due to questions about standards: • Choice of different standards for same purpose • Quality, reliability, and implementability • Inadequate coordination between those with different purposes (e.g. terminology vs. information model) ICBO, 30th July 2011
Need for information model ICBO, 30th July 2011 • Clinical statements require an information model • The simplest information model is just • “put your information here: __________________” • This is absurd, especially for data that ordinarily goes into fields such as : • Name, ID, Date of visit
Balance, overlaps, gaps TMTOWTDI Consider how to record the fact that the patient’s blood type is “RH positive”. What is the information model (field in the record)? What is the terminology (value put in the field)?
Balance, overlaps, gaps Record the fact that “malignant mesothelial cells were found in a pleural fluid aspirate”:
Balance, overlaps, gaps TMTOWTDI There is no single best way to split assertions between the information model and the terminology model (or between the observables and the other values in the terminology!) The best we can do is recognize equivalence The best foundation for representing reality is formal ontology The best tools for recognizing equivalence (by machine) are logic-based Therefore, a formal ontology that supports a logic-based model of semantics is the foundation not just for the terminology but also for the vast majority of uses of the combination of terminology and record elements
Clinical Statements • Are the basis for a common view of patient record structure. • The electronic medical record can be viewed as a collection of statements • A faithful record of what clinicians have heard, seen, thought, and done • Referring to / about the patient and the clinical reality • Other requirements for a medical record, follow naturally from this view • that it be attributable and permanent • that it is possible for statements to be wrong, probabilistic, controverted by other parties, etc. Rector ,Nowlan & Kay (1991) Foundations for an electronic medical record. Methods Inf Med. 30:179-186, 1991
The central role of narrative in clinical documentation ICBO, 30th July 2011 Clinicians communicate by telling stories. Not everything in the stories is easily formalized. That should not stop us from gathering data for decision support. It also should not divert us into thinking NLP will solve the problem (it won’t).
Clinical statement illustrationUnstructured view “February 5, 2008. Mr. Harvey Q. Patient was seen at Community Health Clinic by Dr. Smith.He complained of pain in the right calf. The doctor examined the right leg, which showed swelling and tenderness over gastrocnemius. Doppler ultrasonography revealed a proximal DVT. He was prescribed LMW heparin 70mg SC bid, and plan to RTC in 2 days to begin warfarin therapy. A request and specimen for INR, ATIII and Protein C level were sent to PathLab laboratory".
Mr Harvey Q. Patient Community Health Centre Identifying classes, context, values Dr Smith • Mr Harvey Q. Patient 5-Feb-2008 • 01) visit to Community Health Centre, seen by Dr Smith • 02) Complained of pain the right calf • 03) Swelling and tenderness over right gastrocnemius • 04) Doppler ultrasonographydone • 05) Diagnosis of right proximal deep venous thrombosis • 06) Prescription • 07) Supply request – low molecular weight heparin syringes x 4 • 08) Recommendadminister – low molecular weight heparin 70 mgsubcutaneouslytwice a day for 2days • 07) Return to clinic in 2days to begin warfarin therapy • 08) Test request for International Normalized Ratio (INR), Antithrombin 3and Protein C sent to PathLab Heparin syringes PathLab
Bottom line: ICBO, 30th July 2011 • The world is not made up of questions and answers • Users design their data systems with fields (questions) that take values (answers). • But a question-answer view impedes data re-use and standardization • Ontologies hold (one) key to solving that problem • by giving a common interoperable model of the referents of the statements in the record