Peter L. Elkin

Automated identification of postoperative complications within an electronic medical record using natural language processing.

CONTEXT Currently most automated methods to identify patient safety occurrences rely on administrative data codes; however, free-text searches of electronic medical records could represent an additional surveillance approach. OBJECTIVE To evaluate a natural language processing search-approach to identify postoperative surgical complications within a comprehensive electronic medical record. DESIGN, SETTING, AND PATIENTS Cross-sectional study involving 2974 patients undergoing inpatient surgical procedures at 6 Veterans Health Administration (VHA) medical centers from 1999 to 2006. MAIN OUTCOME MEASURES Postoperative occurrences of acute renal failure requiring dialysis, deep vein thrombosis, pulmonary embolism, sepsis, pneumonia, or myocardial infarction identified through medical record review as part of the VA Surgical Quality Improvement Program. We determined the sensitivity and specificity of the natural language processing approach to identify these complications and compared its performance with patient safety indicators that use discharge coding information. RESULTS The proportion of postoperative events for each sample was 2% (39 of 1924) for acute renal failure requiring dialysis, 0.7% (18 of 2327) for pulmonary embolism, 1% (29 of 2327) for deep vein thrombosis, 7% (61 of 866) for sepsis, 16% (222 of 1405) for pneumonia, and 2% (35 of 1822) for myocardial infarction. Natural language processing correctly identified 82% (95% confidence interval [CI], 67%-91%) of acute renal failure cases compared with 38% (95% CI, 25%-54%) for patient safety indicators. Similar results were obtained for venous thromboembolism (59%, 95% CI, 44%-72% vs 46%, 95% CI, 32%-60%), pneumonia (64%, 95% CI, 58%-70% vs 5%, 95% CI, 3%-9%), sepsis (89%, 95% CI, 78%-94% vs 34%, 95% CI, 24%-47%), and postoperative myocardial infarction (91%, 95% CI, 78%-97%) vs 89%, 95% CI, 74%-96%). Both natural language processing and patient safety indicators were highly specific for these diagnoses. CONCLUSION Among patients undergoing inpatient surgical procedures at VA medical centers, natural language processing analysis of electronic medical records to identify postoperative complications had higher sensitivity and lower specificity compared with patient safety indicators based on discharge coding.

A controlled trial of automated classification of negation from clinical notes

BackgroundIdentification of negation in electronic health records is essential if we are to understand the computable meaning of the records: Our objective is to compare the accuracy of an automated mechanism for assignment of Negation to clinical concepts within a compositional expression with Human Assigned Negation. Also to perform a failure analysis to identify the causes of poorly identified negation (i.e. Missed Conceptual Representation, Inaccurate Conceptual Representation, Missed Negation, Inaccurate identification of Negation).Methods41 Clinical Documents (Medical Evaluations; sometimes outside of Mayo these are referred to as History and Physical Examinations) were parsed using the Mayo Vocabulary Server Parsing Engine. SNOMED-CT™ was used to provide concept coverage for the clinical concepts in the record. These records resulted in identification of Concepts and textual clues to Negation. These records were reviewed by an independent medical terminologist, and the results were tallied in a spreadsheet. Where questions on the review arose Internal Medicine Faculty were employed to make a final determination.ResultsSNOMED-CT was used to provide concept coverage of the 14,792 Concepts in 41 Health Records from Johns Hopkins University. Of these, 1,823 Concepts were identified as negative by Human review. The sensitivity (Recall) of the assignment of negation was 97.2% (p < 0.001, Pearson Chi-Square test; when compared to a coin flip). The specificity of assignment of negation was 98.8%. The positive likelihood ratio of the negation was 81. The positive predictive value (Precision) was 91.2%ConclusionAutomated assignment of negation to concepts identified in health records based on review of the text is feasible and practical. Lexical assignment of negation is a good test of true Negativity as judged by the high sensitivity, specificity and positive likelihood ratio of the test. SNOMED-CT had overall coverage of 88.7% of the concepts being negated.

Use of Complementary and Alternative Medical Therapies by Patients Referred to a Fibromyalgia Treatment Program at a Tertiary Care Center

OBJECTIVE To evaluate the frequency and pattern of complementary and alternative medicine (CAM) use in patients referred to a fibromyalgia treatment program at a tertiary care center. PATIENTS AND METHODS Patients referred to the Mayo Fibromyalgia Treatment Program between February 2003 and July 2003 were invited on their initial visit to participate in a survey regarding CAM use during the previous 6 months. An 85-question survey that addressed different CAM domains was used. RESULTS Of the 304 patients invited to participate, 289 (95%) completed the survey (263 women and 26 men). Ninety-eight percent of the patients had used some type of CAM therapy during the previous 6 months. The 10 most frequently used CAM treatments were exercise for a specific medical problem (48%), spiritual healing (prayers) (45%), massage therapy (44%), chiropractic treatments (37%), vitamin C (35%), vitamin E (31%), magnesium (29%), vitamin B complex (25%), green tea (24%), and weight-loss programs (20%). CONCLUSION CAM use is common in patients referred to a fibromyalgia treatment program.

VA national drug file reference terminology: A cross-institutional content coverage study

BACKGROUND Content coverage studies provide valuable information to potential users of terminologies. We detail the VA National Drug File Reference Terminologys (NDF-RT) ability to represent dictated medication list phrases from the Mayo Clinic. NDF-RT is a description logic-based resource created to support clinical operations at one of the largest healthcare providers in the US. METHODS Medication list phrases were extracted from dictated patient notes from the Mayo Clinic. Algorithmic mappings to NDF-RT using the SmartAccess Vocabulary Server (SAVS) were presented to two non-VA physicians. The physicians used a terminology browser to determine the accuracy of the algorithmic mapping and the content coverage of NDF-RT. RESULTS The 509 extracted documents on 300 patients contained 847 medication concepts in medication lists. NDF-RT covered 97.8% of concepts. Of the 18 phrases that NDF-RT did not represent, 10 were for OTCs and food supplements, 5 were for prescription medications, and 3 were missing synonyms. The SAVS engine properly mapped 773 of 810 phrases with an overall sensitivity (precision) was 95.4% and positive predictive value (recall) of 99.9%. CONCLUSIONS This study demonstrates that NDF-RT has more general utility than its initial design parameters dictated

Comparison of Natural Language Processing Biosurveillance Methods for Identifying Influenza From Encounter Notes

BACKGROUND An effective national biosurveillance system expedites outbreak recognition and facilitates response coordination at the federal, state, and local levels. The BioSense system, used at the Centers for Disease Control and Prevention, incorporates chief complaints but not data from the whole encounter note into its surveillance algorithms. OBJECTIVE To evaluate whether biosurveillance by using data from the whole encounter note is superior to that using data from the chief complaint field alone. DESIGN 6-year retrospective case-control cohort study. SETTING Mayo Clinic, Rochester, Minnesota. PARTICIPANTS 17,243 persons tested for influenza A or B virus between 1 January 2000 and 31 December 2006. MEASUREMENTS The accuracy of a model based on signs and symptoms to predict influenza virus infection in patients with upper respiratory tract symptoms, and the ability of a natural language processing technique to identify definitional clinical features from free-text encounter notes. RESULTS Surveillance based on the whole encounter note was superior to the chief complaint field alone. For the case definition used by surveillance of the whole encounter note, the normalized partial area under the receiver-operating characteristic curve (specificity, 0.1 to 0.4) for surveillance using the whole encounter note was 92.9% versus 70.3% for surveillance with the chief complaint field (difference, 22.6%; P < 0.001). Comparison of the 2 models at the fixed specificity of 0.4 resulted in sensitivities of 89.0% and 74.4%, respectively (P < 0.001). The relative risk for missing a true case of influenza was 2.3 by using the chief complaint field model. LIMITATIONS Participants were seen at 1 tertiary referral center. The cost of comprehensive biosurveillance monitoring was not studied. CONCLUSION A biosurveillance model for influenza using the whole encounter note is more accurate than a model that uses only the chief complaint field. Because case-defining signs and symptoms of influenza are commonly available in health records, the investigators believe that the national strategy for biosurveillance should be changed to incorporate data from the whole health record. PRIMARY FUNDING SOURCE Centers for Disease Control and Prevention.

Exploiting ebXML registry semantic constructs for handling archetype metadata in healthcare informatics

Using archetypes is a promising approach in providing semantic interoperability among healthcare systems. To realise archetype based interoperability, the healthcare systems need to discover the existing archetypes, based on their semantics; annotate their archetypes with ontologies; compose templates from archetypes and retrieve corresponding data from the underlying medical information systems. In this paper, we describe how ebXML Registry semantic constructs can be used for annotating, storing, discovering and retrieving archetypes. For semantic annotation of archetypes, we present an example of an archetype metadata ontology and describe the techniques to access archetype semantics through ebXML query facilities. We present a GUI query facility and describe how the stored procedures, which we introduce, move the semantic support beyond what is currently available in ebXML registries. We also address how archetype data can be retrieved from clinical information systems by using ebXML web services. A comparison of web service technology with the ebXML messaging system is provided to justify the reasons for using web services.

Embedded Structures and Representation of Nursing Knowledge

Nursing Vocabulary Summit participants were challenged to consider whether reference terminology and information models might be a way to move toward better capture of data in electronic medical records. A requirement of such reference models is fidelity to representa- tions of domain knowledge. This article discusses embedded structures in three different approach- es to organizing domain knowledge: scientific reasoning, expertise, and standardized nursing lan- guages. The concept of pressure ulcer is presented as an example of the various ways lexical ele- ments used in relation to a specific concept are organized across systems. Different approaches to structuring information—the clinical information system, minimum data sets, and standardized messaging formats—are similarly discussed. Recommendations include identification of the poly- hierarchies and categorical structures required within a reference terminology, systematic evalua- tions of the extent to which structured information accurately and completely represents domain knowledge, and modifications or extensions to existing multidisciplinary efforts. J Am Med Inform Assoc. 2000;7:539-549.

The introduction of a diagnostic decision support system (DXplain™) into the workflow of a teaching hospital service can decrease the cost of service for diagnostically challenging Diagnostic Related Groups (DRGs)

BACKGROUND In an era of short inpatient stays, residents may overlook relevant elements of the differential diagnosis as they try to evaluate and treat patients. However, if a residents first principal diagnosis is wrong, the patients appropriate evaluation and treatment may take longer, cost more, and lead to worse outcomes. A diagnostic decision support system may lead to the generation of a broader differential diagnosis that more often includes the correct diagnosis, permitting a shorter, more effective, and less costly hospital stay. METHODS We provided residents on General Medicine services access to DXplain, an established computer-based diagnostic decision support system, for 6 months. We compared charges and cost of service for diagnostically challenging cases seen during the fourth through sixth month of access to DXplain (intervention period) to control cases seen in the 6 months before the system was made available. RESULTS 564 cases were identified as diagnostically challenging by our criteria during the intervention period along with 1173 cases during the control period. Total charges were

Exploring the frontier of electronic health record surveillance: the case of postoperative complications.

1281 lower (p=.006), Medicare Part A charges

Guideline and quality indicators for development, purchase and use of controlled health vocabularies.

1032 lower (p=0.006) and cost of service