Biosurveillance evaluation of SNOMED CT's terminology (BEST Trial): coverage of chief complaints

The current United States Health Information Technology Standards Panel's interoperability specification for biosurveillance relies heavily on chief complaint data for tracking rates of cases compatible with a case definition for diseases of interest (e.g. Avian Flu). We looked at SNOMED CT to determine how well this large general medical ontology could represent data held in chief complaints. In this experiment we took 50,000 records (Comprehensive Examinations or Limited Examinations from primary care areas at the Mayo Clinic) from December 2003 through February 2005 (Influenza Season). Of these records, 36,097 had non-null Chief Complaints. We randomly selected 1,035 non-null Chief Complaints and two Board-certified internists (one Infectious Diseases specialist and one general internist) reviewed the mappings of the 1,035 chief complaints. Where the reviewers disagreed, a third internist adjudicated. SNOMED CT had a sensitivity of 98.7% for matching clinical terms found in the chief complaint section of the clinical record. The positive predictive value was 97.4%, the negative predictive value was 89.5%, the specificity was 81.0%, the positive likelihood ratio was 5.181 and the negative likelihood ratio was 0.016. We conclude that SNOMED CT and natural language parsing engines can well represent the clinical content of chief complaint fields. Future research should focus on how well the information contained in the chief complaints can be relied upon to provide the basis of a national strategy for biosurveillance. The authors recommend that efforts be made to examine the entire clinical record to determine the level of improvement in the accuracy of biosurveillance that can be achieved if we were to incorporate the entire clinical record into our biosurveillance strategy.