Using a Multi-Level Case Definition for Opioid-Related Death Surveillance with Medical Examiner Data – Multnomah County, Oregon

BACKGROUND:  Timely and complete data are crucial for effective monitoring of opiate-related death (ORD) trends. Timely surveillance at the state level may mean complete data within a few months of case investigation, while national public health agencies may be satisfied with a year or more delay. At the local level, access to frequently updated data sources allows for daily or weekly updates that are used to inform effective public health response.  Multnomah County Health Department (MCHD) in Portland, Oregon, uses medical examiner (ME) data for ORD surveillance. The ME database receives preliminary data within 48 hours of a fatality, while fully updated information may take up to six weeks. MCHD uses manual surveillance and analysis which delays availability of data for public health use. We hypothesized that an automated surveillance system using a tiered case definition would allow more comprehensive surveillance of ORDs, while improving timeliness and addressing the issue of incomplete information in recently reported ME cases.  

METHODS:  A multi-level case definition for ORDs (confirmed, presumptive, suspect, unlikely, not a case) was applied using SAS 9.3 to all 2012 ME data for Multnomah County, Oregon. We manually reviewed additional ME data unavailable for export from the ME database, including the form used to populate death certificates, to confirm the diagnoses of non-confirmed cases. 

RESULTS:  Manual surveillance using the historic method required an hour to multiple days, depending on the complexity of the cases and the number of analysts involved. The automated method, requiring only one analyst, shortened this to fewer than 10 minutes. The sensitivity of the system was improved by including the presumptive case definition. PPV was also improved after manual review of suspect cases and inclusion of true ORDs. The automated system found an additional six true ORDs that were not identified by the manual method. 

CONCLUSIONS:  Local health departments benefit from near-real-time surveillance to inform public health response, program design, and policy development. However, some data sources are updated frequently, warranting caution during early analysis. To ensure system reliability, the use of preliminary data may require adjustment of case definitions to account for pending or inaccessible information. A multi-level definition, modeled on infectious disease case definitions, effectively ranks the likelihood that potential ORDs are in fact opiate-related. Further exploration of matching between data sources like ME data, ambulance records, and vital statistics may reveal more opportunities to improve surveillance sensitivity and timeliness.