Background In the development of administrative data case-definitions for HIV, it is important to understand the contribution of each data source to prevalence estimates, especially as it pertains to generalizability of methods.

Methods HIV case-definitions were constructed from four population-based databases available in Manitoba: physician claims, hospital discharge, pharmaceutical dispensations, and provincial laboratory tests. Performance was assessed using sensitivity, specificity, positive/negative predictive value (PPV & NPV), and Youden’s index (YI). Cases identified by HIV case-definitions, and those reported to public health surveillance were compared using annualized incidence. The distribution of those flagged as HIV-positive was compared by database.

The best performing case-definition (YI 0.71) was two or more HIV diagnoses in two years in physician claims, or in hospital discharge abstracts; or 14 or more HAART dispensations in two years; or one positive HIV laboratory. Sensitivity, specificity, PPV and NPV was 82.3% (95%CI: 79.1%-85.5%), 86.8% (95%CI: 84.9%-88.7%), 74.1% (95%CI: 70.6%-77.6%), 91.4% (95%CI: 89.8%-93.1%), respectively. Annualized incidence (2009–2015) calculated from this case-definition was 7.4/100,000 persons (95%CI: 6.8–8.1)]; annualized incidence calculated from surveillance data was 7.7/100,000 persons (95%CI: 7.1–8.3). Approximately 76% of cases would have been flagged through a positive laboratory; 43% through pharmaceutical claims; 34% through physician claims; and 11% through hospital abstracts. 95% of cases would have been flagged through the combination of laboratory and pharmaceutical databases. Only 4% of cases were flagged in all four data sources.

Conclusion Although the combination of four databases produced the most complete prevalence snapshot, laboratory data was the most important contributor. The combination of laboratory and pharmaceutical databases would have identified the predominant majority of cases in our sample. Findings can be used to inform the construction of administrative data cohorts where the availability of population-based data sources may be more limited.

Disclosure No significant relationships.