Abstract
BACKGROUND: Lyme disease (LD) is the most common vector-borne disease in the United States, though traditional LD surveillance underestimates the burden of disease. We validated algorithms for early localized and disseminated LD, with and without LD-specific diagnosis codes, in states with high incidence and their neighboring states with low LD incidence. METHODS: We identified cohorts of potential incident LD cases in administrative insurance claims data, October 2015-October 2023, in 1 national and 1 regional insurer. Three algorithms were studied: a primary algorithm of an LD-specific diagnosis code and indicated antibiotic and 2 secondary algorithms for disseminated LD requiring a non-LD-specific musculoskeletal or neurologic diagnosis code, an antibiotic, and an LD diagnostic test. We included individuals from high LD-incidence states and neighboring low LD-incidence states. We validated the algorithms using medical records for a sample of potential cases, classifying them according to modified surveillance case definitions. We calculated positive predictive values (PPVs) for each algorithm. RESULTS: Overall, we identified 9483 potential LD cases in claims data and reviewed 841 medical records. The PPVs for the primary algorithm were 90.7% and 81.3% in high-incidence and neighboring states, respectively, when suspect, probable, and confirmed cases were included; they were 76.6% and 28.0% when only confirmed and probable were included. For confirmed and possible cases, the secondary musculoskeletal algorithm PPVs were 12.9% and 4.1%, and the secondary neurologic algorithm PPVs were 6.2% and 1.8% in high-incidence and neighboring states, respectively. CONCLUSIONS: This study found that claims-based algorithms requiring diagnosis codes for LD or for related symptoms, in addition to other criteria, can identify cohorts of true LD cases. These algorithms, adjusted for PPV, can be used to estimate LD incidence in the United States.