(254) Enhancing US Lyme Disease Detection in Claims Data: Identifying Acute and Disseminated Disease

Background: Lyme disease (LD) is the most common vector-borne disease in the US. It has long been recognized that traditional surveillance data underestimate the burden of LD.

Objectives: We previously validated a claims-based algorithm to identify acute LD based on LD diagnoses and antibiotics using data from a regional insurer in one endemic state. The present work will validate the acute algorithm in another endemic state and non-endemic states bordering endemic states, as well as new algorithms for disseminated musculoskeletal (MS) and nervous system (NS) LD based on symptom-related diagnoses, antibiotics, and evidence of diagnostic testing.

Methods: We performed this retrospective analysis on claims data from one national and one regional insurer. We identified patients from October 2015 – December 2021 that met the acute LD algorithm (an LD diagnosis code and LD-indicated antibiotic within 30 days) or the MS or NS algorithm (all of the following within 60 days: a non-LD diagnosis code of interest, dispensing of an LD-indicated antibiotic, and a procedure code for an LD-specific diagnostic test). We describe the patients identified by these algorithms stratified by home state LD endemicity.

Results: We identified 10,170 and 456 patients who met the acute LD algorithm in LD endemic and non-endemic states, respectively. In endemic states, we additionally identified 2,086 with the MS algorithm and 2,625 with the NS algorithm. In non-endemic states, we identified 134 with the MS algorithm and 221 with the NS algorithm. Case detection by the acute LD algorithm increased prominently in summer months, compared to subtle upticks in summer through fall for the MS and NS algorithms, reflecting known LD seasonal trends.

Across all algorithms, the median age was higher in non-endemic states. The most common antibiotic was similar in endemic and non-endemic states but differed by indication; among patients meeting the acute algorithm, doxycycline was most common (78-79%); for those meeting disseminated algorithms, doxycycline was less common (42-55%) and amoxicillin was more common (34- 44%). Most patients meeting the acute algorithm had no signs or symptoms on their claims (56-60%), but among those who did, the most common were fatigue (15-18%) and joint pain (14-16%). The most frequent algorithm-qualifying diagnosis codes for the disseminated LD algorithms were pain in knee for MS and radiculopathy and polyneuropathy for NS.

Conclusions: These claims-based algorithms identified a sizeable cohort of potential LD cases. Validation of 800 cases associated with these algorithms is on-going. Future work will apply these algorithms, with methods to adjust for misclassification as appropriate, to estimate the burden of LD in the US.