Estimating the Burden of Chlamydia Trachomatis in the US Army – Use of Three Passive Surveillance Systems and Capture-Recapture Methods to Identify Incident Cases

Background: Incidence of Chlamydia trachomatis in US Army soldiers is typically estimated through case reports of notifiable conditions. The degree to which under-reporting affects approximations is unknown. This study was conducted to assess the burden of chlamydia infections and compare case capture across multiple data systems.

Methods: Incident infections among non-deployed Army soldiers for calendar years 2008-2012 were identified from three sources: Health Level 7 laboratory records, Military Health System Data Repository direct care medical records, and reports of notifiable conditions from the Disease Reporting System-internet and its predecessor the Reportable Medical Event System. A thirty day gap in care rule was used to define incident cases in the consolidated database (related records within 30 days of a prior entry were considered part of the same incident). Capture-recapture (CR) methods were applied to estimate undocumented infections.

Results: A total of 52,813 incident chlamydia infections were identified over the 5 years; 9,494 infections occurred in 2012, reflecting an incidence rate (IR) of 19.3 infections per 1000 person-years. A steady decline in incidence was observed, from a high of 26.0 infections per 1000 person-years in 2008. Case capture by source during the 5-year study was 79% for notifiable case reports, 70% for lab records, and 33% for medical records. A marked decrease in laboratory identified infections occurred in 2011 and continued into 2012 when case capture dropped to 62%. CR analysis revealed an additional 1,706 probable infections during 2012, totaling 11,200 infections (IR: 22.8 per 1000 person-years).

Conclusions: Declines in chlamydia incidence were observed, attributable in part to decreased case detection through laboratory records. Declines in lab case detection require further investigation.  Assessment of multiple databases and incorporation of advanced statistical modeling may provide a better approximation of disease burden. Estimates generated likely remain conservative given the asymptomatic nature of infection.