Optimizing Tuberculosis Case Detection through a Novel Diagnostic Device Placement Model: The Case of Uganda

Background Xpert MTB/RIF (Xpert) is being widely adopted in high TB burden countries. Analysis is needed to guide the placement of devices within health systems to optimize the tuberculosis (TB) case detection rate (CDR). Methods We used epidemiological and operational data from Uganda (139 sites serving 87,600 individuals tested for TB) to perform a model-based comparison of the following placement strategies for Xpert devices: 1) Health center level (sites ranked by size from national referral hospitals to health care level III centers), 2) Smear volume (sites ranked from highest to lowest volume of smear microscopy testing), 3) Antiretroviral therapy (ART) volume (sites ranked from greatest to least patients on ART), 4) External equality assessment (EQA) performance (sites ranked from worst to best smear microscopy performance) and 5) TB prevalence (sites ranked from highest to lowest). We compared two clinical algorithms, one where Xpert was used only for smear microscopy negative samples versus another replacing smear microscopy. The primary outcome was TB CDR; secondary outcomes were detection of multi-drug resistant TB, number of sites requiring device placement to achieve specified rollout coverage, and cost. Results Placement strategies that prioritized sites with higher TB prevalence maximized CDR, with an incremental rate of 6.2-12.6% compared to status quo (microscopy alone). Diagnosis of MDR-TB was greatest in the TB Prevalence strategy when Xpert was used in place of smear microscopy. While initial implementation costs were lowest in the Smear Volume strategy, cost per additional TB case detected was lowest in the TB prevalence strategy. Conclusion In Uganda, placement of Xpert devices in sites with high TB prevalence yielded the highest TB CDR at the lowest cost per additional case diagnosed. These results represent novel use of program level data to inform the optimal placement of new technology in resource-constrained settings.