Geographical targeting of active case finding for tuberculosis in Pakistan using artificial intelligence software (SPOT-TB): a pragmatic stepped wedge cluster randomized control trial.

Background Community-wide active case-finding (ACF) is being increasingly implemented as a tuberculosis (TB) elimination intervention. However, conventional site selection strategies may result in low yields from screening. We evaluated whether an artificial intelligence (AI) software guided targeting strategy could improve detection of TB during screening activities (called camps) relative to routine approaches to site selection in the programmatic setting in Pakistan. Methods We conducted a stepped-wedge cluster-randomised trial embedded within Global Fund supported ACF activities implemented by Pakistan s National TB Program and private sector partners. Thirty mobile X-ray van teams operating in 68 districts were randomly assigned to transition from routine site selection approaches (based on field-staff experience and historical data) to an AI-guided targeting strategy, using the software MATCH-AI. We assessed the effect of the intervention on the primary outcome, Camp Positivity Yield, defined as the number of individuals diagnosed with bacteriologically confirmed TB per camp, using generalised linear mixed models. The primary analysis was by intention to treat. Camps conducted within a 5-km radius of the AI selected locations were included in a validated per-protocol analysis. We conducted several district-level subgroup analyses. This trial is registered, number NCT06017843. Findings Between August 2023 and September 2024, 3,936 screening camps were conducted (2,046 control, 1,890 intervention), screening 269,254 individuals. In the intention-to-treat analysis, Camp Positivity Yield was 7% higher in the intervention group relative to the control group, however this difference was not statistically significant (adjusted risk ratio [RR] 1.07, 95% CI: 0.94 -1.22). In the validated per-protocol analysis, Camp Positivity Yield was 32% higher in the intervention group relative to the control group (adjusted RR 1.32, 95% CI: 1.12-1.54). Yields were highest in districts that had moderate baseline yields of 0.5-1% per population screened prior to the trial (adjusted RR: 1.57, 95% CI: 1.13 - 2.18) and in rural districts (adjusted RR 1.43, 95% CI: 1.23 -1.65). Interpretation The use of an AI-guided targeting strategy significantly increased detection of bacteriologically confirmed TB during active case-finding in the validated per-protocol analysis, relative to conventional site-selection approaches employed by field-staff. This software may be considered as a supportive tool to improve the efficiency of community-based TB case-finding interventions in other high burden countries.