Abstract
Symptom-based tuberculosis screening usually depends on self-reported cough, which is subjective. Passive smartphone cough monitoring offers a more objective alternative. This study assessed the diagnostic accuracy and implementation feasibility of smartphone-based cough frequency monitoring for identifying likely TB cases in Uganda. Adults (≥15 years) screened in community settings or tested at Kampala health facilities underwent microbiological testing and 48 hours of cough monitoring with the Hyfe Research app. TB status was determined by Xpert MTB/RIF Ultra and culture, with inverse probability weighting to adjust for differential enrollment. Cough frequency was compared between participants with and without TB, and accuracy assessed using weighted ROC curves. Staff interviews, analyzed thematically, explored provider-side implementation challenges. Of 884 enrolled participants, 197 had both valid recordings and TB status (101 community, 96 facility). Median cough frequency was higher among those with TB: 2.2 coughs/hour (IQR 0.8–6.1) vs 0.9 (0.4–2.0) in the community, and 6.7 (2.6–27.5) vs 2.4 (0.9–4.8) in facilities (Wilcoxon P < 0.0001 for both). AUCs were 0.69 (95% CI 0.58–0.79) in the community and 0.76 (95% CI 0.60–0.88) in facilities. For self-reported cough at 90% sensitivity, specificities were 55% (95% CI 29–88%) and 36% (95% CI 17–68%), respectively. Recorded cough frequency correlated moderately with self-reported cough severity, Saint George’s Respiratory Questionnaire scores, and staff-observed coughs. Staff cited device visibility, stigma, and security concerns as barriers. Smartphone-recorded cough frequency was associated with TB status but did not meet accuracy thresholds for stand-alone screening or triage, and implementation problems limited data collection. Addressing these operational barriers will be critical to future development and deployment of cough monitoring tools for TB screening.