Artificial Intelligence Assisted Innovative Low-cost Skin Testing Model to Accelerate Person-centered TB Prevention Services in Tanzania, Mozambique and Uganda (TEST4TPT)

TEST4TPT is a type 2 hybrid effectiveness-implementation study designed to evaluate the effectiveness and implementation of an integrated approach to tuberculosis infection (TBI) testing and tuberculosis preventive treatment (TPT) delivery in Mozambique, Tanzania, and Uganda. The study comprises four complementary sub-studies. Sub-study 1 evaluates the performance and usability of an artificial intelligence (AI)-assisted approach for interpretation of the SIILTIBCY skin test using the SmartPatch and TBConnect application. Sub-study 2 is a pragmatic cluster-randomised trial comparing different models of TBI testing and TPT delivery, including facility- and home-based approaches. Sub-study 3 is a mixed-methods implementation evaluation examining the implementation of these approaches, including their feasibility and acceptability. Sub-study 4 comprises economic and impact assessments to evaluate the costs and broader potential impact of the approaches evaluated in the study.

The technological approach evaluated in Sub-study 1 combines the SIILTIBCY skin test with the SmartPatch and AI-driven TBConnect application. The SmartPatch is an adhesive patch applied to the participant's forearm at the site of SIILTIBCY administration. The patch contains a unique QR code used to link the participant to their record within the TBConnect platform and includes a removable circular section defining the area where the skin test is administered and subsequently read. The SmartPatch remains in place between administration and interpretation of the test.

Sub-study 1 consists of two complementary components: validation of the AI-assisted SIILTIBCY reading approach and an assessment of the usability and user experience of the SmartPatch and TBConnect system. The validation component assesses the performance and agreement of AI-assisted SIILTIBCY interpretation compared with standard interpretation by a trained healthcare worker. The usability component examines use of the technology by healthcare workers and community health workers, as well as participants' experiences of wearing the SmartPatch. A total of 330 household contacts of people with TB, aged five years or older, will participate across Mozambique, Tanzania, and Uganda.

At the follow-up visit, images of the SIILTIBCY administration site are captured through the TBConnect application for AI-assisted interpretation. The AI-assisted approach estimates the size of the induration and provides a corresponding test classification. The SmartPatch therefore supports standardised identification and imaging of the test site, while TBConnect provides the platform through which the images and AI-assisted interpretation are managed.

For validation, the AI-assisted interpretation is compared with a standard SIILTIBCY reading conducted independently by a trained healthcare worker. Both readings are performed on the same skin test and are conducted independently and under blinded conditions. The comparison evaluates the ability of the AI-assisted approach to correctly classify SIILTIBCY results and its agreement with the healthcare-worker reference reading, including agreement in the measured size of induration. During this validation sub-study, participant management and clinical decisions are based on the standard healthcare-worker reading rather than the AI-assisted result.

Sub-study 1 also assesses whether the SmartPatch and TBConnect approach is practical and acceptable to its intended users. Healthcare workers and community health workers with experience using the system are asked about its usability and its integration into routine activities, while participants provide information about their experience of wearing the SmartPatch. These findings complement the validation results by assessing not only whether the AI-assisted approach performs adequately, but also whether it can be feasibly used within the intended implementation context. Findings from Sub-study 1 will inform whether the AI-assisted approach is incorporated into the home-based testing approach evaluated in Sub-study 2.

Sub-study 2 is the main effectiveness evaluation and consists of a pragmatic cluster-randomised trial. Participating health facilities are allocated to three service-delivery approaches: standard-of-care facility-based assessment; facility-based SIILTIBCY testing and reading; or home-based SIILTIBCY testing and reading delivered by community health workers, where permitted by national guidelines. The home-based arm may incorporate AI-assisted interpretation using SmartPatch and TBConnect if this approach demonstrates adequate performance in Sub-study 1. Health facilities are used as clusters to allow consistent implementation of each approach within facility workflows and minimise contamination between study arms.

Sub-study 3 evaluates the implementation of the TEST4TPT approaches using an adapted RE-AIM (Reach, Effectiveness, Adoption, Implementation and Maintenance) framework, complemented by constructs from the Implementation Outcomes Framework. The evaluation considers reach, effectiveness and clinical utility, adoption, implementation, maintenance, feasibility and acceptability. A convergent mixed-methods approach combines implementation data generated during the study with qualitative interviews involving contact persons, community health workers, healthcare workers and TB/National Tuberculosis Programmes (NTP) policymakers or programme managers. This component is intended to provide an understanding of how and why the different TBI testing and TPT delivery approaches work across different settings, alongside the effectiveness results generated in Sub-study 2.

Sub-study 4 comprises the economic and impact assessments of the TEST4TPT approaches. This component evaluates the economic implications of the different TBI testing and TPT delivery strategies and their potential impact, complementing the effectiveness and implementation findings from the other sub-studies.