Adaptive bandit algorithms increase efficiency of mobile tuberculosis screening programs.

Area of Science:

• Public Health
• Infectious Disease Epidemiology
• Machine Learning Applications

Background:

• Community-based tuberculosis screening with mobile X-ray units enhances case detection by overcoming access barriers.
• Optimizing the placement of mobile screening units is crucial for efficient resource allocation in tuberculosis control.

Purpose of the Study:

• To evaluate the effectiveness of the multi-armed bandit (MAB) framework for optimizing mobile tuberculosis screening locations.
• To compare the performance of two MAB algorithms (Exp3 and LinUCB) against traditional placement strategies.

Main Methods:

• Simulations were conducted over three years for two mobile units serving 95 sites in Lima, Peru.
• Two MAB algorithms, Exp3 and LinUCB, were compared with historical case-rate-driven and random placement strategies.
• LinUCB incorporated local socioeconomic indicators, while both MAB algorithms adapted site selection based on observed screening yields.

Main Results:

• MAB algorithms significantly reduced the average screenings needed per tuberculosis detection: 112 for Exp3 and 79 for LinUCB, versus 152 (random) and 143 (historical).
• LinUCB demonstrated superior performance, achieving 20% increased detection efficiency by week 16 and 50% by week 40 compared to historical placement.
• Both MAB algorithms showed improved tuberculosis screening yields, highlighting the benefits of adaptive, data-driven approaches.

Conclusions:

• Multi-armed bandit algorithms effectively optimize mobile tuberculosis screening site selection, enhancing detection efficiency.
• Adaptive machine learning models offer a promising strategy for improving resource allocation in high-burden tuberculosis settings.
• The findings support the integration of data-driven approaches into public health screening programs for infectious diseases.