Integrating testing volume into bandit algorithms for infectious disease surveillance

Area of Science:

• Epidemiology and Public Health
• Biostatistics
• Infectious Disease Surveillance

Background:

• Mobile testing services are vital for active surveillance of infectious diseases, especially in hard-to-reach populations.
• Identifying infected individuals is key to mitigating disease transmission.
• Multi-armed bandit sampling approaches have been used to maximize positive test yields.

Purpose of the Study:

• To investigate the impact of variable testing volumes across sites on multi-armed bandit sampling effectiveness.
• To develop and evaluate extended sampling frameworks that account for testing volume variability.

Main Methods:

• Extended existing multi-armed bandit sampling frameworks to incorporate testing volume variability.
• Conducted a simulation study using human immunodeficiency virus (HIV) characteristics in Congo-Brazzaville.
• Applied the new methods to COVID-19 testing data from Connecticut.

Main Results:

• The newly developed methods demonstrated improved short- and long-term performance compared to existing approaches.
• The extended frameworks effectively managed variability in testing volumes across different sites.
• The methods maintained computational tractability.

Conclusions:

• The developed methods offer enhanced performance for active infectious disease surveillance when testing volumes vary.
• These computationally tractable approaches are recommended for practical implementation in public health settings.
• Accounting for testing volume variability is crucial for optimizing disease surveillance strategies.