Optimizing Pool Testing for Epidemic Surveillance

 Abstract:  

Testing is one of the key tools in public health surveillance. Often testing resources are limited, and pooled testing emerged as a viable strategy during the COVID outbreak, for early detection of the outbreak or clearing the most number of individuals (maximum ``welfare'').

Here, we study the problem of selecting pools for testing, which minimize detection delay or maximize welfare. However, these problems are very challenging optimization problems because the infection status of individuals can be correlated. Prior work on choosing pools has ignored network correlations.

We design efficient approximation algorithms for both these problems, using techniques from stochastic and combinatorial optimization: sample average approximation, linear programming, and randomized rounding. We further speed up our algorithms using techniques for combinatorially solving the linear program. We evaluate our methods on multiple networked datasets, including one derived from a hospital, and show significant benefits in modeling network correlations.

Committee:

  • Jundong Li, Committee Chair (CS, ECE/SEAS, SDS/UVA)
  • Anil Vullikanti, Advisor (CS/Biocomplexity Institute, UVA)
  • Madhav Marathe (CS/Biocomplexity Institute, UVA)
  • Afsaneh Doryab (SE/CS, SEAS, UVA)