I delivered this talk at the INFORMS conference in San Diego in October, 2009.

We model geographic vector borne disease movement as a random walk that is informed by the underlying geographic area's suitability for the disease vectors and reservoirs. We use Markov decision process design to compute disease intervention locations with the aim of preventing the disease from reaching critical, high-population areas. We illustrate our methods by computing Leishmaniasis intervention strategies in Texas, an area where Leishmaniasis has historically spread northward.

You can also read the related paper on designing Markov decision processes.

These are two additional visualizations for the specific problem of Leishmaniasis spread in Texas discussed in the talk above,

Pictured above is the spread of Leishmaniasis in the study area, as given by our random walk model. Darker, colder colors indicate a larger disease concentration while lighter, warmer colors indicate a lower concentration. Notice that the roughly circular, local spread around the initial condition is influenced by the underlying disease suitability, resulting in asymmetric disease spread,

Pictured above are the solutions computed by the MDP heuristic (left) and the Greedy submodular optimization algorithm (right). The colors indicate the probability the disease will reach the protected boundary, given that it starts at a specific location. As evidenced by the first few placements, the two algorithms select different solutions. However, the performance of the solutions is similar. At about 100 spray locations, both solutions achieve essentially zero probability of reaching the boundary, given that the disease starts in the center of the infected region.

- Budgeted Disease Interdiction (55%)
- MDP Design (55%)
- Models of Smuggler Movement (45%)
- Influenza Surveillance Design (45%)
- Interdicting Nuclear Smugglers (36%)

## Add A Comment