Inverse Problem in Legionella Outbreakes: From direct mobility to inference

Legionnaires’ (or Legionella) Disease (LD) is a type of pneumonia that people catch by inhaling small droplets of water suspended in the air containing the Legionella bacterium. There is no evidence of person-to-person transmission. Outbreaks occur from purpose- built water systems where temperatures are warm enough to encourage growth of the bacteria, e.g. in cooling towers and evaporative condensers. Thirty different outbreaks were officially registered in the world between 1976 and 2017 [1]. In total 3178 people were affected by the LD, 236 of whom died. The fatality rate ranged between 0.8% and 75% depending on the outbreak. A LD outbreak happened in Palmanova, a touristic neighborhood part of the municipality of Calvià, Spain [2], between September and October 2017. The people affected were 27: one local worker and 26 tourists. One tourist died. After a long investigation, the main source was identified to be a whirlpool spa (also known as Jacuzzi) on the rooftop of a hotel: the droplets fell down on the surrounding streets and people inhaled them while walking [3].
When an LD outbreak occurs, the health institutions followed the standard epidemio- logical protocol that consists in asking the infected people about their displacement and then checking one by one the potential sources of the disease. The protocol requires time and money. In addition, it usually forces a temporary closure of the buildings where the water systems are. In the Palmanova case most of them were hotels, obliging the hosts to find another accommodation for their guests and to not accept any new tourist. The consequence is a big economic trouble.
The IFISC1 institute in Palma de Mallorca was contacted by the Spanish Health In- stitutions to collaborate and try to find a new approach to the problem. For this reason we build a computational (agent based) model capable of placing the source of the disease into a city, simulating people moving through the network of roads and getting infected when staying close to the source, and computing the epidemic outbreak informations. We repeat the simulation for many realizations and for different positions of the source. This comparison of the obtained data with the real ones allows us to infer the best parameters of the model and finally the source position. This will be obtained by creating a probability heat-map of the region, telling the more probable locations of the source with the help of different coefficients. The model does not intend to be an alternative to the epidemiological protocol, but an extra instrument in the toolbox of the institutions to fasten the process of finding the source and obtain both health and economic benefits for everyone.