Workshop on Modelling of Infectious Diseases Dynamics

Mosquito-borne diseases have been emerging and re-emerging in the Americas, causing millions of human illnesses. Recent examples include Zika virus and chikungunya. In order to quantify the impact of past outbreaks and predict the course of future outbreaks, it is necessary to merge models with heterogeneous data streams. We present both statistical and mechanistic modeling for mosquito-borne disease spread coupled with data including demographics, internet data, human case counts, weather, and satellite to predict risk. Wehighlight the relative usefulness of our data streams and models depending on the question we are answering and its scale.

Brazil accounts for ~80% of dengue (DENV) cases in the Americas. DENV outbreaks vary spatially within a season and temporally across seasons due to cocirculating serotypes, coexistence of competent vectors, and five distinct climatic zones, making prediction efforts challenging. Identifying specific environmental and sociodemographic factors that explain this variation is key to improving such prediction efforts.  Ecological proxy data (i.e., indirect data sources)  can be used to capture the complex interactions between the host, pathogen, and environment and provide information for local conditions in the absence of granular statistics collected on the ground. Here, we use a combination of 36 satellite imagery, weather, clinical, and census data streams to characterize DENV outbreak shape, seasonal timing, and pairwise correlations in magnitude and onset at the Brazilian state (N=27) and meso region (N=137) levels. Using cross-validated regularized regression models to analyze weekly DENV case data from 2010-2016, we find a parsimonious set of ecological data that explains each outbreak property and measure how well each model captures the property’s variation through several performance metrics. From 2010-2016, the seasonal onset in Para (northern region) was on average 6.74 weeks ahead of other states, positioning it as a potential sentinel state. At the state level, our models explained 62.8% of the variation in outbreak shape, 51.1% of pairwise correlation in outbreak timing,  48.5% of seasonal onset, and 13.0% of pairwise correlation in outbreak magnitude. Outbreak properties were generally better explained at the state level rather than the meso region level. The normalized burn ratio (NBR) had the strongest effect on outbreak shape, while the mean daily temperature range most impacted the state seasonal onset.  The pairwise correlation in outbreak timing between states was best predicted by distance, while the pairwise correlation in outbreak magnitude was best predicted by the similarity in population density. Overall, our results highlight the utility of diverse and disparate ecological data streams for understanding the complex mechanisms that drive DENV transmission dynamics across geographic regions, and how this knowledge can be used to improve the formulation of spatially linked forecasting models of DENV activity.

Pertussis or whooping cough is a highly contagious respiratory disease whose incidence has been increasing for the past two decades in several countries despite the highly extended vaccination. The reasons for this resurgence are a matter of discussion and, while trying to understand this complex problem, new strategies that include more boosters for adolescents and adults have been adopted in many countries in an attempt to improve the control of the disease. However, the impact of these measures on infants (the risk age group) is not clear. In this context mathematical models are being increasingly used to study the dynamics of the disease transmission and to estimate the impact of different control strategies. In this talk I will review our contributions to understand the dynamics of pertussis transmission and the effect of several control measures. Through mathematical modeling we have assessed the impact of the adolescent and pregnant women boosters introduced in Argentina in 2009 and 2012 respectively [1,4] and also have estimated the impact of reducing delays in the administration of the primary dose [2,3]. I will discuss how mathematical modeling (in spite of the uncertainties in parameters values) allows quantitative comparison between different measures and help to suggest where efforts should be directed. And, on the other hand, how comparison of model results with epidemiological data allows to check different assumptions about the transmission process, even when strong and age dependent underreporting could be present. Modeling pertussis transmission to evaluate the effectiveness of an adolescent booster in Argentina [1]   G.Fabricius, P.Bergero,  M.Ormazabal, A.Maltz and D. Hozbor.         Epidemiology and Infection 141, 718-734 (2013). Mathematical modeling of delayed pertussis vaccination in infants [2]     P. Pesco, P.Bergero, G. Fabricius and D. Hozbor.   Vaccine 33, 5475-5480 (2015)[3] Potential impact of changes in the schedule for primary DTP immunization as control strategy for pertussis. P. Bergero,  G. Fabricius,  D. Hozbor, H. Theeten and N. Hens.  The Pediatric Infectious Disease Journal. 37(2):e36-“e42 (2018) [4] Pertussis epidemiology in Argentina: trends after the introduction of maternal immunization. G. Fabricius, P. Martin Aispuro P, P. Bergero, M. Gabrielli, D. Bottero, and D.Hozbor.       Epidemiology and Infection 146, 858-866 (2018).

Zoonotic Visceral Leishmaniasis (ZVL) is one of the world's deadliest and neglected infectious diseases, according to World Health Organization. In this sense, it is important to optimize the control strategy activities. In this work, it was evaluated five control strategies (positive dog elimination, insecticide impregnated dog collar, dog vaccination, dog treatment, and sandfly population control), considering disease control results and cost-effectiveness. It was elaborated a mathematical model based on a set of differential equations in which three populations were represented (human, dog, and sandfly). Humans and dogs were divided into susceptible, latent, clinically ill, and recovery categories. Sandflies were divided into noninfected, infected, and infective. According to our model, the insecticide impregnated dog collar was the strategy that presented the best combination between disease control and cost-effectiveness.

Healthcare-associated infections cause significant patient morbidity and mortality, and contribute to growinghealthcare costs, whose effects may be felt most strongly in developing countries. Active surveillance systems,hospital staff compliance, including hand hygiene, and a rational use of antimicrobials are among the importantmeasures to mitigate the spread of healthcare-associated infection within and between hospitals. Klebsiellapneumoniae is an important human pathogen that can spread in hospital settings, with some forms exhibitingdrug resistance, including resistance to the carbapenem class of antibiotics, the drugs of last resort for suchinfections. Focusing on the role of patient movement within and between hospitals on the transmission andincidence of enterobacteriaceae producing K. pneumoniae Carbapenemase (KPC, an enzyme that inactivatesseveral antimicrobials), we developed a metapopulation model where the connections among hospitals are madeusing a theoretical hospital network based on Brazilian hospital sizes and locations. The pathogen reproductivenumber, R_0 that measures the average number of new infections caused by a single infectious individual, wascalculated in different scenarios defined by both the links between hospital environments (regular wards andintensive care units) and between different hospitals (patient transfer). Numerical simulation was used to il-lustrate the infection dynamics in this set of scenarios. The sensitivity of R_0 to model input parameters, such ashospital connectivity and patient-hospital staff contact rates was also established, highlighting the differentialimportance of factors amenable to change on pathogen transmission and control.

In this talk, we propose an approach based on machine learning algorithms to predict epidemic processes in complex networks. Specifically, we show that it is possible to estimate the outbreak size starting from a single node in a network and determine which networks properties are the most related to the spreading dynamics. Our approach is general and can be applied to any dynamical process running on top of complex networks. Likewise, our work constitutes an important step towards the application of machine learning methods to unravel dynamical patterns emerging in complex networked systems.

Quantitative methods are unquestionably valuable tools for disease propagation models, and have trackedattention of mathematicians, physicists, and statisticians for decades. Mathematical models for diseasepropagation have become more and more sophisticated, incorporating different sets of heterogeneities inthe hope to be as accurate as possible while still being mathematically and computationally tractable. Sociodemographic characteristics such as age distribution, study and work activities, and local and long-range mobility, have been key to study historical infection data and devise scenario assessment. Recent outbreaks of global concern such as the influenza H1N1pdm09, SARS, MERS-CoV, Ebola, Zika, and thecurrent COVID-19 have tested our ability to produce information that is fundamental for preparedness:reproductive number estimation, importation case probability, short-term predictions of disease evolution,and so on. On the other hand, endemic diseases ask for a complementary approach to disease propagationmodels, which is that of disease surveillance. How do we properly separate baseline activity,characterized by stochastic fluctuations and isolated case bursts, from epidemic activity? In other words,how many weekly cases we need to say that the current season of “disease A” have started? Is that levelthe same everywhere? Does influenza-like illnesses always start in the winter? As for real-timesurveillance, are the available data delivered in a timely manner or is there significant delay between caseoccurrence and official report? In this talk we will go over both approaches highlighting the nuances ofeach, their importance to provide relevant information to public health actions, and, hopefully, spark yourinterest in joining the field.

Authors: K. Laneri*, B. Cabella*, P.I. Prado, R.M. Coutinho**, R.A. Kraenkel I'll quickly introduce the Convergent Cross-Mapping (CCM) criterion to investigate causality between two time series. Than I'll present an analysis of the potential environmental drivers of malaria cases in Northwestern Argentina, and discuss plausible interpretations of these results. We have inspected causal links between malaria and climatic variables, based on 12 years of weekly malaria /P. vivax/ cases in Tartagal, Salta, Argentina—at the southern fringe of malaria incidence in the Americas—together with humidity and temperature time-series spanning the same period. Our results show that there are causal links between malaria cases and both maximum temperature, with a delay of five weeks, and minimum temperature, with delays of zero and twenty two weeks. Humidity is also a driver of malaria cases, with thirteen weeks delay between cause and effect. Furthermore we also determined the sign and strength of the effects. These results might be signaling processes operating at short (below 5 weeks) and long (over 12 weeks) time delays, corresponding to effects related to parasite cycle and mosquito population dynamics respectively. The non-linearities found for the strength of the effect of temperature on malaria cases make warmer areas more prone to higher increases in the disease incidence. Moreover, our results indicate that an increase of extreme weather events could enhance the risks of malaria spreading and re-emergence beyond the current distribution. Both situations, warmer climate and increase of extreme events, will be remarkably increased by the end of the century in thishot spot of climate change. * joint first authors** speaker