Abstract
ObjectiveWe seek to integrate multiple streams of geo-coded information withthe aim to improve public health surveillance accuracy and efficiency.Specifically for vector-borne diseases, knowledge of spatial andtemporal patterns of vector distribution can help early prediction ofhuman incidence. To this end, we develop joint modeling approachesto evaluate the contribution of vector or reservoir information on earlyprediction of human cases. A case study of spatiotemporal modelingof tularemia human incidence and rodent population data from Finnishhealth care districts during the period 1995-2013 is provided. Resultssuggest that spatial and temporal information of rodent abundance isuseful in predicting human cases.IntroductionAn increasing number of geo-coded information streams areavailable with possible use in disease surveillance applications.In this setting, multivariate modeling of health and non-health dataallows assessment of concurrent patterns among data streams andconditioning on one another. Therefore it is appropriate to considerthe analysis of their spatial distributions together. Specifically forvector-borne diseases, knowledge of spatial and temporal patternsof vector distribution could inform incidence in humans. Tularemiais an infectious disease endemic in North America and parts ofEurope. In Finland tularemia is typically mosquito-transmitted withrodents serving as a host; however a country-wide understanding ofthe relationship between rodents and the disease in humans is stilllacking. We propose a methodology to help understand the associationbetween human tularemia incidence and rodent population levels.MethodsData on rodent population levels are collected around the countryby the Finnish Natural Resources Institute. Human Tularaemia casesare recorded as laboratory-confirmed and reported to the NationalInfectious Disease Register (NIDR). Human cases and rodent datawere aggregated to match the 20 Finnish health districts over the period1995-2013 [1]. We develop our methodology in a Bayesian setting.The counts of human cases for each health district in a given yearare assumed to follow a Poisson distribution and the rodent data areassumed to have a categorical likelihood. The linear predictors linkedto the human and rodent likelihood functions are then decomposedadditively into spatial, temporal, and space-time interaction randomeffects. We then link the two likelihoods via the interaction term byassuming that the human spatiotemporal variation is dependent on therodent activity with one-year lag. In the case of the rodent data, wealso included two additional spatial and non-spatial contextual termsto better model ecological effects associated with rodent populationlevels as described before [2]. We then finally develop indicators, onthe scale 0 to 1, to quantify the association between human incidenceand a rodent vector.ResultsResults suggest that spatial and temporal information of rodentabundance is useful in predicting human cases.ConclusionsFuture modeling directions are recommended to includeenvironmental and epidemiological factors. To the best of ourknowledge, this is the first time that rodent data, captured for non-health related purposes, is used to better inform the human risk oftularemia in Finland.