The impact of climatic factors on tick-related hospital visits and borreliosis incidence rates in European Russia

Abstract

Tick-borne diseases are among the challenges associated with warming climate. Many studies predict, and already note, expansion of ticks' habitats to the north, bringing previously non-endemic diseases, such as borreliosis and encephalitis, to the new areas. In addition, higher temperatures accelerate phases of ticks' development in areas where ticks have established populations. Earlier works have shown that meteorological parameters, such as temperature and humidity influence ticks' survival and define their areas of habitat. Here, we study the link between climatic parameters and tick-related hospital visits as well as borreliosis incidence rates focusing on European Russia. We have used yearly incidence rates of borreliosis spanning a period of 20 years (1997-2016) and weekly tick-related hospital visits spanning two years (2018-2019). We identify regions in Russia characterized by similar dynamics of incidence rates and dominating tick species. For each cluster, we find a set of climatic parameters that are significantly correlated with the incidence rates, though a linear regression approach using exclusively climatic parameters to incidence prediction was less than 50% effective. On a weekly timescale, we find correlations of different climatic parameters with hospital visits. Finally, we trained two long short-term memory neural network models to project the tick-related hospital visits until the end of the century, under the RCP8.5 climate scenario, and present our findings in the evolution of the tick season length for different regions in Russia. Our results show that the regions with an expected increase in both tick season length and borreliosis incidence rates are located in the southern forested areas of European Russia. Oppositely, our projections suggest no prolongation of the tick season length in the northern areas with already established tick population.

Fig 1. The geographical distribution of the clusters created based on the tick species found in them.

In green, the Northern cluster of federal regions, where I. persulcatus dominates the tick population and in blue, the Southern federal regions where the I. ricinus dominates the tick population. Lastly, the federal regions where both species are found are shown in orange. Voids between the clusters mark the regions where data sets on hospital visits are not available. The map was created using Cartopy in Python and the Natural Earth raster and vector map data, which are freely available through the public domain [39].

Fig 2

The yearly normalised (with respect to the mean) borreliosis incidence rates for the federal regions investigated (top panels) and normalised (with respect to the mean) weekly tick related hospital visits (bottom row of panels). The Northern cluster of federal regions is shown on the left column of panels, the Southern cluster in the middle column and the mixed cluster on the right column. The dashed line shows the average incidence rate for the federal regions in each region, whereas the dash-dot line indicates the mean for each cluster.

Fig 3

Time lagged cross correlation analysis for the temperature at 2m (left panel), evaporation (middle panel) and snow depth (right panel) variables with respect to the weekly hospital visits due to tick bites. Each panel shows the averaged TLCC curve for the three clusters of federal clusters, according to the tick species inhabiting them, investigated in this study.

Fig 4

TLCC analysis for the low (left panel) and high (middle panel) vegetation indices and precipitation (right panel) variables with respect to the weekly tick-related hospital visits for the three federal region clusters investigated in this study. Each panel shows the averaged TLCC curve for the three clusters.

Fig 5

The training accuracy and loss metrics (top and bottom rows of panels, respectively) logged during the training procedures of the two LSTM neural networks, for the Northern (left column) and Southern clusters (right column) of federal regions. The jumps observed in both metrics indicate the point in time in which the callback function interrupted training on one federal region and carried on to the next.

Fig 6

Model prediction and classified tick related hospital visits time series for the Northern cluster (top panel), Southern cluster (middle panel) and mixed cluster (bottom panel).

Fig 7

Tick season length predicted by the LSTM model for the Northern cluster (top left panel), Southern cluster (top right panel) and mixed cluster (bottom panels). For the mixed cluster, since both tick species are found there, both models were used to make predictions.

Fig 8. The start (blue) and end (red) of tick season projected by the LSTM models until the end of the century for the three federal region clusters investigated.

The Southern cluster, where I. ricinus is located, is shown on the top right panel, the Northern cluster, where I. persulcatus is found, on the top left panel, and the mixed cluster, where both species of ticks are found, on the bottom row of panels. The dashed lines show the LSTM models’ projection in terms of week number and the solid lines a linear regression model fit to the time-series to visualise the trend.