Artificial intelligence to predict West Nile virus outbreaks with eco-climatic drivers

Abstract

Background: In Europe, the frequency, intensity, and geographic range of West Nile virus (WNV)-outbreaks have increased over the past decade, with a 7.2-fold increase in 2018 compared to 2017, and a markedly expanded geographic area compared to 2010. The reasons for this increase and range expansion remain largely unknown due to the complexity of the transmission pathways and underlying disease drivers. In a first, we use advanced artificial intelligence to disentangle the contribution of eco-climatic drivers to WNV-outbreaks across Europe using decade-long (2010-2019) data at high spatial resolution.

Methods: We use a high-performance machine learning classifier, XGBoost (eXtreme gradient boosting) combined with state-of-the-art XAI (eXplainable artificial intelligence) methodology to describe the predictive ability and contribution of different drivers of the emergence and transmission of WNV-outbreaks in Europe, respectively.

Findings: Our model, trained on 2010-2017 data achieved an AUC (area under the receiver operating characteristic curve) score of 0.97 and 0.93 when tested with 2018 and 2019 data, respectively, showing a high discriminatory power to classify a WNV-endemic area. Overall, positive summer/spring temperatures anomalies, lower water availability index (NDWI), and drier winter conditions were found to be the main determinants of WNV-outbreaks across Europe. The climate trends of the preceding year in combination with eco-climatic predictors of the first half of the year provided a robust predictive ability of the entire transmission season ahead of time. For the extraordinary 2018 outbreak year, relatively higher spring temperatures and the abundance of Culex mosquitoes were the strongest predictors, in addition to past climatic trends.

Interpretation: Our AI-based framework can be deployed to trigger rapid and timely alerts for active surveillance and vector control measures in order to intercept an imminent WNV-outbreak in Europe.

Funding: The work was partially funded by the Swedish Research Council FORMAS for the project ARBOPREVENT (grant agreement 2018-05973).

Keywords: Climate adaptation; Culex vectors; Early warning systems; Emerging infectious disease; Europe; Outbreaks management; Preparedness; SHAP; West Nile virus; XGBoost; forecasting.

Figure 1

(A) Annual frequency of WNV affected NUTS3 regions and the total human infections cases in Europe, 2010-2019. NUTS3 regions from 28 countries within the EU/EEA were included in this study. Annual count of NUTS3 regions affected by WNV transmission, in Europe (blue bars, left y-axis). The right y-axis shows the annual WNV cases (black curve). (B) Comparison of WNV transmission by NUTS3, 2010-2017 and 2018. B1 represents the cumulative number of years of NUTS3 regions with WNV transmission, 2010-2017, and B2 for 2018.

Figure 2

(A) Logloss score of all models by quarters (Q1-Q4) of the year, Europe 2010-2019. The logloss metrics of all the models are shown for the training data (green circle) and both test data sets, i.e., the year 2018 (light blue) and 2019 (red) of the WNV-outbreaks in Europe. The model-Q2, had the minimum logloss score for both test data sets, hence the best model. (B) Performance metrics of model-Q2 on test data sets. For three classification thresholds shown on the x-axis, the accuracy, balanced accuracy, and specificity score remained more consistent for the test data sets than the other three, i.e., F1 score, precision, and sensitivity.

Figure 3

(A) Summary plot: Top-10 most important SHAP predicted features of the WNV model, Europe 2010-2019. The y-axis indicates the variable name in order of importance from top to bottom. The value next to each is the mean absolute SHAP value, the higher the value, the more important the feature. The x-axis represents the SHAP values showing the change in log-odds. Gradient color indicates the original values feature. Each point corresponds to a feature value of the original data. The most noteworthy feature is the bio10-the mean temperature of the warmest quarter from the preceding year, followed by the maximum temperature of the 2nd quarter of the same year (max_temp_02) and the temperature seasonality (bio4) of the preceding year. (B) Ranking of the feature classes. Summation and ranking of each feature class from the training period estimated from the European-wide SHAP score of each feature. The horizontal axis represents the SHAP estimated aggregation of features per class. The vertical axis represents the feature class.

Figure 4

Comparison of observed data of top SHAP predicted features. The boxplot and the mean value (large solid circle) of top-6 most influential features predicted by SHAP for the NUTS regions with a WNV presence in 2018 but not in the preceding year. The regions are compared using the values of the observed features during 2017 (green color) with the 2018 data (light blue).