Table 1.
Wild bird mortalities tested for HPAI in the Netherlands per year, as a part of the passive surveillance program.
Table 2.
Description of primary data sources.
Fig 1.
Overview of primary data and processed data. White boxes represent primary data, while grey boxes represent processed data. If a variable is used in the analysis, its abbreviation is indicated within parentheses.
Fig 2.
Monthly temporal Patterns in the primary and processed variables.
The top panels display nationwide temporal patterns in primary data, while the bottom panels show processed data. A) Monthly nationwide counts of HPAI incidents in poultry and wild animals (including mammals). B) Nationwide reports of wild bird mortalities across 190 species of interest. Processed data was calculated for each wild bird mortality individually. C) The spatio-temporal rate measures the rate of HPAI incidents occurring in space and time around each wild bird mortality. D) Wild bird mortalities were also counted within a ± 10-day window and a 25 km radius around each wild bird mortality. The bottom panels present monthly means and ranges, with filled circles representing the mean and bars indicating the minimum and maximum values.
Table 3.
Models tested. Numbers indicate: 1, variable included; 0, variable excluded. Shading shows the grouping of the models based on the inclusion of spatio-temporal rate (ST) and wild bird mortality (n). The other variables are: human density (H), sampling density for wild bird mortalities (D) and wild bird density (B).
Table 4.
Ranking of the 10 best performing models by AUC score. The mean AUC is given with the range across the cross validation folds within parenthesis.
Fig 3.
Spatial distribution of the variables.
For temporal ranges, see Table 2. A) HPAI incident density, B) Mean log10 spatio-temporal rates for wild bird mortalities, C) Mean log10 wild bird mortality counts over 10 days within a 25 km radius, D) log10 human population density, E) log10 sampling density for wild bird mortalities, F) log10 summed wild bird density for 190 species of interest. Note the non-linear Fisher scaling of the legends. Abbreviations: A, Amsterdam; G, Groningen; H, Den Haag; R, Rotterdam; U, Utrecht; Z, Zutphen. Boundary data: Statistics Netherlands (CBS) and Kadaster Nederland; via the R package tmap.
Fig 4.
Standardized log10 values of the variables in the case-control dataset.
Note that human, sampling, and bird densities are static variables and thus identical in both case and control situations. The spatio-temporal rate and wild bird mortality counts are elevated in the case situation.
Fig 5.
Mean performance measures from the ten-fold cross validation for all models.
The highest AUC and specificity values were found in the a and c models. Sensitivity was higher in the b models. See Table 3 for model details.
Fig 6.
Posterior means and 95% compatibility intervals for the intercept (⍺) and slopes of the best performing model.
The strongest effects is for the wild bird mortality counts (βN).
Fig 7.
Spatial visualisation of model predictions for the test subset of the case-crossover data.
As expected from the model sensitivity, cases are correctly marked as positive about half of the time. The majority of controls are marked correctly. Abbreviations: FN, false negative; FP, false positive; TN, true negative; TP, true positive. Boundary data: Statistics Netherlands (CBS) and Kadaster Nederland; via the R package tmap.
Fig 8.
Yearly model predictions of HPAI in wild bird mortalities.
For each year, a pair of figures is shown: model predictions on the left using a yellow to red colour gradient and HPAI incident counts on the right in green. Note that no HPAI incidents were reported in 2019. Result were aggregated using a 5x5 km grid. Boundary data: Statistics Netherlands (CBS) and Kadaster Nederland; via the R package tmap.
Fig 9.
Monthly aggregated HPAI infection predictions for wild bird mortalities over the years 2016-2022.
The number at the top of each panel indicates the month. Note that the series starts at the start of the HPAI season, October. Areas of higher risk are found from October to March and are low for the remainder of the year. Boundary data: Statistics Netherlands (CBS) and Kadaster Nederland; via the R package tmap.