layout: true <div class="my-footer"><span> FAST - Regional Restitution Meeting, 07 février 2022 </span></div> <!-- this adds the workshop footer to all slides, depends on my-footer class in css--> <!-- Determine background page as a function of params$style --> <style type="text/css"> .remark-slide-content { background-image: url(img/bg_page_eufmd.png); } </style> --- class: title-slide count: false <!-- Title-page specific background for the Astre style --> # Validation of Qualitative and Cartographic Risk Analyses ## Facundo Muñoz, Renaud Lancelot <img src="data:image/png;base64,#img/Cirad-ASTRE_Fr.png" style="width: 25%" /> ### European Commission for the Control of Foot-and-Mouth Disease (EuFMD) --- # Validation of risk-maps with field data - When we observe the __locations of epidemic outbreaks__ - When we analyse results from a __serological survey__ What does it tell about the __accuracy__ of the risk maps - of Introduction? - of Exposition? --- # 2 case studies 1. Foot-and-Mouth disease in Tunisia 2. Rift-Valley fever in Senegal --- class: middle, center # Case 1. Foot-and-mouth disease in Tunisia --- # FMD in Tunisia case-study .pull-left[ ![:scale 90%](data:image/png;base64,#img/TUN_risk_maps.png) ] .pull-right[ [![](data:image/png;base64,#img/tbed_aqcr.png)](https://dx.doi.org/10.1111/tbed.13920) ] --- # Validation of the risk-of-introduction - Introduction site(s) not always feasible to identify - Only one introduction event is not __enough__ --- # Validation of the risk-of-exposition .pull-left[_Ex-ante_ diffusion risk map and observed outbreaks] .pull-right[ ![:scale 55%](data:image/png;base64,#img/TUN_cases-and-risk-1.png) ] --- # Outbreaks by risk-category .pull-left-third[ Compare to total __surface area__ exposed at each risk level: if risk was constant -> \#cases `\(\propto\)` area ] .pull-right-twothirds[ ![](data:image/png;base64,#img/TUN_risk_freq.png) ] --- # Statistical modelling Take into account: 1. the sampling variability 2. the stpatial clustering caused by tranmissibility .small[ Due to the small number of observations at the most extreme categories we have re-grouped them into 2: __non-exposed and exposed__. ] --- layout: true # Statistical modelling --- ![:scale 90%](data:image/png;base64,#img/validation_model0.png) --- ![:scale 90%](data:image/png;base64,#img/validation_model1.png) --- ![:scale 90%](data:image/png;base64,#img/validation_model2.png) --- ![:scale 90%](data:image/png;base64,#img/validation_model3.png) --- ![:scale 90%](data:image/png;base64,#img/validation_model4.png) --- ![:scale 90%](data:image/png;base64,#img/validation_model5.png) --- ![:scale 90%](data:image/png;base64,#img/validation_model6.png) --- ![:scale 90%](data:image/png;base64,#img/validation_model7.png) --- ![:scale 90%](data:image/png;base64,#img/validation_model8.png) --- layout: false # Results: effect of exposition ![:scale 70%](data:image/png;base64,#img/TUN_effect_exposition.png) --- # Results: spatial effects ![:scale 55%](data:image/png;base64,#img/TUN_spatial-effects.png) --- # Results: predicted risks ![:scale 70%](data:image/png;base64,#img/TUN_pred_risks.png) --- # Results: predicted foci distribution ![:scale 70%](data:image/png;base64,#img/TUN_predicted_foci.png) --- # Conclusions - The validation of diffusion risk-maps is doable with an appropriate __statistical model__ - The procedure can __hardly be done automatically__ (the model itself must be _validated_) and requires the intervention of a statistician. - The __descriptive__ representation of the share of outbreaks by risk-level with respect to the surface area can be used as suggestive and for communication __if the epidemy has sufficiently propagated__. - Data from seroprevalence surveys can be used provided that they are interpretable in terms of disease prevalence (depends on the disease, age of animals, vaccination, etc.) and the use of an appropriate model. --- class: middle, center # Case 2. Rift-valley fever in Senegal --- # MCDA for SBR Serological data: seroprevalence by commune ($m$ positive animals out of `\(n\)` sampled). A total of 138 communes were visited visited (sampling fraction of 32%), and 2,000 AC RVFV ELISA test results were available, of which 199 were positive, giving an individual seroprevalence of `\(\simeq 10\%\)`. --- # Seroprevalence after the 2013-14 epidemic .pull-left[ What is the __predictive value__ of the MCDA index for the risk of infection with RVFV E.g., can high (above median) deciles of the MCDA index be used to focus RVF surveillance on these regions? ] .pull-right[ ![](data:image/png;base64,#EcoPPR/figs/mcda.png) ] --- # Significance of the MCDA index - Calculation of the relative risks `\(RR_i\)` associated with the intervals between deciles of this MCDA index above the median. - The seroprevalence in the interval between the minimum and the median of the MCDA is taken as the __common denominator__ of the `\(RR_i\)`. .small[ - In practice : - Calculation of quantiles 0% (min), 50% (median), 60%, 70%,..., 100% (max) = bounds of the MCDA index classes, - For each class `\(i\)` (50-60%,..., 90-100%), calculate the average of the seroprevalences `\(P_i\)` of the communes in this class - Calculate `\(RR_i = P_i / P_{0-50\%}\)`. ] --- # Bootstrap for confidence intervals .pull-left[ - Draw 500 pseudo-samples of communes, with replacement, of the same size as the initial sample - Re-calculate the `\(RR_i\)` for each pseudo-sample - CI bounds: quantiles 2.5% and 97.5% of each set of 500 `\(RR_i\)` values. ] .pull-right[ ![](data:image/png;base64,#img/table_RR.png) ] --- ## Conclusions How to optimise a monitoring programme using the results of a multi-criteria analysis? .small[ - __Collective work:__ multi-disciplinary group with disease experts to define risk factors and validation data - __Data quality__: relevant to the questions asked, exhaustive (no missing data), and with good spatio-temporal resolution. - Start with the simplest possible MCDA model: Occam's razor principle "_Shave away what is unnecessary_". - __NB__ specific results for RVF __and the occurrence of Tabaski in the rainy season__, which will soon no longer be the the case. This change will have consequences for the epidemiology of RVF and other infectious diseases of small ruminants. It will be necessary to redo the risk assessment for all these diseases. ] --- # Synthesis - __Validation__ of the MCDA index crucial to show the relevance of using this index to guide surveillance and control activities - __Field data__ either from observed outbreaks or from serological surveys. - Using the subjective risk assessment as an __explanatory variable__ for the observations in a statistical model. - The results allow __quantifying the predictive capacity__ of the subjective assessment on disease spread. --- class: middle # .fancy[Thank you!] ![](data:image/png;base64,#img/Cirad-ASTRE_Eng.png) --- layout: false background-image: url(data:image/png;base64,#img/bg_page_eufmd_final.png) background-size: cover class: title-slide