模拟气候风险对死亡率的影响

Modelling theimpactofclimate risksonmortality Detailed case studiesonexposuretohigh temperatures,alternativemodelling forvector-bornediseases Alexandre BoumezouedAmal ElfassihiValentin GermainEve-Elisabeth Titon Climate change, partly caused by human activities, is already havingan impact on our society. According to the latestIntergovernmentalPanel on Climate Change (IPCC)reportof August 2021,1theconcentration of CO2in the atmosphere was at its highest levelin2019in the last 2,000 years. An increase in Earth's surface temperature will cause several problems such as extreme changes in weatherevents (heat waves, heavy precipitation, droughts etc.) by increasing their frequency and intensity. Two main impacts can be identified: TheFrench Insurance Federation(FFA)has published a report2in which the impacts of climate change onthe insurance sector in France are studied for 2040.The increase in the cost of claims due to climate changeisestimated at 21 billion euros over the period 2014to2039 compared to 8 billion euros over the period1988to2013. The study predicts that claims caused by natural hazards will reach 92 billion euros in 2040(an increase of 90%).Climate change will also have an impact on health and mortality.According to a report3published by the WorldHealthOrganisation(WHO),between 2030 and 2050climate change is expected to result in nearly 250,000additional deaths per yeargloballydue tochildhood undernutrition, malaria, diarrhea and heat stress. As these events will affect the whole world, (re)insurers will need to improve their models to cope with climatechange: by having a better understanding of the climate phenomena and their consequences and by takingintoaccount projection assumptions. Executive Summary For the main climatic causes of death,i.e.,those for which there are a significant number of deaths, a Lee-Cartertype of modelling can be applied. Theobjective of this paper is to propose a modelto capture the impact of climate risks on mortality. The modelconstructedisderived from a Lee-Carter modelandis adapted to capture theimpact of a specific cause onoverall mortality rates. Theriskconsidered inthe followingcase studiesisthe"exposure to hightemperatures" inFranceandinthe US(at the state level: in this paper we will present the results obtained for the state ofOklahoma).The objective is to develop a model and make projections for mortality shock values. Theselectedclimate models for the differentgeographical areasstudied include climate variables related totemperature.These variables explain the observed peaks in deaths due to summer heat waves.The globalmodelincludes a term capturing the global mortality without climatic causes, and a term modelling only themortality due to high temperatures. The resulting modelperforms wellinpredicting observedmortality rates. Moreover, itperformsbetter than aclassic Lee-Carter model according to the𝑅2andtheAkaike information criterion(AIC) andBayesian informationcriterion(BIC). The impact ofhigh temperaturesis clearly observable at high ages (over 65 years). Themortalityshocks—calculated according to theEuropean Insurance and Occupational Pensions Authority (EIOPA)methodology—incorporating the modelling of the impact of high temperatures are on average 6.12% larger than theconventional shocks provided by a classic Lee-Carter model. For climatic causeswith fewer deaths, such asvector-borne diseases, it is not possible to apply the modeldeveloped earlier. Specific modelling needs to be explored: for mosquito-borne diseases,arefinedSusceptible/Infected/Recovered (SIR)-type model could be appropriated. Scopeanddata SCOPE For this paper, we choseto focus on France anda specific state in theUSwithone specific climatic cause:exposure tohigh temperatures. In France, there have been significant heat waves: since 1947, 41 heatwaves hit France with different intensities,with a major heat wave in 2003 (more than 12,000 deaths). The climate inthe US is not homogeneous throughout the countryasnot all states are subjectto heat waves.The areaextending over several states in the southwest of the US is the one presenting heat waves.As our model does notproduceconsistent results for statesoutside ofthis region,the state of Oklahoma, which is prone to heat waves,isselected to present the results. DATA This study combines the use of three databases: one for the mortality linked to the specific cause (GlobalHealthData), one for theglobal national mortality (HumanMortalityDatabase) and one for the climate variables. GlobalHealth Data (GHD):This database is published by the Institute for Health Metrics and Evaluation(IHME). GHD is constructed with the death numbers classified bydifferent parameters (age, territory,years etc.) and particularly by the cause of death. The selection of death numbers relative to one specificrisk or cause is possible on this database. On the“pollution”cause,for instance, we can find all thedeaths caused by pollution,including