Climate change may cause shifts in geographical range, prevalence, and/or severity of some infectious diseases [1
], including tularemia [6
], a dangerous zoonotic disease caused by the intracellular bacterium Francisella tularensis
] and widely prevalent in Europe, Asia, and America [8
]. Transmission of Tularemia is usually caused by contact with infected rodents and hares, or by arthropod vectors [9
]. In Europe, there is also a strong association between F. tularensis subsp. holarctica
(Type B) and water conditions, with many humans reported to have contracted the disease around lakes and rivers [10
]. Europe as a whole does not have a clear trend of tularemia outbreaks in recent decades, but rather a pattern of repeated local emergence and re-emergence throughout most countries [11
]. In Sweden, however, the nationwide incidence of tularemia increased during the period 1984-2012 and the disease now occurs over a larger geographical area [12
Previous studies have reported relationships between hydroclimatic factors and tularemia outbreaks [10
], with some also evaluating future change scenarios [10
]. However, perspectives and conclusions regarding future tularemia changes vary. For example, for tularemia in Sweden, Palo et al. [16
] concluded that warming should not increase the frequency of tularemia outbreaks, whereas Rydén et al. [10
] addressed a future scenario of an approximately 2 °C increasinge, concluding that increase in monthly summer temperature should be expected to increase the duration of tularemia outbreaks in Sweden, and Ma et al. [6
] showed generally high tularemia sensitivity to hydroclimatic variability and change.
In view of the quite limited investigations so far, and their different perspectives and conclusions, this study aims to more comprehensively consider future climate change projections and assess their implications for tularemia incidence. This is done with focus on change trends in disease outbreaks along the steep climatic gradient spanned by different Swedish sites (counties) with relevant data and previously established statistical disease models, which are here combined with the latest outputs of a multi-model ensemble of global climate models (GCMs) from phase six of the Coupled Model Intercomparison Project (CMIP; [18
4. Discussion and Conclusions
Results for the six Swedish counties show large tularemia sensitivity to relatively small hydroclimatic change trends, and large inter-GCM uncertainty levels for disease projections compared to those for the underlying hydroclimatic variables. High sensitivity to the power-law disease scaling characteristics is also evident in the widely different disease projection results under more or less similar hydroclimatic change trends among the counties. Among counties, the relatively southern counties of Örebro and Gävleborg (Ockelbo) exhibit periodic or overall tularemia declines, respectively, while the most northern counties Norrbotten and Jämtland, but also the southern Värmland, exhibit large increases, and the intermediate Dalarna and Gävleborg (Ljusdal) exhibit intermediate trends of mostly increases (depending on climate scenario) until 2100.
In general, projected long-term trends in endemic levels are closely related to the power-law exponent
for scaling of Tul with Tullag
. With Norrbotten having superlinear
and Värmland and Jämtland having sub- but near-linear
= 0.99 and 0.93, respectively, associated tularemia and endemic level impacts tend to be enhanced by projected shifts in hydroclimatic (scale factor A) conditions. The differences in projected tularemia cases among counties and some mixed results for the three climate scenarios challenge possible notions that climate change (and higher emission scenarios) will generally lead to (higher) increases of disease incidence. The different best-fit
values for the different counties may then implicitly reflect geographic differences in, e.g.,: (1) demographics; (2) risk of pathogen exposure; (3) other local conditions/measures affecting vulnerabity/resilience to disease; (4) and interactions among hydroclimatic factors. This notion has also been discussed in other research using statistical disease modeling to project future disease burden under various climate scenarios. For example, Hales et al. (2002) [25
] estimated that climate change would lead to 50–60% of the global population being at risk of dengue transmission by 2085, compared with 35% without the projected climate change. For falciparum malaria, however, Rogers and Randolph (2000) [26
] projected that, by 2050, 23 million hosts would be gained in previously uninfected regions while 25 million human hosts would be lost in areas no longer suitable for transmission, which would lead to little net disease change in total.
With regard to the latter, the high-emissions scenario (SSP5-8.5) led to similar increases in summer temperature across the counties, but warming-related changes in precipitation and runoff affected tularemia results most. Other studies have found that human-related factors may play a more important disease role than climate, e.g., superior healthcare infrastructure might lead to net lowering of disease impacts even if climate change enhances pathogen ranges [19
]. The history of widely studied diseases (e.g., malaria, yellow fever, dengue fever) also shows that human activities and their impacts on local ecology may affect disease spreading more significantly than climate change [27
]. In addition to external drivers, internal complexity of climate-disease interactions also affects disease risk. For example, an increase in temperature may increase mosquito biting rates, parasite replication within mosquitoes, and mosquito development, but also increase mosquito mortality, making disease outcomes difficult to determine [28
This study also has several limitations. First, in Sweden, high tularemia incidence usually appears in low- population areas [14
]. So, although an area can have high outbreak rate projections (such as Norrbotten and Värmland in this study), local population levels may set lower upper limits for outbreaks. Second, the considered disease models do not explicitly take human behavioral factors into account, such as time spent on outdoor activities, which is an important factor for exposure of humans to the disease. Moreover, the statistical tularemia models have been rather inaccurate in estimating the magnitude of recent outbreaks, especially for the two most northern counties (Norrbotten and Jämtland) [14
]. Thus, the present results for how hydroclimatic changes may impact future outbreaks should be used with caution, as comparative indications rather than in terms of absolute disease outbreak projections.
Uncertainties are inevitable in and among projected results of different climate models, and in all studies using climate model outputs in other types of models. The latter, propagated type of climate-related uncertainties are here shown to be amplified in disease models with high climate sensitivity, such as the tularemia models for Norrbotten, Jämtland, and Värmland in this study. Bring et al. (2019) [29
] found relatively good model-data agreement for ensemble mean outputs of runoff and temperature in the Nordic-Arctic region, but worse agreement for precipitation outputs, and considerable doubt still remains about realism and accuracy of hydroclimatic results from individual GCMs. Contradictions inevitably emerge in disease projections for localized transmission routes [14
] and future work needs to continue exploration of opportunities to improve projection realism and accuracy.
Well-archived data of infectious diseases in clinics and laboratories, along with adequately-recorded climate, hydrological, and other environmental as well as socio-economic data in recent years has made it feasible to develop statistical disease models, which can further be used in combination with related projected hydroclimatic and other types of data to quantify scenarios of possible future disease evolution (Figure 2
). The available historical records along with forthcoming new scenario-projected model data and model-coupling methods will surely benefit more accurate large-scale assessments of future disease pressures and risks. In addition, advancements in mechanistic disease modeling are needed to bridge gaps and overcome weakness of statistical models, which may not be as relevant for other locations and new hydroclimatic and other environmental and societal conditions than the ones they were fitted to.
In conclusion, this study has quantified the implications of scenario-projected future hydroclimatic trends for possible future disease evolution, using site-specific, established, and parameterized statistical tularemia models. Results show highly divergent disease change trends and fluctuation levels around these for future climate change scenarios among Swedish counties, with scenarios of steeper future climate warming not necessarily leading to steeper disease increases. The directions of future tularemia change trends are robust in some counties, as seen from results across various future climate scenarios and their representations by different GCMs. Such robust change-trend projections are essential to identify and useful in pointing out needs for policy and management measures to avoid clear negative directions of future disease evolution, even though uncertainties about absolute future disease numbers may be large.