Climate change is at the forefront of changes in ecosystems, and while the insurance industry has studied its impact on property and casualty losses extensively for many years, we are only just beginning to understand climate change’s effects on morbidity and mortality.

In previous generations, the spread of rodents and other animals paved the way for infectious diseases and pandemics such as the infamous Black Death (bubonic plague). This arguably came about in part due to climate, with the “so‑called mediaeval climatic optimum”.¹ We (a group of researchers from the Society of Actuaries, Columbia University and Gen Re) researched the complex relationship between climate variables and the spread of one infectious disease, Lyme disease (LD), in the U.S. This blog provides a summary on some of our findings.

Lyme disease is a well-known disease particularly in the eastern U.S., and is also endemic in other geographies. Its transmission to humans occurs via the bite of ticks infected with certain types of Borrelia (a genus of bacteria). While by no means a major contributor to causes of death, it can be a debilitating and long-term disease, particularly if left untreated.

Changes in temperature or precipitation trends as well as corresponding changes in land usage have the potential to spread not only LD but other vector-borne illnesses to new geographies. Changes in geographic footprint are a dangerous phenomenon, in part because lack of awareness can lead to untreated disease and complications.

The following graph² shows the shifting geography of LD in the U.S.:

To better understand the determinants of LD, we combined data on the LD incidence (the number of reported LD cases in a county per 100,000 inhabitants) with data on temperature, precipitation, land cover, and socioeconomic factors for 3,104 U.S. counties over the period 2000‑2018. The following graph illustrates how average minimum temperature and temperature range have changed in the U.S. over this period, showing a similar pattern to the changes in the LD incidence.³

We also performed extended feature engineering over the period of a tick’s life span to account for time lag effects of precipitation and temperature on the incidence. A favourable environment in one year improves the survival of ticks in the initial stages of their life cycle, thereby increasing the following year’s host-seeking (and potentially Lyme-transmitting) tick population.

To estimate different types of effects of climatic, land cover, and socioeconomic factors on the LD incidence, a fixed effects spatial autoregressive model with autoregressive disturbances, the SARAR model, was employed, with county and time effects.⁴ Nonspatial linear models rule out spillover effects of predictor variables;⁵ that is, a change in precipitation or temperature in one county is assumed to have no effect on the LD incidence in any other county, including its neighbours.

However, in the context of infectious diseases, it might be reasonable to expect spillover effects. Improved microclimate or the expansion of tick habitats in one county may increase tick density in that county, and more ticks might be able to migrate to neighbouring counties via attaching to their hosts. The SARAR model allows for such spillover effects as well as for spatial correlation in unobserved factors captured by the error term. Estimation results for climatic factors are presented next.

Temperature

The effects of the average minimum temperature in spring and summer of a given and preceding year on the LD incidence were found to be positive, indicating that increasing temperatures may lead to more LD. The effects of spring, summer, and winter temperature ranges in both years were found to be negative. Temperature range variables capture the difference between temperature extremes in different seasons. Long-term exposure to temperature extremes may increase tick mortality rates due to overheating and dehydration in summer or freezing in winter.⁶

Precipitation

The effects of total precipitation in all seasons, and in given and preceding years, on the LD incidence were found to be of low magnitude and negative. Precipitation may have contradictory effects on the incidence. High humidity is generally beneficial for tick survival and host-seeking,⁷ but humans tend to stay indoors or use rain protection on rainy days, which protects them from tick bites. The negative estimates of the effects of precipitation in a given year on the LD incidence may indicate that the negative impact of precipitation on humans is greater than its positive impact on ticks.

Extreme weather

Extreme weather events are represented by two variables: frost, which captures below zero temperatures in the (near) absence of snow cover, and drought, which captures high temperatures in the (near) absence of precipitation. The effects of drought and frost on LD incidence were found to be negative. These results are consistent with previous findings that extreme weather events, especially if they repeat over time, may reduce tick survival⁸ and host-seeking capacities.⁹

In addition to finding that climate has a material if complex relationship with incidence of LD in the U.S., the report adds to existing literature on the topic by:

• Linking disease to healthcare costs and complications
• Offering an international scan (Germany, Finland, and Japan)
• Offering the advanced regression model as a framework for other infectious diseases and their link to climate

For further details, see the full report published by the Society of Actuaries Research Institute: climate-impact-tick-borne-illness-report.pdf (soa.org)

The author’s gratitude goes to Yaryna Kolomiytseva and Sara Goldberg who reviewed this blog, as well as Jan Kasperek, intern and data scientist at Gen Re, whose graphics were used above to illustrate the findings in the report.

Endnotes

1. Henson 2002, The Rough Guide to Weather. Rough Guides 383
2. Society of Actuaries Research Institute 2022, Climate Impact on Tick-Borne Illnesses, https://www.soa.org/4a1211/globalassets/assets/files/resources/research-report/2022/climate-impact-tick-borne-illness-report.pdf, p.7.
3. Ibid., p.17.
4. Kelejian and Prucha 1998, The Journal of Real Estate Finance and Economics 17: 99‑121. doi:https://doi.org/10.1023/A:1007707430416 100; Anselin, Gallo and Jayet 2008, The Econometrics of Panel Data. Advanced Studies in Theoretical and Applied Econometrics, 625‑660. doi:https://doi.org/10.1007/978-3-540-75892-1_19 640; Millo and Piras 2012, Journal of Statistical Software 47 (1): 1‑38. doi:https://doi.org/10.18637/jss.v047.i01 10‑11
5. LeSage and Pace 2009, Introduction to Spatial Econometrics. Boca Raton: Taylor & Francis Group 34‑37
6. Eisen, et al. 2016, Journal of Medical Entomology 53 (2): 250‑261. doi:https://doi.org/10.1093/jme/tjv199 252
7. Ibid; Rodgers, Zolnik and Mather 2007 Journal of Medical Entomology 44 (2): 372‑375. doi:https://doi.org/10.1093/jmedent/44.2.372
8. Stafford 1994 Journal of Medical Entomology 31 (2): 310‑314. doi:https://doi.org/10.1093/jmedent/31.2.310; Eisen, et al. 2016 Journal of Medical Entomology 53 (2): 250‑261. doi:https://doi.org/10.1093/jme/tjv199
9. Schulze, Jordan and Hung 2001 Journal of Medical Entomology 38 (2): 318‑324. doi:https://doi.org/10.1603/0022-2585-38.2.318

All endnotes last accessed on 6 July 2022.