Skip to main content
White Paper

Driver-based modelling

The mortality impacts of climate change - part 3 of a 5-part series

Climate Quantified|Insurance Consulting and Technology
Climate Risk and Resilience|Climate and Resilience Hub|Insurer Solutions

By Keziah Baskerville-Muscutt and Richard Marshall | March 30, 2021

This paper investigates the liability-side impact of climate change. The third chapter describes our modelling approach in detail.

What do we mean by a ‘driver-based model’ of mortality?

Unlock More

About this series

'The mortality impacts of climate change' investigates the liability-side impact of climate change, focusing on the effects on UK life insurers. It considers the mechanisms through which climate can affect mortality, models and quantifies some of those effects and discusses the implications for the life insurance industry.

A driver-based model of mortality (or “cause-of-death” model) considers the response of overall population mortality to changes in the levels of individual drivers of mortality. Using a traditional example, if it is believed that there will be a continued trend towards decreasing smoking prevalence, then by understanding the link between smoking and mortality from a range of different causes, the model will predict a change in mortality which reflects this change in behaviour. At its most basic, such a model could predict a change in all-cause mortality as a function of a single driver of mortality. However, much of the value of a driver-based approach to modelling is in its ability to distinguish between the effects of a single driver on a range of causes of death, and to compound the effects of multiple drivers on each single cause of death. For instance, alcohol consumption might have a significant impact on deaths from liver disease but may have little or no impact on deaths from lung cancer; on the other hand, both smoking and obesity would be expected to materially affect the risk of cardiovascular disease mortality.

There is no fixed choice of drivers in the model.

  • If a holistic view of changes in mortality is sought, different actuaries (and those experts advising them) will likely have differing views on which are the key drivers underlying the mortality improvements of the future.
  • Models developed for other purposes will require different drivers to be included. For instance, a model of the effects of exposure to ultraviolet light on mortality would include a measure of exposure to ultraviolet radiation, perhaps sunbed sessions per person per day, which would not be considered sufficiently material for inclusion in a holistic model.

The set of drivers in the model need not be complete if the purpose of the model is to investigate the effect of a change between two different outlooks for those drivers. The difference between the mortality projections will be fairly stable with respect to the addition of other drivers (whose outlooks remain the same in both scenarios being considered).

Irrespective of the drivers chosen, the model is typically parameterised using a wide range of results from academic papers, overlain with the actuary’s own expert judgement.

Willis Towers Watson’s driver-based model of mortality and calibration approach

In this model, the relationship between each short-listed climate driver and cause-specific mortality (see Chapter 2) was calibrated to a wide range of results from academic papers, overlain with the actuary’s own expert judgement. Table 1 shows the relative risk measures which have been used to estimate the impact of a given unit of change in a driver of cause-specific mortality for each of several causes.

Table 1. Calibration of the cause of death model: relative risks for death (by cause) per driver in 2019
Driver Cause of death Source(s)
Cancer IHD CVD Respiratory Infectious Other Medical Non-Medical
Extreme heat
% of year at 21 °C or above
1.04 1.06 1.06 1.1 1.13 1.04 1.17 (Ishigami, et al., 2008)
(Page, et al., 2007)
Extreme cold
% of year at 1°C of below
1.02 1.03 1.03 1.09 1.02 1.02 1.02 (Hajat, et al., 2007)
(Analitis, et al., 2008)
Snowfall
% of year with at least 3cm snowfall
1 1.2 1.15 1.2 1 1 0.95 (Gorjanc, et al., 1999)
Rainfall
% of year with daily rainfall of 10mm or greater
1 1 1 1 1 1 1.06 (Gorjanc, et al., 1999)
Storms
% of year during which a named storm is over the UK.
1 1.3 1.3 1.3 1 1 1 (Yan, et al., 2018)
Air Pollution
% of year during which:
· PM10 is above 35µg/m3, or
· O3 is above 42.5 ppb
1.03 1.11 1.11 1.12 1.03 1.03 1.03 (Rooney, et al., 1998)

Relative risk is not static and may change over time with a changing baseline climate. For example, under a worst-case scenario long-term increases in levels of air pollution can lead to an increase in both the percentage of year during which PM10 is above 35µg/m3 O3 is above 42.5 ppb and an increase in the mean level of background air pollution. To take account of the change in risk that accompanies a change in background air pollution we adjust relative risk linearly over time, using actuarial expert judgement.

The assumed change in relative risk for the best-case, best-estimate and worst-case scenarios for each driver in 2078 is shown in Table 2.

The calibrated links between drivers and cause-specific mortality can be combined with indicators of how these drivers have changed over the recent past and best-estimate projections (e.g. from academic studies or NHS or Public Health England publications) of how these drivers might continue to change in the short- to mid-term future to estimate the evolution of overall mortality rates. The next sub-section sets out how the best-estimate projections (or scenarios) have been developed for this model.

Table 2. Assumed change in relative risk for each cause of death by driver by 2078
Driver Best-case Best-estimate Worst-case
Extreme heat
% of year at 21°C or above
3% 7% 10%
Extreme cold
% of year at 1°C or below
0% 0% 0%
Snowfall
% of year with at least 3cm snowfall
0% 0% 0%
Rainfall
% of year with daily rainfall of 10mm or greater
0% 0% 0%
Storms
% of year during which a named storm is over the UK
0% 1% 7%
Air Pollution
% of year during which:
· PM10 is above 35µg/m3, or
· O3 is above 42.5 ppb
0% 2% 20%

Scenario development – from IPCC scenarios to climate-variable projections

The changes in each climate driver under three future climate scenarios were calculated, again using available academic research, data and actuarial expert judgement. The scenarios aim to be a realistic representation of possible policy, technology and physical risk developments under different temperature pathways. They therefore represent plausible climate-aware, real-world scenarios rather than climate stress-test scenarios, i.e. they are not potential worse case scenarios.

The data and assumptions used to construct these scenarios are detailed in Table 3, but at a high level the scenarios cover:

  • Best-estimate scenario (baseline) – aligning approximately with the RCP 4.5 scenario defined in the 2014 IPCC report (IPCC, 2014).
  • Worst-case scenario - aligning approximately with the RCP 8.5 scenario defined in the 2014 IPCC report (IPCC, 2014).
  • Best-case scenario - aligning approximately with the RCP 2.6 scenario defined in the 2014 IPCC report (IPCC, 2014).

For each scenario, expectations of life were calculated at five-yearly intervals from age 20 to age 90 (as at 1 Jan 2018) for males or females. The proportional change in expectation of life was calculated relative to the base scenario for both the best- and worst-case scenarios. The proportional change in deferred expectation of life from age 65 was also calculated relative to the base scenario for both the best- and worst-case scenarios at each five-yearly interval from 2023 until 2053.

Table 3. Data sources and assumptions used to construct historic and future (2018-2078) projections for each climate driver
Driver Scenario Projection (2018-2078) Assumptions Source
Extreme heat
% of year at 21 °C or above
Historic Based on historical (1950-2005) daily-modelled temperature series (Guo, et al., 2018)
(ISIMIP, 2020)
Best-estimate (baseline) Increase from 1% to 6% Based on daily-modelled temperature series for future (2006-2100) periods under the RCP 4.5 scenario defined by the IPCC As above
Worst-case scenario (RCP 8.5) Increase from 1% to 9% Based on daily-modelled temperature series for future (2006-2100) periods under the RCP 8.5 scenario defined by the IPCC As above
Best-case scenario (RCP 2.6) Increase from 1% to 4% Approximated based on the relationship between RCP 8.5 and 4.5 scenarios As above
Extreme cold
% of year at 1°C of below
Historic Based on historical (1950-2005) daily-modelled temperature series (Guo, et al., 2018)
(ISIMIP, 2020)
Best-estimate (baseline) Decrease from 4.4% to 1.6% Based on daily-modelled temperature series for future (2006-2100) periods under the RCP 4.5 scenario defined by the IPCC As above
Worst-case scenario (RCP 8.5) Decrease from 4.4% to 1% Based on daily-modelled temperature series for future (2006-2100) periods under the RCP 8.5 scenario defined by the IPCC As above
Best-case scenario (RCP 2.6) Decrease from 4.4% to 2.3% Approximated based on the relationship between RCP 8.5 and 4.5 scenarios As above
Rainfall
% of year with daily rainfall of 10mm or greater
Historic Based on historical (1980-2010) daily-modelled temperature series (Met Office, 2019)
Best-estimate (baseline) Increase from 3.9% to 4.5% Based on daily-modelled temperature series for future (2011-2100) periods under the RCP 4.5 scenario defined by the IPCC As above
Worst-case scenario (RCP 8.5) Increase from 3.9% to 5.3% Based on daily-modelled temperature series for future (2011-2100) periods under the RCP 8.5 scenario defined by the IPCC As above
Best-case scenario (RCP 2.6) Increase from 3.9% to 4.2% Approximated based on the relationship between RCP 8.5 and 4.5 scenarios As above
Snowfall
% of year with at least 3cm snowfall
Historic Historic data on number of days with snow in the UK at Heathrow Airport (WeatherOnline, 2020)
Best-estimate (baseline) Decrease from 0.8% to 0.64% Based on UK Climate Projections 18 factsheet (Met Office, 2019)
Worst-case scenario (RCP 8.5) Decrease from 0.8% to 0% As above As above
Best-case scenario (RCP 2.6) No change
As above As above
Storms
% of year during which a named storm is over the UK
Historic Assume only named storms will cause death(s) (Met Office, 2018)
Best-estimate (baseline) No change Based on results of Kunkel et. Al, 2013 (Kunkel, et al., 2013)
Worst-case scenario (RCP 8.5) Increase from 2.5% to 8% Assume 1% increase every 30 years As above
Best-case scenario (RCP 2.6) No change Based on results of Kunkel et. Al, 2013 As above
Air Pollution
% of year during which:
· PM10 is above 35µg/m3, or
· O3 is above 42.5 ppb
Historic Based on historical observations of ozone concentrations (Defra, 2020)
Best-estimate (baseline) Decrease from 4.9% to 4.5% Approximated based on Doherty et. al. 2017 (Doherty, et al., 2017)
Worst-case scenario (RCP 8.5) Increase from 4.9% to 19.7% As above As above
Best-case scenario (RCP 2.6) Decrease from 4.9% to 4.4% As above As above

Consistency with existing mortality projections

Whilst it would be reasonable to assess the difference between the best-estimate and each of the best- and worst-case scenarios without considering other drivers and improvements in mortality, for consistency between the best-estimate scenario and an instance of the CMI_2017 model, the best-estimate scenarios for the climate drivers were considered to be contributing to the overall improvements generated by the CMI_2017 model.

The effect of all other possible drivers of mortality was therefore the improvements obtained by netting off the improvements due to climate drivers. These background improvements were included, unchanged, in both the best-estimate and the stressed (best- or worst-case) scenarios. This has the effect of contextualising the modelled change in mortality improvements within the overall outlook for mortality in the UK.


Next time

The fourth chapter looks at the individual and combined effects of stresses to the climate variables. Also considered are the potential impacts of other unmodelled consequences of climate change, and of climate volatility in individual years rather than long-term trends.

Upcoming chapters

  • Impacts of climate change
  • Implications for life insurers and pension schemes

Footnotes:

Analitis, A. et al., 2008. Temperature effects on mortality: Potential confounding by air pollution and possible interactions within the PHEWE project.. Epidemiology, 19(1), p. 214.
Defra, 2020. UK air: data archive. [Online] Available at: https://uk-air.defra.gov.uk/data/ [Accessed July 2020].
Doherty, R. et al., 2017. Multi-model impacts of climate change on pollution transport from global emission source regions.. Atmospheric Chemistry and Physics, 17(23), pp. 14219-14237.
Eisenberg, D. & Warner, K., 2005. Effects of snowfalls on motor vehicle collisions, injuries, and fatalities.. American journal of public health, 95(1), pp. 120-124.
Gorjanc, M. et al., 1999. 1999. Effects of temperature and snowfall on mortality in Pennsylvania.. American Journal of Epidemiology, 149(12), pp. 1152-1160.
Guo, Y. et al., 2018. Quantifying excess deaths related to heatwaves under climate change scenarios: A multicountry time series modelling study. PLoS medicine, 15(7).
Hajat, S., Kovats, R. & Lachowycz, K., 2007. Heat-related and cold-related deaths in England and Wales: who is at risk?. Occupational and environmental medicine, 64(2), pp. 93-100.
IPCC, 2014. Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. In: R. Pachauri & L. Meyer, eds. Geneva, Switzerland: Intergovernmental Panel on Climate Change. 
Ishigami, A. et al., 2008. An ecological time-series study of heat-related mortality in three European cities.. Environmental Health, 7(1), p. 5.
ISIMIP, 2020. The Inter-Sectoral Impact Model Intercomparison Project. [Online] Available at: https://www.isimip.org/ [Accessed November 2020].
Kunkel, K. et al., 2013. Monitoring and understanding trends in extreme storms: State of knowledge.. Bulletin of the American Meteorological Society, 94(4), pp. 499-514.
Met Office, 2018. UK storm season 2017/18. [Online] Available at: https://www.metoffice.gov.uk/weather/warnings-and-advice/uk-storm-centre/uk-storm-season-2017-18 [Accessed August 2020].
Met Office, 2019. UK Climate Projections User Interface. [Online] Available at: https://ukclimateprojections-ui.metoffice.gov.uk/ui/home [Accessed July 2020].
Met Office, 2019. UKCP18 Factsheet:, s.l.: s.n.
Page, L., Hajat, S. & Kovats, R., 2007. Relationship between daily suicide counts and temperature in England and Wales.. The British Journal of Psychiatry, 191(2), pp. 106-112.
Rooney, C., McMichael, A., Kovats, R. & Coleman, M., 1998. Excess mortality in England and Wales, and in Greater London, during the 1995 heatwave.. Journal of Epidemiology & Community Health, 52(8), pp. 482-486.
WeatherOnline, 2020. Heathrow Airport - snow depths. [Online] Available at: https://www.weatheronline.co.uk/weather/maps/city?WMO=03772&CONT=ukuk&LAND=AC&ART=SNW&LEVEL=150 [Accessed July 2020].
Yan, M. et al., 2018. September. Tropical Storms and Associated Risk to All-Cause, Accidental, Cardiovascular, and Respiratory Mortality in 78 United States Communities. ISEE Conference Abstracts.

Authors

Keziah Baskerville-Muscutt
Risk Analyst, Insurance Consulting and Technology

Richard Marshall
Director

Richard Marshall is a Director in Willis Towers Watson’s Insurance Consulting and Technology business and leads the development of mortality and demographic risk models for our UK business.


Related Solutions

Contact Us