Skip to main content

Shining more light on medium-term views of mortality improvements

A new framework for future mortality – part 4 of a 7-part series

By Richard Marshall | October 21, 2019

Part 4 in our series discusses medium-term improvements and looks at how medical modelling, driver-based modelling and expert judgements combine to form a coherent view of mortality improvements over this time period.
Insurance Consulting and Technology
Insurer Solutions

The medium term is perhaps the most neglected period for mortality improvement modelling. Many approaches combine some form of initial improvement assumptions or short-term projections with a long-term target rate of improvement. The medium term can then become simply a by-product of the shape and period of transition from the former to the latter.

Characterising the ‘medium term’ for improvement modelling

Back in the second post of this series, we characterised the medium term as the time over which we can still forecast a range of drivers of mortality based on current information, suggesting this might extend out to around 10 to 15 years into the future. This information could include:

  • Pharmaceutical and technological innovations in development, for which we can form a view on the scale and timing of their ultimate impact.
  • Political trajectories in respect of planned health and social care expenditure.
  • The current approximate distribution of diseases or other medical and lifestyle factors relevant to morbidity and mortality risk within the population.

Whilst changes which have taken place in the past are still of relevance to the medium term, their effects make a progressively smaller contribution beyond, say, five years into the future, a feature distinguishing the medium term from the short term. The medium term is more reliant upon what is going to change in the future than what has changed in the past.

A non-stationary population and medical modelling

The specific changes that are particularly relevant in the medium term are population dynamics and medical trends.

Unlock More

About the series

In this series of articles, we discuss the impediments to understanding mortality improvements and the dangers of herding and group-think in this process. In response to these issues, we introduce a new framework for the development of mortality improvement assumptions.

Since mortality improvements are usually expressed as a proportional reduction in mortality for individuals at a fixed age in successive calendar years, the comparison is between different cohorts of lives. This means that differences in the composition of the population at each age under 60 now will affect the improvements observed in mortality at age 60 at some point in the future; the improvements will reflect the relative health of the different cohorts.

This means that one way to investigate mortality improvements is to investigate the current and projected future prevalence of chronic diseases and medical and lifestyle factors within the population.

Suppose we have an approximate breakdown of the population for males aged 50 and 60 (say) into groups defined by the presence or absence of certain chronic diseases, smoker status and BMI range. By modelling the incidence of those diseases in the healthy population and all-cause mortality from each group over a period of 10 years, we can arrive at a projected breakdown of the population at the end of that 10-year period for those who will then be aged 60. This could indicate a higher or lower prevalence of each condition, or a change in the distribution of individuals between smoker statuses and BMI ranges. This is what we mean when we describe the population as being ‘non-stationary’.

A non-stationary population is effectively the reason for the existence of a ‘cohort effect’, typically incorporated into models of mortality. The difference between current mortality rates for those aged 60 and those which we would expect to see 10 years from now depends on a combination of:

  • the change in the prevalence of chronic conditions, and
  • the change in all-cause mortality in the presence of each condition.

A healthier cohort of lives (i.e. one with a lower prevalence of chronic conditions and a lower propensity to smoke than those cohorts preceding it) reaching age 60 would give the appearance of a higher rate of improvements at that age; the opposite would be true of a cohort with a higher prevalence of chronic conditions and a higher proportion of current smokers.

We can estimate the changes in all-cause mortality by using driver-based modelling, informed by the opinions of medical experts within whose fields of expertise the chronic conditions are to be found. The change in the distribution of conditions can be approximated using a disease-based model, such as the Willis Towers Watson PulseModel, our medically-informed model of morbidity and mortality.

Using Willis Towers Watson PulseModel to project improvements

Figure 1: Example of mortality reduction factors derived from PulseModel outputs
Figure 1: Example of mortality reduction factors derived from PulseModel outputs

Allowing for variation in the distribution of conditions across different ages means that cohort effects emerge naturally within the mortality projections. The first couple of years of the projection will be the most sensitive to the accuracy of our initial conditions, with changes in the mortality reflecting a (fairly rapid – particularly at old ages) reversion to the proportion of the population with each condition implied by the combined incidence and mortality rates. After converting to annual improvement rates, it is possible to remove (or place very low weight on) these initial years, combining them with the short-term view, the development of which was discussed in the previous post.

Figure 2: Example of mortality improvements surface
Figure 2: Example of mortality improvements surface

Driver-based modelling to inform morbidity and mortality transition rate evolution

Medical modelling does not make driver-based models obsolete - far from it. An understanding of the evolution of drivers of mortality will allow the evolution of both the incidence of given diseases and all-cause mortality in the presence of given diseases to be estimated for the purpose of the medical model projection.

As an example, if we believed that the imminently available pharmaceuticals for the treatment of Alzheimer’s disease represented negligible change from the status quo, but that these would strengthen over the next 10 years to the point where progression in (say) 25% of Alzheimer’s cases could be halted, this would have implications for mortality improvements for those with current neurological conditions and (separately) those who do not currently have neurological conditions.

Similarly, projected trends in behavioural factors, such as obesity or prevalence of smoking, would be expected to have an impact on both morbidity and mortality rates for a range of different groups of diseases.

Medical expertise

In setting those projected trends, and in assessing the overall improvements in morbidity and (all-cause) mortality rates per disease group, the advice of medical experts in relevant fields of practice will be invaluable. In the Willis Towers Watson PulseModel, the advice of a panel of such medical experts was used as the basis for the rates of evolution of the transition probabilities over time, making the projection genuinely ‘forward-looking’ rather than purely extrapolative.

Next time

In the next instalment, we will consider how firms can form their views of mortality improvements in the long-term and how they can blend their views over the short-, medium- and long-terms. We will discuss:

  • Fundamental drivers of mortality versus horizon scanning
  • Long-term international trends
  • Long-term average rates of improvement over varying periods of history
  • The ‘cost’ of future improvements as a limiting factor

Upcoming articles

  • Consistency with an Internal Model or economic capital model
  • Potential impacts on financial reporting


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.

Contact Us