Project Background:

This project evaluated the success of current policies and plan for health and social services and pensions, estimates of past trends and forecasts of future mortality and life expectancy are essential. These forecasts were completed for one or more countries and in order to plan and set priorities at the subnational level, local mortality forecasts were required.


Study Aims:

The aim of the project was to develop, test and apply statistical methods for estimating both historical and projected trends in total and cause-specific mortality.

We developed a family of Bayesian space-time-age-birth cohort models which leverage relatedness both within and between these dimensions in order to make robust and coherent estimates. For cause-specific data we took into account relatedness between disease groups. In particular, we modelled the way in which mortality from some diseases may either increase or decrease together because they have common drivers, or may change in opposite directions due to competing causes. Model outputs are estimates and forecasts of death rates and where appropriate life expectancy. We evaluated the performance of each model by withholding data from recent years, and evaluated how the model estimates the known-but-withheld data.

Health data:

ONS mortality. We applied the methods to deaths both from England and Wales’ districts and from US counties. The England and Wales mortality, population and geographical data are held by SAHSU.

Benefits to Public:

Consistent forecasts for all subnational units within a country are very rare, even though mortality and life expectancy vary substantially within countries, both geographically and in relation to social class.This project overcomes a methodological gap in making estimates of past trends and future projections of mortality by age group, location, and medical cause of death.


Bennett JE, Li G, Foreman K, Best N, Kontis V, Pearson C, Hambly P, Ezzati M. The future of life expectancy and life expectancy inequalities in England and Wales: Bayesian spatiotemporal forecasting. Lancet April 2015.