Stochastic models for cause-specific mortality

EPSRC Research Area Statistics and applied probability
This project aims to develop a Bayesian framework to estimate and forecast cause-specific mortality rates for multiple countries taking into account of incomplete data. Each recorded death requires a medical certificate of cause of death to be completed, which is then coded using the International Statistical Classification of Diseases and Related Health Problems (ICD) maintained by the World Health Organisation (WHO). Comparability of cause-specific mortality rates requires these codes to be constant across time - ICD codes have, however, changed roughly every decade. The use of death registration data from a range of countries in addition to the records for England and Wales will allow us to borrow information on mortality trends and correct for inconsistencies in the way causes of death were recorded at various points in time. A missing data framework will be incorporated into the Bayesian model specifications to account for the effects of changing ICD codes as well as different coding practices that have significantly affected deaths due to respiratory diseases. Bayesian methods will also enable us to incorporate prior information (e.g. trends in the quality of healthcare, expert opinion) and uncertainty in a natural manner. To estimate and forecast cause specific mortality rates, we will develop a Lee Carter type model in the presence of competing risks. Generalised additive models can be used to investigate how cause specific mortality rates vary smoothly with parameters such as age and cohort. The predictive model for the cause specific mortality developed during this project could be of interest to numerous stakeholders, including governments for social and health care planning; public and private pension schemes for use in statutory funding valuations; and the life insurance industry to support accurate reserving and pricing of their products.