Modelling cause-of-death mortality

Abstract

The prime objective of this thesis is to investigate recent trends in cause-specific mortality in England and Wales. The recent slowdown in mortality rate improvements is of particular interest. Various modelling processes for the cause-specific mortality are explored so as to determine the appropriate models for this purpose. Three main datasets are used for the purpose of modelling the cause-specific mortality trends in England and Wales. An initial exploratory data analysis is performed on each dataset in order to investigate any obvious trends in the mortality experience. Following the exploratory data analysis of the cause-specific mortality data, Lee-Carter type models are compared against generalised linear models based on the cause-specific data from 1950 to 2018. These models are applied to cancers, circulatory system diseases, external causes of mortality, and infectious diseases in order to assess their ability and appropriateness for modelling the trends in certain causes of death over a longer period of time. Generalised linear models under the Poisson and negative binomial distributions are compared to the Lee-Carter models with the same distributional assumptions. The generalised linear models under the negative binomial distribution are found to have better performance when modelling the chosen causes of death and are used to model the cause-specific mortality trends in the subsequent works. This thesis provides enhanced insight into the quantification of cause-specific contributions to the slowdown in all-cause mortality rate improvements in England and Wales for the period from 2001 to 2018. The causes considered herein include cancers, circulatory system diseases, digestive system diseases, endocrine system diseases, external causes of mortality, genitourinary system diseases, infectious diseases, mental and behavioural illnesses, musculoskeletal system diseases, nervous system diseases, respiratory system diseases, and deaths from all other causes. Breakpoint analysis techniques are used for the purpose of determining the location at which mortality improvements begin to slow down. Cause-specific contributions to the slowdown in improvements are determined by comparing improvements in various scenarios. This work is also extended to include future projections of cause-specific mortality scenarios. The corresponding results provide insights into causes of death that are drivers of life expectancy improvements. Finally, generalised linear models are used to model the cause-specific mortality experience in England from 2001 to 2016 when additional information about the geographical region as well as the income deprivation decile for the considered population is included. Following the work performed on cause-specific contributions to mortality improvement slowdowns, knowledge about breakpoints in cause-specific mortality trends is applied in the resulting models. The results of these analyses enable a better understanding of cause-specific mortality trends in England in relation to income deprivation deciles and geographical regions. The research in this thesis makes valuable and applicable advancements in the field of cause-of-death mortality. The findings are of benefit to researchers, policy-makers and insurance professionals