Exploring statistical models for estimating remaining life years & loss in life expectancy for cancer patients

As advancements in cancer prognosis and treatment continue for many cancer types, there is a growing interest in estimating survival over longer time periods. The assessment of the years remaining after a cancer diagnosis, or the life expectancy following a cancer diagnosis (LEC) and the years lost due to cancer or Loss in Life Expectancy (LLE) has become an appealing quality providing insights into the enduring impact of cancer on an individual’s life beyond diagnosis. These estimates contribute to both epidemiological and clinically important research questions in the cancer survival area. Despite the appeal of LEC and LLE, they are not commonly used, mainly due to the complexity of their estimations. This PhD project aimed to enhance statistical models and methods for estimating LEC and LLE for cancer patients, utilising registry-based data in a relative survival setting. Within this context, general population mortality rates serve as expectations for the mortality rates in the cancer-free population. They are derived from external mortality tables, often treated as known (fixed) and measured without uncertainty. Study I evaluated existing approaches for estimating standard errors of 5-year relative survival, the most often used metrics in the cancer survival field, and LLE, revealing situations where uncertainty in expected mortality must be considered. In Study II, a methodology incorporating uncertainty from expected measures while estimating LEC and LLE was proposed, using existing Stata software. The research project delved into the complexity of extrapolation in estimating LEC and LLE, employing flexible parametric relative survival models. Study III focused on estimating LEC and LLE for individuals diagnosed with MyeloProliferative Neoplasms (MPN), revealing a decreased life expectancy in individuals with MPN relative to the general population. Study IV explored geographical patterns of LEC and LLE in Queensland, Australia, using a spatial flexible parametric relative survival model within a Bayesian ...