Development and Bayesian calibration of a microsimulation model for lung cancer: natural history, screening and treatment
Lung cancer is one of the most common cancers and most deadly cancer in New Zealand and can affect anyone. In order to reduce the mortality rate from lung cancer and increase the survival rate of patients with lung cancer, early detection through screening is crucial. However, no national lung cancer screening program has been established in New Zealand to date. Consequently, microsimulation studies are an alternative method of analyzing the performance of screening processes at an individual level.
This thesis presents research on statistical methods for the development and evaluation of microsimulation models (MSM) for the natural history, screening, and treatment of lung cancer in New Zealand (NZ). Initially, we adapted a continuous-time MSM to describe the natural history of lung cancer and used it as a tool for the implementation of the calibration process and assessment of predictive accuracy. We performed a Bayesian calibration through the Hamiltonian Markov Chain (HMC) Monte Carlo approach. Bayesian calibration employs Bayesian reasoning to incorporate prior beliefs on model parameters, and information from various sources about lung cancer, to derive posterior distributions for the calibrated parameters. We then used the calibrated parameters to validate the natural history model based on the lung cancer data collected from the Ministry of Health, New Zealand. Furthermore, we studied the survival rate of lung cancer in NZ and evaluated how cancer screening affects mortality reduction. According to our knowledge, no research studies have been conducted in New Zealand on the development of microsimulation models for lung cancer. Our primary objective was to develop a microsimulation lung cancer screening model to analyze the mortality associated with lung cancer on an individual basis. Also, simulation algorithms were developed to simulate data and hence compare the performance of screening in terms of early detection. The entire methodology was implemented in R.4.0.4. Then, we considered a microsimulation model development with a treatment component to analyze the effect of treatment on individual data based on early detection through screening. Two treatment options, chemotherapy and gefitinib were considered in this study.
This thesis work contributes to a better understanding of the feasibility of developing microsimulation models to predict lung cancer survival in New Zealand, as well as a detailed analysis of the importance of screening at the individual level. We also identified several limitations and issues with model calibration due to the lack of data available in lung cancer data records in New Zealand.