Sequential drug decision problems in long-term medical conditions: a case Study of Primary Hypertension Eunju Kim ba, ma, msc

1.5Thesis structure

The rest of the thesis is organised as follows. In the following chapter, a mathematical description of SDDPs is developed to gain an understanding of the nature of SDDPs. As the mathematical definition of SDDPs can be structured in different ways depending on the structure of underlying evaluation model, key cost-effectiveness modelling techniques are introduced first, followed by their advantages and disadvantages in using for the underlying evaluation model of SDDPs. Other issues, which need to be considered to structure SDDPs, are also discussed with the comparison between SDDPs and representative combinatorial optimisation problems. Potential sources of computational complexity of SDDPs are explained based on the mathematical description of SDDPs.

In Chapter 3, a systematic review is presented to identify approximate optimisation methods to solve an SDDP. The theoretical background and methodological application of the potential optimisation methods, which were identified in the systematic review, are explained. A simplified SDDP case-study is used to test the feasibility of the selected methods within the context of SDDPs.

Chapter 4 is the overview of modelling an SDDP for primary hypertension. Hypertension and the pharmacologic management are explained based on the clinical guidelines and textbooks. A literature review on previous cost-effectiveness modelling studies in primary hypertension is summarised. A conceptual framework of the hypertension SDDP model is detailed based on the elements used for the mathematical definition of SDDPs.

The hypertension SDDP model consists of two closely linked models - an evaluation model and an optimisation model. Chapter 5 describes the population characteristics, model structure, data used and key assumptions to populate the underlying evaluation model. Chapter 6 explains how the suggested optimisation methods - enumeration, SA, genetic algorithm (GA) and reinforcement learning (RL) – works with the underlying evaluation model using the pseudo-codes.

Chapter 7 presents how the hypertension SDDP model was validated internally and externally. The experimental results obtained from different approximate optimisation methods in various settings of key parameters are investigated regarding the quality of solution and computational efficiency.

Chapter 8 summarises what this study achieved and discusses the applicability of the proposed methods. This chapter also lists the limitations of the hypertension SDDP model and suggests the direction of future research.

Chapter 9 is the conclusion of this study.