Extrapolating survival with applications to health technology assessment

<p dir="ltr">Survival extrapolation is fundamental to health economic modelling, addressing the task of projecting long-term outcomes from short-term survival data. Clinical studies often have limited follow-up periods, yet accurate projections of survival curves are essential in order to estimate restricted mean survival time, life expectancy, and quality-adjusted life years for health technology assessment (HTA). These projections allow decision-makers to compare the costs and effectiveness of new and existing interventions, helping them decide whether to adopt new healthcare technologies.</p><p dir="ltr">To evaluate the accuracy of survival extrapolation methods, we systematically compared standard parametric models (SPMs) and flexible parametric models (FPMs) in Study I. Using data from the Swedish Cancer Register, this study compared the all-cause survival framework (ASF) with the relative survival framework (RSF) to predict survival at 10-year and lifetime horizons. Results showed that FPMs generally outperformed SPMs in predicting 10-year survival. When extrapolating to a lifetime horizon, RSF models provided more accurate predictions than ASF models, especially when using FPMs. Notably, ASF models tended to overestimate survival from limited follow-up data, whereas RSF models showed a trend toward underestimation.</p><p dir="ltr">In Study II, we developed a novel multistate framework integrating relative survival ex- trapolation with mixed time scales for cost-effectiveness analysis. We demonstrated this approach using an irreversible illness-death model applied to clinical trial data, extending and comparing it with previous cost-effectiveness analyses. This framework enables the in- corporation of external long-term mortality data into survival extrapolation from short-term follow-up data within a multistate model.</p><p dir="ltr">The multistate framework was applied to model the natural history of chronic myeloid leukaemia (CML) treatment in Study III. We estimated life expectancy and quality-adjusted life expectancy for patients with chronic-phase CML (CP-CML) diagnosed in Sweden between 2007 and 2017, comparing these with the general population. Results showed proportional losses in life expectancy of 9-15% and in quality-adjusted life expectancy of 29-33% across age-sex strata, indicating modestly low losses in life expectancy but greater losses in quality- adjusted life expectancy among patients with CP-CML.</p><p dir="ltr">The natural history model from Study III was extended to a cost-of-illness study through estimation and projection of the economic burden for CML in Sweden from 2015 to 2030 in Study IV. Results showed that prevalent cases are estimated to nearly double from 1205 (95% CI: 1014-1397) in 2015 to 2120 (95% CI: 1916-2325) by 2030, total direct healthcare expenditures declined from USD 40.04 million (95% CI: 33.70-46.40) to USD 30.67 million (95% CI: 27.70-33.64). This trend indicates that decreasing treatment costs may mitigate the economic burden on the Swedish healthcare system.</p><p dir="ltr">In conclusion, in this thesis I have worked collaboratively to systematically evaluate and advance survival extrapolation methods for HTA. A key methodological innovation is the integration of relative survival extrapolation with multistate modelling, combining short-term survival data with long-term external information. Together with collaborators, I applied these methods in two CML studies to demonstrate how such advanced statistical modelling can provide practical insights to support clinical understanding and inform health policy.</p><h3>List of scientific papers</h3><p dir="ltr">This thesis is based on the following scientific articles, which are referred to by Roman numerals throughout and presented in full at the end.</p><p dir="ltr">I. <b>Enoch Yi-Tung Chen</b>, Yuliya Leontyeva, Chia-Ni Lin, Jung-Der Wang, Mark S. Clements, and Paul W. Dickman. Comparing Survival Extrapolation within All-Cause and Relative Survival Frameworks by Standard Parametric Models and Flexible Parametric Spline Models Using the Swedish Cancer Registry. Medical Decision Making. 2024;44(3):269-282.<br><a href="https://doi.org/10.1177/0272989x241227230" rel="noreferrer" target="_blank">https://doi.org/10.1177/0272989x241227230<br><br></a></p><p dir="ltr">II. <b>Enoch Yi-Tung Chen</b>, Paul W. Dickman, and Mark S. Clements. A Multistate Model Incorporating Relative Survival Extrapolation and Mixed Time Scales for Health Technology Assessment. PharmacoEconomics. 2025;43:297-310.<br><a href="https://doi.org/10.1007/s40273-024-01457-w" rel="noreferrer" target="_blank">https://doi.org/10.1007/s40273-024-01457-w<br><br></a></p><p dir="ltr">III. <b>Enoch Yi-Tung Chen</b>, Torsten Dahlén, Leif Stenke, Magnus Björkholm, Shuang Hao, Paul W. Dickman, and Mark S. Clements. Loss in Overall and Quality-Adjusted Life Expectancy for Patients with Chronic-Phase Chronic Myeloid Leukemia. European Journal of Haematology. 2025;114(2):334-342.<br><a href="https://doi.org/10.1111/ejh.14328" rel="noreferrer" target="_blank">https://doi.org/10.1111/ejh.14328<br><br></a></p><p dir="ltr">IV. <b>Enoch Yi-Tung Chen</b>, Paul W. Dickman, Fabrizio Di Mari, Torsten Dahlén, Leif Stenke, Magnus Björkholm, Mark S. Clements, and Shuang Hao. Empirical and Projected Economic Burden of Chronic Myeloid Leukaemia in Sweden from 2015 to 2030: a Population-Based Study. [Submitted]</p>