Modelling Techniques in Economic Evaluation of Bleeding Disorders: A Comparative Overview

Inherited bleeding disorders (iBDs), such as haemophilia A and B, von Willebrand disease (VWD), and Glanzmann thrombasthenia, present unique challenges for health technology assessment (HTA). Treatments are often lifelong, costly, and evolving rapidly, especially with the advent of gene therapies and non-factor replacement options like emicizumab. Economic evaluations (EEs) are essential to guide reimbursement and policy decisions, and modelling plays a central role in these assessments.

This article explores the main modelling techniques used in economic evaluations of bleeding disorders, drawing on the systematic review by Prameyllawati et al. (2025), which analysed 47 model-based EEs, and other key literature.

Markov Models

Markov models simulate disease progression through defined health states over time, typically using cycles (e.g., yearly). They are well-suited for chronic conditions with recurring events. Markov models were the most commonly used technique (57% of studies) in Prameyllawati et al. (2025), especially for severe haemophilia A and B. These models included recurrent bleeding events, joint damage, and mortality over a lifetime horizon.

Markov models are ideally suited to handling long-term disease progression, and are therefore suitable for chronic conditions with recurring events, incorporating quality-adjusted life years (QALYs) into their assessments. However, this modelling type may oversimplify complex pathways if health states are too few and requires robust transition probabilities and utility data. Overall, Markov models are best suited to evaluating long-term treatment options such as long-term prophylaxis (e.g., factor concentrates, emicizumab) and gene therapies.

Decision Trees

Decision trees model short-term outcomes by mapping decisions and chance events in a branching structure. They are simple and intuitive. This technique was used in 19% of studies in Prameyllawati et al. (2025), primarily for short-term interventions like bypassing agents (BPAs) or immune tolerance induction (ITI).

Decision trees are easy to construct and interpret, and ideal for short-term, one-off decisions. However, they are less well suited to chronic conditions or recurrent events due to their limited ability to model long-term outcomes. Decision tree modelling is best suited to acute treatments with short time horizons, such as managing bleeding episodes in patients with inhibitors.

Microsimulation and Individual-Level Models

These models simulate individual patient pathways, allowing for heterogeneity in characteristics and treatment responses. This technique was used in 11% of studies in Prameyllawati et al. (2025), especially for gene therapies and pharmacokinetic-guided dosing strategies.

These techniques can be beneficial as they capture individual variability, making them suitable for personalised medicine and complex dosing. However, they are data-intensive and computationally demanding, requiring detailed individual-level data. Microsimulation and individual-level models are best suited to evaluating gene therapies (e.g., valoctocogene roxaparvovec, etranacogene dezaparvovec) and personalised prophylaxis.

Markov Decision Processes and Hybrid Models

These combine Markov models with decision trees or incorporate probabilistic decision-making over time. These techniques are used in a minority of studies to model complex treatment strategies, such as ITI protocols or switching therapies.

They offer a flexible structure for complex decisions, with the ability to model both short-term and long-term outcomes. However, they are complex to design and validate and are harder to interpret, and as such are less commonly used. They are best suited to evaluating multi-step treatment strategies or mixed populations.

Key Considerations for Model Selection

In choosing an appropriate model type, there are a number of aspects to consider:

1. Time Horizon: Most studies used a lifetime horizon, appropriate for chronic conditions like haemophilia.

2. Cycle Length: One-year cycles were common, though shorter cycles (e.g., weekly) were used for gene therapy and acute treatments.

3. Health States: Bleeding events, especially joint bleeds, were central to most models. Some included joint damage, surgery, or Pettersson scores.

4. Clinical Outcomes: QALYs were the most reported outcome, often derived from literature. Few studies collected primary utility data.

Overall, Markov models are ideal for chronic, recurrent conditions and decision trees for short-term interventions. However, for personalized therapies, microsimulation can be considered. Model choice should be justified based on disease characteristics and treatment complexity.

Conclusion:

Modelling techniques in economic evaluation must be tailored to the nature of bleeding disorders and the interventions being assessed. While Markov models dominate, individual-level approaches are gaining relevance with the rise of personalised medicine. Transparent, well-justified modelling is essential to support HTA and ensure equitable access to innovative therapies.

References:

- Prameyllawati DM, Lingsma HF, Cnossen MH, ten Ham RMT. A Systematic Review of Modelling Approaches in Economic Evaluations of Treatments for Inherited Bleeding Disorders. Appl Health Econ Health Policy. 2025. https://doi.org/10.1007/s40258-025-00996-3

- Barton P, Bryan S, Robinson S. Modelling in the economic evaluation of healthcare. J Health Serv Res Policy. 2004;9(2):110–8.

- Briggs A et al. Decision Modelling for Health Economic Evaluation. Oxford University Press; 2006.