Presentation (CTI)

OBJECTIVES: The National Institute for Health and Care Excellence (NICE) published clinical guidelines for diabetic retinopathy (DR) and diabetic macular oedema (DMO) in 2024 (NG242) and age-related macular degeneration (AMD) in 2018 (NG82). We examined the economic models used to inform these guidelines.
METHODS: We reviewed the modelling approaches taken in NG82 and NG242, considering potential alternatives.
RESULTS: NG82 used a patient-level simulation (PLS) with a two-eye Markov health state structure and NG242 presented a cohort model using a one-eye Markov health state structure. While PLS structures modelling vision in each eye as a continuous variable have been published, previous NICE cost-effectiveness Technology Appraisals have typically also used a one- or two-eye Markov cohort structure. However, the cohort Markov state transition approach has limitations. We argue that a Markov structure introduces unnecessary complexity by grouping vision into health states. This requires transformation of reported trial and/or indirect treatment comparison endpoints into transition probabilities, often imposing artificial limits on possible transitions. Further, the number of health states required is significantly increased when modelling two eyes. Whilst considered more complex, we believe a PLS approach modelling vision in each eye as a continuous variable presents a more realistic representation of the disease whilst reducing the model complexity. Reported primary trial outcomes can be used directly in statistical analyses to estimate changes in vision on a continuous scale within the model, capturing the full benefit of treatment for the patient without manipulation. It is also easier to visualize and test alternative efficacy assumptions when considering vision on a continuous scale than within a Markov health state structure.
CONCLUSIONS: When modelling retinal conditions, a PLS based on continuous vision in each eye provides the most accurate model of the condition, whilst providing a flexible framework to test alternative assumptions.