Is Less More? Aggregating Health States in State Transition Models
Author(s)
Jack Ettinger, MSc1, Johan Maervoet, MBA, PharmD, PhD2.
1Health Economics Associate, PAREXEL International - Access Consulting, London, United Kingdom, 2PAREXEL, Wavre, Belgium.
1Health Economics Associate, PAREXEL International - Access Consulting, London, United Kingdom, 2PAREXEL, Wavre, Belgium.
OBJECTIVES: State transition models (STMs) are commonly used for cost-effectiveness modelling. STMs are structured around a set of mutually exclusive and collectively exhaustive health states capturing the key features of the disease and treatment. Modelling guidelines suggest finding a balance between simplicity and complexity in modeling health states. We aimed to investigate the impact of aggregating health states on modeling outcomes in cardiovascular disease (CVD).
METHODS: Three CVD STMs were built and run over a 10-year time horizon. The first had 10 health states: healthy, ischaemic stroke (IS), haemorrhagic stroke (HS), hospitalized heart failure (HF), non-hospitalised HF, post-IS, post-HS, post-hospitalized HF, post-non-hospitalised HF, and dead. The event health states were tunnel states. The second had six health states: grouping the two stroke and two HF-type states. The last had four health states, grouping strokes and HFs together. A synthetic dataset of 10,000 UK adult patients mirroring routinely collected primary care data on heart attacks and strokes over 10 years was used to derive transition probabilities assuming a linear effect. One-off costs for treatment of acute events were randomly generated from gamma distributions. Post-event costs, health state utility values (HSUVs), and mortality rates were given fixed, plausible values. In the six and four health state STMs, event costs were averaged. Weighted averages were calculated for HSUVs and mortality rates. The model, reporting, and all analyses were programmed in R.
RESULTS: The undiscounted costs and QALYs produced by each of the three models were very similar: £5,674.83 and 9.10 QALYs for the 10-health state STM, £5,674.79 and 9.10 QALYs for the six-health state STM, and £5,674.03 and 9.10 QALYs for the four-health state STM.
CONCLUSIONS: This research supports the argument that having a good understanding of underlying data might be more important than developing more disaggregated model structures to produce accurate results.
METHODS: Three CVD STMs were built and run over a 10-year time horizon. The first had 10 health states: healthy, ischaemic stroke (IS), haemorrhagic stroke (HS), hospitalized heart failure (HF), non-hospitalised HF, post-IS, post-HS, post-hospitalized HF, post-non-hospitalised HF, and dead. The event health states were tunnel states. The second had six health states: grouping the two stroke and two HF-type states. The last had four health states, grouping strokes and HFs together. A synthetic dataset of 10,000 UK adult patients mirroring routinely collected primary care data on heart attacks and strokes over 10 years was used to derive transition probabilities assuming a linear effect. One-off costs for treatment of acute events were randomly generated from gamma distributions. Post-event costs, health state utility values (HSUVs), and mortality rates were given fixed, plausible values. In the six and four health state STMs, event costs were averaged. Weighted averages were calculated for HSUVs and mortality rates. The model, reporting, and all analyses were programmed in R.
RESULTS: The undiscounted costs and QALYs produced by each of the three models were very similar: £5,674.83 and 9.10 QALYs for the 10-health state STM, £5,674.79 and 9.10 QALYs for the six-health state STM, and £5,674.03 and 9.10 QALYs for the four-health state STM.
CONCLUSIONS: This research supports the argument that having a good understanding of underlying data might be more important than developing more disaggregated model structures to produce accurate results.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
EE562
Topic
Economic Evaluation, Methodological & Statistical Research
Disease
Cardiovascular Disorders (including MI, Stroke, Circulatory)