OBJECTIVES

State transition models (STM) are commonly used in medical decision making. Transitions between health states (in subsequent time periods) in such models are determined by transition probabilities (TPs). In estimating TPs, analysts face many data related issues, e.g. data may be missing for specific transitions, data may need to be extrapolated, and data from multiple sources may need to be combined. Due to these issues, TPs may be estimated inappropriately, introducing bias into the STM; such bias may lead to erroneous cost-effectiveness results and inappropriate adoption decisions. There is currently little guidance on how to address common issues associated with TP estimation.

METHODS

A review of NICE Technology Appraisals (TAs) was conducted to understand how TPs were estimated in recent models, and to identify issues that commonly arise during TP estimation. TAs from 1st January 2019 to 27th May 2020 were included if they conducted cost-effectiveness analysis using STMs or multi-state models.

RESULTS

Twenty-eight models were included in the review. Several methods for estimating TPs were identified: survival analysis (n=11); count method (n=8); multi-state modelling (n=7); logistic regression (n=2); negative binomial regression (n=1); simulation (n= 1); calibration (n= 1), and expert elicitation (n=1). Evidence Review Groups found several issues with how TPs were estimated, including: important transitions excluded from the model (n=5); potential selection bias when estimating TPs for post-randomisation health states (n=2); issues associated with the use of multiple sources of data (n=4); potential biases due to mis-matching populations (n =2), and assumptions around extrapolation (n =3). These issues were not resolved in almost every instance.

CONCLUSIONS

Issues associated with TP estimation are common. Failing to address these issues may bias model results and lead to sub-optimal decisions. Further research is required to address these methodological problems.