Developing Markov models from real-world data: A case study of heart failure modeling using administrative data

Objectives Markov models characterize disease progression as specific health states based on clinical or biological measures. However, these measures are not always collected outside clinical trials. In this article, an alternative approach is presented that uses real-world data to define the health states and to model transitions between them, specific to a local setting, to estimate the cost-effectiveness of telemonitoring (TM) versus no TM for heart failure. Methods The incidence of hospitalization for usual care was estimated from hospital episode statistics (HES) data in the United Kingdom and converted into a monthly transition matrix with 5 health states (4 states are defined based on the number of hospitalizations in the previous year and death) to estimate the cost-effectiveness of TM in a local UK primary care trust (PCT) using probabilistic sensitivity analysis from a healthcare perspective. Results Geographical variation in hospitalization rates were present, which led to different health state transition matrices in different localities. In the PCT that was evaluated, TM accrued mean additional costs of £3610 and 0.075 additional quality-adjusted life-years (QALYs) compared with usual care per patient, resulting in a mean incremental cost effectiveness ratio of £48 172/QALY. Conclusions The use of administrative data to define health states and transition matrices based on health service events is feasible, and TM was not cost-effective in our analysis. Given the increasing emphasis on using real-world evidence, it is likely that these approaches will be used more in the future