OBJECTIVES: We investigated the cost effectiveness of maribavir for post-transplant cytomegalovirus (CMV) infection in solid organ and hematopoietic stem cell transplant (SOT/HSCT) patients refractory (with or without resistance [R/R]) to conventional therapies. Initial model results lacked face validity. This study evaluated the use of real-world data (RWD) to enhance model design and outcomes.

METHODS: The cost-effectiveness model (CEM) comprised two consecutive Markov models. A three-state CMV Markov was used for 12 months, capturing transitions between clinically significant CMV (cs-CMV), non-clinically significant CMV (n-csCMV), and dead health states. From 12 months until the lifetime horizon, patients moved to a two-state, alive/dead model. Transition probabilities remained constant in the first 12-months and were informed by the SOLSTICE trial (NCT02931539). Based on expert review of the model, the number of patients occupying the cs-CMV health state at 12 months was considered implausibly high. RWD from a multinational retrospective CMV outcomes study were used to inform enhancements to the CEM structure.

RESULTS: RWD demonstrated that HSCT and SOT recipients may experience recurrences 82 and 155 weeks, respectively, following the first R/R CMV episode, justifying an extended CMV Markov duration of 18 months. RWD revealed the risk of recurrence diminished with time since clearance. Tunnel states and time-dependent transition probabilities were incorporated, capturing the relationship between time since clearance and recurrence. These structural modifications reduced the number of patients occupying the csCMV health state at the end of the CMV Markov from 42% to 10% in the maribavir arm and 39% to 13% in the investigator-assigned treatment arm. The new approach using RWD resulted in a reduction in the ICER from £65,884 to £20,163.

CONCLUSIONS: Face validity is a crucial component of model validation. RWD provide developers with an important frame of reference and an opportunity to reduce model uncertainty.