OBJECTIVES: Treatment sequencing models are vital in health technology assessments (HTAs) since they simulate the progression of patients through multiple therapy lines, mirroring real-world treatment pathways. These models evaluate the entire treatment pathway and support policy guidelines that optimize cost-effectiveness. However, time-dependency challenges arise in sequencing modelling with three plus therapeutic lines using common semi-Markov structures. Patients simultaneously enter and exit intermediate lines, making modelling time-dependent events like treatment up-titration, stopping rules or time-specific utilities an approximation without the extensive use of tunnel states. This study examines the impact of approximating time-dependent events on an intermediate treatment line in a semi-Markov sequencing model.

METHODS: An ‘approximation’ and an ‘exact’ three-line sequencing modelling simulations were compared to track a 4-state semi-Markov model with ‘good’, ‘moderate’, ‘poor’ and ‘death’ states over a 3-years horizon. The base-case simulation modelled a time-dependent stopping rule for patients in the ‘poor’ state after 8 therapy cycles in the first and second lines. The exact simulation used tunnel states to precisely model every cycle for each health state whereas the approximation method followed patients using OFFSET to track the proportion of patients from therapy-line entry to the stopping rule.

RESULTS: No differences were observed in the first line. The two methods estimated 0.04(1.5%) life-years total difference for time resided in the second and third lines. The relative differences in second-line health state occupancy for good, moderate, poor, and death states between methods were 37%, 53%, 11%, and 3%, respectively. There was no difference in mortality. The approximation and exact methods required 1,131 and 7,607 active cells, respectively.

CONCLUSIONS: Over three years, the ‘approximation’ model significantly misassigned life-years relative to the ‘exact’ model. This imprecision could lead to the misinterpretation of cost-effectiveness. A patient-level simulation model is recommended to overcome the challenge of time-dependent events when sequence modeling three plus treatments.