OBJECTIVES : Partitioned survival modelling (PSM) and semi-Markov modelling (SMM) are commonly used in cost-effectiveness (CE) analyses of oncology treatments. However, survival estimates from the two approaches may differ, a key source of uncertainty which is not always explored when developing a model.

The objective is to estimate long-term survival for 1^st line extensive stage small cell lung cancer treatment with durvalumab plus platinum–etoposide (D+EP) and platinum–etoposide (EP) using PSM and SMM.

METHODS : PSM and SMM models were estimated from parametric overall survival (OS), progression-free survival (PFS), and time-to-progression curves derived from the CASPIAN trial’s patient-level data and selected using goodness-of-fit statistics. The models consist of three health states: progression-free, progressed, and death. PSM health state occupancy is based on the area under the OS and PFS curves. SMM health state occupancy is based on patient movement between health states, estimated using transition probabilities derived from the PFS, OS and time-to-progression curves.

RESULTS : Median OS estimates in the D+EP and EP groups respectively were: 13.0 and 10.3 months (CASPIAN); 12.7 and 10.6 months (PSM), and 12.2 and 10.2 months (SMM). Estimated PSM and SMM based mean survival differences for D+EP vs EP over 5 and 15 year horizons respectively were 3.8 and 6.0 months (PSM) and 5.1 and 10.0 months (SMM).

CONCLUSIONS : Estimated survival differences for D+EP vs EP using PSM and SMM were small but would likely impact a CE assessment. This study highlights the importance of assessing structural assumptions and uncertainty in model development.