Presentation (CTI)

OBJECTIVES: Algorithms for the digitisation of survival curves by Hoyle et al. in 2011 and Guyot et al. in 2012 have helped make health technology assessment of cancer treatments one of the most published types of economic evaluations with more than 16,000 peer-reviewed papers published over the last 15 years. Unfortunately, these types of analyses are complex and it can be a challenge for non-technical reviewers to assess whether the data is correct. This case study aims to present a discrepancy arising from what appears to be a misalignment of the initial coordination of the digitisation algorithm.
METHODS: The overall survival (OS) curves from a pivotal trial for a next wave PD1-inhibitor called serplulimab were reconstructed from the pivotal phase III trial (Astrum-005) and compared to a recent published cost-effectiveness analysis by Zheng et al. 2024 (Zheng). Survival analysis examining the difference between the original Astrum-005 OS data and the reconstructed data in Zheng were compared.
RESULTS: The hazard ratio between the Zheng OS curve and the Astrum-005 curve was 0.974 [95% CI 0.845, 1.122]. There was a difference in the median OS reported by Zheng (not reached) vs Astrum-005 trials (15.4mth). The fitted hazard rate over time peaked at around 0.06 around 8 months for the Zheng data vs 0.8 at 11 months for the Astrum-005 data. For both datasets, AIC/BIC statistics favoured the loglogistic distribution for extrapolation. Over a 10-year time period, this resulted in a 9% overestimation of the survival when using Zheng compared to the Astrum-005.
CONCLUSIONS: It is recommended that validation of the digitisation is performed by replicating the statistics reported for the original clinical data (median survival, hazard ratio, survival at different time points etc). Moreover, it is recommended that journals enlist reviewers with appropriate expertise when considering manuscripts containing cancer modelling.