OBJECTIVES: Patients with cancer are likely to have a higher mortality risk than the general population. For HTA purposes, survival of patients observed in clinical trials is extrapolated based on parametric distributions such as the Weibull and Loglogistic. Another option to extrapolate survival is to assume the mortality hazards of these patients are proportionally higher compared to those of the general population. This study aims to assess the impact of assuming proportional hazards (PH) versus general population mortality (GPM) in a PH GPM model.

METHODS: The PH GPM model was fitted to a dataset in multiple myeloma (MM), where the maximum follow-up was 35 months. The mean [Δ incremental] survival results of the PH GPM model were compared to corresponding results of a Weibull parametric model assuming PH (combined fit) and not assuming PH (individual fit). The extrapolations were compared to a more mature (78 months follow-up) data-cut of the MM trial.

RESULTS: Mean survival of the PH GPM model, Weibull combined fit, and Weibull individual fit were 5.7 versus 4.1 years [Δ 1.6], 8.9 and 5.7 years [Δ 3.2] and 12.4 and 4.6 years [Δ 7.8]. The visual fit of the PH GPM model on the 78 months validation data-cut was best. Both the Weibull individual and combined fits predicted hazards over time that were below the GPM hazards for the active treatment, which is not clinically realistic.

CONCLUSIONS: In case the PH assumption of disease-specific mortality in the trial compared to GPM holds, the PH GPM approach is an appropriate methodology to extrapolate survival. The advantage is that the extrapolations do not rely on a statistical distribution but on GPM. Therefore, the predicted hazards are unlikely to go below the GPM hazards. If the PH assumption compared to GPM does not hold, other extrapolation techniques are more appropriate.