Approximation of Long-Term Survival with Polatuzumab Vedotin Plus Bendamustine and Rituximab for Patients with...

To inform healthcare professionals, payers and health technology organisations of estimated survival benefits of new treatments, statistical methods can be used to model the projected clinical benefits versus costs of new interventions. This is particularly relevant for new treatments where data describing all progression events are incomplete and long-term survival outcomes are not yet established. In patients with the fast-growing B-cell cancer, diffuse large B-cell lymphoma (DLBCL), heterogeneous clinical efficacy outcomes are observed with the presence of both 'cured' (long-term survivors [LTS]) and 'non-cured' patients. Mixture cure rate models represent an alternative approach to traditional standard parametric survival models as they capture this heterogeneity. The aim of this analysis was to use progression-free survival (PFS) as an intermediate endpoint to estimate long-term survival with polatuzumab vedotin (Pola) + bendamustine (B) + rituximab (R) treatment (Pola+BR); these survival estimates will be utilised to inform future economic analyses.

Using data from the Phase II randomised cohort of the GO29365 trial (NCT02257567), we estimated the overall survival (OS) benefit and proportion of LTS with Pola+BR versus BR alone in patients with relapsed/refractory (R/R) DLBCL. Alongside standard parametric survival models, a mixture cure rate model was evaluated for each treatment arm, exploring both OS and OS informed by PFS.

The estimated mean OS was 3.78 years for Pola+BR versus 1.07 years for BR using standard parametric methods and 4.00 years versus 1.02 years using a mixture cure rate model (OS informed by PFS). The proportion of LTS using the mixture cure rate model was 23.0% (95% confidence interval: 9.3, 45.36) for Pola+BR versus 0% for BR (assuming a generalised gamma distribution). Of the extrapolation methods tested, mixture cure rate model predictions were best aligned with the observed survival data in GO29365.

These models suggest that compared with BR alone, Pola+BR is associated with a higher proportion of LTS ranging from 22.0 to 26.6%, depending on the distribution assumed. However, the upper and lower limits of the confidence intervals of the point estimates are reaching from 9 to 45%.