PF389 PROGRESSION‐FREE SURVIVAL PREDICTS OVERALL SURVIVAL IN FRONTLINE CLL

Background: Many phase II and III trials in oncology utilise the surrogate endpoint progression‐free survival (PFS) as the primary endpoint; however, the relationship between a surrogate endpoint and true clinical benefit in terms of overall survival (OS) or quality of life must be demonstrated. Aims: This research aimed to explore and quantify the relationship between PFS and OS in frontline CLL using two independent modelling approaches. Methods: A systematic literature search was conducted March 2017 to identify data on efficacy and safety outcomes of treatments for frontline CLL. PFS and OS Kaplan‐Meier curves were manually digitised using GetData Graph Digitizer (version 2.26). Potential treatment modifying variables were also extracted. Seventeen publications included data for PFS at 24 months (PFS24) and OS at 36 (OS36) and 60 months (OS60). Each treatment arm of each trial was considered as an independent observation. Pearson's correlation was used to calculate the correlation between OS36 or OS60, and treatment modifying variables and PFS24. Significantly correlated variables were then utilised in the robust ordinary least squares (OLS) regression analyses to develop models that could predict OS from PFS. Using the same data, a second analysis was performed with Cox semi‐parametric models. First patient‐level OS and PFS data were recreated using the Guyot algorithm. Next, to inform the modelling strategy, the validity of proportional hazards (PH) assumptions were checked using three methods: log log plot graphical observation, Cox model check, and Schoenfeld's residuals from the Cox model. Hazard ratios (HRs) were then derived from the models. Results: Using the seventeen identified publications, OS36 was found to be significantly correlated with elevated B microglobulin (R2 = −0.65; p = 0.064), del17p (R2 = −0.50; p = 0.016), and PFS (R2 = 0.38; p = 0.021). Similarly, OS60 was found to be significantly correlated with elevated B microglobulin (R2 = −0.71; p = 0.033), del17p (R2 = −0.48; p = 0.085), and PFS (R2 = 0.39; p = 0.062). Eight publications (16 treatment arms) included OS36, del17p, and elevated B2 microglobulin data, whilst five publications (nine treatment arms) included OS60, del17p, and B2 microglobulin data. The OLS regression models incorporating PFS24 and the additional covariates del17p and elevated B microglobulin were significant predictors of OS36 and OS60, with #R2 values of 79.7% and 90.8%, respectively. To assess regression model accuracy, PFS24, del17p, and elevated B2 data were identified from recent phase II/ III publications and entered into the models. Six of the seven OS36 regression model predictions were conservative, with differences between observed and predicted values ranging from −5.78 to 0.09%. Four of the seven OS60 regression model predictions were conservative, with differences between observed and predicted values ranging from −12.69 to 7.65%. PH assumptions were deemed acceptable in all treatment arms. HRs (OS vs. PFS) derived from the Cox models (mean: 0.25; range: 0.10 to 0.49) were significant ( p  < 0.05) in all treatment arms except one. Summary/Conclusion: The regression models and Cox model‐derived HRs can predict OS from PFS in this dataset of frontline CLL trials. These models, therefore, support the use of PFS as a surrogate endpoint for OS.