OBJECTIVES: Survival data are common outcomes oncology indications. Researchers often fit parametric survival functions to reported Kaplan-Meier (KM) curves to extrapolate estimates of survival beyond the duration of the study. Additionally, both overall survival (OS) and progression-free survival (PFS) are often needed, but clinical trials may report only PFS due to limited durations of follow up and relatively fewer OS events. The Weibull is a common survival function that can analyze OS and PFS data simultaneously, if the hazards of each are assumed to be proportional, allowing PFS to be a surrogate measure for OS. This analysis evaluates these two assumptions for 9 blood cancer indications: Weibull survival distribution and OS/PFS hazard proportionality.

METHODS: Clinical trials reporting OS and PFS KM curves were identified for first-line (L1) and relapsed/refractory (R/R) follicular lymphoma (FL), diffuse large B-cell lymphoma (DLBCL), acute myeloid leukemia (AML), multiple myeloma (MM), and R/R chronic lymphocytic leukemia (CLL). KM curves were digitized and ln(time) vs ln(cumulative Hazard) plots were created to graphically evaluate the Weibull PH assumptions. If the lnTime-lnHaz plots are linear, then the Weibull distribution is appropriate. If the plots of OS and PFS are parallel, then the proportional hazard assumption is appropriate.

RESULTS: 44 studies examining 66 treatments were evaluated. Visual inspection of the lnTime-lnHaz plots indicated that both assumptions (Weibull and PH) held well for the majority of L1 MM (n=13), L1 DLBCL (n=3), R/R CLL (n=16), R/R MM (n=11), R/R FL (n=3) studies. Among the 3 R/R DLBCL studies, one study did not exhibit linearity. No studies of L1 FL (n=3), L1 AML (n=10) or R/R AML (n=4) met the criteria for the Weibull distribution.

CONCLUSIONS: This evaluation suggests appropriateness of the Weibull and OS and PFS hazard proportionality assumptions for analysis of survival data in several blood cancers.