Use Of External Data To Guide Long-Term Survival Extrapolations Of Trial Data For Chronic Lymphocytic Leukemia.

Presented at: ISPOR 17th Annual European Congress, November 8-12, 2014, Amsterdam, The Netherlands • Our example analysis extrapolated overall survival data from the COMPLEMENT-1 trial4 • Double-blind, multicenter clinical trial of first-line treatment for chronic lymphocytic leukemia (CLL) • Patients randomised to ofatumumab + chlorambucil (OCHL; n=221) or chlorambucil (CHL; n=226) and followed for a median 29 months • The survival rate was >70% at 4 years in both arms • Patient-level data from both trial arms were used in the present study • The external long-term follow-up study used to guide the extrapolation was the C9011 trial5 • Patients randomised to CHL, fludarabine, or fludarabine + CHL for first-line treatment of CLL • Overall survival was reported for a period of about 18 years • The CHL arm (n=193) of trial C9011 was used for the present study to guide the extrapolation of the CHL arm of the COMPLEMENT-1 trial • The data could not be used directly as the overall survival rate over the first 4 years was markedly lower than observed for CHL in COMPLEMENT-1 • The disparity could be due to differences in patient population characteristics, the CHL dosing regimen, and/or general improvements in outcomes for cancer patients since the C9011 study cohort was enrolled (1990-1994) • The published Kaplan-Meier curve was digitised, and the graphical coordinates, together with the numbers at risk, were used to estimate a patient-level data set by the method of Guyot et al (2012)6 • Long-term extrapolation for the overall survival in the COMPLEMENT-1 trial, guided by the C9011 data, was estimated using a staged approach • Stage 1: Parametric survival functions were fitted for the CHL arm in both studies with an indicator for study • Stage 2: The average treatment effect of OCHL vs CHL was estimated by survival analysis including data for both arms of COMPLEMENT-1 and the C9011 CHL arm, and including indicators for treatment and study. The coefficient for treatment from this model, which considered the shape of long-term survival for the CHL arm but allowed for study differences, provided an estimate of the treatment effect • Stage 3: Long-term survival associated with OCHL was estimated by applying the treatment effect from Stage 2 to the estimated survival for CHL from Stage 1 (using the same type of function) • Several different functions were considered • Exponential • Weibull • Log-logistic • Log-normal • Gompertz • Gamma • The fits of the parametric survival functions to the data were assessed graphically and by comparison of Akaike information criterion (AIC) and Bayesian information criterion (BIC) statistics