Leukemia-free survival as a surrogate end point for overall survival in the evaluation of maintenance therapy for patients with acute myeloid leukemia in complete remission

Abstract

Background: In trials designed to evaluate new therapies for hematologic malignancies, end points such as leukemia-free survival are often used as surrogates for overall survival in acute leukemia. We aimed to assess whether leukemia-free survival is an acceptable statistical surrogate for overall survival when applied to remission maintenance therapy for acute myeloid leukemia.

Design and methods: Data were analyzed from a randomized Phase III trial of remission maintenance immunotherapy with histamine dihydrochloride plus low-dose interleukin-2 versus no treatment in adults with acute myeloid leukemia. A two-stage surrogate validation model was applied in which correlations between Kaplan-Meier estimates of leukemia-free survival and overall survival, and between log hazard ratios reflecting treatment effects were analyzed. Country of patient enrollment was the unit of analysis.

Results: Kaplan-Meier estimates of overall survival at 36, 48, and 60 months and leukemia-free survival at 24 months were reasonably correlated (R(2) ranging from 0.44 to 0.84) both for the overall (n=320) and first complete remission (n=261) populations. The effects of histamine dihydrochloride/interleukin-2 on log hazard ratios for leukemia-free survival and overall survival were well correlated (R(2)=0.88-0.93).

Conclusions: The significant correlations between overall survival and the surrogate end point (leukemia-free survival) and between the effect of histamine dihydrochloride/interleukin-2 on leukemia-free survival and overall survival satisfy the two-stage surrogate validation model.

Trial registration: ClinicalTrials.gov NCT00003991.

Figure 1.

Two-stage validation model for LFS as a surrogate for OS in a trial of HDC/IL-2 as remission maintenance therapy in AML.

Figure 2.

Kaplan-Meier (K-M) estimates of OS at 36 months versus LFS at 24 months for HDC/IL-2-treated AML patients and controls. Weighted linear regression analyses (WLRA) were performed to test Condition #1 in the surrogate validation model (see main text and Figure 1). The WLRA reflect the overall (A) and CR1 (B) populations grouped by treatment and country. Circle size is proportional to the number of patients in each country. US=United States; AU=Australia (including New Zealand); SW=Sweden; GE=Germany; FR=France; IS=Israel; CA=Canada. The WLRA equations (see main text) revealed good correlations between OS and LFS for both the overall and CR1 populations, satisfying Condition #1 of the model.

Figure 3.

Logarithms of OS hazard ratios versus logarithms of LFS hazard ratios for patients grouped by country. Weighted linear regression analyses (WLRA) were performed to test Condition #2 in the surrogate validation model (see main text and Figure 1). The WLRA reflect the overall (A) and CR1 (B) populations grouped by treatment and country. Circle size is proportional to the number of patients in each country. US = United States; AU = Australia (including New Zealand); SW = Sweden; GE = Germany; FR = France; IS = Israel; CA = Canada. The WLRA equations (see main text) revealed good correlations between the OS and LFS hazard ratios for both the overall and CR1 populations, satisfying Condition #2 of the model.

Figure 4.

“Leave-one-out” cross-validation analysis. Gray circles correspond to the predicted country-specific hazard ratios for overall survival using the observed country-specific hazard ratios for leukemia-free survival and the surrogate model built on all other countries, gray vertical lines to 95% prediction intervals and black squares to the observed country-specific hazard ratios for overall survival. US=United States; AU=Australia (including New Zealand); SW=Sweden; GE=Germany; FR=France; IS=Israel; CA=Canada.