Assessment of the consistency and robustness of results from a multicenter trial of remission maintenance therapy for acute myeloid leukemia

Abstract

Background: Data from a randomized multinational phase 3 trial of 320 adults with acute myeloid leukemia (AML) demonstrated that maintenance therapy with 3-week cycles of histamine dihydrochloride plus low-dose interleukin-2 (HDC/IL-2) for up to 18 months significantly improved leukemia-free survival (LFS) but lacked power to detect an overall survival (OS) difference.

Purpose: To assess the consistency of treatment benefit across patient subsets and the robustness of data with respect to trial centers and endpoints.

Methods: Forest plots were constructed with hazard ratios (HRs) of HDC/IL-2 treatment effects versus no treatment (control) for prospectively defined patient subsets. Inconsistency coefficients (I²) and interaction tests (X²) were used to detect any differences in benefit among subsets. Robustness of results to the elimination of individual study centers was performed using "leave-one-center-out" analyses. Associations between treatment effects on the endpoints were evaluated using weighted linear regression between HRs for LFS and OS estimated within countries.

Results: The benefit of HDC/IL-2 over controls was statistically consistent across all subsets defined by baseline prognostic variables. I² and P-values of X² ranged from 0.00 to 0.51 and 0.14 to 0.91, respectively. Treatment effects were statistically significant in 14 of 28 subsets analyzed. The "leave-one-center-out" analysis confirmed that no single center dominated (P-values ranged from 0.004 to 0.020 [mean 0.009]). The HRs representing the HDC/IL-2 effects on LFS and OS were strongly correlated at the country level (R² = 0.84).

Limitations: Small sample sizes in some of the subsets analyzed.

Conclusions: These analyses confirm the consistency and robustness of the HDC/IL-2 effect as compared with no treatment. LFS may be an acceptable surrogate for OS in future AML trials. Analyses of consistency and robustness may aid interpretation of data from multicenter trials, especially in populations with rare diseases, when the size of randomized clinical trials is limited.

Trial registration: ClinicalTrials.gov: NCT00003991.

Figures

Figure 1

Forest plots of leukemia-free survival (LFS) hazard ratios (HR) and their confidence intervals (CI) by baseline characteristics. O/N = event rate per arm where O is the number of observed events (relapse or death) and N is the sample size. HR = hazard ratio, CI = confidence interval, CR = complete remission, CR1 = first complete remission, WBC = white blood cell, SWOG = Southwest Oncology Group, AML = acute myeloid leukemia, HiDAC = high-dose cytosine arabinoside.

Figure 2

Forest plots of leukemia-free survival (LFS) hazard ratios (HR) and their confidence intervals (CI) by country. O/N = event rate per arm where O is the number of observed events (relapse or death) and N is the sample size.

Figure 3

Distribution of P-values for the treatment effect on leukemia-free survival in a "leave-one-center-out" cross-validation. N = sample size.

Figure 4

Correlation between treatment effects on leukemia-free survival and overall survival (R2 = coefficient of determination). The size of each circle is proportional to the number of patients in the corresponding country.