Subgroup analyses are pervasive in clinical trials. In a positive trial, they may be used to characterize patients who may benefit from therapy versus those who may not. An example of a positive trial is ENZAMET1, in which patients with mHSPC receiving enzalutamide derived an overall survival benefit compared to those receiving standard of care (HR 0.67, 95% CI 0.52-0.86). Looking at the Forest plot of the subgroup analyses, Dr. Halabi highlights that in general, all subsets of patients benefited from enzalutamide, however, those with early planned docetaxel may not (HR 0.90, 95% CI 0.62-1.31), with a significant p-value for interaction (p=0.04). This suggests that there are significant differences over time between the planned early docetaxel yes or no groups.
In a negative trial, subgroup analyses may be used to identify at least some patients with a treatment benefit. An example of a negative trial is PROSTVAC2, in which patients with mCRPC were randomly assigned to PROSTVAC (Arm V; n = 432), PROSTVAC plus granulocyte-macrophage colony-stimulating factor (Arm VG; n = 432), or placebo (Arm P; n = 433). Neither active treatment had an effect on median OS (Arm V, 34.4 months; HR 1.01, 95% CI, 0.84-1.20; p = 0.47; Arm VG, 33.2 months; HR 1.02, 95% CI 0.86-1.22; p = 0.59; Arm P, 34.3 months). Looking at the Forest plot of the subgroup analyses, patients stratified by age <=71 (HR 0.74, 95% CI 0.57-0.95) appear to derive a survival benefit compared to those >71 (HR 1.38, 95% CI 1.08-1.76; p = 0.0006) who appear to derive a significant survival disadvantage. Taking these findings into appropriate clinical context is what can make interpretation of survival analysis challenging.
Dr. Halabi’s overall warning for subgroup analyses is that there can be a machine for producing false-negative and false-positive results. Specifically, the type I error rate can be very high in subgroup analyses, and appropriate power for detecting meaningful differences in these analyses can be misleading. According to Dr. Halabi, a common mistake that should be avoided is an incorrect inference that a subgroup effect is presently based on separate tests of treatment effects within each level of the characteristics of interest, that is, to compare one significant and one non-significant p-value.
Dr. Halabi lists the following criteria for assessing credibility of subgroup analyses:
- Can chance explain the apparent subgroup effect?
- Is treatment effect consistent?
- Was the subgroup hypothesis one of a small number of hypotheses developed a priori with direction specified?
- Is there strong preexisting biological support?
- Is the evidence supporting the effect based on within or between study comparisons?
- There should be greater emphasis on the overall result than on a subgroup result
- Test of treatment subgroup interaction rather than treatment effect within subgroups
- Interpret the results in the context of other trial principles of biological rationale and coherence
- Generally, the number of subgroups tested should be minimized
- A statistical test of treatment-subgroup interaction and/or a subgroup stratification variable should be included
Presented by: Susan Halabi, PhD, Duke University Medical Center, Durham, NC
Written by: Zachary Klaassen, MD, MSc – Assistant Professor of Urology, Georgia Cancer Center, Augusta University/Medical College of Georgia, Twitter: @zklaassen_md at the 2019 Advanced Prostate Cancer Consensus Conference (APCCC) #APCCC19, Aug 29 - 31, 2019 in Basel, Switzerland
1. Davis ID, Martin AJ, Stockler MR, et al. Enzalutamide with Standard First-Line Therapy in Metastatic Prostate Cancer. N Engl J Med 2019 Jul 11;381(2):121-131.
2. Gulley JL, Borre, Vogelzang NJ, et al. Phase III Trial of PROSTVAC in Asymptomatic or Minimally Symptomatic Metastatic Castration-Resistant Prostate Cancer. J Clin Oncol 2019 May 1;37(13):1051-1061.
Further Related Content: A Subgroup Analysis of mCRPC Clinical Trials Presentation