Recent research finds that a new tool in biomarker discovery may be inferred through the analysis of Kaplan-Meier (KM) curves for time-to-event data from clinical trials. Using clinical trial simulations and trial data reanalysis, researchers investigated how survival results from clinical trials could be impacted by previously undiscovered responder subpopulations.1
Brian M. Alexander, MD, department of radiation oncology, Harvard Medical School, Boston, Massachusetts, who was a study author, told Cancer Therapy Advisor that the study “was initially conducted in response to frustration with the way clinical trials are commonly summarized: ‘The drug only improved survival by 2 months, on average.’ This likely refers to median survival or some other summary metric that might not be indicative of how the drug was actually working.”
“The theory was that if there were subgroups that were going to do better with a drug, and that those subgroups weren’t randomly distributed throughout the survival distribution, that we would see different kinds of survival curve shapes,” Dr Alexander explained.
He said that biomarkers that were both prognostic in the control arm and predictive of response to a drug would result in a characteristic KM curve appearance. “We thought we could find an example of this, but then also argue that you could start with the curves and use them, along with some knowledge of various biomarker prevalence data and prognostic associations, to rank potential predictive hypotheses.”
According to Dr Alexander, the theory was that there were subgroups that were going to do better with a drug, and that those subgroups weren’t randomly distributed throughout the survival distribution.
Aatur Singhi, MD, PhD, surgical pathologist at the University of Pittsburgh Medical Center, Hermitage, Pennsylvania, and assistant professor of pathology at the University of Pittsburgh, who was not involved in the study, said, “In the absence of a well-defined biomarker that is known to predict response to a specific therapy, clinical trials for these therapies have to be broad and, therefore, accept an ‘all-comer’ design.”
Dr Singhi elaborated: “However, cancers are heterogeneous and it’s likely that a specific form of therapy may not be effective in a certain population. Defining patient populations that will respond to a specific therapy is obviously important, and upon completion of trials a retrospective analysis is often performed to identify clinicopathologic features that may predict response.”