For a concrete example of time-dependent treatment effects resulting from a prognostic and predictive biomarker, data from published clinical trial from the Radiation Therapy Oncology Group (RTOG 9402) were reanalyzed. RTOG 9402 was designed to evaluate the role of procarbazine, lomustine, and vincristine (PCV) chemotherapy with radiation therapy (RT) for patients with anaplastic oligodendrogliomas and anaplastic oligoastrocystomas.2
This study served as a representative example in which certain biomarker-specific effects led to time-varying treatment effects and represented a discordant proportional hazard compared with those that were observed across the overall trial population.
They evaluated the presence of time-varying treatment effects due to unknown predictive and prognostic factors using a Grambsch-Therneau test. Patient-level data were used to estimate the hazard ratio (HR) between experimental and control arms using the Cox proportional hazards model.
Researchers then “demonstrated a quantitative framework to fit survival data from clinical trials using statistical models incorporating known estimates of biomarker prevalence and prognostic value to prioritize their predictive biomarker hypotheses.[”1
The findings show a Cox model with treatment-time interaction term led to a statistically significant negative treatment-time interaction (-0.080, P <.001), which researchers determined was indicative of an increasing treatment effect. The Grambsch-Therneau test “also supported evidence of nonproportionality of HRs when the investigators compared treatment and control arms (P <.001).”
The discovery of a particular genetic aberration in oligodendrogliomas, 1p/19q codeletion, changed the researchers’ interpretation of RTOG 9402: the tumors of 48% of the patients who had undergone the requisite testing were found to harbor this chromosomal loss, and it was determined that there was a statistically significant benefit seen in these patients when PCV was added. Those without the 1p/19q codeletion did not benefit to the same degree.
On the results of the study, Joe Ensor, PhD, cancer biostatistician, Houston Methodist Research Institute, Texas, said, “Predictive and prognostic factors could also be tested by fitting a Cox proportional hazards model to the data. But the proportional hazards assumption is key and the authors used the Grambsch-Therneau test, which is based on the Schoenfeld residuals of the model, to assess proportionality.”Markers Successfully Prioritized
Deviation from proportional hazards by the Grambsch-Therneau test and through the use of Cox models with treatment-time interaction terms could provide evidence for time-varying treatment effects and could help in the identification of novel biomarkers.
Researchers also demonstrated that simulation of several hypothetical clinical trials scenarios can show the predictive or prognostic nature of certain biomarkers; this information could be gleaned just by looking at the survival curves of patients included in the study.
Dr Alexander said, “We found an example in glioma with 1p/19q codeletion and IDH mutation and response to PCV chemotherapy that fit the description of what we were trying to show. We also showed that we could effectively prioritize known markers in another example in glioblastoma and then showed some examples for future exploration.”
But this new method of identifying biomarkers of response comes with some caveats. “The drawbacks for the use of the methods suggested in this paper are the same for all methods of discovery; findings need to be validated,” Dr Ensor cautioned. “Statistical methodologies such as cross-validation — either exhaustive or nonexhaustive — should be used to more accurately estimate the performance of any model derived through this method of discovery.”
- Rahman R, Ventz S, Fell G, Vanderbeek AM, Trippa L, Alexander BM. Divining responder populations from survival data [published online March 12, 2019]. Ann Oncol. doi: 10.1093/annonc/mdz087
- Intergroup Radiation Therapy Oncology Group Trial 9402, Cairncross G, Berkey B, Shaw E, et al. Phase III trial of chemotherapy plus radiotherapy compared with radiotherapy alone for pure and mixed anaplastic oligodendroglioma: Intergroup Radiation Therapy Oncology Group Trial 9402. J Clin Oncol. 2006;24(18):2707-2714.