Research Highlight · AACR Drug Discovery and Development 2026
In-silico Phase III trial of avelumab plus axitinib versus sunitinib in advanced renal cell carcinoma
Presented as Abstract A026 at AACR Drug Discovery and Development 2026, this work tests whether a model-transfer framework can reproduce observed randomized trial behavior and transport treatment-effect estimates to independent cohorts.
Overview
Randomized clinical trials provide the strongest evidence for treatment benefit, but they are costly and cannot evaluate every clinically relevant treatment question. This project asks whether machine-learning models trained in a randomized trial can create a useful in-silico framework for testing alternative treatment scenarios.
The study focuses on advanced renal cell carcinoma and compares avelumab plus axitinib with sunitinib. Its central goal is not to replace a randomized trial, but to assess whether a carefully constructed model-transfer analysis can reproduce the treatment-effect pattern observed in trial data.
Clinical Motivation
Therapy development often requires deciding which treatments should progress from an early signal to larger randomized studies, or which established regimens may warrant testing in a new clinical setting. A computational framework that can stress-test expected benefit using existing trial and molecular data could make those decisions more evidence-informed.
For renal cell carcinoma, RNA-seq profiles may capture tumor biology that is relevant to treatment outcomes. The challenge is to use these data without overstating a model-derived contrast as causal proof; the model must first recover known randomized-trial behavior and then be evaluated across independent cohorts.
Study Design
Machine-learning models were trained with RNA-seq data from the JAVELIN Renal 101 randomized clinical trial and internally evaluated for progression-free survival and overall survival within the avelumab plus axitinib and sunitinib arms.
The analysis then applied each treatment-arm model to patients in the opposite arm to generate counterfactual survival predictions under the alternative treatment. Treatment effects were summarized using Kaplan-Meier curves and restricted mean survival time at 24 months.
Counterfactual Trial Framework
The model-transfer step creates a structured comparison: for each patient profile, it estimates the expected outcome under the observed regimen and under the alternative regimen. Aggregating those predictions yields a virtual treatment contrast that can be compared with the direction and magnitude of the randomized-trial result.
This is a validation-oriented use of counterfactual modeling. The framework is most credible when it reproduces known trial-level effects, is evaluated with appropriate uncertainty, and is tested outside the development dataset rather than treated as a substitute for prospective randomization.
What We Found
In the available JAVELIN Renal 101 data, avelumab plus axitinib showed a progression-free survival benefit over sunitinib, while overall survival was not significantly different. The in-silico analysis preserved the progression-free survival pattern favoring avelumab plus axitinib and likewise did not identify an overall-survival benefit.
For progression-free survival, the observed trial comparison showed median PFS of 13.6 versus 8.1 months and a 24-month restricted mean survival time difference of 2.68 months. The counterfactual analysis produced median PFS of 12.4 versus 8.3 months and a restricted mean survival time difference of 2.29 months.
External Cohorts
The treatment-transfer framework was also applied to independent cohorts with available RNA-seq data, including CheckMate 025, TCGA-KIRC, and HCRN-Nivolumab. These analyses showed a similar overall treatment-effect pattern, providing an important check on whether the result was tied only to the original trial cohort.
External application does not convert the analysis into a new randomized trial, but it is a meaningful test of transportability. It asks whether the model-derived result remains directionally consistent when evaluated in cohorts that differ in treatment context and data provenance.
Key Takeaways
- The work uses JAVELIN Renal 101 RNA-seq data to construct an in-silico Phase III trial framework.
- The counterfactual analysis reproduced the observed progression-free survival pattern favoring avelumab plus axitinib.
- Overall survival was not significantly different in either the observed or counterfactual comparison.
- Independent cohorts were used to assess whether the treatment-effect pattern transported beyond the development trial.
- The framework is intended to support hypothesis generation and trial prioritization, not replace prospective randomized evidence.