gms | German Medical Science

Background: Overall survival (OS) is the most important endpoint in oncology clinical trials, however, its use is limited by sufficient sample size and follow-up times. Progression-free survival (PFS) is an established surrogate for OS, based on the reliable detection of disease progression through imaging techniques. Here, the tumor growth is captured via the sum of largest diameters (SLD) of selected target lesions. Censoring due to limited follow-up, as well as dichotomizing tumor response lead to a loss of information and are known limitations for PFS estimation. We therefore reason that direct modeling of the tumor mass via the SLD may lead to improved predictions for patient prognosis and response to treatment.

Methods: The Vol-PACT group recently proposed a set of models to describe tumor shrinkage/ growth over time: A cluster of four models is used to describe the percentage change of the tumor quantity (e.g. SLD, PSA-level) after the start of a new therapy. The curves, derived from these models, can model many different shapes of tumor-growth. E.g. an exponential growth from the beginning on, an exponential decline (tumor response) or a decline at the beginning that changes into a regrowth (tumor progress) after some time. The estimated curve of each patient is then used to calculate the patient's time-to-progression (TTP). The latter is a simplified version of the PFS, where an event is defined by an increase in tumor mass by 20% compared to the smallest measured tumor-size (RECIST criteria). From this derived TTP, the survival quantities median TTP and the hazard ratio (HR) for treatment for a two-arm trial scenario are calculated. Based on input from a real trial scenario in lung cancer, we performed extensive simulation studies with varying numbers of patients where we varied the maximal follow-up time. Operating characteristics and performance of the Vol-PACT approach were evaluated in detail. Tumor growth rates, median TTP and the HR that are derived from the models were compared to the true underlying values and also to the classical “observed” estimations.

Results: The application of the models on the percentage change of the tumor-size does not fulfill the usual assumptions of the non-linear regression and derived a suggestion for improvement. We applied the models to a 2-arm study scenario with 500 patients in each arm. We cut the patient's data at 210 days or at the time of their progression. Ergo the patients had maximal 6 measurements. The predicted median TTP was 100 days in the control-arm and 153 in the treatment-arm. The true values were 112 days and 176 days. The predicted HR was 0.81 vs. the true HR of 0.83.

Conclusion: Direct modeling of tumor growth is a promising approach to support the decision-making processes in drug development. Modeling allows to derive good estimates for time-to-event endpoints like TTP-HR but they potentially underestimate median TTP.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.