A model-based approach to optimize detection of treatment effects in early oncology trials.
e13508 Background: Tumor size change from baseline (RECIST) is often used to assess antitumor activity of investigational agents in phase 1 trials, but this measure does not take into account the tumor growth rate (TGR) prior to treatment. TGR could be highly variable in a phase 1 ‘all-comers’ patient population, with a high TGR potentially masking a meaningful treatment effect (TE). Assessing the change in TGR (Mehrara et al, BJC 2011) using historical tumor burden assessments (Gomez-Roca et al, Eur J Cancer 2011) may provide higher sensitivity to true TE (TTE). The objectives of the current study are to formalize a Growth Rate Based Method (GRBM) and to compare, using simulated patients, the ability of GRBM and RECIST assessments to detect and quantify TTE. Methods: The exponential tumor growth model (Claret et al, J Clin Oncol. 2009) was used to simulate the sum of the longest diameters (SLD) individual time courses of 2000 virtual patients under different TGR scenarios: slow, medium, fast and ‘all comers’ (highly variable TGR as often encountered in phase 1). Different sampling designs were simulated wherein tumor assessments with measurement errors were obtained, ranging from 16 to 4 weeks prior to treatment initiation (TI), immediately before TI, and 8 and 16 weeks after TI. TTE was defined as the difference between the simulated SLDs at 16 weeks with and without treatment. GRBM response was defined as the model-predicted difference between the SLDs with and without treatment, as estimated from the simulated samples. Sensitivity (Se) of RECIST or GRBM was defined as the probability of classifying a patient as a RECIST or GRBM >30% reduction when the TTE >30%, and specificity (Sp) as the probability of RECIST or GRBM reduction <30% when TTE <30%. Results: RECIST Se was consistently inferior to GRBM, notably in the all-comers TGR (37% vs 74-89%) scenario. GRBM maintained Se (71-89%) irrespective of TGR whereas RECIST Se degraded significantly with increasing TGR. RECIST Sp was high (97-100%) while GRBM Sp was lower (55-80%) depending on sampling design and TGR. Conclusions: Incorporation of an additional pretreatment tumor assessment by GRBM may augment RECIST by increasing sensitivity to TTE, particularly in a heterogeneous TGR phase 1 setting.