A Systematic Comparative Analysis of Tumor Size Models Based on Erlotinib Clinical Data in Advanced NSCLC

Early assessment of efficacy and dose optimization remain critical challenges in the development of anticancer therapies. Empirical models of solid tumor size dynamics-a key prognostic biomarker-have played a central role in addressing these challenges. However, a systematic comparison of commonly used tumor size models, in terms of descriptive and predictive performance as well as generalizability within a population framework, has not been conducted to date. The present research sought to develop a methodological framework for the optimization of tumor models, offering a basis for more accurate predictions of tumor dynamics. The corresponding modeling workflow was practically tested against clinical data of erlotinib, a treatment administered to patients with advanced NSCLC. Five widely used tumor size models were evaluated, of which only three-the Bi-Exponential (BiExp), the Linear-Exponential (LExp), and Claret's Tumor Growth Inhibition (TGI) model-demonstrated reproducibility of the base model during a repeated cross-validation approach. Among these, the TGI model exhibited superior descriptive and predictive performance. However, a thorough literature search showed that erlotinib clinical data in NSCLC have been analyzed using only the BiExp and LExp models. Furthermore, extrapolation from 3 to 16 months revealed outlier predictions for the BiExp and TGI models, while the LExp model showed higher consistency, suggesting that models utilizing an exponential growth function may have a more limited extrapolation range than those assuming linear growth. Despite a clear ranking of models based on descriptive and predictive performance, no hierarchy emerged with respect to discriminatory ability. All three models showed high accuracy in distinguishing RECIST-based objective responders, while accuracy in predicting the emergence of acquired resistance remained uniformly low. Trial Registration: Clinical trial number: NCT00364351.