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

Area of Science:

• Pharmacometrics and mathematical modeling in oncology.
• Clinical trial data analysis and interpretation.
• Biomarker development for cancer therapy assessment.

Background:

• Accurate tumor size models are vital for assessing anticancer drug efficacy and optimizing dosage.
• Existing tumor size models lack systematic comparison regarding performance and generalizability.
• Previous analyses of erlotinib in NSCLC primarily used Bi-Exponential and Linear-Exponential models.

Purpose of the Study:

• To establish a methodological framework for optimizing tumor growth models.
• To systematically compare commonly used tumor size models for descriptive and predictive accuracy.
• To evaluate model generalizability within a population framework using clinical data.

Main Methods:

• Developed and applied a modeling workflow to clinical data from erlotinib treatment in advanced NSCLC patients.
• Evaluated five established tumor size models using repeated cross-validation.
• Assessed descriptive, predictive performance, generalizability, and extrapolation capabilities of the models.

Main Results:

• Three models (Bi-Exponential, Linear-Exponential, Claret's TGI) showed reproducibility; Claret's TGI model performed best descriptively and predictively.
• The Linear-Exponential model demonstrated better consistency for long-term extrapolation compared to Bi-Exponential and TGI models.
• All evaluated models accurately identified objective responders but showed low accuracy in predicting acquired resistance.

Conclusions:

• The Claret's TGI model offers superior performance for analyzing erlotinib in NSCLC, despite not being previously used.
• Linear growth models may offer more reliable extrapolation than exponential models in certain contexts.
• Further research is needed to improve models for predicting acquired resistance in anticancer therapy.