Background: The epidermal growth factor receptor (EGFR) is frequently mutated in patients with lung adenocarcinoma (LUAD), leading to its constitutive activation and subsequent uncontrolled cell proliferation. To counteract this, tyrosine kinase inhibitors (TKIs) targeting EGFR have been developed. However, the emergence of metastases and resistance mutations often undermines the durability of the treatment response. Knowledge-based mechanistic models that replicate existing clinical trial outcomes, tailored to specific population characteristics, provide invaluable assistance in designing future clinical trials. Methods: We have developed physiologically based pharmacokinetic (PBPK) models for gefitinib and osimertinib, two EGFR TKIs, to accurately simulate the drugs' distribution within the primary tumor and metastatic sites following administration. These models were integrated with a mechanistic model of EGFR-mutant LUAD, enabling the representation of the mechanism of action of gefitinib and osimertinib. This model outputs the evolutions of the primary tumor and each metastasis, facilitating the evaluation of patient progression following RECIST criteria. Notably, the model encapsulates the heterogeneity within the tumor, representing various subclones defined by unique mutation profiles, thereby reflecting differential responses to treatment. Calibration of the model was achieved using publicly available data from the NEJ002, FLAURA and AURA3 clinical trials. Rigorous visual predictive checks and statistical tests were employed to ensure the proper behavior of the model. Results: The model adeptly replicated the time to progression in EGFR-mutant LUAD patients receiving gefitinib or osimertinib, whether as first-line or second-line therapy. Furthermore, it accurately mirrored the progression causes as per the RECIST criteria, including the emergence of new metastases in the lung, brain, liver, and bone. With a consistent virtual population as a reference, the combined model facilitated a comparative analysis of the efficacy of both treatments, thereby underscoring its utility in evaluating therapeutic strategies. Conclusions: The faithful replication of real-world data significantly enhances the credibility of our model, rendering it a promising tool for deriving pertinent insights and informing treatment strategies. Following its consecutive prospective validations in the FLAURA2 and MARIPOSA trials, the model is poised to facilitate the generation of synthetic control arms in future clinical trials. This advancement promises a more nuanced analysis of covariate relationships, especially when comparing investigational treatments to established standards of care. Additionally, it potentially reduces the requisite patient cohort size in such trials, optimizing resource utilization.