Interlesional Heterogeneity of EGFR Mutations: A Systematic Review and Meta-analysis

Activating mutations in the epidermal growth factor receptor (EGFR) are a well-established oncogenic driver in non–small cell lung cancer (NSCLC) and several other solid tumours, predicting high initial response rates and progression-free survival with targeted tyrosine kinase inhibitors (TKIs) [1]. In addition to their predictive value, EGFR mutation subtypes also carry prognostic relevance, with meta-analytic evidence demonstrating significant associations between specific EGFR variants and overall survival in NSCLC [2]. Globally, approximately one-third of NSCLC harbour sensitising EGFR mutations, although the prevalence varies markedly by region, reaching nearly 50% in East Asian populations but only around 14% in Western cohorts [3, 4]. Recent molecular epidemiology studies further highlight this geographic variability, revealing distinct mutational spectra and hotspot distributions in EGFR across European and Asian populations [5, 6].

Clinical trials have consistently demonstrated that first- and second-generation EGFR-TKIs significantly outperform chemotherapy, doubling the median progression-free survival and improving overall survival, as well as quality of life, for patients with EGFR-mutant NSCLC [7‐9]. Although substantially better tolerated than cytotoxic regimens, EGFR-TKIs still produce characteristic adverse effects, particularly dermatological and gastrointestinal toxicities, which, while typically manageable, remain essential considerations in clinical decision-making [10]. Consequently, international guidelines now recommend routine EGFR genotyping for all newly diagnosed advanced NSCLC, with EGFR-TKIs firmly established as the standard of care [11].

Yet, despite this therapeutic paradigm shift, reliably capturing a patient’s true EGFR mutation status remains a challenge. A single solid tumour biopsy represents only a snapshot of a genetically heterogeneous disease [12]. Intra- and inter-tumoural subclonal diversity means that mutations detected in one lesion may be absent in others [13, 14]. Consistent with this, prior syntheses report discordant EGFR mutation results between primary lung tumours and paired distant metastases in ~10% of patients [15]. Similarly, comparisons of tissue versus liquid (blood-based) assays typically report concordance rates of only 60–80% [16]. Clinically, such uncertainty can complicate treatment selection and resistance monitoring.

To provide a clear conceptual framework, we distinguish between interlesional heterogeneity and EGFR mutation discordance. Interlesional heterogeneity refers to true biological differences in EGFR genotype between distinct tumour lesions within the same patient, for example, between the primary tumour and metastases or across metastatic sites. In contrast, EGFR mutation discordance is an empirical, test-level observation, defined here as a binary mismatch in detectable EGFR mutation status between paired specimens from the same patient (mutated vs wild-type), rather than as quantitative variation such as differences in variant allele fraction. Discordance is therefore not synonymous with interlesional heterogeneity. However, in routine clinical practice, it represents a pragmatic and clinically observable indicator that may reflect underlying spatial heterogeneity. At the same time, discordance can also arise from non-biological sources, including sampling limitations, temporal separation between specimens, variable circulating tumour DNA (ctDNA) shedding, pre-analytical handling, and differences in assay sensitivity and analytical thresholds. Accordingly, throughout this review, we interpret discordance as a context-dependent signal shaped by both biological and methodological factors, rather than as a direct measurement of lesion-to-lesion genetic divergence.

Liquid biopsy, typically the detection of ctDNA in blood or other body fluids, offers a minimally invasive approach to address some of these limitations [17, 18]. Serial ctDNA analysis enables the timely tracking of tumour evolution and the emergence of resistance mutations [19]. Current expert consensus supports ctDNA analysis as a practical complement to tissue genotyping, especially when repeat or multisite tumour sampling is needed [20]. Plasma ctDNA testing is now routinely integrated into NSCLC care pathways and has gained regulatory approval as an alternative when tumour tissue is insufficient or inaccessible [17, 21, 22]. Beyond plasma, other fluids such as cerebrospinal fluid (CSF), urine, pleural fluid, and even saliva can harbour tumour-derived DNA, expanding the clinical utility of liquid biopsy in detecting EGFR mutations in specific scenarios [23‐25].

Crucially, systematic evidence is lacking on when and why EGFR mutation results differ between specimens across solid tumours. Most prior reviews have focused on a single type of comparison (e.g. primary vs metastatic tissue [15] or plasma vs tissue [26]) and rarely examine multiple biospecimen types. It remains unclear how discordant results should be interpreted: Do they simply reflect sampling error, or do they signal meaningful spatial heterogeneity or the emergence of drug resistance?

Existing meta-analyses have generally been restricted to a single comparison, most commonly between primary and metastatic tissues or between tissue samples and plasma ctDNA. They rarely synthesise discordance across multiple liquid compartments within one framework. As a result, there is limited systematic evidence on how discordance changes when the sampling strategy shifts from tissue to liquid, or from one liquid compartment to another, and on which clinical contexts most strongly modulate disagreement. To address this evidence gap, we conducted a comprehensive pan-cancer systematic review and meta-analysis to date, covering over 150 studies, to comprehensively quantify EGFR mutation discordance rates across tissue–tissue, tissue–liquid, and liquid–liquid comparisons, spanning diverse fluids and clinical settings. Beyond estimating an overall discordance rate, we characterised how discordance varies with sampling strategy, treatment exposure, resistance development, and metastatic site under real-world testing conditions, with the goal of informing context-aware biopsy strategies and molecular monitoring in EGFR-mutant disease.

Our meta-analysis shows that EGFR mutation discordance is common, and likely reflects a combination of tumour heterogeneity, sampling effects, temporal evolution, and assay-related factors. Because discordance is a measurable disagreement between test results rather than a direct biological readout, it cannot, on its own, distinguish true interlesional heterogeneity from sampling, temporal, and technical contributors. The pooled discordance across all studies was approximately 16%, indicating that roughly one in six paired samples yielded conflicting EGFR results. While this overall estimate is broadly in line with earlier focused meta-analyses, the contribution of the present study is to characterise the heterogeneity around that average and to identify settings in which discordance is substantially higher. By integrating tissue–tissue, tissue–liquid, and liquid–liquid comparisons and synthesising evidence across multiple biofluids, we show that discordance is not a single fixed quantity but varies markedly by sampling strategy and biological compartment. This is reinforced by the considerable between-study heterogeneity and wide prediction interval, which suggests that the true discordance rate in a given clinical scenario can deviate substantially from the pooled mean.

Our pooled discordance rate is substantially higher than the approximately 10% discordance reported in earlier meta-analyses that focused only on primary versus metastatic tissue pairs [15]. We believe this increase is likely due to our expansion of scope to include both tissue and liquid biopsies. By comparing liquid–liquid and tissue–liquid samples alongside tissue–tissue pairs, our study adds to the literature by emphasising the real-world complexity of mutational heterogeneity in the age of advanced molecular testing. Even with high-quality assays, concordance/discordance rates can vary [32‐34], reinforcing that this cannot be explained by technical noise alone and likely reflects a combination of tumour heterogeneity, sampling effects, and assay-related variability. These discrepancies in sampling results have tangible clinical implications. In several cohorts, EGFR discordance rates reached 20–40% when comparing tissue biopsies and blood samples [35‐38]. Crucially, discordant findings can be actionable, as a liquid biopsy may reveal mutations that are missed by tissue sampling. Accumulating evidence indicates that such discrepancies are unlikely to be explained by technical artefacts alone and may reflect underlying tumour heterogeneity alongside assay-related and sampling factors. Multi-region sequencing and cfDNA analyses have shown that many EGFR alterations are present only in subclonal fractions of the tumour, explaining why different sampling modalities capture non-overlapping variants [39, 40]. Such findings suggest that discordant results are consistent with subclonal variation, alongside assay-related and sampling factors.

Comparing different sampling approaches, we found that tissue–tissue and tissue–liquid pairs exhibited similar discordance rates (16.8% vs 15.5%), whereas liquid–liquid pairs showed substantially higher discordance rates (34.0%). This pattern suggests that liquid specimens can access distinct, partially non-overlapping tumour reservoirs. We hypothesise that this stems from both spatial heterogeneity and variable ctDNA shedding dynamics. For example, lesions in the CNS shed minimal DNA into peripheral blood, while CSF directly samples the brain microenvironment [41, 42]. A plasma test alone may therefore miss mutations confined to brain metastases, whereas CSF can detect these subclones [43‐45]. Our results support the strategy of using multiple sampling modalities (e.g. tissue plus liquid or various liquids), particularly in clinically challenging scenarios such as CNS involvement or acquired resistance.

The specific fluid source also influenced discordance rates. Plasma, the most commonly tested source, yielded a pooled discordance of 16.5%, whereas serum showed lower discordance (10.9%). These results likely reflect differences in cfDNA abundance and processing, as plasma generally provides higher-yield and higher-quality ctDNA than serum, which can be contaminated by genomic DNA released during clotting [46, 47]. Similarly, urine-based cfDNA analyses produced intermediate discordance but captured mutations missed by tissue and plasma in some cases. Urine cfDNA NGS has been shown to detect key resistance mutations, including T790M, with sensitivities comparable to those of plasma [23]. Taken together, these findings highlight that sampling distinct biological compartments can reveal otherwise hidden clonal diversity.

Clinical factors further modulated discordance. Patients with prior TKI exposure showed notably higher discordance rates than treatment-naive patients, consistent with the clonal evolution of resistance mutations under therapeutic pressure. Tran et al. reported that discordance rates rose from 20% at diagnosis to 60% at the time of TKI progression [48]. Such findings suggest that acquired resistance subclones can emerge in some sites but not others, indicating that relying on a single sample at progression may miss actionable mutations. In this context, discordance is not simply a measurement issue but a clinically observable marker consistent with tumour adaptation and spatial heterogeneity in the context of treatment pressure.

Technical differences across studies also influenced the observed discordance. Studies using PCR-based assays generally showed slightly higher discordance than those using NGS, reflecting the limited sensitivity of older methods. However, even with highly sensitive assays, substantial discordance remained, indicating biological complexity alongside technical and methodological variability. This is consistent with recent reviews demonstrating that EGFR testing technologies have evolved substantially over the past decade, with wide variation in sensitivity, specificity, and clinical adoptions across regions, which contributes to the heterogeneity observed in published EGFR results [49]. We also noted regional differences: studies from East Asia (where EGFR-mutant tumours are more common [50]) reported significantly higher EGFR discordance rates compared to Western cohorts. These variations may reflect population-specific tumour biology, environmental factors such as smoking patterns, and region-specific testing or differences in study design. Importantly, because more than 80% of the included articles in our meta-analysis originated from Asian populations, the pooled discordance estimates should be interpreted with caution when extrapolating to Western settings, where EGFR prevalence and diagnostic pathways may differ.

These geographic differences should also be interpreted in light of regional variation in molecular testing pathways and platform uptake. Across settings, EGFR is assessed using assays with markedly different analytical sensitivities and target coverage, ranging from hotspot PCR-based methods to broader NGS panels. Liquid biopsy performance is further shaped by pre-analytical handling and laboratory thresholds for calling low-allele-fraction variants. As a result, the observed discordance in a given region is not purely a biological signal but rather a composite of tumour heterogeneity, clinical sampling practices (e.g. re-biopsy patterns at progression), and the sensitivity and specificity profile of the local testing approach. In our synthesis, the predominance of Asian cohorts and the diversity of assay strategies likely contribute to the wide prediction interval and substantial residual heterogeneity, and provide a plausible explanation for why discordance estimates vary across studies even when the nominal comparison type is the same. Our findings, therefore, support a context-aware interpretation of discordant EGFR results, in which both biology and the regional diagnostic ecosystem are considered when selecting between tissue, plasma, or other compartments.

We interpret EGFR discordance as a pragmatic indicator of spatial inter-tumoural (and intra-tumoural) heterogeneity, with critical translational implications. First, the presence of discordance justifies more comprehensive testing approaches. Wan et al. proposed a ‘blood-first’ strategy, using plasma EGFR testing initially, with reflex tissue testing if the results are negative [16]. Our findings support this approach: using multiple sampling sites and methods maximises the likelihood of detecting all actionable mutations. Second, discordance itself may serve as a prognostic marker. A patient whose liquid and tissue results conflict may carry subclones poised for therapeutic resistance, requiring closer monitoring or early combination therapies. Real-time discordance tracking through serial ctDNA sampling could enable truly adaptive treatment, allowing clinicians to anticipate and intercept emerging resistance clones. Third, these data underpin the rationale for combining spatial genomics, such as multi-region sequencing or advanced imaging, with liquid biopsy surveillance, to map and monitor tumour evolution in real time.

From a clinical perspective, these findings have direct implications for EGFR-targeted therapy selection and molecular testing strategies. In the first-line setting, discordance between tissue and liquid biopsies may lead to false-negative results if a single sampling modality is used, potentially delaying initiation of EGFR-TKI therapy. At disease progression, the higher discordance observed after prior TKI exposure highlights the risk of missing resistance mutations when relying on a single-site biopsy, supporting the use of repeat and complementary sampling approaches, including liquid biopsy, to guide sequential therapy decisions.

These findings are aligned with current international recommendations from organisations such as the International Association for the Study of Lung Cancer (IASLC) and the College of American Pathologists/Association for Molecular Pathology (CAP/AMP), which emphasise the complementary role of tissue and liquid biopsy and recognise tumour heterogeneity as a key challenge in molecular diagnostics [51].

While the results provide valuable insights, several limitations should be noted. Significant inter-study heterogeneity resulted from differences in patient populations, sampling methods, and reporting standards. To address this, we used random-effects models, conducted extensive subgroup analyses, and performed sensitivity analyses to assess the robustness of our findings. We also report both between-study variance (τ²) and prediction intervals to emphasise the context dependence of discordance. Notably, meta-regression explained little of the between-study variance, with pseudo-R² values below 5% in univariate analyses and no reduction in τ² in the multivariable model, indicating substantial residual heterogeneity due to unmeasured moderators. Several plausible technical contributors could not be evaluated systematically because they were inconsistently reported, including the time interval between paired samplings, platform-specific analytical sensitivity (for example, limits of detection, panel coverage, sequencing depth, and variant calling thresholds), and pre-analytical handling of liquid biopsies (for example, collection tubes, processing delays, centrifugation, storage conditions, and cfDNA extraction protocols). Publication bias is probably present, as negative or null discordant findings may be underreported. Moreover, most studies did not consistently specify which EGFR mutations were discordant, nor whether the mutations had an activating or resistance-conferring role, which prevented mutation- or exon-level analyses. As a result, our pooled estimates reflect overall EGFR mutation discordance rather than mutation-specific biological heterogeneity. Additionally, few studies directly linked discordance to clinical outcomes, which limits causal inferences about progression or resistance. The definitions of ‘discordance’ also varied across studies (some included any mismatch, while others distinguished between gains and losses); we therefore pooled these for analysis. Because discordance was defined as a binary mismatch (mutated vs wild type), study-level EGFR mutation prevalence is mathematically related to the expected discordance rate; we therefore treat prevalence in meta-regression as a case-mix descriptor rather than a mechanistic driver, and this dependence should be considered when interpreting moderator analyses. Lastly, most data originated from high-volume academic centres, which may not reflect community practice. However, by including studies from diverse geographic regions and practice settings, we aimed to maximise the generalisability of our results. Despite these limitations, the consistent findings across different settings strengthen our conclusions.

In conclusion, EGFR mutation discordance is a common and clinically meaningful phenomenon, reflecting the complex interplay between tumour spatial heterogeneity, sampling strategy, and technical factors in NSCLC. Importantly, discordant results should not be dismissed as assay errors or inconsistencies; rather, they represent informative signals that highlight uncertainty in molecular profiling and the potential presence of spatially or temporally distinct tumour subclones. While the substantial methodological heterogeneity across studies limits immediate clinical implementation, acknowledging discordance may help contextualise negative or conflicting test results and reduce the risk of missed therapeutic opportunities, inappropriate treatment selection, or delayed identification of resistance. Our findings support increased awareness of discordance in clinical decision-making and provide a rationale for integrated and adaptive biopsy strategies, including multi-site and liquid biopsy approaches. Future prospective studies are needed to evaluate whether systematic consideration of discordance, through approaches such as real-time monitoring, multi-site sampling, or adaptive testing strategies can be optimally incorporated into clinical workflows and improve patient outcomes. In the evolving era of precision oncology, careful attention to mutation discordance may help ensure that molecular testing strategies better reflect tumour complexity in real-world practice.