四种二代测序检测的真实世界分析一致性：检测方法学与生物信息学对临床决策的影响

Background: Broad tumor genomic profiling is increasingly required for selecting targeted therapies, yet analytical concordance across next-generation sequencing (NGS) assays used in routine practice remains uncertain. We conducted a real-world comparison of four NGS platforms to evaluate variability in key clinically relevant metrics, including tumor mutational burden (TMB), driver SNVs, and driver CNVs. Methods: Ten patients provided four matched FFPE eppendorfs each. Each aliquot contained at least 10 FFPE sections of 5-7 μm thickness, and all blocks met a ≥20% tumor cell requirement, ensuring uniform pre-analytic quality across all submitted material. All aliquots were analyzed in parallel across four laboratories: FMI (reference CGP), Private Lab 1, Private Lab 2, and Private Lab 3. For the primary comparison, assays were evaluated for patient-level reporting success (≥1 analyzable aliquot per patient), tumor mutational burden (TMB), driver SNVs, and driver CNVs. Results: FMI and Private Lab 1 successfully generated complete reports for 10/10 patients, Private Lab 3 reported 9/10. Private Lab 2 produced reports for 2/10 patients, an analyzability rate insufficient for inclusion in comparative metrics. TMB was reported for 10/10 patients by FMI and Private Lab 1 and 9/9 evaluable patients by Private Lab 3. For clinically actionable SNVs, FMI identified 6 variants; concordance was Private Lab 3 100% (6/6) and Private Lab 1 67% (4/6). For driver SNVs, FMI detected 31 variants; Private Lab 3 showed strong concordance across analyzable patients (83-100%), whereas Private Lab 1 showed markedly lower concordance (0-33%). For driver CNVs, FMI detected 14 alterations; Private Lab 3 identified 6/14, while Private Lab 1 detected none. Even within the two patients analyzed by Private Lab 2, concordance with FMI was inconsistent. Driver SNVs showed partial overlap, and several FMI-confirmed alterations were not detected. Driver CNVs were not identified in either patient, despite their presence in the reference assay. Conclusions: Concordance across NGS assays was suboptimal even among aliquots that were successfully processed, indicating substantial assay-to-assay variability. The markedly limited analyzability of the fourth platform further demonstrates that test performance in real-world settings is driven not only by the biological material but also by pre-analytic workflow, laboratory methodology, and the robustness of bioinformatic pipelines. Our findings show that results from different assays can diverge significantly, which may directly affect therapeutic decisions for patients. These observations underscore the need for independent analytical validation of every NGS assay before its results are used to guide clinical management.