组织活检的病理亚型分类是肺癌（LC）诊断的金标准，但在出现不确定的组织活检或无法到达肿瘤的情况下会变得复杂。利用液体活检（LBx）中的蛋白肿瘤标记物（TMs）和循环肿瘤DNA（ctDNA）对LC的诊断可进行最小侵入性的支持。本研究评估了基于LBx的诊断支持算法在诊断LC和分型为小细胞肺癌（SCLC）和非小细胞肺癌（NSCLC）方面的表现，旨在直接影响临床实践。在这项多中心前瞻性研究（NL9146）中，分析了1096名疑似LC患者的血液中八种蛋白TMs（CA125、CA15.3、CEA、CYFRA 21-1、HE4、NSE、proGRP和SCCA）和EGFR、KRAS和BRAF的ctDNA突变。通过评估预先规定的阳性预测值（PPV）≥95%或≥98%的逻辑回归模型来建立识别LC、NSCLC或SCLC的单个和组合TMs的表现。通过递归特征消除选择多参数模型中包括的信息最丰富的蛋白TMs。单个TMs在预先规定的PPVs下，可分别以46%、25%和40%的敏感性识别LC、NSCLC和SCLC患者。结合多个TMs和ctDNA的多参数模型显著提高了敏感性，分别为65%、67%和50%。在疑似LC的患者中，LBx基于的决策支持算法可确定约三分之二的LC和NSCLC患者以及一半的SCLC患者。因此，这些模型具有临床价值，尤其在无法进行病理亚型分类的患者中，可支持LC诊断。版权©2023作者。由Elsevier B.V.出版。保留所有权利。

Pathologic subtyping of tissue biopsies is the gold standard for the diagnosis of lung cancer (LC), which could be complicated in cases of e.g. inconclusive tissue biopsies or unreachable tumors. The diagnosis of LC could be supported in a minimally invasive manner using protein tumor markers (TMs) and circulating tumor DNA (ctDNA) measured in liquid biopsies (LBx). This study evaluates the performance of LBx-based decision-support algorithms for the diagnosis of LC and subtyping into small- and non-small-cell lung cancer (SCLC and NSCLC) aiming to directly impact clinical practice.In this multicenter prospective study (NL9146), eight protein TMs (CA125, CA15.3, CEA, CYFRA 21-1, HE4, NSE, proGRP and SCCA) and ctDNA mutations in EGFR, KRAS and BRAF were analyzed in blood of 1096 patients suspected of LC. The performance of individual and combined TMs to identify LC, NSCLC or SCLC was established by evaluating logistic regression models at pre-specified positive predictive values (PPV) of ≥95% or ≥98%. The most informative protein TMs included in the multi-parametric models were selected by recursive feature elimination.Single TMs could identify LC, NSCLC and SCLC patients with 46%, 25% and 40% sensitivity, respectively, at pre-specified PPVs. Multi-parametric models combining TMs and ctDNA significantly improved sensitivities to 65%, 67% and 50%, respectively.In patients suspected of LC, the LBx-based decision-support algorithms allowed identification of about two-thirds of all LC and NSCLC patients and half of SCLC patients. These models therefore show clinical value and may support LC diagnostics, especially in patients for whom pathologic subtyping is impossible or incomplete.Copyright © 2023 The Authors. Published by Elsevier B.V. All rights reserved.