Evaluating the value of circulating miR-21, miR-210 and miR-942 in the diagnosis of early-stage lung adenocarcinoma Authors: Rong Li, Guangmei Chen, Yue Shao, Xiaohan Jin, Ziyi Zhang, Wei Wu, Mengnan Sun, Lichuan Zhang Published: 2025 ABSTRACT BackgroundThe incidence of lung adenocarcinoma has been gradually increasing in recent years. Its early diagnosis remains challenging. Dysregulation of miRNAs is involved in the development of many malignant tumors, miRNAs have shown potential promise in the early diagnosis of tumors. This study aimed to evaluate the value of serum miR-21, miR-210, and miR-942 expression in the early diagnosis of lung adenocarcinoma(LUAD).MethodsPreoperative peripheral blood was collected from 155 patients with suspected lung nodules who were due to be operated on, and the serum levels of miR-21, miR-210, and miR-942 were determined by real-time fluorescence quantitative PCR (RT-qPCR). Based on the postoperative pathology results, 130 patients with stage I-II early-stage LUAD were selected as the study subjects, and 80 healthy people over the same time period were selected as the control group. An early diagnostic model of LUAD with the combination of the 3 miRNAs was established. The diagnostic performance was evaluated using a receiver operating characteristic (ROC) curve. ResultsCompared with the control subjects, the expression of the 3 miRNAs was significantly upregulated in the LUAD patients in both the training and testing sets. Based on the logistic regression model, the AUC value for diagnosing early-stage LUAD using the 3-miRNA panel in the training set is 0.909, with an accuracy of 87.1%. In the testing set, the AUC is 0.890, and the accuracy is 82.9%. The combined AUC for both the training and testing sets is 0.901, with an accuracy of 84.3%. Serum miRNA-21, miRNA-210, and miRNA-942 all showed higher diagnostic efficacy in comparison with the conventional tumor marker Cyfra21-1 (AUC: 0.554 vs. 0.790, 0.856, and 0.621, respectively). Subgroup analysis based on clinical features showed that the 3-miRNA panel has better predictive performance for lung nodules that appeared solid in imaging findings, with an AUC of 0.903, a sensitivity of 90.0%, and a specificity of 80.0%.ConclusionsThe combination of serum miR-21, miR-210, and miR-942 could be employed as potential serum markers for early diagnosis of LUAD.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12885-025-15201-w. Introduction Lung cancer remains the most common cause of cancer-related deaths worldwide, and the patients’ survival depends largely on the stage at which the tumor is detected [1]. Compared with the dismal prognosis of advanced non-small cell lung cancer (NSCLC), the 5-year overall survival (OS) in early-stage NSCLC (stage I-II) can reach 61–88% [2]. NSCLC at stage I-II can even be cured by prompt surgery or stereotactic radiotherapy (SABR) [3]. Therefore, early diagnosis plays a crucial role in improving lung cancer survival. Although low-dose computed tomography (LDCT) can decrease mortality, the specificity of LDCT is limited, especially in the evaluation of indeterminate pulmonary nodules such as ground-glass opacities (GGOs), which are common in early-stage lung adenocarcinoma (LUAD). Invasive procedures are often required for diagnosis, which may lead to unnecessary surgical resection of benign lesions and increased healthcare burden [4]. To achieve early prediction and diagnosis of lung cancer, minimally invasive auxiliary methods are urgently needed as a complement to LDCT. Additionally, relatively easy-to-obtain biomarkers from body fluids, especially from blood, could prove to be the optimal predictors in the context of precision medicine. However, it may be affected by the tumor burden, and some early tumors may not release enough tumor information, making it difficult to detect. This leads to low specificity and sensitivity of conventional tumor markers, with the positive rate of a single tumor marker for early lung cancer being less than 30% [5]. Even if CEA, NSE, CYFRA21-1, ProGRP, and SCCA are jointly detected for early lung cancer, the positive rate is only 35.29% [6]. The emergence of some new markers is expected to improve the positive predictive value of early lung cancer diagnosis [7]. Circulating tumor DNA (ctDNA) and Circulating tumor cells (CTCs), although they can be used to assist in lung cancer screening [8, 9], are limited in their promotion due to the immaturity of detection technology, high cost, and low positive rate in early-stage lung cancer [10]. Meanwhile, serum miRNA has the advantages of high sensitivity and stability, and can avoid the heterogeneity of tumors [11], which may be more advantageous in assessing the individual risk of early-stage lung cancer in uncertain nodules detected by LDCT imaging [12]. MicroRNAs (miRNAs) are small non-coding RNAs. A large body of evidence suggests that miRNAs are capable of regulating a wide range of biological functions in the human body. miRNA dysregulation has been observed in a variety of malignant tumors, such as chronic lymphocytic leukemia [13, 14], glioblastoma [15], thyroid-like carcinoma [16], hepatocellular carcinoma [17], and lung carcinoma [18], where even subtle changes in miRNAs expression levels can significantly affect the disease outcome [19]. Research has found that miRNA can provide personalized risk assessments for uncertain lung nodule imaging [20]. miR-21 is a relatively early and most studied miRNA, and was shown to be upregulated in most malignant tumors, including lung cancer [21]. High expression of miR-21 can indicate poor prognosis in NSCLC [22]. miRNA-210 is another miRNA of interest, and is involved in tumorigenesis and development by affecting cell proliferation, apoptosis, and metabolism [23]. Its expression level is significantly elevated in LUAD tissues, and hypoxia can promote its expression [24]. miR-942 has been less well studied in lung cancer, but its overexpression promotes proliferation and metastasis, and affects the tumor TNM staging [25]. The above studies suggest that miR-21, miR-210, and miR-942 could be potential targets for the diagnosis and treatment of NSCLC. NSCLC accounts for more than 80% of lung cancers, of which LUAD has gradually increased in incidence to become the most common pathologic subtype of NSCLC, accounting for about 40–45% [26]. In particular, LUAD tends to predominate in the Asian non-smoking female population, whose imaging features usually manifest as lung nodules [27]. The current 5-year survival rate for LUAD is less than 20%, with late diagnosis contributing to poor clinical outcomes [28]. On the other hand, this is also partly due to insufficient knowledge of molecular diagnosis and tumor gene expression profiles. Early diagnosis and treatment of LUAD has entered a plateau period, and intensive research on molecular-assisted diagnosis will be helpful in changing this situation [29]. However, relatively few studies have specifically addressed early-stage LUAD—particularly those presenting as subsolid nodules—which remain diagnostically challenging due to their indolent imaging appearance and low tumor burden.Therefore, we aimed to develop and validate a serum-based diagnostic model using miR-21, miR-210, and miR-942 for the auxiliary diagnosis of early-stage LUAD in clinically evaluated patients with suspicious pulmonary nodules. The model was constructed using logistic regression and evaluated using ROC curves. To provide biological context, we also analyzed performed pathway enrichment analysis to explore the relevance of the selected miRNAs to lung cancer pathogenesis. Methods Study subjects A total of 155 patients who were hospitalized in the Affiliated Zhongshan Hospital of Dalian University from June 1, 2021 to January 31, 2022 for surgery due to clinically or radiologically suspected lung nodules were included, and clinical baseline data were collected. Inclusion criteria were: a single chest CT–detected pulmonary nodule < 30 mm in diameter, no prior history of malignant tumors, no pleural effusion, and no preoperative neoadjuvant chemotherapy or radiotherapy. Lung cancer was confirmed in all patients by postoperative histopathology. Early-stage lung cancer was defined as stage I–II according to the AJCC/UICC 9th edition TNM classification [30]. To reduce potential confounding from inflammatory and chronic lung conditions, we excluded individuals with a documented history of pulmonary tuberculosis, clinician-diagnosed chronic obstructive pulmonary disease (COPD), asthma, pulmonary fibrosis, pneumoconiosis, or any prior malignant tumor. We also excluded participants with an acute respiratory infection within 4 weeks prior to phlebotomy; no radiographic pneumonia was present at the time of sampling. Healthy volunteers (n = 80) during the same period underwent low-dose chest CT scans, which showed no radiologically suspicious pulmonary nodules. Minor findings such as benign-appearing subcentimeter nodules (≤ 4 mm) or calcified granulomas were not excluded. None of the participants had a history of malignancy. All individuals provided written informed consent prior to blood draw.The baseline demographics, clinical characteristics, and pathological information (pathological type and degree of differentiation) of the enrollees are shown in Table 1. This study was approved by the Research Ethics Committee of Zhongshan Hospital, Dalian University (Approval No. 201733), and written informed consent was obtained from all participants.Table 1Characteristics of subjects in the training and testing setTraining set(n = 140)Testing set(n = 70)LUAD(n = 90)Healthy control(n = 50)p-value (LUAD vs.Healthy control)LUAD(n = 40)Healthy control(n = 30)p-value (LUAD vs.Healthy control)Age (years)Mean ± SD58.9 ± 10.856.7 ± 9.10.21262.6 ± 7.560.1 ± 6.60.148Gender0.2390.121 Male34(37.8%)24(48%)8(20%)11(36.7%) Female56(62.2%)26(52%)32(80%)19(63.3%)Nodule diameter(mm), Median ± IQR15.5 ± 6.67 ± 2.2 ≤ 1015(16.7%)38(95%) 11–2054(60%)2(5%) 21–3018(20%)0 > 303(3.3%)0Nodule density Solid16(17.7%)4(10.0%) Sub-solid24(26.7%)7(17.5%) Ground glass50(55.6%)29(72.5%)Differentiation High29(32.2%)21(52.5%) Moderate57(63.3%)11(27.5%) Poor4(4.5%)8(20.0%)Pathological type AIS04(10%) MIA40(44.4%)26(65%) ICA50(55.6%)10(25%)Image characteristic Lobulation18(20%)1(2.5%) Spiculation22(24.4%)1(2.5%) Pleural indentation23(25.6%)1(2.5%) Air bubble12(13.3%)7(17.5%) Vessel29(32.2%)7(17.5%) Calcification4(4.4%)1(2.5%)TNM stage 05(5.6%)0 Ⅰ85(94.4%)35(87.5%) Ⅱ05(12.5%)AIS Carcinoma in-situ, MIA Microinvasive adenocarcinoma, ICA Invasive adenocarcinoma Table 1Characteristics of subjects in the training and testing setTraining set(n = 140)Testing set(n = 70)LUAD(n = 90)Healthy control(n = 50)p-value (LUAD vs.Healthy control)LUAD(n = 40)Healthy control(n = 30)p-value (LUAD vs.Healthy control)Age (years)Mean ± SD58.9 ± 10.856.7 ± 9.10.21262.6 ± 7.560.1 ± 6.60.148Gender0.2390.121 Male34(37.8%)24(48%)8(20%)11(36.7%) Female56(62.2%)26(52%)32(80%)19(63.3%)Nodule diameter(mm), Median ± IQR15.5 ± 6.67 ± 2.2 ≤ 1015(16.7%)38(95%) 11–2054(60%)2(5%) 21–3018(20%)0 > 303(3.3%)0Nodule density Solid16(17.7%)4(10.0%) Sub-solid24(26.7%)7(17.5%) Ground glass50(55.6%)29(72.5%)Differentiation High29(32.2%)21(52.5%) Moderate57(63.3%)11(27.5%) Poor4(4.5%)8(20.0%)Pathological type AIS04(10%) MIA40(44.4%)26(65%) ICA50(55.6%)10(25%)Image characteristic Lobulation18(20%)1(2.5%) Spiculation22(24.4%)1(2.5%) Pleural indentation23(25.6%)1(2.5%) Air bubble12(13.3%)7(17.5%) Vessel29(32.2%)7(17.5%) Calcification4(4.4%)1(2.5%)TNM stage 05(5.6%)0 Ⅰ85(94.4%)35(87.5%) Ⅱ05(12.5%)AIS Carcinoma in-situ, MIA Microinvasive adenocarcinoma, ICA Invasive adenocarcinoma Characteristics of subjects in the training and testing set Age (years) Mean ± SD Nodule diameter (mm), Median ± IQR AIS Carcinoma in-situ, MIA Microinvasive adenocarcinoma, ICA Invasive adenocarcinoma Sample collection and quantification of serum MiRNA using the RT-qPCR method A 5 ml specimen of preoperative fasting peripheral blood was collected from the study subjects, centrifuged for 10 min in a low-speed centrifuge, and 2 ml of supernatant serum was aspirated and placed in a −80℃ refrigerator to be stored for use. Total RNAs were isolated from serum using a miRcute miRNA Kit, and the reverse transcription reaction was completed using a miRcute Enhanced miRNA cDNA First Strand Synthesis Kit. The U6 small nuclear RNA (snRNA), a commonly used endogenous reference in serum-based RT-qPCR studies, was used for normalization [31]. Finally, the miRNA expression level was detected by real-time fluorescence quantitative polymerase chain reaction (RT-qPCR) using a miRcute Enhanced miRNA qPCR Kit (SYBR Green). miRNA relative expression was calculated using the Eq. 2-ΔCT. Quantitative real-time PCR (RT-qPCR) was conducted using the miRcute Enhanced miRNA qPCR Kit (SYBR Green) (Tiangen, China) on a 96-well platform.Relative expression was calculated using the 2^−ΔCt method, where ΔCt = Ct(miRNA) − Ct(U6). All reactions were run in technical triplicates with non-template controls; melting-curve analysis confirmed single-peak amplicons. Primer sequences and qPCR program Forward primers targeting mature miRNAs were:hsa-miR-21: UAGCUUAUCAGACUGAUGUUGAhsa-miR-210: AGCCCCUGCCCACCGCACACUGhsa-miR-942: UCUUCUCUGUUUUGGCCAUGUG hsa-miR-21: UAGCUUAUCAGACUGAUGUUGA hsa-miR-21: UAGCUUAUCAGACUGAUGUUGA hsa-miR-210: AGCCCCUGCCCACCGCACACUG hsa-miR-210: AGCCCCUGCCCACCGCACACUG hsa-miR-942: UCUUCUCUGUUUUGGCCAUGUG hsa-miR-942: UCUUCUCUGUUUUGGCCAUGUG A universal reverse primer supplied by the kit was used. qPCR thermal profile: Initial enzyme activation/denaturation at 95 °C for 15 min; followed by 40–45 cycles of 64 °C for 20 s (denaturation) and 60 °C for 34 s (annealing/extension). Bioinformatics analysis The target genes of serum miRNAs were determined using the miRWalk and mirDIP databases. The overlapping targets from both databases were selected for further analysis. Gene ontology (GO) enrichment analysis was conducted to ascertain the molecular functions, cellular composition, and biological processes associated with the target genes, and the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis was employed to ascertain the biological pathways associated with the target genes. Radiologic assessment Thin-slice chest CT at baseline was reviewed to categorize nodule density as pure ground-glass opacity nodule (pGGN), part-solid nodule (PSN), or solid nodule (SN) according to standard radiologic definitions. Maximum nodule diameter (mm) at baseline was recorded. A concise imaging summary of the malignant subgroup (diameter and density) will be provided in Table 1, while that of the benign subgroup will be included in Supplementary Table S1. Statistic analysis Continuous variables were assessed for normality and are presented as mean ± standard deviation (SD) if normally distributed, or as median (interquartile range, IQR) if skewed. Between-group comparisons used Student’s t-test for normally distributed variables and the Mann–Whitney U test for non-normal data; categorical variables were compared using the χ² test or Fisher’s exact test, as appropriate. Diagnostic performance was evaluated by receiver operating characteristic (ROC) analysis with area under the curve (AUC), sensitivity, specificity, and the optimal cut-off determined by Youden’s index (YI = sensitivity + specificity − 1) [32]. Logistic regression was used to construct the 3-miRNA model. A 95% confidence interval (CI) was used as a measure of the precision of the regression coefficients, and the presence of statistical significance was determined by a P-value [33]. Values of P < 0.05 were considered statistically significant. Analyses were performed using SPSS 23.0 (IBM, Armonk, NY, USA) and GraphPad Prism 8.0 (GraphPad Software, CA, USA). Results General clinical data collection of enrolled patients A total of 155 patients were enrolled in this study. 136 diagnoses were of lung cancer and 19 were benign. In the lung cancer group, 130 cases of early-stage lung cancer were adenocarcinoma, of whom 88 were female and 42 were male, and there were a total of 6 cases of middle- and late-stage lung cancer, including 3 cases of adenocarcinoma, 1 case of squamous cell carcinoma, and 2 cases of small cell lung cancer. Since the proportion of patients with advanced lung cancer was relatively small in this study, and all early lung cancers were adenocarcinomas, 130 patients with LUADs were selected as the study subjects. Among the 19 patients ultimately diagnosed with benign disease after surgical resection, pathological entities included (but were not limited to) hamartoma (n = 3), Nontuberculous granuloma(n = 4), tuberculous granuloma (n = 2), and non-granulomatous inflammatory nodules (n = 10). These lesions were resected because preoperative clinical and/or radiologic assessment suggested malignancy. A concise summary of benign nodule diameter and density will be provided in Table S1 to aid interpretation of differential diagnosis. A 3-miRNA combination model for early diagnosis of LUAD was established in 130 early-stage LUAD patients and 80 healthy controls: the training set (n = 140) consisted of 90 lung cancer patients and 50 healthy subjects, and the testing set (n = 70) consisted of 40 lung cancer patients and 30 healthy subjects. The characteristics of the subjects in the training and testing sets are shown in Table 1. miR-21, miR-210, and miR-942 expression in the training set To determine whether the expression levels of miRNAs in serum samples from lung cancer patients could be distinguished from those of healthy individuals, we first quantified the expression of three candidate miRNAs in serum samples from the training set (n = 140) by RT-qPCR. Of interest, in order to better investigate the detection efficacy of these miRNAs in early-stage lung cancer, we intentionally included 5 patients in the training set with stage 0 LUAD and 85 patients with stage I LUAD. The results suggested that miR-21, miR-210, and miR-942 expression showed significant upregulation in the lung cancer group compared with healthy subjects (Fig. 1A).Fig. 1Expression of three serum miRNAs in early stage lung adenocarcinoma group and healthy control group (Figure1A: Training set; Figure1B: Testing set) Expression of three serum miRNAs in early stage lung adenocarcinoma group and healthy control group (Figure1A: Training set; Figure1B: Testing set) miR-21, miR-210 and miR-942 expression validation in the testing set The results showed that miR-21, miR-210, and miR-942, which were differentially expressed in the lung cancer group in the training set, were similarly significantly up-regulated in their expression levels in the test set compared with the healthy control group (Fig. 1B). Thus, it was demonstrated that the expression of these three miRNAs could distinguish lung cancer patients from the cancer-free population. A 3-miRNA panel optimized as serum-based biomarkers for early-stage LUAD diagnosis To assess the diagnostic reliability of the 3 miRNAs, we performed a univariate logistic regression analysis for each individual miRNA in the training set (n = 140) and testing set (n = 70). The regression coefficient (β) was a valid parameter indicating that the odds of an outcome (i.e., developing lung cancer) increased with increasing exposure (i.e., serum miRNA levels). The analysis showed that miRNA-21, miRNA-210 and miRNA-942 were all significantly associated with lung cancer in the training set (P < 0.05; Table 2), and miRNA-210 and miRNA-942 were found to be significantly associated with lung cancer in the testing set (P < 0.05; Table 2). Regardless of whether in the training set or the testing set, miRNA-210 exhibits a consistently high β value, indicating that the elevation in miRNA-210 expression levels significantly influences the increased incidence of lung cancer.To assess the diagnostic ability of the combined detection of the 3 miRNAs in the training set (n = 140), we first performed a ROC analysis for each miRNA comparing early stage LUAD patients and healthy controls. The AUC values of these 3 miRNAs ranged from 0.601 to 0.871. When the optimal diagnostic cut-off for each miRNA was determined using Youden’s index (YI), the sensitivities of these 3 miRNAs ranged from 78.9% to 95.6%, and the specificities ranged from 36.0% to 76.0% (Fig. 2A; Table 3). Subsequently, ROC analysis was performed on the testing set (n = 70) to compare early-stage LUAD patients with cancer-free controls. We found that these three miRNAs had similar diagnostic abilities in the testing set and the training set, with AUC values of 0.644–0.741, sensitivities of 70.0–95.0%, and specificities of 56.7–76.7% (Fig. 2B; Table 3). Similar results were obtained by combining the training and testing sets(n = 210). When early-stage LUAD patients(n = 130) were compared with healthy controls(n = 80), the AUC values, sensitivities, and specificities of these there miRNAs in the combined group were 0.621–0.856, 86.2–94.6%, and 38.7–76.2%, respectively (Fig. 2C; Table 3). Overall, these data suggest that these three individual miRNAs are very promising specific markers to assist in the early diagnosis of cancer in the serum of NSCLC patients.Fig. 2ROC curve of miRNA-21, miRNA-210 and miRNA-942 detection and 3-miRNA detection for early diagnosis of lung adenocarcinoma:(Figure 2A: Training set; Figure 2B: Testing set; Figure 2C:Training set+Testing set)Table 2Univariate logistic regression analysis on the training and testing set for 3 individual MiRNAsmiRNATraining set (n = 140)Testing set (n = 70)Regressioncoefficient (β)p-valueExp(β)Regressioncoefficient (β)p-valueExp(β)miRNA-212.1540.0238.6181.6580.3455.248miRNA-2102.2920.0039.8983.1190.02022.634miRNA-942−0.4400.0440.644−1.1560.0210.315Table 3ROC analysis of differentially expressed Mirnas was performed in the training set, testing set and combination set. ROC curve analysis of Cyfra21-1 and Cyfra21-1 combined with 3-miRNA in early stage lung adenocarcinoma patientsmiRNAAUC95% CISensitivity(%)Specificity(%)Training set (TR)miRNA-210.8180.745–0.89278.974.0miRNA-2100.8710.812–0.93087.876.0miRNA-9420.6010.495–0.70795.636.03-miRNA0.9090.859–0.95987.886.0Testing set (TS)miRNA−210.7410.622–0.86095.056.7miRNA−2100.8230.723–0.92382.576.7miRNA−9420.6440.504–0.78470.063.33-miRNA0.8900.811–0.96985.083.3Two setscombined(TR + TS)miRNA-210.7900.728–0.85294.651.2miRNA-2100.8560.805–0.90786.276.2miRNA-9420.6210.536–0.70593.138.73-miRNA0.9010.858–0.94385.485.0Cyfra21-10.5870.511–0.66335.487.5Cyfra21-1 + 3-miRNA0.9150.877–0.95487.785.0 ROC curve of miRNA-21, miRNA-210 and miRNA-942 detection and 3-miRNA detection for early diagnosis of lung adenocarcinoma:(Figure 2A: Training set; Figure 2B: Testing set; Figure 2C:Training set+Testing set) Table 2Univariate logistic regression analysis on the training and testing set for 3 individual MiRNAsmiRNATraining set (n = 140)Testing set (n = 70)Regressioncoefficient (β)p-valueExp(β)Regressioncoefficient (β)p-valueExp(β)miRNA-212.1540.0238.6181.6580.3455.248miRNA-2102.2920.0039.8983.1190.02022.634miRNA-942−0.4400.0440.644−1.1560.0210.315 Univariate logistic regression analysis on the training and testing set for 3 individual MiRNAs Table 3ROC analysis of differentially expressed Mirnas was performed in the training set, testing set and combination set. ROC curve analysis of Cyfra21-1 and Cyfra21-1 combined with 3-miRNA in early stage lung adenocarcinoma patientsmiRNAAUC95% CISensitivity(%)Specificity(%)Training set (TR)miRNA-210.8180.745–0.89278.974.0miRNA-2100.8710.812–0.93087.876.0miRNA-9420.6010.495–0.70795.636.03-miRNA0.9090.859–0.95987.886.0Testing set (TS)miRNA−210.7410.622–0.86095.056.7miRNA−2100.8230.723–0.92382.576.7miRNA−9420.6440.504–0.78470.063.33-miRNA0.8900.811–0.96985.083.3Two setscombined(TR + TS)miRNA-210.7900.728–0.85294.651.2miRNA-2100.8560.805–0.90786.276.2miRNA-9420.6210.536–0.70593.138.73-miRNA0.9010.858–0.94385.485.0Cyfra21-10.5870.511–0.66335.487.5Cyfra21-1 + 3-miRNA0.9150.877–0.95487.785.0 ROC analysis of differentially expressed Mirnas was performed in the training set, testing set and combination set. ROC curve analysis of Cyfra21-1 and Cyfra21-1 combined with 3-miRNA in early stage lung adenocarcinoma patients Two sets combined (TR + TS) In order to improve the sensitivity and specificity of the early prediction of lung cancer, we proposed to combine these 3 miRNAs to evaluate their combined diagnostic efficiency. Logistic regression was performed on all 3 miRNAs in the training set. By combining the 3 miRNAs, we built a prediction model, In(p/1-p) = −1.165 + 2.154miRNA-21 + 2.292miRNA-210-0.440miRNA-942. ROC analysis based on the algorithm showed that the 3-miRNA panel in the training set produced an AUC value of 0.909 [95% confidence interval (CI):0.859–0.959] (Fig. 2A), which was significantly higher than the AUC values of the individual miRNAs (Table 3). When the Jordon index was used to define the optimal cut-off value (0.738), the accuracy, sensitivity, and specificity of the training set were 87.1%, 87.8%,86.0%, respectively (Table 3). By using the same logistic regression algorithm, we found that the 3-miRNA panel was equally effective in distinguishing between lung cancer patients and cancer-free controls in the testing set and the joint set. The corresponding AUC accuracy, sensitivity, and specificity were 0.890 [95% confidence interval (CI):0.811–0.969], 82.9%, 85.0%, and 83.3% (Fig. 2B; Table 3) for the testing set, and 0.901 [95% confidence interval (CI):0.858–0.943], 84.3%, 85.4%, and 85.0% for the combined set (Fig. 2C; Table 3). These results suggest that the combination of 3-miRNA assays is a significant and reliable tumor-specific serum marker for early diagnosis of NSCLC patients. Diagnostic evaluation of conventional tumor markers CEA and Cyfra21-1 in early-stage LUAD Since the conventional markers, CEA and Cyfra21-1, have a relatively high sensitivity for LUAD, we evaluated the value of their expression in early-stage LUAD. In this study, the expression of CEA in early-stage LUAD did not differ from that of healthy controls(P = 0.58), whereas the expression of Cyfra21-1 in patients with early-stage LUAD was significantly higher than that of healthy controls༈P = 0.034༉. Furthermore, ROC analysis showed that the AUC value of Cyfra21-1 for diagnosing early stage LUAD was 0.587, with a sensitivity of 35.4% and a specificity of 87.5%. When Cyfra21-1 was combined with the 3-miRNA panel to diagnose lung cancer, the AUC was 0.915, with a sensitivity of 87.7% and a specificity of 85.0% (Table 3). miRNA expression in subgroup stratification of different clinical features To explore the relationship between miRNA expression and each clinical feature, all 130 early-stage LUAD patients were categorized according to clinical features (gender, smoking history, nodule size, nodule density, differentiation degree, and stage). The relative expression of the three miRNAs in each subgroup of clinical features was analyzed. The results showed that miRNA-21 and miRNA-210 differed according to nodule size and stage, whereas other factors (gender, smoking history, nodule densityand degree of differentiation) were not associated with the relative expression of the three miRNAs(Table S2). Considering the difficulty in clinical judgment of pulmonary nodules less than 10 mm, We separately analyzed the expression of miRNAs in 53 patients with stage 0-IA1 LUAD with nodule diameter ≤ 10 mm. The results showed that the expression of miRNA-21, miRNA-210, and miRNA-942 were higher in the lung cancer group than in the healthy control group, and the difference was statistically significant (P < 0.05,Fig. 3A and Table S3). We carried out further ROC analysis on it and discovered that for pulmonary nodules with a diameter of ≤ 10 mm, the diagnostic capacity of miRNA-210 was superior to that of miRNA-21 and miRNA-942. The AUC of the 3-miRNA set was 0.884 (sensitivity and specificity of 83.0% and 87.5%, respectively) in the combined assay.(Fig. 4A; Table 4).Fig. 3The expression levels of serum miRNA-21, miRNA-210 and miRNA-942 in subtypes of nodules with a diameter of ≤10mm, nodule density and nodule differentiation. (A) The expression of all three miRNAs according to nodule diameter; (B) The expression of all three miRNAs according to nodule density; (C) The expression of all three miRNAs according to nodule differentiationFig. 4ROC curve of early diagnosis of miRNA-21, miRNA-210, miRNA-942 and 3-miRNA detection in the subgroups of lung nodule diameter less than 10mm, nodule density and nodule differentiation. (A) ROC curve for the early diagnosis of the 3-miRNA panel based on nodule diameter; (B) ROC curve for the early diagnosis of the 3-miRNA panel based on nodule density; (C) ROC curve for the early diagnosis of the 3-miRNA panel based on nodule differentiationTable 4ROC curve of early diagnosis of miRNA-21, miRNA-210, miRNA-942 and 3-miRNA detection in the subgroups of lung nodule diameter less than 10 mm, nodule density and nodule differentiation:NodulecharacteristicsmiRNAAUC95% CISensitivity(%)Specificity(%)d ≤ 10 mmmiRNA-210.7900.713–0.8670.9430.537miRNA-2100.8620.796–0.9270.8680.750miRNA-9420.6420.549–0.7360.9250.4253-miRNAs0.8840.821–0.9470.8300.875Solid nodulesmiRNA−210.8330.700−0.9670.9000.667miRNA−2100.8300.687–0.9730.9000.737miRNA−9420.7120.521–0.9021.0000.4673-miRNA0.9030.802−1.0000.9000.800Sub-solid nodulesmiRNA-210.7940.663–0.9240.9350.650miRNA-2100.8740.774–0.9740.9030.800miRNA-9420.7160.563–0.8690.8060.6503-miRNA0.8770.781–0.9740.8390.800Highly differentiated nodulesmiRNA-210.7440.637–0.8510.5400.867miRNA-2100.8120.718–0.9060.7800.800miRNA-9420.6330.497–0.7690.8800.4673-miRNA0.8320.743–0.9210.7400.900Moderately differentiated nodulesmiRNA-210.8310.749–0.9130.9260.650miRNA-2100.8810.815–0.9470.8970.775miRNA-9420.6140.495–0.9330.9710.3503-miRNA0.9440.904–0.9840.8820.875 The expression levels of serum miRNA-21, miRNA-210 and miRNA-942 in subtypes of nodules with a diameter of ≤10mm, nodule density and nodule differentiation. (A) The expression of all three miRNAs according to nodule diameter; (B) The expression of all three miRNAs according to nodule density; (C) The expression of all three miRNAs according to nodule differentiation ROC curve of early diagnosis of miRNA-21, miRNA-210, miRNA-942 and 3-miRNA detection in the subgroups of lung nodule diameter less than 10mm, nodule density and nodule differentiation. (A) ROC curve for the early diagnosis of the 3-miRNA panel based on nodule diameter; (B) ROC curve for the early diagnosis of the 3-miRNA panel based on nodule density; (C) ROC curve for the early diagnosis of the 3-miRNA panel based on nodule differentiation Table 4ROC curve of early diagnosis of miRNA-21, miRNA-210, miRNA-942 and 3-miRNA detection in the subgroups of lung nodule diameter less than 10 mm, nodule density and nodule differentiation:NodulecharacteristicsmiRNAAUC95% CISensitivity(%)Specificity(%)d ≤ 10 mmmiRNA-210.7900.713–0.8670.9430.537miRNA-2100.8620.796–0.9270.8680.750miRNA-9420.6420.549–0.7360.9250.4253-miRNAs0.8840.821–0.9470.8300.875Solid nodulesmiRNA−210.8330.700−0.9670.9000.667miRNA−2100.8300.687–0.9730.9000.737miRNA−9420.7120.521–0.9021.0000.4673-miRNA0.9030.802−1.0000.9000.800Sub-solid nodulesmiRNA-210.7940.663–0.9240.9350.650miRNA-2100.8740.774–0.9740.9030.800miRNA-9420.7160.563–0.8690.8060.6503-miRNA0.8770.781–0.9740.8390.800Highly differentiated nodulesmiRNA-210.7440.637–0.8510.5400.867miRNA-2100.8120.718–0.9060.7800.800miRNA-9420.6330.497–0.7690.8800.4673-miRNA0.8320.743–0.9210.7400.900Moderately differentiated nodulesmiRNA-210.8310.749–0.9130.9260.650miRNA-2100.8810.815–0.9470.8970.775miRNA-9420.6140.495–0.9330.9710.3503-miRNA0.9440.904–0.9840.8820.875 ROC curve of early diagnosis of miRNA-21, miRNA-210, miRNA-942 and 3-miRNA detection in the subgroups of lung nodule diameter less than 10 mm, nodule density and nodule differentiation: The 130 cases of early-stage LUAD were divided into six subgroups according to the imaging characteristics of lung nodules (solid, sub-solid, ground glass) and the degree of pathological differentiation (highly differentiated, moderately differentiated, poorly differentiated). Analysis revealed that the expression of all 3 miRNAs was significantly higher than that of healthy individuals in solid and subsolid nodules as well as in moderately differentiated and poorly differentiated nodules; in ground-glass nodules and highly differentiated nodules, only the expression of miRNA-21 and miRNA-210 were higher than that of healthy individuals, whereas the expression of miRNA-942 was not significantly different between lung cancer patients and healthy individuals (Fig.3B-3C and TableS2). Since the prediction model in this study was composed of the combination of miRNA-21, miRNA-210, and miRNA-942, only four subgroups of lung cancer patients in which the expression of miRNAs was significantly different from that of healthy people (solid nodules, sub-solid nodules, highly differentiated, and moderately differentiated nodules) were selected for the subsequent analysis. Univariate logistic regression analyses and ROC analyses were both performed for each miRNA between these four subgroups of lung cancer patients and healthy controls. It can be seen that all three miRNAs combined have the strongest diagnostic ability for solid nodules and moderately differentiated nodules. Similarly, by using the same logistic regression algorithm in the training set, we further evaluated the diagnostic ability of the 3-miRNA panel in these four subgroups, and the different diagnostic abilities of the 3-miRNA panel for the subgroups were found to be in agreement with the results for individual miRNAs, with solid nodules (AUC = 0.903; accuracy, sensitivity, and specificity of 85.7%, 90.0%, and 80.0%, respectively) higher than sub-solid nodules (AUC = 0.877; accuracy, sensitivity, and specificity of 80.4%, 83.9%, and 80.0%, respectively), and moderately differentiated nodules(AUC = 0.944; accuracy, sensitivity, and specificity of 87.0%, 88.2%, and 87.5%, respectively) higher than highly differentiated nodules(AUC = 0.832; accuracy, sensitivity, and specificity of 77.5%,74.0%, and 90.0%, respectively(Fig. 4B C and Table 4). Gene analysis One miRNA can target multiple genes and one gene can be regulated by multiple miRNAs. In order to understand the potential biological processes regulated by these three miRNAs (miR-21-3p, miR-210-3p, miR-942-5p), we first screened these three miRNAs through the websites miRWalk (http://129.206.7.150/) and mirDIP (http://ophid.utoronto.ca/mirDIP/) to identify the target genes of these three miRNAs, respectively, and then analyzed the biological processes and pathways of these genes through the Database for Annotation, Visualization and Integrated Discovery (DAVID) pathway website (https://www.example.com). We found that the target genes of all three miRNAs were involved in the regulation of many lung cancer-associated oncogenic pathways. MiR-21-3p was involved in JAK-STAT, P53, human cytomegalovirus infection, and Kaposi’s sarcoma-associated herpesvirus infection in the map. miR-942-5p was involved in MAPK, P53, Ras, ErbB, PI3K-Akt, Hippo signaling pathways and cellular cytosolic signaling pathways(Fig. 4). MiR-210-3p was involved in transcriptional deregulation and cAMP signaling pathways in cancer(Fig. 5), which are consistent with the characteristics of the cancer-specific biomarkers in early-stage LUAD in our current study. In addition, we found that the most regulated molecular biological function of miR-21-3p, miR-210-3p, and miR-942-5p was protein binding (Fig.S1A-S1C). MiR-942-5p is not only involved in composing the actin cytoskeleton, but also plays an important role in the development and differentiation of the nervous system (Fig.S1C).Fig. 5KEGG pathway analysis (A:miR21-3p KEGG pathway analysis; B:miR210-3p KEGG pathway analysis; C:miR942-5p KEGG pathway analysis) KEGG pathway analysis (A:miR21-3p KEGG pathway analysis; B:miR210-3p KEGG pathway analysis; C:miR942-5p KEGG pathway analysis) Discussion LUAD has a rich blood vessel supply and secretory function, and has a higher tendency to metastasize [34], so early diagnosis is crucial for improving its survival. An increasing number of miRNAs have been identified to be strongly associated with LUAD. Rani et al. first identified a group of six miRNAs with significantly higher levels than healthy controls after comparing serum samples from 40 patients with LUAD (comprising stages I, II, III, and IV) with those of healthy subjects (miR-30c-1, miR-616, miR-146b-3p, miR-566, miR-550, miR-939), while the levels of miR-339-5p and miR-656 were much lower compared with the control group [35]. The differences in miRNA profiles provide further evidence that serum miRNAs have the potential to be used as diagnostic biomarkers for LUAD, however, due to the limited sample size, this study did not further explore the predictive ability of serum miRNA for early stages of the disease. Another study found in the plasma miRNAs of 75 lung cancer patients (stage II-IV, including 18 cases of LUAD) and the healthy control group that miR-19b and miR-183 could be used to identify lung cancer. Especially miR-183 was more effective in distinguishing LUAD from healthy individuals, with an AUC of 0.924, a sensitivity of 100%, and a specificity of 81%. In addition, the combined AUC value of the two was 0.990, and the sensitivity and specificity were 94.7% and 95.2% respectively. They were regarded as strong predictors of lung cancer [36]. Similarly, Sun et al. verified that in 28 cases of LUAD (IA-IIIB) miR-21 could distinguish LUAD patients from healthy controls with high AUCs (0.900 and 0.880, respectively), sensitivity (0.821 and 0.821, respectively), and specificity (0.929 and 0.964, respectively). When miR-21 was detected in combination with miR-339-5p, the AUC value was significantly higher (0.963), sensitivity (0.929) and specificity (0.929) [37]. However, the sample groups in the aforementioned studies contained relatively fewer patients with early-stage LUAD. In the present study—including 130 stage I–II LUADs and 80 healthy controls—we observed that a 3-miRNA panel (miRNA-21/miRNA-210/miRNA-942) achieved high diagnostic accuracy across training and testing sets, with high AUC values (0.890–0.909), sensitivity (85.0% −87.8%), and specificity (83.3% − 86.0%), and were significantly superior to the individual detection of miRNA-21, miRNA-210, and miRNA-942 (Fig. 2). Importantly, our intended use is not population screening but preoperative auxiliary diagnosis/triage among patients who are already under clinical evaluation for CT-suspected malignant nodules. This positioning reflects our cohort structure (all cancers surgically confirmed; limited benign resections driven by preoperative suspicion) and is aligned with real-world decision points where a blood-based adjunct may help reduce unnecessary procedures. Notably, most LUADs were very early stage (stage ≤ I for the vast majority), and imaging often showed subsolid or GGO features that are diagnostically challenging. Accordingly, we interpret our panel as a candidate adjunct for preoperative risk stratification of indeterminate nodules rather than a stand-alone screening assay.This is consistent with many studies. Gao et al. conducted a multicenter cohort analysis and finally developed a 4-miRNA combination consisting of miR-450b-5p, miR-3615, miR-106b-3p, and miR-125a-5p, secreted by extracellular vesicles (EVs); this combined assay had a good diagnostic performance for early-stage LUAD. It can even separate AIS and MIA patients from non-cancer control areas, with AUC values of 0.846 (95% CI: 0.764 to 0.944) and 0.92 (95% CI: 0.86 to 0.978), respectively [38]. Compared with prior miRNA panels and meta-analyses, our observed AUCs are broadly comparable; however, cross-study benchmark should be interpreted cautiously given platform, population, and modeling differences. In this study, we found that Cyfra21-1 expression was up-regulated in patients with early stage LUAD, which is consistent with the results of some studies in China and abroad [39]. The CEA expression level, however, was not significantly different from that of the healthy population, which could be attributed to the fact that the rate of increase and concentration level of CEA increase with the development of the tumor, and its diagnostic sensitivity is related to the tumor stage. Cyfra21-1 alone showed only modest discrimination in our cohort, and combining it with the 3-miRNA panel did not materially outperform the miRNA panel alone. This might be associated to some extent with differences in sample size, nodule size and stage, as well as variations in detection reagents and detection methods. When analyzed for clinicopathological factor correlations, miRNA-21 and miRNA-210 differed between nodule size, nodule density, and stage, whereas the three miRNAs were not associated with other clinical correlates (gender, smoking history). Hetta et al. found that miR-21, miR-210, miR-150, miR-192, miR-197, and miR-320 were more highly expressed in lung cancer patients who smoked [40]. While in this study, most of the LUAD group were non-smoking women, we did not observe any correlations with age, sex, or smoking history (Table S2), although miRNA-21 and miRNA-210 varied with nodule size and pathological stage, consistent with biological plausibility. The prognosis for patients with LUAD is closely related to nodal density. Solid nodules have a higher content of tumor cells than sub-solid or ground-glass nodules. It has been shown that the higher the content of solid component, the more pronounced the lymphatic infiltration of NSCLC, and a consolidation-to-tumor ratio(CTR) >75% is an independent prognostic factor for disease-free survival (DFS) [41]. Pure solid nodules confer a worse prognosis than partially solid nodules of the same nodule diameter with ground glass opacity (GGO) [42]. In subgroup analyses, our 3-miRNA combination performed best in solid nodules and in moderately differentiated tumors, and retained good performance for very small lesions (≤ 10 mm). These findings suggest potential utility for triaging patients when imaging alone is indeterminate. It is essential to identify genetic biomarkers and regulatory factors as well as signaling pathways, as these can reveal the molecular mechanisms of lung adenocarcinogenesis and progression, and have guiding significance for targeted therapy, drug resistance studies, and prognostic assessment of lung cancer. DAVID pathway analysis showed that all three miRNAs were involved in the regulation of many lung cancer-related signaling pathways, such as JAK-STAT, P53, MAPK, Ras, ErbB, PI3K-Akt, Hippo, etc. (Fig. 3). Studies have shown that overexpression of miR-21 can lead to invasion, metastasis, and gemcitabine resistance of lung cancer [43], and can also activate the PI3K/AKT signaling pathway and increase the cisplatin resistance of LUAD A549 cells, therefore, reduction of the expression in combination with cisplatin chemotherapy may be effective in improving therapeutic efficacy in patients with NSCLC [44]. MiR-210 can elevate the expression of some pro-angiogenic factors by activating the JAK2/STAT3 signaling pathway [45]. In addition, miR-210 secreted by cancer-associated fibroblasts can promote epithelial-mesenchymal transition by targeting the target gene UPF1 and activating the PTEN/PI3K/AKT pathway, which promotes NSCLC migration and invasion [46]. Both in-vivo and ex-vivo studies have demonstrated that circFBXW7 attenuates malignant progression of LUAD via miR-942-5p, which represents a novel mechanism of cancer progression and brings the study of potential therapeutic targets and prognostic biomarker exploration of LUAD a step closer [47]. As our pathway analysis relied on database-based enrichment rather than experimental validation, these signals should be regarded as hypothesis-generating and will require orthogonal confirmation in future studies. However, there are some limitations to this study.This single-center study included only 19 benign resections, all of which were preoperatively suspected as malignant; thus, specificity against a broader spectrum of benign mimics (e.g., granulomas, hamartomas) remains to be fully established. The cohort comprises East Asian patients; external validation in multicenter, geographically diverse populations is required. We excluded clinician-diagnosed COPD and recent respiratory infection (≤ 4 weeks) to reduce inflammatory confounding, and controls had negative CT; nonetheless, residual or subclinical inflammation cannot be entirely ruled out. Furthermore, it should be noted that our study was designed as a diagnostic accuracy evaluation of circulating miRNAs rather than an imaging–biomarker integration study. Healthy controls had negative chest CT and benign resections were limited. Under this structure, nodule-level CT variables are not available for controls and the benign sample size is insufficient to develop and validate integrated models. Future prospective cohorts of CT-indeterminate nodules should quantify incremental value using nested models (CT features ± miRNAs), DeLong tests, NRI/IDI, calibration, and decision-curve analysis to define actionable clinical benefit. While U6 was used as a commonly adopted endogenous reference for serum miRNA RT-qPCR, alternative normalization strategies (e.g., exogenous spike-ins or endogenous miR-16) merit evaluation in future validations. Conclusions The combination of serum miRNA-21, miRNA-210, and miRNA-942 demonstrates robust diagnostic performance for early-stage LUAD in a surgically confirmed cohort and may serve as a preoperative adjunct to aid the triage of CT-indeterminate pulmonary nodules, including small lesions (≤ 10 mm). These findings support further prospective, multicenter studies integrating imaging and biomarker features, with external validation and health-economic modeling to define clinical utility. Supplementary Information Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2.