A Systematic Review and Meta-Analysis of a Liquid Biopsy-Based Multimodal Approach for the Diagnosis of Pulmonary Nodules

By Ren Shuang Cao¹, Tian Jiao Huang², Zhi Heng Lin³, Fan Yang⁴, Fei Liu¹, Feng Gao¹

Affiliations

1. Department of Respiratory, Wangjing Hospital of China Academy of Chinese Medical Sciences, Beijing, China
2. The First School of Clinical Medicine, Heilongjiang University of Traditional Chinese Medicine, Harbin, China
3. College of Traditional Chinese Medicine, Shanghai University of Traditional Chinese Medicine, Shanghai, China
4. College of Traditional Chinese Medicine, Beijing University of Chinese Medicine, Beijing, China

doi: 10.29271/jcpsp.2026.08.1054

ABSTRACT
Liquid biopsy demonstrates clinical utility as a non-invasive tool for discriminating malignant tumours from benign pulmonary nodules, as evidenced by recent research. This systematic review compared the diagnostic performance of current clinical modalities for pulmonary nodules, aiming to generate evidence-based solutions to address the underdiagnosis of malignant nodules and the overdiagnosis of benign lesions. A Bayesian bivariate random-effects model was employed to perform meta-analysis across diagnostic categories, followed by meta-regression and subgroup analyses. Results were visualised using summary receiver operating characteristic curves. Subgroup analysis revealed that liquid biopsy-derived multimodal diagnostic models achieved superior performance (sensitivity: 90.2%; specificity: 83.6%) compared to CT-based evaluation (81.5%, 67.9%) and risk prediction models (68.7%, 70.6%). These findings indicate that multimodal models integrating liquid biopsy, radiological features, and risk prediction significantly outperform single-modality approaches, thereby potentially addressing diagnostic imbalances in pulmonary nodule assessment.

Key Words: Liquid biopsy, Pulmonary nodules, Multimodal diagnostic model, Comparative diagnostic test accuracy studies.

INTRODUCTION

A pulmonary nodule is a discrete radiographic entity measuring less than 3 centimetres,  manifesting as a focal, discernible density surrounded by the intricate landscape of lung tissue. Around 1.6 million people in the US are diagnosed with pulmonary nodules each year, which are detected in approximately 30% of computed tomographic (CT) chest images. According to the National Lung Screening Trial (NLST), 24% of screened patients had a concerning pulmonary nodule; however, only 4% of those nodules were ultimately determined to be malignant, even in this high-risk population.¹ Although approximately 96% of all pulmonary nodules identified in the NLST among people at high risk of lung cancer who underwent low-dose computed tomography (LDCT) were benign, recent studies have reported that the overall probability of malignancy for incidental pulmonary nodules measuring 7–10 mm in diameter is 1.7%.
 

However, among individuals with risk factors for lung cancer, the probability of malignancy may increase to approximately 10% for a lung nodule measuring 7 mm in diameter that exhibits associated imaging features such as irregular or spiny margins or an upper lobe location.² Similarly, for pulmonary nodules that initially exhibit the high-risk features described in the guidelines, the risk of malignancy can be underestimated due to insufficient evidence if the nodules have a high probability of malignancy.

The evaluation of a larger solid pulmonary nodule (8-30 mm) should begin with an estimate of the probability that the nodule is malignant. The current management paradigm for pulmonary nodules >8 mm is based on pretest probability of malignancy. Nodules with intermediate risk (5-65%), which account for almost half of pulmonary nodules identified by chest CT, require further diagnostic evaluation. This may include additional imaging, bronchoscopy, percutaneous biopsy, or surgi- cal biopsy. Even minimally invasive procedures carry significant risks and anxiety for patients, and the cost of diagnostic evaluation increases 28-fold when a biopsy is performed.^3,4 Patients with intermediate-risk pulmonary nodules require additional risk-stratification tools to identify candidates who need aggressive evaluation. Current malignancy probability estimation relies on clinician-radiologist experience or validated prediction models, which frequently lead to overesti- mation or underestimation.

Existing models depend on patient-specific data with limited population representativeness and incorporate only a limited number of variables, including lung cancer risk factors (such as age, smoking history, personal/family history of cancer, and the presence of COPD) and nodule charashuangc-teristics associated with malignancy (including larger size, upper-lobe location, partially solid density, irregular margins, fewer total nodules, increased fluorodeoxyglucose-positron emission tomography (FDG-PET) uptake, and rapid volume doubling). Malignancy probability in 8-30 mm solid nodules ranges from <1% to >70%, varying with patient risk factors and imaging features (such as size, location, margin charac-teristics [smooth, irregular, lobulated, spiculated], and the presence of calcification). Optimal pulmonary nodule assess- ment must rapidly identify malignancies to expedite treatment while minimising unnecessary investigations for benign cases, balancing timely intervention with avoidance of excessive diagnostics.

Recent research has increasingly focused on liquid biopsies for diagnosing malignant pulmonary nodules. These analyses detect tumour-derived components released from lung cancer or precancerous lesions, including circulating tumour cells (CTCs), circulating genetically abnormal cells (CACs), cell-free DNA (cfDNA), and methylated circulating tumour DNA (ctDNA). Numerous lung cancer biomarkers are in development, with ctDNA methylation showing promising performance in initial clinical validation. Liquid biopsies offer non-invasive, reproducible testing that may integrate with existing and emerging risk stratification profiles to categorise intermediate-risk pulmonary nodules, potentially reducing unnecessary investigations, patient anxiety, and healthcare costs. However, comprehensive systematic reviews of compa-rative diagnostic test accuracy (DTA) studies for pulmonary nodules remain scarce. This systematic review and meta-analysis evaluated the diagnostic efficacy and application characteristics of pulmonary nodule screening modalities, including liquid biopsies, clinician-radiologist expertise, and validated risk prediction models. The study aimed to provide evidence-based guidance for balancing concerns about underdiagnosis with the risk of overdiagnosis in lung cancer screening.

METHODOLOGY

This systematic review was conducted in accordance with the Preferred Reporting Items for a Systematic Review and Meta-analysis of Diagnostic Test Accuracy Studies (PRISMA-DTA) statement.⁵ The study protocol was registered with the International Prospective Register of Systematic Review PROSPERO (CRD42023464553). The scope of the database search and the process of literature selection, including study inclusion and exclusion, are shown in the flowchart (Figure 1). The Quality Assessment of Diagnostic Accuracy Studies-Comparative (QUADAS-C) tool was used to assess the risk of bias in comparative DTA studies.

Figure 1: Study flowchart.

A comprehensive search was performed of the bibliographic databases PubMed, Embase, the Cochrane Library, and Web of Science up to 1^st October 2023. Search terms included controlled terms (such as MeSH in PubMed and Emtree in Embase) and free-text terms. The following terms were used (including synonyms and closely related words) as index terms or free-text words: pulmonary nodule, liquid biopsies and diagnostic techniques, and the respiratory system. A search filter was used to limit the results to humans and adults. Clinical Trials Registries were included in the search, as shown in Figure 1. Conference abstracts were included to identify models that were published in full text elsewhere but were not reviewed here. Title summarisation, full-text screening, and reference checking were independently carried out by two review authors (ZHL and FY). Discre-pancies were resolved by an independent data scientist (RSC), and the reasons for article exclusion were recorded. TJH extracted the data, and RSC verified the data. Both authors reviewed and resolved any remaining discrepancies to ensure the accuracy of the database.

The study selection used original diagnostic studies eva- luating liquid biopsy and imaging models for pulmonary nodules and provided detailed reporting of biomarkers, including CTCs, CACs, cfDNA,  ctDNA, multimodal models (combining liquid biopsies with clinical models), LDCT, CT, positron emission tomography-computed tomography (PET-CT), and radiologic-clinical feature models. Eligible studies were also required to report patient baseline characteristics, pathological diagnoses, diagnostic performance metrics (including 2×2 contingency tables), and diagnostic value indicators. Two independent reviewers (FY and BW) piloted QUADAS-C assessments to measure inter-rater agreement. This tool evaluated bias risk in comparative DTA studies through 14 items categorised into four domains: patient selection, index tests, reference standard, and flow/timing. Overall, the risk of bias for a study was classified as high risk if one or more domains were assessed as high risk, and as unclear if any domain received an unclear assessment.

This study followed the guidance in the Cochrane Handbook and recommended the use of hierarchical models,⁶ Bayesian methods, and a combination of the bivariate model and the hierarchical summary receiver operating characteristic (HSROC) model. The HSROC model was used to perform the DTA meta-analysis and was implemented using the Bayesian model-fitting software Stan. For the diagnostic meta-analysis, using R/RStudio and the lme4 package, the confusion matrix values for the research queue (TP/FN/TN/FP) were used to build a bivariate random-effects model. For the pooled logit sensitivity and logit specificity, a normal distribution with a mean of zero and a standard deviation (SD) of 1.5 (N (0, 1.5)) was used, which is equivalent to a 95% prior interval (the interval formed by the 2.5% and 97.5% quantiles of the prior distribution) ranging from 0.05 to 0.95 on the probability scale. For the between-study SDs, a zero-truncated normal with a mean of zero and a unit SD (N ≥₀[0, 1]) was used. Bivariate models were used for subgroup analyses and meta-regression. Categorical meta-regression was then performed to compare the diagnostic accuracy of the different pulmonary nodule assessment moda- lities using estimates derived from each collected or reconstructed 2 × 2 contingency table. A random-effects model with hierarchical random-effects was used to obtain summary estimates of sensitivity and specificity for studies that used the same reference standard and the same definition of pulmonary nodules. The HSROC plots were generated to present the pooled estimates, 95% confidence intervals (CIs), and pre-diction regions for each detection modality, along with the sensitivity and specificity estimates from each study, weighted according to study size. Study weights—the amount each study contributes to the overall sensitivity and specificity estimates—were calculated using a method based on the decomposition of Fisher’s information matrix that generalises to bivariate meta-regression.⁷

RESULTS

A total of 832 studies with English titles or abstracts were retrieved. Only fully matched/randomised controlled trials were considered. Ultimately, 13 studies analysing the diagnostic precision of pulmonary nodules were incorporated (Figure 1; Table I).^8-20 The evaluation of bias risk in compa-rative DTA studies was undertaken via QUADAS-C. The diagnostic accuracy reported in the included studies was assessed independently by two reviewers. The two reviewers independently assessed inter-reviewer agreement and resolved any discrepancies through discussion or, if necessary, by consulting a third reviewer. All studies adhered to identical inclusion criteria and utilised two or more methods to forecast the probability of malignancy in pulmonary nodules. Across all included studies, the number of patients included in the diagnostic accuracy analyses ranged from 18 to 728. The study populations comprised adult men and women. The comparative DTA studies that met the network evidence review (NER) criteria included a variety of pulmonary nodule malignancy risk assessment modalities and were categorised according to the detection or prediction modality: CT and PET-CT-based pulmonary nodule risk interpretation (7 studies, 53.8%); clinical model (risk prediction; 6 studies, 46.2%); ctDNA methylation (6 studies, 46.2%); CACs/CTCs (5 studies, 38.5%); combined clinical, imaging, and liquid biopsy biomarkers models were defined as mixed models (6 studies, 46.2%). Assays involving liquid biopsy biomarkers were defined as liquid biopsy (LB)-based models (11 studies, 84.6%), including ctDNA methylation, CACs/CTCs, and mixed models.

Table I: Characteristics of the included studies.
 

Study	Diagnostic method	Simple	TP	FN	TN	FP	AUC	Youden index	Sensitivity	Specificity
He et al.8	PulmoSeek Plus (combined model)	541	414	9	59	59	0.9	0.48	0.88	98
	PulmoSeek (ctDNA methylation)	541	393	30	59	59	0.85	0.43	83.5	0.93
	Clinical model (CIBM)	541	389	34	59	59	0.85	0.42	82.8	0.92
Liang et al.9	PulmoSeek (ctDNA methylation)	220	155	5	25	35	0.84	0.39	81.8	92.9
	Clinic model (Mayo)	140	13	87	37	3	0.59	0.06	35.7	92
Gilbert et al.14	PET- CT	312	139	52	99	22	0.77	0.55	76.3	96.9
	DCE-CT	312	182	9	36	85	0.62	0.251	69.9	13
Yi et al.20	CT	119	64	15	37	3	0.68	0.74	84.9	72.8
	PET/CT	119	76	3	35	5	0.71	0.84	93.2	95.3
Liang et al.16	Mayo Clinic model	35	4	23	110	8	0.67	0.08	78.6	81
	ctDNA methylation (tissue)	58	33	4	17	4	0.75	0.7	86.2	96.2
	ctDNA methylation (plasma)	66	31	8	23	4	0.72	0.65	81.8	14.8
Tahvilian et al.19	LungLB™ (CAC)	151	86	26	28	11	0.78	0.52	75.5	89.2
	Mayo Clinic model	151	26	86	35	4	0.52	0.13	40.4	79.5
	PET-CT	79	26	13	18	22	0.57	0.12	55.7	66.7
Zeng et al.10	ctDNA (balf)	48	22	5	17	4	0.82	0.63	81.2	81.5
	ctDNA (plasma)	48	18	9	15	6	0.68	0.38	68.8	66.7
Xing et al.18	Clinic model (Mayo)	100	47	10	39	4	0.83	0.73	86	82.5
	ctDNA methylation	100	50	7	39	4	0.91	0.78	89	87.7
	Mixed model (ctDNA and clinic)	100	51	6	41	2	0.95	0.85	92	89.5
Ye et al.11	AI-CT	727	391	138	129	69	0.74	0.39	71.5	73.9
	CAC	727	414	115	145	53	0.77	0.52	76.9	78.3
	Mixed model (CAC and clinic)	728	475	55	161	37	0.88	0.71	87.4	89.6
Zhang et al.13	CTC	120	76	13	26	5	0.84	0.69	85	85.4
	Mixed model (CTC and clinic)	120	80	9	26	5	0.92	0.74	88.3	89.9
Zhong et al.12	CTC	18	9	0	6	3	NA	0.75	83.3	75
	Mixed model (CTCs and CA125)	18	10	0	6	2	NA	0.83	88.9	83.3
Xu et al.15	Mixed model (CAC and clinic)	47	22	4	20	1	0.96	0.89	79.8	0.85
	CAC	47	19	7	17	4	0.77	0.54	76.6	73.1
	CT	47	23	3	13	8	0.77	0.5	76.6	88.5
Liu et al.17	ctDNA methylation	29	12	5	9	3	0.72	0.44	72.4	80
	Mixed model (ctDNA and clinic)	29	12	3	12	2	0.86	0.66	82.8	80
	PET-CT	61	45	5	2	9	0.65	0.08	77.05	90
TP: True positive;  FN: False negative; TN: True negative; FP: False positive; AUC: Area under the receiver operating characteristic (ROC) curve.

Table II: Summary of diagnostic performance for assessment models.

Model	Sensitivity	Specificity	LR+ (95% CI)	LR− (95% CI)	DOR (95% CI)
CACs/CTCs	0.723 (0.555-0.839)	0.805 (0.644-0.903)	3.65 (2.02-7.44)	0.35 (0.20-0.57)	10.8 (4.0-28.3)
Clinical model	0.687 (0.528-0.814)	0.706 (0.549-0.834)	2.33 (1.50-3.98)	0.45 (0.27-0.69)	5.32 (2.23-12.5)
CT-based evaluation	0.815 (0.720-0.890)	0.679 (0.518-0.801)	2.52 (1.70-3.95)	0.28 (0.17-0.42)	9.34 (4.44-18.8)
ctDNA	0.840 (0.737-0.901)	0.772 (0.633-0.869)	3.66 (2.29-6.25)	0.21 (0.13-0.33)	17.6 (8.32-38.7)
LB-based	0.824 (0.785-0.889)	0.810 (0.732-0.868)	4.45 (3.20-6.28)	0.19 (0.14-0.26)	23.0 (14.0-37.2)
Mixed model	0.902 (0.826-0.949)	0.836 (0.704-0.914)	5.50 (3.08-10.2)	0.12 (0.06-0.21)	46.8 (19.5-111)
LR+: Positive likelihood ratio; LR−: Negative likelihood ratio; DOR: Diagnostic odds ratio; CI: Credible interval; LB: Liquid biopsy. Data are presented as posterior median (95% credible interval).

Figure 2: Meta-analysis of SROC model: (A) CT-based evaluation; (B) Clinic model; (C) CACs/CTCs; (D) mixed model; (E) ctDNA methylation;
(F) LB-based model. Figure 3: Subgroup analysis of the HSROC model. (A) Meta-regression of screening modalities; (B) Subgroup analysis of diagnostic performance of screening modalities; (C) Accuracy of screening modalities.

The sensitivity and specificity ranges for CT and PET-CT were as follows: 71.5-84.9%, 61.9-92.5%, 66.7-96.2%, and 18.2-87.5%, respectively. The summary receiver operating characteristic (SROC) revealed the pooled 95% CIs and prediction regions, and the diagnostic odds ratio (DOR) was 9.38 (2.94, 28.55). Nonetheless, the specificity of CT-based evaluation remained unstable, even after calculating the study weights (Figure 2A). Additionally, the forest plot illustrated significant differences in specificity across studies. CT demonstrated solid reliability with an accuracy range of 71.5-84.9%, compared with PET-CT, which displayed greater variability with an accuracy range of 55.7-93.2%. Additionally, the Youden index for CT and PET-CT ranged from 0.39 to 0.74 in diagnostic accuracy evaluations. However, variations of 0.39-0.74 and 0.08-0.84 indicated significant instability when diagnosing malignant pulmonary nodules.

Similar to CT-based evaluation, the clinical model demons-trated unstable diagnostic accuracy for the diagnosis of malignant pulmonary nodules. Additionally, the sensitivity of clinical models was concerning. According to the results, the Mayo Clinic model demonstrated sensitivity ranging from 13.0% to 82.5%, specificity ranging from 89.7% to 93.2%, and a Youden index ranging from 0.06 to 0.73. The sensitivity, specificity, and Youden index of the CIBM and AI models were 92%, 50%, 0.42, and 73.9%, 65.2%, 0.39, respectively. The forest plot revealed substantial variation in the sensitivity of the three clinical models, which was further explored by subsequent meta-regression analysis of the clinical models (Figure 2B). The AUC of the clinical model, particularly the Mayo Clinic model, remained relatively stable (0.52-0.83). This stability can be attributed to the model's well-performing and consistent specificity. Addi-tionally, SROC, which incorporates study weights, indicates that the pooled estimates, 95% reliability, and prediction region of the clinic model were highly dispersed. It is important to note that Figure 2B, including study weights, indicates significant dispersion in the pooled estimates, 95% reliability, and prediction regions of the clinic model.

CACs/CTCs demonstrated higher diagnostic sensitivity, ranging from 73.1% to 78.3% and 75.0% to 85.4%, respectively, with corresponding Youden indices ranging from 0.52 to 0.54 and 0.69 to 0.83. CACs/CTCs also showed consistently good diagnostic performance, with accuracy exceeding 75% across studies. The SROC plot showed that the pooled estimates, 95% confidence region, and prediction region for CACs/CTCs were relatively concentrated, particularly for sensitivity (73.1-85.4%), whereas specificity was concentrated between 71.8% and 100% (Figure 2C). The pooled estimates of sensitivity and false-positive rate were 0.78 (95% CI: 0.75-0.81) and 0.25 (95% CI: 0.21-0.31), respectively, with no significant heterogeneity observed.

The ctDNA methylation assay demonstrated strong diag-nostic performance, with serum samples showing a sensi-tivity range of 66.7-96.9% and a Youden index of 0.39–0.78, while bronchoalveolar lavage fluid (BALF) samples exhibited a sensitivity of 81.5-89.2% and a Youden index of 0.63–0.70. Specificity for serum samples varied widely, whereas BALF samples showed consistent specificity (Table I). Notably, the three-gene ctDNA methylation panel in serum achieved high diagnostic accuracy, with sensitivity, specificity, and Youden index of 87.7%, 90.7%, and 0.78, respectively, along with an AUC of 0.91 and a diagnostic OR of 69.6, indicating superior diagnostic efficacy. The forest plot indicated an overall sensitivity of 0.86 (95% CI: 0.76-0.92) and a false positive rate of 0.26 (95% CI: 0.15-0.41). The SROC curve presented the pooled estimate, 95% credible interval, and prediction region after incorporating study weights, demonstrating greater stability in sensitivity than in specificity, with consi-stently high overall sensitivity (Figure 2E).

As a combined model incorporating clinical, imaging, and liquid biopsy biomarkers, the mixed model achieved a DOR of 46.8 (95% CI: 19.5-111), reflecting greater sensitivity of liquid biopsy and high specificity of the clinical model and CT-based evaluation (Table II). The HSROC curve indicates that the 95% confidence interval and prediction regions of the weighted mixed model are near the upper left corner, demonstrating strong diagnostic accuracy in both sensitivity and specificity (Figure 2D). Furthermore, integrating the clinical model with ctDNA methylation and CAC/CTC improved the specificity of these biomarkers to varying degrees (Figure 2E). The mixed model showed promising performance in diagnosing malignant pulmonary nodules, with an AUC ranging from 0.86 to 0.96 and a Youden index ranging from 0.48 to 0.85. The HSROC analysis also showed that the liquid biopsy-based model—defined as an assay using liquid biopsy biomarkers—had a tightly clustered 95% confidence and prediction region, with good overall sensi-tivity (0.82) and specificity (0.81; Figure 2F; Table II).

Meta-regression analyses were performed for each of the six assay categories, illustrating sensitivity, specificity, corres-ponding 95% confidence intervals, and prediction regions (Figure 3A). The accuracy-by-subgroup plot provided a visual comparison of the pooled diagnostic performance across subgroups, highlighting variations in sensitivity and specificity (Figure 3C). In the subgroup analysis, the mixed model demonstrated the highest sensitivity (90.2%) and specificity (83.6%), with the 95% credible and prediction regions confirming robust diagnostic accuracy (Figure 3B). This suggests that the mixed model effectively combines the high sensitivity of liquid biopsy approaches with the stable specificity of clinical models, particularly the Mayo Clinic model. The HSROC analysis further confirmed that the weighted mixed model achieved the best overall diagnostic performance (Figure 3B). Among liquid biopsy-based assays, ctDNA methylation and CAC/CTC showed higher specificity than CT-based evaluation, which is known for its high false positive rate. In terms of sensitivity, ctDNA methylation outperformed CT-based evaluation (84.0% vs. 81.5%, respectively), whereas CAC/CTC exhibited lower sensitivity (72.3%; Table II). The HSROC curves indicated that ctDNA methylation and liquid biopsy-based models demonstrated substantially better diagnostic efficacy than CT-based evalu-ation for the diagnosis of malignant pulmonary nodules; in contrast, CAC/CTC demonstrated comparable performance to CT-based evaluation due to limited sensitivity. Although the clinical model demonstrated relatively high specificity (70.6%), its poor sensitivity resulted in the lowest overall diagnostic performance among the methods assessed (Figure 3B).

DISCUSSION

This study investigated the diagnostic efficacy of different testing modalities for detecting benign/malignant lung nodules. Although earlier studies demonstrated the significant potential of liquid biopsy for specific lung cancer staging and post-chemotherapy prognostic assessment, no study explicitly discussed the diagnostic efficacy of liquid biopsy and its derived multimodal lung cancer diagnostic models for assessing benign/malignant lung nodules or compared them with routinely recommended CT and validated risk prediction models. Meta-analysis of various testing modalities indicated that liquid biopsy and multimodal diagnostic models demonstrated superior diagnostic efficacy for benign/malignant pulmonary nodules compared to CT or clinical risk prediction models alone. To summarise, CT-based evaluations exhibit a high false-positive rate, as evidenced by recently published guidelines. Consequently, longer surveillance intervals (e.g., 6 or 12 months) are now recommended instead of 3 months to address concerns regarding over-medication, increased patient costs, and radiation exposure risks. A recent large-scale study reported a consistent association between CT radiation dose and the risk of haematological malignancies, including lymphoid and myeloid malignancies and acute leukaemia (AL), even at low doses (10–15 mGy).²¹ Further-more, FDG-PET imaging, bronchoscopy, and transthoracic needle biopsy were considered unlikely to improve the detection of malignancy in patients with small solid pulmonary nodules.²² Nodules <8 mm have lower spatial resolution on FDG-PET imaging, posing challenges for localisation during bronchoscopy or transthoracic needle biopsy. The higher radiation dose and increased healthcare costs associated with PET-CT further complicate risk–benefit assessments in the management of pulmonary nodules. Differentiating inflammatory from malignant lesions remains difficult, and the low metabolic activity of indolent malignant tumours increases the likelihood of false-negative PET-CT results in patients with pulmonary nodules. A high-quality meta-analysis assessing the discriminatory accuracy of PET imaging for pulmonary nodules reported a pooled sensitivity of 89% (95% CI, 86-91%) and specificity of 75% (95% CI, 71-79%), with notable heterogeneity in specificity and a 16% reduction in specificity observed in regions with high pre-valence of fungal infections,²³ which partly also suggests that CT assessment of pulmonary nodules had significant shortcomings in controlling for false-positive and false-negative rates. It is worth noting that a multivariate analysis revealed that inappropriate radiologist recommendations were the most significant predictor of guideline-inconsistent care. These findings suggest that CT imaging assessment of pulmonary nodules based on radiologist interpretation introduces some degree of uncertainty into pulmonary nodule malignancy risk prediction.²⁴

Clinical models estimate the probability of pulmonary nodule malignancy by combining CT imaging features with clinical risk factors, including advanced age, a high smoking index, and a family history of lung cancer. Although clinical models have relatively poor sensitivity, they partially compensate for the low specificity of CT imaging features alone in pre-dicting malignancy. This may explain why clinical models, particularly the Mayo Clinic model, can be effectively com-bined with advanced liquid biopsy techniques to improve the diagnostic efficacy of benign/malignant pulmonary nodules. Such integration may also help reduce the false-positive rate associated with CT-based evaluation. Although validated risk prediction models provide a standardised approach to the assessment and management of pulmonary nodules that can assist in clinical decision-making, these models rely heavily on patient-specific data, which may not always be available or representative of the majority of patients. It is difficult to incorporate subgroup types into the calculation formula, leading to overestimation or underestimation of actual malignancy risk.²⁵ The results of this meta-analysis suggest that risk prediction models are indeed better suited as clinical aids to interpretation than stand-alone diagnostic tools, which future direction might involve refining these models with additional variables, enhancing accuracy, and broadening their applicability across diverse patient populations.

The past decade has seen the development of sensitive assays for tumour-derived substances or cancer-specific analytes in the circulatory system, and liquid biopsies have attracted considerable attention in lung cancer screening and diagnosis. The multimodal lung cancer diagnostic model, integrating multi-omics and multidisciplinary data—including radiological and clinical characteristics—has significantly improved the efficiency and accuracy of early diagnosis compared to single-image interpretation or risk prediction models alone, offering substantial benefit to patients with malignant pulmonary nodules and low tumour burden, whether managed surgically or with drug therapy. Numerous studies have demonstrated that abnormal DNA methylation contributes to tumorigenesis through mechanisms including global hypomethylation, focal hypermethylation at CpG islands, and mutagenesis at methylated cytosines. DNA methylation changes have been detected in the develop-ment of lung adenocarcinoma, even preceding atypical adenomatoid hyperplasia. ctDNA analysis offers a non-invasive approach that minimises sampling bias and the risk of missing information due to tumour heterogeneity, while enabling repeated sampling for longitudinal monitoring. As a highly promising biomarker for malignant pulmonary nodules, ctDNA methylation has shown strong diagnostic performance, particularly when combined with validated clinical risk prediction models for pulmonary nodule assessment, highlighting its potential for integration into multimodal diagnostic frameworks. CTCs are cancer cells shed into the bloodstream from primary or metastatic tumours and are associated with tumour metastasis and proliferation. However, CTC levels decline with decreasing lesion size, resulting in low sensitivity for detecting malignant solitary pulmonary nodules, which remains a major limitation for their use in early cancer detection. Compared with CT imaging alone, the integrated model demonstrated superior performance in reducing false-positive diagnoses of malignant lung nodules. This suggests that a multimodal strategy combining ctDNA methylation, CAC/CTC-based liquid biopsy, and a clinical model incorporating imaging and patient characteristics may reduce overdiagnosis in lung cancer screening, thereby alleviating patient anxiety and avoiding unnecessary investigations. However, overdiagnosis remains a concern, particularly in cases of inert cancers that may never result in morbidity or mortality if left undetected. This raises two additional concerns. First, diagnostic results must be communicated to patients in a manner that minimises anxiety, given that liquid biopsy, whether used alone or integrated with a clinical model, cannot achieve 100% specificity within a short timeframe. Second, negative results should not lead patients to overestimate the absence of lung cancer risk, as 100% sensitivity is likewise unattainable. These issues present challenges for the use of these approaches in the benign/malignant diagnosis of pulmonary nodules.

This study has several limitations that warrant conside-ration. The meta-analysis incorporated data from only 13 studies. While comprehensive searching was undertaken, this relatively small number may limit the generalisability of the findings to the broader spectrum of pulmonary nodule diagnostic research. Secondly, significant heterogeneity was observed among the included studies. Variations in methodologies, participant populations, and outcome measurement techniques across these studies likely contributed to this heterogeneity and may affect the robust-ness and interpretation of the pooled estimates. Thirdly, the potential for publication bias must be acknowledged. Studies reporting statistically significant or positive findings are more likely to be published, which could introduce bias if relevant non-significant or negative studies were not identi-fied or included.

In the future, the potential for diagnosing benign or malignant pulmonary nodules, whether through CT-based evaluation, risk prediction, clinical models, or multimodal mixed models, may extend beyond stratifying patients into different risk categories for determining follow-up strategies in low-risk pulmonary nodule populations or treatment options in high-risk pulmonary nodule populations. Integrating radiological, clinical, and liquid biopsy features is essential to maximise true-positive and true-negative rates in patients with intermediate-risk pulmonary nodules, thereby enabling earlier identification of malignant nodules while minimising unne-cessary overdiagnosis. This study systematically evaluated the testing and screening tools currently used in pulmonary nodule diagnosis, analysing the characteristics and summary comparisons of various testing modalities in terms of diagnostic efficacy. The excellent performance of multi-histology, multi-modal diagnostic models based on liquid biopsy, imaging histology features, and clinical risk factors in malignant lung nodule diagnosis and screening was highlighted, alongside future clinical challenges that require resolution. Only comparative DTA studies were included to construct a Bayesian hierarchical bivariate random-effects model. Although heterogeneity among testing modalities was substantially reduced, potential bias due to strict inclusion criteria cannot be denied. Furthermore, given the wide variety of liquid biopsy approaches, establishing a more systematic and standardised evaluation framework is necessary to improve further the reliability and broad applicability of their diagnostic performance. Such an approach could help address the current diagnostic imbalance between the lack of early lung cancer diagnosis and potential overdiagnosis of benign pulmonary nodules.

CONCLUSION

This study systematically evaluated the diagnostic and screen-ing tools currently used for pulmonary nodule assessment and analysed and compared the diagnostic efficacy of different testing modalities. It also highlighted the potential of multi-modal diagnostic models integrating liquid biopsy, radiological features, and clinical risk factors for the diagnosis and screen-ing of malignant pulmonary nodules. The study also identified future clinical challenges that require further investigation. On the other hand, only comparative DTA studies were included in this meta-analysis to construct a Bayesian hierarchical bivariate random-effects model. Although the heterogeneity among the testing modalities was substantially reduced, potential bias due to the strict inclusion criteria cannot be denied. In addition, due to the wide variety of liquid biopsy approaches, it is necessary to establish a more systematic and standardised evaluation system in the future to improve further the reli-ability and broad applicability of their diagnostic performance, to address the current imbalance between inadequate early diagnosis of lung cancer and the potential overdiagnosis of benign pulmonary nodules.

FUNDING:
This study was funded by the high-level Traditional Chinese Medicine Hospital Construction Project of Wangjing Hospital of China Academy of Traditional Chinese Medicine-Special Project for Clinical Evidence-Based Research on Traditional Chinese Medicine (WJYY-XZKT-2023-11) and 2024 Capital Health Development Research Special Project (WJCC-202320).

PATIENTS’ CONSENT:
There were no human participants in this article; thus, informed consent was not required.

COMPETING INTEREST:
The authors declared no conflict of interest.

AUTHORS’ CONTRIBUTIONS:
RSC, TJH: Designed the study, analysed the data, interpreted the results, drafted the manuscript, and critically reviewed the manuscript.
ZHL, FY, FL: Piloted the QUADAS-C guideline questions inde-pendently to assess inter-reviewer agreement, interpreted the results, and critically reviewed the manuscript.
FG: Obtained funding, interpreted the results, and critically reviewed the manuscript.
All authors approved the final version of the manuscript to be published.

REFERENCES

1. National Lung Screening Trial Research Team; Aberle DR, Adams AM, Berg CD, Black WC, et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med 2011; 365(5):395-409. doi: 10.1056/NEJMoa 1102873.
2. National Lung Screening Trial Research Team; Church TR, Black WC, Aberle DR, Berg CD, Clingan KL, et al. Results of initial low-dose computed tomographic screening for lung cancer. N Engl J Med 2013; 368(21):1980-91. doi: 10.1056/ NEJMoa1209120.
3. Lokhandwala T, Bittoni MA, Dann RA, D’Souza AO, Johnson M, Nagy RJ, et al. Costs of diagnostic assessment for lung cancer: A medicare claims analysis. Clin Lung Cancer 2017; 18(1):e27-34. doi: 10.1016/j.cllc.2016.07.006.
4. Gareen IF, Duan F, Greco EM, Snyder BS, Boiselle PM, Park ER, et al. Impact of lung cancer screening results on participant health‐related quality of life and state anxiety in the national lung screening trial. Cancer 2014; 120(21): 3401-9. doi: 10.1002/cncr.28833.
5. McInnes MDF, Moher D, Thombs BD, McGrath TA, Bossuyt PM, and the PRISMA-DTA Group, et al. Preferred reporting items for a systematic review and meta-analysis of diagnostic test accuracy studies: The PRISMA-DTA statement. JAMA 2018; 319(4):388. doi: 10.1001/jama.2017.19163.
6. Cerullo E, Sutton AJ, Jones HE, Wu O, Quinn TJ, Cooper NJ. MetaBayesDTA: Codeless Bayesian meta-analysis of test accuracy, with or without a gold standard. BMC Med Res Methodol 2023; 23(1):127. doi: 10.1186/s12874-023-01910-y.
7. Burke DL, Ensor J, Snell KIE, Van Der Windt D, Riley RD. Guidance for deriving and presenting percentage study weights in meta‐analysis of test accuracy studies. Res Synth Methods 2018; 9(2):163-78. doi: 10.1002/jrsm.1283.
8. He J, Wang B, Tao J, Liu Q, Peng M, Xiong S, et al. Accurate classification of pulmonary nodules by a combined model of clinical, imaging, and cell-free DNA methylation biomarkers: A model development and external validation study. Lancet Digit Health 2023; 5(10):e647-56. doi: 10.1016/S2589- 7500(23)00125-5.
9. Liang W, Chen Z, Li C, Liu J, Tao J, Liu X, et al. Accurate diagnosis of pulmonary nodules using a noninvasive DNA methylation test. J Clin Invest 2021; 131(10):e145973. doi: 10.1172/JCI145973.
10. Zeng D, Wang C, Mu C, Su M, Mao J, Huang J, et al. Cell-free DNA from bronchoalveolar lavage fluid (BALF): A new liquid biopsy medium for identifying lung cancer. Ann Transl Med 2021; 9(13):1080. doi: 10.21037/atm-21-2579.
11. Ye M, Tong L, Zheng X, Wang H, Zhou H, Zhu X, et al. A classifier for improving early lung cancer diagnosis incor-porating artificial intelligence and liquid biopsy. Front Oncol 2022; 12:853801. doi: 10.3389/fonc.2022.853801.
12. Zhong M, Zhang Y, Pan Z, Wang W, Zhang Y, Weng Y, et al. Clinical utility of circulating tumour cells in the early detection of lung cancer in patients with a solitary pulmonary nodule. Technol Cancer Res Treat 2021; 20: 15330338211041465. doi: 10.1177/15330338211041465.
13. Zhang W, Duan X, Zhang Z, Yang Z, Zhao C, Liang C, et al. A combination of CT and telomerase-positive circulating tumour cells improves diagnosis of small pulmonary nodules. JCI Insight 2021; 6(11):e148182. doi: 10.1172/jci. insight.148182.
14. Gilbert FJ, Harris S, Miles KA, Weir-McCall JR, Qureshi NR, Rintoul RC, et al. Dynamic contrast-enhanced CT compared with positron emission tomography CT to characterise solitary pulmonary nodules: The SPUtNIk diagnostic accuracy study and economic modelling. Health Technol Assess 2022; 26(17):1-80. doi: 10.3310/WCEI8321.
15. Xu L, Zhou JY, Wang LL, Yang RN, Li TX, Zhu XL. A multidimensional model incorporating radiomics score and liquid biopsy for the prediction of malignant and benign pulmonary nodules. Transl Cancer Res TCR 2022; 11 (11):4009-18. doi: 10.21037/tcr-22-1755.
16. Liang W, Zhao Y, Huang W, Gao Y, Xu W, Tao J, et al. Non-invasive diagnosis of early-stage lung cancer using high-throughput targeted DNA methylation sequencing of circulating tumour DNA (ctDNA). Theranostics 2019; 9(7):2056-70. doi: 10.7150/thno.28119.
17. Liu QX, Zhou D, Han TC, Lu X, Hou B, Li MY, et al. A noninvasive multianalytical approach for lung cancer diagnosis of patients with pulmonary nodules. Adv Sci (Weinh) 2021; 8(13):2100104. doi: 10.1002/advs.2021 00104.
18. Xing W, Sun H, Yan C, Zhao C, Wang D, Li M, et al. A prediction model based on DNA methylation biomarkers and radiological characteristics for identifying malignant from benign pulmonary nodules. BMC Cancer 2021; 21(1):263. doi: 10.1186/s12885-021-08002-4.
19. Tahvilian S, Kuban JD, Yankelevitz DF, Leventon D, Henschke CI, Zhu J, et al. The presence of circulating genetically abnormal cells in blood predicts risk of lung cancer in individuals with indeterminate pulmonary nodules. BMC Pulm Med 2023; 23(1):193. doi: 10.1186/s12890-023- 02433-4.
20. Yi CA, Lee KS, Kim BT, Choi JY, Kwon OJ, Kim H, et al. Tissue characterisation of solitary pulmonary nodule: comparative study between helical dynamic CT and integrated PET/CT. J Nucl Med 2006; 47(3):443-50.
21. Bosch de Basea M, Thierry-Chef I, Harbron R, Hauptmann M, Byrnes G, Bernier MO, et al. Risk of haematological malignancies from CT radiation exposure in children, adolescents, and young adults. Nat Med 2023; 29(12): 3111-9. doi: 10.1038/s41591-023-02620-0.
22. Mazzone PJ, Lam L. Evaluating the patient with a pulmonary nodule: A review. JAMA 2022; 327(3):264-73. doi: 10. 1001/jama.2021.24287.
23. Deppen SA, Blume JD, Kensinger CD, Morgan AM, Aldrich MC, Massion PP, et al. Accuracy of FDG-PET to diagnose lung cancer in areas with infectious lung disease: A meta-analysis. JAMA 2014; 312(12):1227-36. doi: 10.1001/jama. 2014.11488.
24. Wiener RS, Gould MK, Slatore CG, Fincke BG, Schwartz LM, Woloshin S. Resource use and guideline concordance in evaluation of pulmonary nodules for cancer: Too much and too little care. JAMA Intern Med 2014; 174(6):871-80. doi: 10.1001/jamainternmed.2014.561.
25. Mattox AK, Bettegowda C, Zhou S, Papadopoulos N, Kinzler KW, Vogelstein B. Applications of liquid biopsies for cancer. Sci Transl Med 2019; 11(507):eaay1984. doi: 10.1126/ scitranslmed.aay1984.