Sputum Liquid Biopsy for Lung Cancer Screening, Diagnosis, Subtyping, Surveillance, Response Prediction, and Prognostication: A Scoping Review

Abstract

Background/Objectives: Liquid biopsy (LB) is transforming cancer care by enabling minimally invasive tumor profiling. While current research and clinical pathways mostly focus on blood LB, sputum represents a non-invasive, readily available respiratory specimen that may offer unique advantages for lung cancer (LC) care. Despite its potential, the maturity, breadth, and clinical applicability of sputum-based LB remain elusive. Methods: We conducted a scoping review to systematically map the existing literature on sputum LB in LC. Electronic databases were searched for studies evaluating sputum-derived biomarkers—cytologic, genomic, epigenetic, transcriptomic, proteomic, metabolomic, metagenomic, and extracellular vesicle–derived products—across the LC care continuum. Study designs, technologies, clinical contexts, and reported outcomes were extracted and synthesized qualitatively. Results: The literature demonstrated substantial heterogeneity in sputum collection, processing, and analytical platforms. Early work focused on cytometry and genetic alterations, while recent studies increasingly explore DNA methylomics, microRNAs, extracellular vesicle-derived products, and multi-omics approaches. The evidence suggests potential utility of sputum biomarkers for early detection and risk stratification, particularly in high-risk populations, with emerging data supporting roles in molecular subtyping, response monitoring, prognostication, and surveillance. However, few studies report prospective validation, direct comparison with blood-based LB, or impact on actual patient outcomes. Conclusions: Sputum LB is a promising yet underdeveloped modality in LC care. This scoping review highlights technological innovations alongside significant methodological heterogeneity and translational gaps. Future research should focus on standardization, prospective validation, impact on patient outcomes, and integration with blood- and other body fluid–based LB, as well as imaging biomarkers. This will enable incorporation of sputum-based LB into actual clinical pathways of LC care.

1. Introduction

Lung cancer (LC) remains the third most commonly diagnosed cancer in adults and is the most common cause of cancer-related deaths worldwide [1,2,3]. The American Cancer Society estimates that, during the year 2026, 229,410 new cases of LC will be diagnosed and 124,990 deaths will be caused by LC—which is more than the cancer-related deaths caused by colorectal cancer and pancreatic cancer combined [2]. Similar trends are reported for European countries: it is estimated that LC will cause 215,300 deaths in European Union countries and 31,600 deaths in the United Kingdom during the year 2026 [3]. Trends for Asian countries are likely worse, based on extrapolation from data from the Global Cancer Observatory (GLOBOCAN) 2022: age-standardized mortality rates for Turkey, China, and East Asian countries were 35.06 (per 100,000), 32 (per 100,000), and 25.13 (per 100,000), respectively [1]. LC survival rates remain poor, despite substantial advances in imaging, molecular diagnostics, and systemic therapies [4].

Early detection, accurate molecular subtyping, longitudinal surveillance, and timely prediction of treatment response represent central priorities in contemporary LC research [5]. Low-dose computed tomography (LDCT) screening demonstrated a mortality benefit in high-risk populations in the National Lung Screening Trial (NLST), the NELSON trial, and the ITALUNG trial [6,7,8]. However, LDCT screening is fraught with false-positive findings, overdiagnosis, cumulative radiation exposure, and restricted accessibility in many healthcare systems [9,10,11]. Although tissue biopsy remains the reference standard for diagnosis and molecular profiling of LC, it is an invasive procedure with significant risk for complications, can be technically challenging at times, and is not feasible for repeated sampling during surveillance or response monitoring [10,12]. These limitations drove growing interest in minimally invasive diagnostic approaches, particularly liquid biopsy [12].

Liquid biopsy (LB) is a non-invasive method of detecting tumor cells, tumor-associated products, or other cargo embedded within extracellular vesicles (EVs) inside bodily fluids [13]. Blood-based LB has been extensively studied in the literature and is already in clinical use for detection of actionable genomic alterations (AGAs), as well as for identifying resistance in patients with non–small cell LC (NSCLC) [14,15,16]. Commercially available blood-based LB tests that can be used for comprehensive biomarker profiling in patients with NSCLC include FoundationOne^® Liquid CDx (Foundation Medicine, Inc.; Boston, MA, USA), Guardant360^® CDx (Guardant Health, Inc.; Palo Alto, CA, USA), Agilent Resolution ctDx^® FIRST (Agilent, Inc.; Santa Clara, CA, USA), and Neo PanTracer^® LBx (NeoGenomics Laboratories, Inc.; Fort Myers, FL, USA). Additionally, a blood-based LB test, Shield^® (Guardant Health, Inc.; Palo Alto, CA, USA), received FDA (Food and Drug Administration) approval for colorectal cancer screening in 2024 [17]. Blood-based LB is invaluable in the care of patients with advanced or metastatic LC, but its role in detecting and evaluating early-stage cancers is often limited, given that shedding of tumor cells and tumor-associated products in the systemic circulation may be minimal [18]. Perhaps this explains why the NHS Galleri trial—the largest clinical trial of multicancer early detection tests—failed to meet its primary clinical endpoint of reducing the incidence of stage III–IV cancers [19], even though the overall results were generally encouraging [20]. Having said this, the Nodify Lung^® (Biodesix, Inc.; Louisville, CO, USA) nodule risk assessment strategy, which combines the results of two blood tests (Nodify CDT^® and Nodify XL2^®; Biodesix, Inc.; Louisville, CO, USA) alongside clinical information, can potentially reduce unnecessary procedures and improve early LC diagnosis [21]. These blood-based proteomics LB tests were validated in the PANOPTIC, FHCC, CLARIFY, and ORACLE cohorts [22].

Sputum-based LB represents a biologically attractive yet comparatively underexplored modality for the care of patients with LC [23]. As a respiratory tract specimen derived directly from the airways, sputum can provide an indirect but up-close snapshot of the tumor microenvironment, potentially better than peripheral blood can [24]. Tumor cells and tumor-derived products from central and peripheral LCs are exfoliated into airway secretions, potentially enabling earlier detection and higher analytical sensitivity [25]. Moreover, sputum sampling affords a cheap, painless, and non-invasive means for LB, which can be monitored longitudinally over time [24,25]. However, sputum poses its own unique challenges as a medium for LB, including potential contamination with oral microbes, the barrier posed by viscous mucus film, variable quality based on collection and processing methods, and heavy biological noise from non-tumor cells [26] (see Figure 1).

Over the preceding two decades, multiple preclinical feasibility and clinical studies have evaluated sputum-based assays for diverse applications across the LC care continuum, including screening, diagnosis, molecular subtyping, prognostication, prediction of therapeutic response, and surveillance for recurrence [27]. Historically, sputum cytology (in conjunction with plain radiography) played a role in LC diagnosis, particularly for centrally located tumors, but its limited sensitivity curtailed widespread adoption [28]. Advances in molecular technologies—including next-generation sequencing (NGS), digital polymerase chain reaction (PCR), DNA methylation profiling, transcriptomics, proteomics, and EV analysis—renewed interest in sputum as a rich source of biomarkers [29].

Despite a growing body of literature, the evidence base for sputum LB has remained fragmented. Studies vary widely in terms of patient populations, sputum collection techniques, analytical platforms, biomarker targets, clinical endpoints, and study design. Furthermore, sputum-based investigations are distributed across multiple disciplines, including pulmonology, oncology, pathology, molecular diagnostics, and bioengineering, complicating a coherent synthesis. Scoping reviews are particularly well suited to emerging and heterogeneous research domains, where the objective is to map available evidence, clarify concepts, and identify gaps, rather than to generate pooled effect estimates [30]. Accordingly, we performed this scoping review to comprehensively map the published literature on sputum LB for LC. By systematically mapping the evidence available on sputum LB across the LC care continuum—screening, diagnosis, subtyping, response prediction, response monitoring, surveillance, and prognostication—we aimed to assess the translational readiness of the available evidence, identify methodological strengths and limitations, reveal gaps in the current literature, and delineate priorities for future research.

2. Materials and Methods

2.1. Scoping Review Methodology and Reporting Framework

This study was conducted as a scoping review of the published literature following guidance from the Joanna Briggs Institute (JBI) for scoping reviews [30] and was reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) [31]. The review protocol was prospectively registered on the Open Science Framework prior to study initiation and is freely available online [32].

2.2. Eligibility Criteria

Eligibility criteria were defined a priori and were aligned with the Population–Concept–Context framework recommended for scoping reviews. All original research studies and conference abstracts that evaluated sputum-based testing for diagnosed or suspected LC were eligible for inclusion. Randomized trials, non-randomized studies, diagnostic accuracy studies, cohort and case–control studies, cross-sectional studies, and feasibility studies were included. Systematic reviews, meta-analyses, and scoping reviews with reproducible methodologies were also included for contextual support. Editorials, commentaries, narrative reviews, and opinion pieces were excluded, although their reference lists were manually screened to identify potentially eligible studies. Our primary focus was on clinical studies performed on sputum samples for screening, diagnosis, subtyping, prognostication, response prediction, surveillance of recurrence, and monitoring of response to therapy for patients with suspected or diagnosed LC. Although we included studies evaluating spontaneously expectorated or induced sputum, we did not include studies pertaining to other respiratory tract samples (such as tracheal aspirate, bronchial washing, bronchial brushings, or bronchoalveolar lavage) unless they included sputum specimens as well. Although our main focus was on studies of human sputum, we also included studies utilizing bronchial cell cultures or LC cell lines if they included testing or experiments utilizing sputum samples. With respect to LB, we planned to include studies testing for any type of products in the sputum, including circulating tumor cells (CTCs), EV/exosomes, circulating tumor DNA (ctDNA), cell-free DNA (cfDNA), microRNAs (miRNAs), tumor metabolites, autoantibodies, peptides, and so forth.

2.3. Information Sources and Search Strategy

A comprehensive literature search was conducted using PubMed/MEDLINE, Scopus, Embase, and Google Scholar, supplemented by searches of grey literature sources including OpenGrey and OAIster. Clinical trial registries searched included CENTRAL, ClinicalTrials.gov, WHO ICTRP, and the EU Clinical Trials Register. Search strategies combined terms related to sputum or respiratory specimens, LC, and LB- or tumor-derived biomarkers, using database-specific controlled vocabulary and free-text terms. The complete search strategies for each major database were predefined and implemented as registered in the online protocol [32].

2.4. Study Selection and Data Charting

Study selection was performed independently by two reviewers following removal of duplicate records. Titles and abstracts were screened, followed by full-text review of potentially eligible studies. Disagreements were resolved by consensus amongst the authors.

Data were extracted using a standardized charting form developed a priori and pilot-tested on a subset of studies. Extracted variables included study identification, study design, population characteristics, definition of cases and controls (if applicable), sputum collection methodology, biomarkers evaluated, analytical platforms, reference standard, histologic subtypes and stages of LC, clinical applications, performance measures (such as area under the receiver operating characteristic curve [AUROC], sensitivity, and specificity), and authors’ conclusions.

2.5. Risk of Bias Assessment and Critical Appraisal

Methodological quality was assessed to contextualize findings. Diagnostic accuracy studies were evaluated using QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies), and prognostic biomarker studies were evaluated using QUIPS (Quality in Prognostic Studies) [33]. Other observational studies were assessed using BIOCROSS (Biomarker-based Cross-Sectional Studies), ROBINS-I, the Newcastle–Ottawa Scale, or other JBI critical appraisal tools, as deemed appropriate [34]. We planned to appraise randomized trials using the RoB 2 tool, although no such studies were included in the final synthesis.

2.6. Data Synthesis

The findings were synthesized descriptively and presented in narrative and tabular formats, stratified by clinical application, biomarker class, and study design. Quantitative meta-analysis was not planned a priori due to anticipated heterogeneity. During the preparation of this manuscript, the authors used ChatGPT version 5.4 (OpenAI, San Francisco, CA, USA) and Google Gemini’s Nano Banana version 2.0 (Google LLC, Mountain View, CA, USA) to assist in generating figures. The authors have reviewed and edited the output and take full responsibility for the content of this publication.

3. Results

3.1. Study Selection and Inclusion

A total of 270 studies were included in this scoping review of sputum LB for screening, diagnosis, subtyping, response prediction, response monitoring, surveillance, and prognostication in LC (see Figure 2).

3.2. Characteristics of Sources

The included literature spanned 1978 to 2026, demonstrating sustained interest over nearly five decades and a marked increase in publications over time (see Figure 3). Only 4 studies were published before 1990, compared with 23 in the 1990s, 83 in the 2000s, 102 in the 2010s, and 58 from 2020 onward.

The evidence base was distributed across nine predefined domains (Figure 4). Methylomics represented the largest category (57/270, 21%), followed by genomics (49/270, 18%), cytopathology (49/270, 18%), proteomics (41/270, 15%), and transcriptomics (29/270, 11%). Smaller but increasingly contemporary bodies of literature were identified in metagenomics (19/270, 7%), metabolomics (10/270, 4%), integromics (9/270, 3%), and sputum processing (7/270, 3%), as shown in Figure 5.

Among included studies, the literature was geographically diverse: China and the United States were the most frequently represented countries, with additional contributions from India, Japan, Italy, France, South Korea, Canada, Spain, Brazil, Iran, and other countries. Most studies were observational in design and were primarily framed as diagnostic biomarker investigations. Cohort sizes were generally modest, with a median reported total sample size of 91 participants.

3.3. Reporting Quality and Risk of Bias

Reporting of diagnostic performance metrics was inconsistent across studies. Explicit diagnostic performance metrics (such as sensitivity, specificity, and AUROC) were reported in 173 studies (64.1%). Positive and negative predictive values were rarely reported, even in prospective cohort studies with an intended screening use-case. The limited and uneven reporting of performance metrics, together with substantial heterogeneity in study populations, assay methods, comparators, and thresholds, precluded a robust cross-study quantitative synthesis. In view of this, we adopted a descriptive evidence mapping approach.

Based on risk-of-bias assessments, 11 (4%), 132 (49%), 61 (23%), and 50 (19%) studies were deemed to be at low, moderate, moderate-high, and high risk of bias, respectively (see Figure 6). Overall, the evidence base was characterized by substantial methodological limitations. Recurrent concerns included selective or clinically enriched patient populations (introducing spectrum bias), non-representative healthy controls, insufficient detail on index test conduct and threshold derivation, incomplete reporting of flow and timing, incomplete reporting of blinding, lack of multivariable adjustment for confounders, and limited external validation. These issues were especially common in case–control–style biomarker discovery studies.

3.4. Evidence Mapping and Synthesis by Biomarker Domain

The predominant clinical application was diagnosis, which was identified in 211/270 studies (78%). Many studies were also classified as relevant to screening (80/270, 30%), detection of AGAs (40/270, 15%), and prognostication (39/270, 14%). Fewer studies addressed response monitoring (17/270, 6%), subtyping (15/270, 5%), and surveillance (5/270, 2%). Because individual studies could contribute to more than one use case, these categories were not mutually exclusive (see Figure 7). Nonetheless, the overall pattern was clear: the sputum LB literature has focused predominantly on early detection and diagnostic classification, whereas applications in longitudinal monitoring, histologic subtyping, and surveillance remain comparatively underdeveloped.

3.4.1. Cytopathology

Cytopathology comprised 49 published studies (see Appendix A.1) and represented one of the oldest streams of literature, with a median publication year of 2007. These studies compromised two major groups: (a) morphologic assessment of exfoliated cells using cytology, flow cytometry, and/or automated DNA cytometry; and (b) FISH (fluorescent in situ hybridization)-based techniques for detection of chromosomal aberrations and microsatellite alterations, such as LOH (loss of heterozygosity) and MSI (microsatellite instability). An overall summary of the results from these studies is presented in Table 1.

Traditional cytology (Papanicolaou smear) formed an important historical foundation for the field and established sputum as a clinically relevant respiratory specimen. However, the literature also reflected longstanding limitations of conventional sputum cytology, including variable sensitivity, dependence on specimen adequacy, and susceptibility to operator- and interpreter-related variability [35].

Sputum-based automated DNA cytometry in conjunction with flow cytometry was validated in multiple case–control and cohort studies [36,37,38,39]. Current clinical trials are exploring the role of sputum-based automated DNA cytometry for screening of LC. In particular, CyPath^® Lung (bioAffinity Technologies, Inc.; San Antonio, TX, USA) is being evaluated in a multicenter trial that started recruitment in early 2026 and is anticipated to be completed in 2029 [40]. CyPath^® Lung (bioAffinity Technologies, Inc.; San Antonio, TX, USA) uses meso-tetra (4-carboxyphenyl)-porphyrin-labeled sputum in conjunction with flow cytometry and artificial intelligence to detect malignant cells in an automated manner [36]. Another study by Qiu et al. explored magnetic-activated cell sorting (MACS) for improving sputum-based cytologic diagnosis of LC [41].

FISH-based approaches for cytopathologic diagnosis of LC in sputum specimens focused on chromosomal aberrations and microsatellite alterations. Sputum microsatellite analyses appeared technically feasible and biologically informative, with signals detectable in some patients with established LC, and occasionally in heavy smokers, but the studies were limited by small numbers, proof-of-concept designs, lack of standardized marker panels, and generally moderate to high risk of bias [42,43,44,45]. Across studies evaluating chromosomal aneusomies and related genomic aberrations in sputum, FISH-based assays generally demonstrated higher sensitivity than conventional sputum cytology, although performance varied substantially by clinical setting, assay platform, and study design [28,46,47]. Beyond targeted FISH assays, Arvanitis et al. [48] showed, using 48 microsatellite markers, that sputum fractional allele loss was approximately 10-fold higher in LC than in controls. Overall, chromosomal aberration assays in sputum consistently outperformed sputum cytology, where compared, and appeared most promising in high-risk or diagnostically enriched populations, but interpretation is tempered by frequent case–control designs, small sample sizes, inadequate-specimen exclusions, and generally moderate to high risk of bias.

Telomerase activity was explored across three prospective diagnostic studies from India [49,50,51]. In all studies, a TRAP (Telomerase Repeat Amplification Protocol) assay was used, and generally favorable but variable performance was reported. However, all of these studies were conducted in clinically suspected, predominantly advanced-stage, populations, rather than in screening cohorts. Moreover, TRAP assay sensitivity was inconsistent across studies, and false-positive results were observed in some non-malignant inflammatory pulmonary conditions.

Table 1. Overall summary of sputum-based cytopathology studies.

Cytopathology Approach	Representative Techniques	Representative Studies	Main Findings	Typical Setting	Key Limitations
Conventional sputum cytology	Papanicolaou smear; morphologic assessment of exfoliated cells	Historical cytology literature, including Payne et al. (1997) [35]	Established sputum as a clinically relevant respiratory specimen; historical foundation for LC detection, but diagnostic yield variable	Historical screening and diagnostic studies	Variable sensitivity; dependence on specimen adequacy; operator- and interpreter-related variability
Automated DNA cytometry/flow cytometry	Quantitative microscopy; automated DNA cytometry; flow cytometry; porphyrin-labeling of sputum; AI-assisted classification	Patriquin et al. (2015) [36] Rebel et al. (2021) [37] Bederka et al. (2022) [38] Bauta et al. (2023) [39]	Promising discrimination between malignant and non-malignant sputum samples; supported contemporary assay development, including CyPath® Lung (bioAffinity Technologies, Inc.; San Antonio, TX, USA)	Case–control and cohort studies, with evolving interest in screening applications	Heterogeneous designs and platforms; limited prospective screening validation; incomplete standardization across studies
Cell enrichment for cytologic diagnosis	Magnetic-activated cell sorting (MACS)	Qiu et al. (2008) [41]	Magnetic enrichment of bronchial epithelial cells appeared to improve recovery of diagnostically relevant cells from sputum	Diagnostic enrichment setting	Limited external validation and sparse follow-up literature
FISH/chromosomal aberration assays	FISH for chromosomal aberrations; aneusomy panels; chromosomal copy number abnormalities	Jia et al. (2000) [44] Baron et al. (2017) [47] Shlomi et al. (2018) [46]	FISH-based chromosomal aberration assays generally demonstrated higher sensitivity than conventional sputum cytology in head-to-head comparisons; appeared especially promising in high-risk populations	High-risk or diagnostically enriched cohorts	Performance varied by assay platform, clinical setting, and study design; small or selected samples; moderate to high risk of bias
Microsatellite alteration assays	Loss of heterozygosity (LOH); microsatellite instability (MSI); fractional allele loss panels	Mao et al. (1994) [42] Miozzo et al. (1996) [43] Castagnaro et al. (2007) [45] Arvanitis et al. (2003) [48]	Microsatellite-based analyses were technically feasible and biologically informative; abnormal signals detected in some patients with LC and occasionally in heavy smokers, with fractional allele loss markedly higher in cases than controls	Mostly proof-of-concept and small case–control studies	Small sample sizes; proof-of-concept designs; non-standardized marker panels; limited reproducibility data; generally moderate-high or high risk of bias
Telomerase-based assays	Telomerase repeat amplification protocol (TRAP) assay	Sen et al. (2001 and 2002) [49,50] Pasrija et al. (2007) [51]	Prospective diagnostic studies reported generally favorable but inconsistent performance for telomerase activity in sputum	Clinically suspected, predominantly advanced-stage, populations	Sensitivity inconsistent across studies; false-positive results in some non-malignant inflammatory conditions

Abbreviations: AI, artificial intelligence; FISH, fluorescence in situ hybridization; LC, lung cancer; LOH, loss of heterozygosity; MACS, magnetic-activated cell sorting; MSI, microsatellite instability; TRAP, telomerase repeat amplification protocol.

Table 1. Overall summary of sputum-based cytopathology studies.

Cytopathology Approach	Representative Techniques	Representative Studies	Main Findings	Typical Setting	Key Limitations
Conventional sputum cytology	Papanicolaou smear; morphologic assessment of exfoliated cells	Historical cytology literature, including Payne et al. (1997) [35]	Established sputum as a clinically relevant respiratory specimen; historical foundation for LC detection, but diagnostic yield variable	Historical screening and diagnostic studies	Variable sensitivity; dependence on specimen adequacy; operator- and interpreter-related variability
Automated DNA cytometry/flow cytometry	Quantitative microscopy; automated DNA cytometry; flow cytometry; porphyrin-labeling of sputum; AI-assisted classification	Patriquin et al. (2015) [36] Rebel et al. (2021) [37] Bederka et al. (2022) [38] Bauta et al. (2023) [39]	Promising discrimination between malignant and non-malignant sputum samples; supported contemporary assay development, including CyPath® Lung (bioAffinity Technologies, Inc.; San Antonio, TX, USA)	Case–control and cohort studies, with evolving interest in screening applications	Heterogeneous designs and platforms; limited prospective screening validation; incomplete standardization across studies
Cell enrichment for cytologic diagnosis	Magnetic-activated cell sorting (MACS)	Qiu et al. (2008) [41]	Magnetic enrichment of bronchial epithelial cells appeared to improve recovery of diagnostically relevant cells from sputum	Diagnostic enrichment setting	Limited external validation and sparse follow-up literature
FISH/chromosomal aberration assays	FISH for chromosomal aberrations; aneusomy panels; chromosomal copy number abnormalities	Jia et al. (2000) [44] Baron et al. (2017) [47] Shlomi et al. (2018) [46]	FISH-based chromosomal aberration assays generally demonstrated higher sensitivity than conventional sputum cytology in head-to-head comparisons; appeared especially promising in high-risk populations	High-risk or diagnostically enriched cohorts	Performance varied by assay platform, clinical setting, and study design; small or selected samples; moderate to high risk of bias
Microsatellite alteration assays	Loss of heterozygosity (LOH); microsatellite instability (MSI); fractional allele loss panels	Mao et al. (1994) [42] Miozzo et al. (1996) [43] Castagnaro et al. (2007) [45] Arvanitis et al. (2003) [48]	Microsatellite-based analyses were technically feasible and biologically informative; abnormal signals detected in some patients with LC and occasionally in heavy smokers, with fractional allele loss markedly higher in cases than controls	Mostly proof-of-concept and small case–control studies	Small sample sizes; proof-of-concept designs; non-standardized marker panels; limited reproducibility data; generally moderate-high or high risk of bias
Telomerase-based assays	Telomerase repeat amplification protocol (TRAP) assay	Sen et al. (2001 and 2002) [49,50] Pasrija et al. (2007) [51]	Prospective diagnostic studies reported generally favorable but inconsistent performance for telomerase activity in sputum	Clinically suspected, predominantly advanced-stage, populations	Sensitivity inconsistent across studies; false-positive results in some non-malignant inflammatory conditions

Abbreviations: AI, artificial intelligence; FISH, fluorescence in situ hybridization; LC, lung cancer; LOH, loss of heterozygosity; MACS, magnetic-activated cell sorting; MSI, microsatellite instability; TRAP, telomerase repeat amplification protocol.

3.4.2. Genomics

The genomics category included 49 published studies with a median publication year of 2008 (see Appendix A.2). This body of literature evaluated DNA-level alterations, including AGAs and related genomic signals, using PCR-based and sequencing-based approaches. Genomic studies were largely directed toward diagnostic discrimination, although some also explored screening and surveillance contexts. Across this category, the literature reflected a transition from single-analyte investigations toward broader panel-based strategies, but reporting remained heterogeneous with respect to patient spectrum, assay standardization, and clinical validation. An overall summary of the results from the genomics studies is given in Table 2.

Earlier work focused on the detection of TP53 and K-RAS mutations in sputum samples from high-risk smokers and patients with suspected LC. These studies were predominantly PCR-based feasibility or diagnostic studies and showed that tumor-related mutations could be detected noninvasively, although performance varied considerably, and specificity was sometimes limited by field cancerization or mutation-positive non-cancer controls. For K-RAS, the earliest proof-of-concept study by Takeda et al. [52] detected a codon 12 mutation in one of five sputum samples from LC patients, establishing technical feasibility. In a case–control study, Ronai et al. [53] found K-RAS mutations in 48.6% (18/37) of NSCLC sputum samples versus 12.5% (5/40) of non-cancer lung disease controls, indicating modest sensitivity and imperfect specificity. Zhang et al. [54] later reported K-RAS codon 12 mutations in 10/22 (45.5%) sputum samples from LC patients, with significant tumor–sputum concordance and identical mutations in nine patients, supporting the biologic validity of sputum mutation analysis. However, Nakajima et al. [55] highlighted important limitations, showing K-RAS mutations in 15–19% of sputum samples from LC patients but also in 20% of non-oncologic controls and observing discordance between sputum and tumor in several paired cases. Chen et al. [56] detected tumor-specific TP53 mutations in only 1 of 10 pre-diagnostic sputum smears from patients with known tumor mutations, indicating low sensitivity with early assays. By contrast, Wang et al. [57] reported TP53 mutations in sputum from 30/54 LC patients (55.6%) and 2/114 benign pulmonary disease controls, corresponding to 55.6% sensitivity and 98.3% specificity, outperforming sputum cytology alone (35.2%). Similarly, Guo et al. [58] found TP53 mutations in 27.6% of post-bronchoscopic sputum samples, which was significantly higher than sputum cytology (6.9%) in patients suspected of LC. Anderson et al. [59] also showed p53 protein overexpression in sputum collected before diagnosis in all six patients whose tumors harbored mutant TP53, but also in two of five patients without tumor TP53 mutations, suggesting that p53-related abnormalities may reflect high-risk field changes rather than strictly tumor-specific events. Overall, sputum detection of K-RAS and TP53 alterations provided early evidence that airway-exfoliated genetic abnormalities could be captured noninvasively and often improved upon cytology, but the literature was limited by small sample sizes, older low-sensitivity assays, heterogeneous designs, and generally moderate to high risk of bias.

Eleven studies focused on the detection of EGFR (epidermal growth factor receptor) mutations, which are considered AGAs in patients with NSCLC. Across studies evaluating EGFR mutations in sputum, assay performance varied markedly in terms of specimen quality, disease stage, and molecular platform, but newer approaches generally showed that sputum can serve as a feasible noninvasive substrate for tumor genotyping in selected patients. A prospective paired study of late-stage LC by Su et al. [60] reported high concordance between sputum and tissue EGFR testing using ARMS-PCR, with a sensitivity of 90.9%, a specificity of 96%, and an overall accuracy of 97.1%, albeit after cytologic enrichment and exclusion of tumor-cell–negative sputum samples. Studies focusing on sputum supernatant cfDNA reported more moderate overall sensitivity but consistently high specificity. For instance, Wang et al. [61] reported 46.2% sensitivity and 100% specificity for EGFR mutations overall, with sensitivity rising from 24.0% in stage I–IIIA disease to 65.0% in stage IIIB–IV disease and reaching 92.9% when malignant cells were seen on sputum cytology. Similarly, Isaka et al. [62] showed that sputum droplet digital PCR (ddPCR) had high sensitivity only in cytology-positive samples (80.0%) but extremely low sensitivity in cytology-negative samples (3.1%), despite 100% specificity in both groups. Diagnostic performance was superior in studies evaluating sputum supernatant cfDNA by NGS [25,63]. Overall, the evidence indicates that sputum can support EGFR mutation detection, particularly in advanced disease and in sputum samples with malignant cells or adequate tumor-derived DNA, but that sensitivity is inconsistent in unselected sputum. Additionally, most studies were limited by small size, selected populations, and moderate to high risk of bias.

Evidence for BRAF mutation and EML4-ALK fusion detection in sputum was sparse and was directed primarily toward molecular profiling of established NSCLC rather than screening or early diagnosis. For BRAF, the only sputum-related study identified was a preclinical method-development investigation by Emaus et al. [64], which used artificial sputum spiked with mutant DNA and showed that selective extraction with ion-tagged oligonucleotides and magnetic ionic liquids improved analytical sensitivity for low-abundance BRAF V600E detection to 0.1% of the mutant allele fraction. However, no patient-derived sputum samples were studied, so no clinical diagnostic inferences could be made. For EML4-ALK, clinical evidence was limited but more substantive: in a large, prospective, multicenter Japanese study, Soda et al. [65] demonstrated that multiplex RT-PCR could detect EML4-ALK fusion transcripts in sputum specimens from patients with confirmed NSCLC, with four positive sputum samples identified and concordance observed with paired tumor or effusion specimens, supporting the technical feasibility of sputum-based fusion testing. Wang et al. [66] further showed, in a retrospective cohort of 1274 patients with advanced NSCLC, that cytological specimens overall yielded ALK detection rates comparable to tissue analysis by RT-PCR, although the sputum subgroup comprised only two samples and therefore provided minimal sputum-specific evidence. Morikawa et al. [67,68] also reported on the feasibility and utility of the Lung Cancer Compact Panel^® (LCCP by DNA Chip Research, Inc.; Kanagawa, Japan) for detection of EML4-ALK gene fusions in cytologically positive sputum specimens. The LCCP LB test was approved by the Ministry of Health, Labor, and Welfare in Japan as a multi-companion diagnostic kit for LC in November 2022. Overall, the sputum literature for BRAF and EML4-ALK remains very limited, with clinical support confined mainly to proof-of-concept ALK fusion detection in confirmed NSCLC and no direct patient-based clinical validation for sputum BRAF mutation testing.

Across studies evaluating multi-gene sputum assays, panel-based approaches generally outperformed single-marker testing and conventional sputum cytology, although performance depended strongly on clinical setting, disease stage, and specimen adequacy. In early-stage disease, Jiang et al. [69] reported that a four-gene sputum mini-chip assay targeting HYAL2, FHIT, p16, and SP-A achieved 70% sensitivity and 92% specificity for stage I NSCLC; when combined with CT, the sensitivity increased to 91%, and the overall accuracy increased to 90%, with particularly favorable performance for central tumors. In a subsequent optimization study, the same group [70] developed a six-gene panel (ENO1, FHIT, HYAL2, SKP2, p16, and 14-3-3ζ) that yielded sensitivities of 81.6% to 83.7% and a specificity of 93.9% for distinguishing stage I NSCLC from healthy smokers or COPD controls, markedly exceeding sputum cytology for early-stage disease (81.4% vs. 41.9% sensitivity). Within a CT-screening context, Carozzi et al. [71] applied a broader molecular profile including LOH/MSI at 12 loci, K-RAS and TP53 mutations, and plasma DNA quantification, finding allelic imbalance in 84.2% of LC cases but also in 73.0% of participants with non-calcified nodules and 28.9% of negative controls, suggesting biologic signal but limited specificity when used alone. More recent multiplex genomic studies in advanced NSCLC extended this concept to larger mutation panels in sputum cfDNA: Qin et al. [24] showed that induced sputum supernatant outperformed sputum sediment for 168-gene profiling and achieved 69.2% concordance with tumor tissue, while Xie et al. [25] reported 77% concordance for sputum supernatant cfDNA using a 520-gene panel, increasing to 90% when combined with plasma. Likewise, in a sputum multi-driver panel study, Wang et al. [72] demonstrated that a 10-gene NGS assay detected hotspot alterations in sputum cfDNA with 81.3% overall sensitivity and 100% specificity relative to paired tumor samples, with sensitivity rising to 87.0% in stage IIIB–IV disease and 94.1% in advanced-stage patients with malignant sputum cytology. Overall, multi-gene sputum panels appear more informative than single-gene assays or cytology alone and may be especially useful when integrated with imaging or other LB sources; however, the evidence remains heterogeneous, spanning early case–control mini-chip studies to advanced-disease NGS profiling, with most studies carrying at least some concern for selection bias or limited generalizability.

Table 2. Overall summary of sputum-based genomics studies.

Domain	Representative Techniques	Representative Studies	Main Findings	Typical Setting	Key Limitations
TP53 and K-RAS mutations	PCR-based mutation detection of K-RAS codon 12 and TP53 alterations	Takeda et al. [52] Zhang et al. [54] Nakajima et al. [55] Chen et al. [56] Guo et al. [58] Anderson et al. [59]	Compared with cytology, molecular detection often improved sensitivity, but specificity was imperfect High-risk controls were mutation-positive (field cancerization)	Mainly high-risk smokers and patients with suspected or established LC; mostly feasibility, diagnostic, or case–control designs	Small sample sizes; older low-sensitivity platforms; heterogeneous pre-analytic methods; occasional tumor-sputum discordance; concern for field cancerization; overall risk of bias generally moderate to high
EGFR mutation	ARMS-PCR, ddPCR, sequencing-based techniques; sputum cell sediment or supernatant cfDNA	Su et al. [60] Wang et al. [73] Isaka et al. [62]	Sputum-based EGFR genotyping is feasible, especially in advanced NSCLC and in cytology-positive or tumor-enriched sputum Specificity was consistently high, whereas sensitivity varied markedly	Paired tumor-sputum samples from suspected or known NSCLC; often, advanced stage; cytology-positive sputum in some studies	Sensitivity inconsistent in unselected sputum; selected populations or exclusion of inadequate specimens; limited external clinical validation
BRAF mutation EML4-ALK fusion	BRAF V600E detection in virtual sputum; multiplex RT-PCR and LCCP for EML4-ALK fusion detection	Emaus et al. [64] Soda et al. [65] Wang et al. [66] Morikawa et al. [68]	Preclinical evidence of BRAF V600E mutation in artificial sputum Proof-of-concept clinical studies supported EML4-ALK fusion detection, but tiny sample size	Established NSCLC undergoing molecular profiling; mostly cytology-positive sputum	Extremely limited evidence base; findings mainly support feasibility
Multi-gene panels	Mini-chip and multiplex assays targeting combinations of genes or alterations	Jiang et al. [69] Jiang et al. [70] Carozzi et al. [71]	Panels consistently outperformed single-gene assays; favorable sensitivity and specificity in early-stage NSCLC, when combined with CT	Case–control and screening-enriched cohorts; several studies emphasized central tumors or CT-screening cohorts	Panel marker composition and thresholds varied considerably across studies; limiting comparability; selection bias; limited external validation
Broad NGS profiling of sputum cfDNA	Large targeted sequencing panels (e.g., 10-gene, 168-gene, or 520-gene panels)	Qin et al. [24] Xie et al. [25] Wang et al. [72]	Sputum cfDNA NGS demonstrated moderate to high concordance with paired tumor tissue Better performance in sputum supernatant than cell sediment Higher sensitivity in advanced disease	Advanced NSCLC; paired tissue comparator studies evaluating molecular profiling	Selected, advanced-disease cohorts; applicability to clinical settings uncertain; heterogeneity due to differences in panel size, specimen processing techniques, and adequacy criteria

Abbreviations: ALK, anaplastic lymphoma kinase; ARMS, amplification refractory mutation system; BRAF, v-raf murine sarcoma viral oncogene, homolog B; cfDNA, cell-free deoxyribonucleic acid; CT, computed tomography; ddPCR, droplet digital polymerase chain reaction; EGFR, epidermal growth factor receptor; EML4, echinoderm microtubule-associated protein-like 4; K-RAS, Kirsten rat sarcoma viral oncogene homolog; LC, lung cancer; LCCP, Lung Cancer Compact Panel^® (DNA Chip Research, Inc.; Kanagawa, Japan); LOH, loss of heterozygosity; NGS, next-generation sequencing; NSCLC, non–small cell lung cancer; p53, tumor protein 53; PCR, polymerase chain reaction; RT, reverse transcription.

Table 2. Overall summary of sputum-based genomics studies.

Domain	Representative Techniques	Representative Studies	Main Findings	Typical Setting	Key Limitations
TP53 and K-RAS mutations	PCR-based mutation detection of K-RAS codon 12 and TP53 alterations	Takeda et al. [52] Zhang et al. [54] Nakajima et al. [55] Chen et al. [56] Guo et al. [58] Anderson et al. [59]	Compared with cytology, molecular detection often improved sensitivity, but specificity was imperfect High-risk controls were mutation-positive (field cancerization)	Mainly high-risk smokers and patients with suspected or established LC; mostly feasibility, diagnostic, or case–control designs	Small sample sizes; older low-sensitivity platforms; heterogeneous pre-analytic methods; occasional tumor-sputum discordance; concern for field cancerization; overall risk of bias generally moderate to high
EGFR mutation	ARMS-PCR, ddPCR, sequencing-based techniques; sputum cell sediment or supernatant cfDNA	Su et al. [60] Wang et al. [73] Isaka et al. [62]	Sputum-based EGFR genotyping is feasible, especially in advanced NSCLC and in cytology-positive or tumor-enriched sputum Specificity was consistently high, whereas sensitivity varied markedly	Paired tumor-sputum samples from suspected or known NSCLC; often, advanced stage; cytology-positive sputum in some studies	Sensitivity inconsistent in unselected sputum; selected populations or exclusion of inadequate specimens; limited external clinical validation
BRAF mutation EML4-ALK fusion	BRAF V600E detection in virtual sputum; multiplex RT-PCR and LCCP for EML4-ALK fusion detection	Emaus et al. [64] Soda et al. [65] Wang et al. [66] Morikawa et al. [68]	Preclinical evidence of BRAF V600E mutation in artificial sputum Proof-of-concept clinical studies supported EML4-ALK fusion detection, but tiny sample size	Established NSCLC undergoing molecular profiling; mostly cytology-positive sputum	Extremely limited evidence base; findings mainly support feasibility
Multi-gene panels	Mini-chip and multiplex assays targeting combinations of genes or alterations	Jiang et al. [69] Jiang et al. [70] Carozzi et al. [71]	Panels consistently outperformed single-gene assays; favorable sensitivity and specificity in early-stage NSCLC, when combined with CT	Case–control and screening-enriched cohorts; several studies emphasized central tumors or CT-screening cohorts	Panel marker composition and thresholds varied considerably across studies; limiting comparability; selection bias; limited external validation
Broad NGS profiling of sputum cfDNA	Large targeted sequencing panels (e.g., 10-gene, 168-gene, or 520-gene panels)	Qin et al. [24] Xie et al. [25] Wang et al. [72]	Sputum cfDNA NGS demonstrated moderate to high concordance with paired tumor tissue Better performance in sputum supernatant than cell sediment Higher sensitivity in advanced disease	Advanced NSCLC; paired tissue comparator studies evaluating molecular profiling	Selected, advanced-disease cohorts; applicability to clinical settings uncertain; heterogeneity due to differences in panel size, specimen processing techniques, and adequacy criteria

Abbreviations: ALK, anaplastic lymphoma kinase; ARMS, amplification refractory mutation system; BRAF, v-raf murine sarcoma viral oncogene, homolog B; cfDNA, cell-free deoxyribonucleic acid; CT, computed tomography; ddPCR, droplet digital polymerase chain reaction; EGFR, epidermal growth factor receptor; EML4, echinoderm microtubule-associated protein-like 4; K-RAS, Kirsten rat sarcoma viral oncogene homolog; LC, lung cancer; LCCP, Lung Cancer Compact Panel^® (DNA Chip Research, Inc.; Kanagawa, Japan); LOH, loss of heterozygosity; NGS, next-generation sequencing; NSCLC, non–small cell lung cancer; p53, tumor protein 53; PCR, polymerase chain reaction; RT, reverse transcription.

3.4.3. Methylomics and Epigenetics

Methylomics and epigenetics was the largest domain (57 studies) and one of the most prominent molecular areas in the sputum literature (see Appendix A.3), with a median publication year of 2012. This category contributed heavily to the screening and diagnosis evidence base and frequently evaluated multi-gene methylation panels rather than isolated loci. The prominence of this domain suggests that methylation-based assays have been especially attractive for sputum applications, likely owing to biological plausibility in early carcinogenesis and technical compatibility with low-input DNA extracted from sputum. An overall summary of the results for epigenetics and methylomics studies is provided in Table 3.

Sputum methylomics represents one of the most mature sputum biomarker domains, with studies spanning early biologic discovery, case-control diagnostic evaluation, prospective risk prediction, and post-resection surveillance. Early work established biologic plausibility by showing that p16 promoter methylation is detectable in sputum and increases across the histologic spectrum from basal cell hyperplasia and squamous metaplasia to carcinoma in situ and invasive squamous cell carcinoma, supporting methylation as an early event in lung carcinogenesis [74]. Subsequent studies demonstrated that methylation abnormalities are also frequent in smokers without LC, consistent with airway field cancerization; for example, p16, DAPK, MGMT, and RASSF1A methylation were detectable in bronchial epithelium and sputum from cancer-free smokers and persisted after smoking cessation, although this background prevalence limited specificity for cancer detection [75].

Multi-gene panels generally outperformed single-gene assays. In a cross-sectional risk study, increasing numbers of methylated genes in sputum were associated with higher LC risk, with greater than or equal to three methylated genes conferring an odds ratio of 6.2. The strongest prospective evidence came from a nested case-control analysis within the Colorado Sputum Screening Cohort [76], in which a six-gene panel (p16, MGMT, DAPK, RASSF1A, PAX5b, GATA5) predicted incident LC in a high-risk cohort up to 6 years before diagnosis; within 18 months of diagnosis, greater than or equal to three methylated genes yielded an adjusted odds ratio of 6.5, with 64% sensitivity and 64% specificity. In established NSCLC, sputum methylation also reflected tumor biology more closely than serum: in stage III disease, sputum methylation frequencies approximated tumor methylation frequencies, and a four-gene panel (p16, DAPK, PAX5b, GATA5) achieved a positive predictive value of 86% for tumor methylation status, whereas serum sensitivity was poor [77]. Broader methylation studies further suggested relevance to smoking-related risk phenotypes, including chronic mucous hypersecretion and recurrence after resection, although not all methylation markers were directly diagnostic for incident LC [78,79].

Novel approaches to methylomics and epigenetics have moved beyond locus-specific promoter methylation to assessment of global DNA methylation topology. Tajbaksh et al. [80] used three-dimensional DNA methylation imaging to differentiate normal respiratory epithelial cells from hypomethylated malignant cells in resected tumors and matching sputum based on in situ immunofluorescence (5-methylcytosine, 4′-6-diamidino-2-phenylindole co-localization). Likewise, Soukiasian et al. [81] utilized DNA methylation topology analysis in a sputum-based clinical assay that detected early-stage NSCLC with 95.8% sensitivity, albeit with limited specificity compared with high-risk benign lung disease. Additionally, Li et al. [82] used droplet digital MSP (methylation-specific PCR) to simultaneously quantify multiple miRNA and DNA methylation sites and develop an integromic signature for detection of LC.

A recent systematic review and meta-analysis by Wen et al. [27] synthesized 15 sputum cfDNA methylomics studies and found substantial heterogeneity in diagnostic performance, with reported sensitivities ranging from 10% to 93% and specificities from 8% to 100%; pooled across all genes, the summary sensitivity was 54.3%, and the specificity was 79.7%, with an HSROC AUC of 0.71, indicating moderate overall discriminatory ability. The review identified RASSF1A, APC, and CYGB as the most frequently studied genes, but their pooled sensitivities were modest, at 39%, 44%, and 47%, respectively, despite generally better specificities, especially for RASSF1A. Notably, two less frequently studied genes, SOX17 and TAC1, showed sensitivities above 85% with specificities above 70% in the most recent studies, suggesting that newer targets and newer platforms such as digital PCR may outperform older methylation panels. Wen et al. [27] also emphasized major sources of variability, including tumor location and stage, spontaneous versus induced sputum collection, whether cellular pellet or supernatant was analyzed, DNA extraction methods, and assay platform, with most studies still using case–control designs and QMSP (Quantitative Methylation Specific PCR)-based methods. Taken together, the available evidence indicates that sputum methylomics is biologically robust and promising for risk stratification and adjunctive early detection, but current performance remains too heterogeneous for stand-alone clinical implementation, and methodological standardization plus validation of high-performing genes will be essential for routine clinical use.

Table 3. Overall summary of sputum-based methylomics and epigenetics studies.

Domain	Representative Techniques	Representative Studies	Main Findings	Typical Setting	Key Limitations
Proof-of-concept	Conventional MSP	Belinsky et al. [74]	Biologic plausibility of sputum methylation as an early event in lung carcinogenesis	Early biologic discovery and translational studies in smokers and LC patients	Primarily mechanistic and observational; limited immediate clinical validation for stand-alone diagnosis
Field cancerization in smokers	Nested MSP, MSRE-PCR	Belinsky et al. [75] Rosell et al. [83]	Methylation abnormalities are frequent, even in cancer-free smokers, and may persist after smoking cessation	High-risk smoking populations without known LC	Background prevalence reduces cancer specificity and complicates interpretation of isolated methylated loci
Multi-gene methylation panels	Multiplex MSP, MSRE-PCR	Belinsky et al. [76] Mohammed et al. [84]	Panel-based assays generally outperformed single-gene testing; increasing numbers of methylated genes correlated with higher lung cancer risk	High-risk smokers and known LC patients; case–control or cohort designs	Moderate sensitivity and specificity; predictive performance varied by time to diagnosis and cohort characteristics
Methylation markers linked to post-resection recurrence	Multiplex MSP, MSRE-PCR, CoBRA-MSP	Belinsky et al. [79] Tessema et al. [85]	Gene methylation in sputum post-resection was associated with odds of recurrence	High-risk screening or surveillance cohorts undergoing longitudinal follow-up	Few methylation markers were specific for recurrence
DNA methylation topology	3D quantitative DNA methylation imaging	Tajbakhsh et al. [80] Soukiasian et al. [81]	Sputum-based methylation topology can detect hypomethylated cancerous cells, potentially detecting early LC	High-risk smokers, LC patients, and COPD patients	Small sample size; limited prospective clinical validation; moderate to high-risk of bias
Systematic review of sputum cfDNA methylomics	QMSP and ddMSP	Wen et al. [27]	Sensitivity and specificity of sputum methylated tumor DNA for LC detection varied considerably; divergence relates to tumor site, sample acquisition, extraction methods, and methylation measurement techniques	Meta-analysis of 15 studies with substantial, but unquantified, heterogeneity	Substantial heterogeneity in study designs, sputum acquisition protocols, and methylation measurement techniques; however, no objective measure of heterogeneity reported

Abbreviations: CoBRA, combined bisulfite modification and restriction analysis; COPD, chronic obstructive pulmonary disease; ddMSP, droplet digital methylation-specific polymerase chain reaction; DNA, deoxyribonucleic acid; LC, lung cancer; MSP, methylation-specific polymerase chain reaction; MSRE, methylation-sensitive restriction enzyme; NSCLC, non–small cell lung cancer; PCR, polymerase chain reaction; QMSP, quantitative methylation-specific polymerase chain reaction.

Table 3. Overall summary of sputum-based methylomics and epigenetics studies.

Domain	Representative Techniques	Representative Studies	Main Findings	Typical Setting	Key Limitations
Proof-of-concept	Conventional MSP	Belinsky et al. [74]	Biologic plausibility of sputum methylation as an early event in lung carcinogenesis	Early biologic discovery and translational studies in smokers and LC patients	Primarily mechanistic and observational; limited immediate clinical validation for stand-alone diagnosis
Field cancerization in smokers	Nested MSP, MSRE-PCR	Belinsky et al. [75] Rosell et al. [83]	Methylation abnormalities are frequent, even in cancer-free smokers, and may persist after smoking cessation	High-risk smoking populations without known LC	Background prevalence reduces cancer specificity and complicates interpretation of isolated methylated loci
Multi-gene methylation panels	Multiplex MSP, MSRE-PCR	Belinsky et al. [76] Mohammed et al. [84]	Panel-based assays generally outperformed single-gene testing; increasing numbers of methylated genes correlated with higher lung cancer risk	High-risk smokers and known LC patients; case–control or cohort designs	Moderate sensitivity and specificity; predictive performance varied by time to diagnosis and cohort characteristics
Methylation markers linked to post-resection recurrence	Multiplex MSP, MSRE-PCR, CoBRA-MSP	Belinsky et al. [79] Tessema et al. [85]	Gene methylation in sputum post-resection was associated with odds of recurrence	High-risk screening or surveillance cohorts undergoing longitudinal follow-up	Few methylation markers were specific for recurrence
DNA methylation topology	3D quantitative DNA methylation imaging	Tajbakhsh et al. [80] Soukiasian et al. [81]	Sputum-based methylation topology can detect hypomethylated cancerous cells, potentially detecting early LC	High-risk smokers, LC patients, and COPD patients	Small sample size; limited prospective clinical validation; moderate to high-risk of bias
Systematic review of sputum cfDNA methylomics	QMSP and ddMSP	Wen et al. [27]	Sensitivity and specificity of sputum methylated tumor DNA for LC detection varied considerably; divergence relates to tumor site, sample acquisition, extraction methods, and methylation measurement techniques	Meta-analysis of 15 studies with substantial, but unquantified, heterogeneity	Substantial heterogeneity in study designs, sputum acquisition protocols, and methylation measurement techniques; however, no objective measure of heterogeneity reported

Abbreviations: CoBRA, combined bisulfite modification and restriction analysis; COPD, chronic obstructive pulmonary disease; ddMSP, droplet digital methylation-specific polymerase chain reaction; DNA, deoxyribonucleic acid; LC, lung cancer; MSP, methylation-specific polymerase chain reaction; MSRE, methylation-sensitive restriction enzyme; NSCLC, non–small cell lung cancer; PCR, polymerase chain reaction; QMSP, quantitative methylation-specific polymerase chain reaction.

3.4.4. Proteomics

The proteomics category included 41 published studies (refer to Appendix A.5) with a median publication year of 2011. Proteomics studies of sputum were comparatively heterogeneous, ranging from early tumor-marker immunoassays and inflammatory protein panels to contemporary mass spectrometry discovery platforms and biomarker-enabled devices (see

Table 4. Overall summary of sputum-based proteomics studies.

Domain	Representative Techniques	Representative Studies	Main Findings	Typical Setting	Key Limitations
Single-marker tumor protein assays	Induced sputum ELISA for CYFRA21-1, CEA, NSE, etc.	Hillas et al. [86]	CYFRA21-1 was ~7× higher in LC than COPD (86% sensitivity, 75% specificity)	Diagnostic discrimination in symptomatic or clinically suspected LC	Older, single-analyte assays; modest sample size; limited external validation
Exploratory combined protein panels	ELISA for multiple proteins e.g., VEGF, ICAM-1, TNFα, etc.	Bar-Shai et al. [87]	Inflammatory and tumor-related proteins differed significantly among LC, COPD, and healthy controls; combined biomarker score improved case discrimination	Pilot, diagnostic enrichment, case–control studies	Combined biomarker score remains exploratory (calibration missing; external validation uncertain); risk of overfitting
Protein expression markers in exfoliated sputum cells	Cell-block IHC for MCM2, MCM7, etc.	Pankkal et al. [88]	MCM2 (80.3% sensitivity and 100% specificity) and MCM7 (92.1% sensitivity and 100% specificity) augmented conventional cytology	Cytology-linked diagnostic workup using exfoliated sputum cells in suspected LC	Single-study evidence (limited sample size); specialized cytologic processing required
High-dimensional discovery proteomics	diaPASEF (MS) profiling across 527 sputum proteins	Arenas-De Larriva et al. [89]	An internally cross-validated sPLS-DA model discriminated LC from controls, with an AUROC of 0.97	Established or suspected LC compared with controls (case–control designs)	Small sample size; case–control designs; limited external validation
Secretome (EV)-linked predictive proteomics	Proteome analysis of NSCLC cell-line secretomes integrated with patient sputum	Böttger et al. [90]	Feasibility of response prediction (34 sputum-detectable proteins associated with response to cisplatin)	Prospective cohort with established NSCLC undergoing chemotherapy	Early phase evidence; clinical utility uncertain
Multiplex biosensor devices	Portable sputum biosensor based on multichannel organic electrochemical transistor technology	Zhang et al. [91]	Excellent performance (AUROC 0.931) in case–control cohort; potential for longitudinal monitoring	Case–control designs (LC cases and heavy smokers at risk)	Case–control, diagnostically enriched populations; real-world effectiveness yet to be established

Abbreviations: AUROC, area under the receiver operating characteristic curve; CEA, carcinoembryonic antigen; COPD, chronic obstructive pulmonary disease; CRP, C-reactive protein; CYFRA21-1, cytokeratin fragment 19; diaPASEF, data-independent acquisition, parallel accumulation, serial fragmentation; ELISA, enzyme-linked immunosorbent assay; ICAM, intercellular adhesion molecule; IHC, immunohistochemistry; LC, lung cancer; MCM, minichromosome maintenance; NSE, neuron-specific enolase; sPLS-DA, sparse partial least squares discriminant analysis; SCLC, small cell lung cancer; TNFα, tumor necrosis factor-alpha; VEGF, vascular endothelial growth factor.

). Collectively, the available evidence suggests that sputum proteins may support LC detection, subtype discrimination, prognostication, surveillance, and possibly treatment-response assessment. The earlier literature mostly focused on single sputum-based biomarkers for diagnostic or classification purposes using immunohistochemistry or ELISA, while the more recent literature focused on high-dimensional quantitative proteomics for prognostication, response assessment, and surveillance.

Earlier work showed that conventional protein tumor markers in induced sputum could have diagnostic utility: CYFRA21-1 was approximately seven-fold higher in LC than COPD and achieved 86% sensitivity and 75% specificity, outperforming other measured markers such as CEA and NSE [86]. Other pilot studies suggested that combining inflammatory, angiogenic, autophagy, and adhesion-related proteins in induced sputum may improve case discrimination, as neutrophils, beclin-1, VEGF, ICAM, and TNFα differed significantly in LC versus COPD and healthy controls [87], although the derived combined score was exploratory and lacked full diagnostic calibration metrics. A separate immunocytochemical study of sputum cell blocks showed particularly strong diagnostic performance for proliferation-associated proteins, with MCM2 yielding 80.3% sensitivity and 100% specificity and MCM7 yielding 92.1% sensitivity and 100% specificity [88], suggesting that protein expression markers in exfoliated sputum cells may substantially augment conventional sputum cytology.

More recent proteomics literature employing high-dimensional quantitative proteomics (such as diaPASEF) demonstrated superior performance compared to traditional immunohistochemical methods. Arenas-De Larriva et al. [89] quantified 527 sputum proteins and identified inflammatory and immune-related differences between cancer and control samples, with CRP and SERPINA1 among the most upregulated proteins; an internally cross-validated sPLS-DA model achieved an apparently excellent AUROC of 0.97 for cancer versus control and very high performance for SCLC, although the study was limited by small size and lack of external validation. In a different clinical role, Böttger et al. [90] identified 34 sputum-detectable secretome-derived proteins linked to cisplatin sensitivity, including UGGT1, COL6A1, and MAP4, supporting the feasibility of sputum proteomics for predictive biomarker development rather than diagnosis alone. Technology-driven approaches also showed promise: a portable sputum biosensor integrating CEA, NSE, and CA125 achieved a combined AUROC of 0.931, 87.0% sensitivity, and 86.5% specificity in a case–control cohort, while also demonstrating potential for longitudinal treatment monitoring [91]. Overall, sputum proteomics appears promising across several potential applications, but the evidence remains mostly early-phase, with many studies limited by case–control designs, modest sample sizes, incomplete external validation, or exploratory model development.

3.4.5. Transcriptomics

Evidence in this domain comprised 29 published studies (Appendix A.4), with a later median year of publication, 2016, indicating a more contemporary wave of sputum biomarker research. These studies encompassed messenger RNA (mRNA) and non-coding RNA (ncRNA) approaches and were commonly framed as diagnostic or subtype-related investigations. Compared with DNA-based methods, transcriptomic assays may capture a broader and more dynamic biological signal, but they also appear to be more vulnerable to pre-analytic variability related to RNA integrity, cellular composition, and sputum handling (see Table 5).

Studies evaluating single mRNA, microRNA (miRNA), small nucleolar RNA (snoRNA), circular RNA (circRNA), and other RNA targets for LC diagnosis were small-scale and heterogeneous, although several showed meaningful incremental diagnostic value beyond conventional cytology. Among mRNA-based assays, survivin mRNA was commonly studied: Chen et al. [92] reported that adding sputum survivin RT-PCR to cytology increased diagnostic sensitivity from 47.1% to 80.2%, while Dong et al. [93] found survivin mRNA in 63.5% of sputum samples from LC patients and in none of the COPD controls, with combined survivin plus cytology improving sensitivity from 37.5% to 78.8% without loss of specificity. A large cross-sectional study by Chen et al. [94] further reported that template-ready PCR detection of hTERT mRNA in sputum achieved 84.2% sensitivity and approximately 96% specificity compared with non-malignant pulmonary disease controls, suggesting that assay innovation can substantially improve transcript-based detection. In the non-coding RNA space, Bagheri et al. [95] found that sputum miR-223 was markedly upregulated in NSCLC and yielded an AUROC of 0.90, with 82% sensitivity and 95% specificity, outperforming the other tested miRNAs and snoRNAs. Bai et al. [96] identified sputum circ_0006949 as a promising circRNA biomarker that was elevated in NSCLC compared with both healthy and lung infection controls and reportedly outperformed conventional serum tumor markers. By contrast, Lacroix et al. [97] showed only limited sputum positivity for preproGRP RT-PCR in SCLC (22%), indicating more modest clinical sensitivity despite strong analytical detection capability.

Across panel-based sputum transcriptomics studies, multi-marker RNA signatures generally outperformed single-analyte assays, although performance varied by platform, intended use case, and study design. Early miRNA panel studies showed promising diagnostic accuracy: Yu et al. [98] developed a four-miRNA panel (miR-486, miR-21, miR-200b, miR-375) that achieved 80.6% sensitivity and 91.7% specificity in the optimization cohort for stage I lung adenocarcinoma and retained moderate accuracy in an independent validation cohort (70.3% sensitivity, 80.0% specificity overall), with better performance for adenocarcinoma and peripheral tumors. For squamous cell carcinoma, Xing et al. [99] identified a three-miRNA panel (miR-205, miR-210, miR-708) with 73% sensitivity and 96% specificity in the optimization set and 72% sensitivity and 95% specificity in an independent validation cohort, with similar sensitivity across stages I–IV. Other sputum miRNA panels yielded comparable but somewhat lower performance, including a five-miRNA panel in Roa et al. [100] with 83.3% sensitivity and 100% specificity in a small double-blind validation cohort, a three-miRNA panel in Razzak et al. [101] with 67% sensitivity/90% specificity for early-stage NSCLC and 64%/100% for advanced disease, and a two-miRNA digital PCR panel in Li et al. [102] with 65.7% sensitivity and 85.0% specificity. Methodologically stronger integrative studies suggested that combining sputum RNA biomarkers with other modalities improved clinical utility. Shen et al. [103] showed that a sputum miR-31/miR-210 panel alone had modest sensitivity (~65%) but high specificity (~90%), and that combining with CT improved specificity from ~84% to ~92% while maintaining sensitivity above 92% across training and validation sets. Similarly, in smokers with CT-detected indeterminate solitary pulmonary nodules, Xing et al. [104] reported that a three-miRNA sputum panel (miR-21, miR-31, miR-210) achieved 82.9% sensitivity and 87.8% specificity in training, with stable performance in both internal and external testing cohorts (80.5–82.1% sensitivity, 86.1–88.4% specificity), indicating potential value for distinguishing malignant from benign nodules. Other panel approaches included a two-snoRNA panel (snoRD66, snoRD78) that yielded reproducible sensitivity and specificity of approximately 75% and 84% in both training and testing cohorts and improved the specificity of CT when combined with imaging [105,106]. A combined five-ncRNA sputum panel also reached 89.1% sensitivity and 89.1% specificity for stage I NSCLC [106], while a three-marker non-coding RNA panel (miR-145, miR-126, miR-7) reportedly achieved 90% sensitivity and 90% specificity [107], though these studies were more vulnerable to overfitting because validation was limited or internal only. Overall, panel-based sputum transcriptomics appeared more diagnostically informative than single-marker RNA assays and added value to LDCT or cytology. However, most included studies used case–control designs with hospital-based populations, and only a minority included robust external validation in clinically relevant nodule evaluation settings, which limits generalizability.

Table 5. Overall summary of sputum-based transcriptomics studies.

Transcriptomic Approach	Representative Biomarkers	Representative Studies	Main Findings	Key Limitations
Single-target mRNA assays	Survivin mRNA RT-PCR; hTERT mRNA template-ready PCR; preproGRP RT-PCR	Chen et al. [92] Chen et al. [94] Dong et al. [93] Lacroix et al. [97]	Single-gene mRNA assays improved diagnostic yield over sputum cytology; hTERT mRNA showed strong specificity compared with benign pulmonary controls; preproGRP showed limited sensitivity for SCLC	Sample size; heterogeneous studies; RNA integrity, sputum handling, and cellular composition likely influenced performance
Single ncRNA markers	miR-223; circ_0006949	Bagheri et al. [95] Bai et al. [96]	Promising discriminatory ability; miR-223 and circ_0006949 showed specificity for NSCLC	Small and exploratory studies; limited external validation; risk of overfitting
Histology-oriented miRNA panels	Four-miRNA adenocarcinoma panel (miR-486, miR-21, miR-200b, miR-375); three-miRNA squamous cell carcinoma panel (miR-205, miR-210, miR-708)	Yu et al. [98] Xing et al. [99]	Panels outperformed single target assays; consistent across training and validation cohorts; potential for sputum-based histologic subtyping	Hospital-based case–control cohorts with diagnostically enriched populations; broad applicability uncertain
General diagnostic miRNA panels	Five-miRNA, three-miRNA, and two-miRNA digital PCR panels	Roa et al. [100] Razzak et al. [101] Li et al. [102]	miRNA panels had high sensitivity and specificity for NSCLC; digital PCR more precise and feasible for sputum-based quantification	Performance varied across platforms, panel composition, and disease stage; small sample size; internal validation only
Integrated RNA plus imaging approaches	miR-31/miR-210 panel with CT; three-miRNA panel for indeterminate SPN	Shen et al. [103] Xing et al. [104]	Sputum RNA biomarkers combined with CT improved sensitivity and specificity for SPN triage and diagnosis	Independent, external validation awaited; impact on clinical outcomes remains to be demonstrated
snoRNA and broader ncRNA panels	snoRD66/snoRD78; five-ncRNA panel; miR-145/miR-126/miR-7 panel	Su et al. [106]	snoRNA and mixed ncRNA panels had moderate to strong diagnostic performance; improved specificity when combined with CT	Small sample size; limited external validation; risk of overfitting

Abbreviations: AUROC, area under the receiver operating characteristic curve; CT, computed tomography; GRP, gastrin-releasing peptide; hTERT, human telomerase reverse transcriptase; LC, lung cancer; LDCT, low-dose computed tomography; mRNA, messenger ribonucleic acid; miRNA, micro-ribonucleic acid; ncRNA, non-coding ribonucleic acid; NSCLC, non–small cell lung cancer; PCR, polymerase chain reaction; RNA, ribonucleic acid; RT-PCR, reverse transcription polymerase chain reaction; SCLC, small cell lung cancer; snoRNA, small nucleolar ribonucleic acid; SPN, solitary pulmonary nodule.

Table 5. Overall summary of sputum-based transcriptomics studies.

Transcriptomic Approach	Representative Biomarkers	Representative Studies	Main Findings	Key Limitations
Single-target mRNA assays	Survivin mRNA RT-PCR; hTERT mRNA template-ready PCR; preproGRP RT-PCR	Chen et al. [92] Chen et al. [94] Dong et al. [93] Lacroix et al. [97]	Single-gene mRNA assays improved diagnostic yield over sputum cytology; hTERT mRNA showed strong specificity compared with benign pulmonary controls; preproGRP showed limited sensitivity for SCLC	Sample size; heterogeneous studies; RNA integrity, sputum handling, and cellular composition likely influenced performance
Single ncRNA markers	miR-223; circ_0006949	Bagheri et al. [95] Bai et al. [96]	Promising discriminatory ability; miR-223 and circ_0006949 showed specificity for NSCLC	Small and exploratory studies; limited external validation; risk of overfitting
Histology-oriented miRNA panels	Four-miRNA adenocarcinoma panel (miR-486, miR-21, miR-200b, miR-375); three-miRNA squamous cell carcinoma panel (miR-205, miR-210, miR-708)	Yu et al. [98] Xing et al. [99]	Panels outperformed single target assays; consistent across training and validation cohorts; potential for sputum-based histologic subtyping	Hospital-based case–control cohorts with diagnostically enriched populations; broad applicability uncertain
General diagnostic miRNA panels	Five-miRNA, three-miRNA, and two-miRNA digital PCR panels	Roa et al. [100] Razzak et al. [101] Li et al. [102]	miRNA panels had high sensitivity and specificity for NSCLC; digital PCR more precise and feasible for sputum-based quantification	Performance varied across platforms, panel composition, and disease stage; small sample size; internal validation only
Integrated RNA plus imaging approaches	miR-31/miR-210 panel with CT; three-miRNA panel for indeterminate SPN	Shen et al. [103] Xing et al. [104]	Sputum RNA biomarkers combined with CT improved sensitivity and specificity for SPN triage and diagnosis	Independent, external validation awaited; impact on clinical outcomes remains to be demonstrated
snoRNA and broader ncRNA panels	snoRD66/snoRD78; five-ncRNA panel; miR-145/miR-126/miR-7 panel	Su et al. [106]	snoRNA and mixed ncRNA panels had moderate to strong diagnostic performance; improved specificity when combined with CT	Small sample size; limited external validation; risk of overfitting

Abbreviations: AUROC, area under the receiver operating characteristic curve; CT, computed tomography; GRP, gastrin-releasing peptide; hTERT, human telomerase reverse transcriptase; LC, lung cancer; LDCT, low-dose computed tomography; mRNA, messenger ribonucleic acid; miRNA, micro-ribonucleic acid; ncRNA, non-coding ribonucleic acid; NSCLC, non–small cell lung cancer; PCR, polymerase chain reaction; RNA, ribonucleic acid; RT-PCR, reverse transcription polymerase chain reaction; SCLC, small cell lung cancer; snoRNA, small nucleolar ribonucleic acid; SPN, solitary pulmonary nodule.

3.4.6. Metabolomics

Sputum metabolomics studies, totaling 10 in number (see Appendix A.6), were predominantly exploratory case–control investigations using mass spectrometry (MS), nuclear magnetic resonance (NMR), or spectroscopy platforms to identify discriminatory metabolic signatures for LC detection, with most evidence still at the biomarker-discovery stage (see

Table 6. Overall summary of sputum-based metabolomics studies.

Study	Sample Size	Techniques	Main Findings	Key Limitations
Lewis et al. (2010)	50 sputum samples (25 LC and 25 controls)	FTIR spectroscopy of sputum cell pellets; fingerprint-region spectral features linked to glycogen, proteins, and nucleic-acid-associated bands	FTIR-based sputum profiling separated LC from control using a small set of discriminatory wavenumbers (metabolic fingerprinting feasibility)	Case–control design with non-representative healthy controls; data-driven feature selection; absence of external validation
Ahmed et al. (2016)	20 sputum specimens (10 NSCLC and 10 benign lung conditions)	1H-MRS/NMR of sputum and EBC to assess low–molecular weight metabolites such as glucose, methanol, acetate, propionate, lysine, and formate	Relative absence of glucose in sputum and lower methanol in EBC noted in patients with NSCLC (biological feasibility)	Tiny sample size; predominantly advanced-stage NSCLC; hypothesis-generating findings
Ardatskaya et al. (2016)	147 patients (60 LC, 21 LC + CAP, and 38 COPD) and 30 healthy controls	Gas-liquid chromatographic analysis of sputum SCFA: acetate, propionate, and butyrate fractions plus anaerobic index	SCFA profiling showed systematic differences across healthy controls, COPD, LC, and LC complicated by CAP	Mixed disease cohorts; diagnostic metrics not reported; lack of external validation; reproducibility unclear
Cameron et al. (2016)	Sputum from 34 suspected LC cases (16 confirmed) and 33 healthy controls	FIE-MS and GC-MS of gangliosides, polyamines, and lipid metabolites	Untargeted MS profiling identified sputum metabolites that distinguished LC from both healthy controls and symptomatic non-cancer patients	Case–control design; non-representative controls; possible confounding bias
O’Shea et al. (2016)	Sputum from 23 LC cases, 11 symptomatic patients, and 33 healthy volunteers	FIE-MS features integrated with ANN classifiers	Secondary modeling of sputum metabolomic data discriminated LC from control with excellent internally cross-validated diagnostic performance	Tiny sample size; case–control design; lack of external validation; potential overfitting
Zhang et al. (2016)	307 sputum samples (167 NSCLC and 140 controls)	ND-EESI-MS lipid fingerprinting of DPPC, PG, PGP, and related phospholipid species	NSCLC sputum showed lower relative abundance of DPPC and higher PG and PGP compared with controls	No diagnostic metrics reported; lack of external validation; reproducibility unclear; potential overfitting
Gao et al. (2018)	100 sputum samples (50 NSCLC and 50 controls)	ND-EESI-MS analysis of spontaneous sputum without extensive pretreatment	ND-EESI-MS identified sputum fingerprints that could differentiate NSCLC patients from healthy controls by PCA	Case–control design with diagnostically enriched population; possible confounding bias
Zheng et al. (2021)	143 spontaneous sputum samples (76 adenocarcinoma and 67 controls)	ND-EESI-MS with PLS-DA or OPLS-DA to assess hydroxyphenyllactic acid, phytosphingosine, N-nonanoylglycine, sphinganine, and S-carboxymethyl-L-cysteine	A five-metabolite sputum panel discriminated lung adenocarcinoma from controls with high accuracy; pathway analysis implicated sphingolipid metabolism, fatty-acid metabolism, carnitine synthesis, and the Warburg effect	Case–control design with diagnostically enriched population; lack of external validation; potential overfitting
Ahmed et al. (2022)	15 sputum specimens (2 squamous cell carcinoma; 13 adenocarcinoma)	Pre- versus post-surgical resection NMR and LC-QTOF-MS of sputum and EBC: lipids, purines, carnitines, glucose, acetate, propionate, AMP, and diacetylspermine	Numerous sputum and EBC metabolites changed after resection; potential utility for treatment-response assessment and recurrence surveillance; sputum changes included glucose, adenosine monophosphate, and N1,N12-diacetylspermine	Tiny sample size; confounding bias from surgery-related physiologic stress; hypothesis-generating findings

Abbreviations: AMP, adenosine monophosphate; ANN, artificial neural network; AUROC, area under the receiver operating characteristic curve; CAP, community-acquired pneumonia; CID, collision-induced dissociation; DPPC, dipalmitoyl phosphatidylcholine; EBC, exhaled breath condensate; FIE-MS, flow infusion electrospray ion mass spectrometry; FTIR, Fourier transform infrared spectroscopy; GC-MS, gas chromatography–mass spectrometry; LC-QTOF-MS, liquid chromatography quadrupole time-of-flight mass spectrometry; MRS, magnetic resonance spectroscopy; ND-EESI-MS, neutral desorption extractive electrospray ionization mass spectrometry; NMR, nuclear magnetic resonance (spectroscopy); NSCLC, non–small cell lung cancer; OPLS-DA, orthogonal partial least squares discriminant analysis; PCA, principal component analysis; PG, phosphatidylglycerol; PGP, phosphatidylglycerol phosphate; PLS-DA, partial least squares discriminant analysis; SCFA, short-chain fatty acid.

). Early feasibility work by Ahmed et al. [108] showed that metabolomic profiling of induced sputum and exhaled breath condensate (EBC) was technically feasible in advanced NSCLC, with reduced methanol in EBC and a possible signal from absent sputum glucose, although the study was very small and limited to stage III–IV disease. Later longitudinal work from the same group [109,110] examined metabolic shifts before and after surgical resection in early-stage NSCLC, identifying postoperative changes in sputum metabolites such as glucose, adenosine monophosphate, and N1,N12-diacetylspermine, thereby supporting biologic tumor-related metabolic reversibility, but these studies were not designed to establish diagnostic accuracy. Cross-sectional discovery studies suggested that sputum metabolic patterns can distinguish cancer from controls: Cameron et al. [111] found clear metabolomic separation between LC cases and healthy controls, with several metabolites, including ganglioside GM1, achieving AUROCs above 0.8. Similarly, Zhang et al. [112] reported altered sputum phospholipids in NSCLC, with lower dipalmitoyl phosphatidylcholine and higher phosphatidylglycerol and phosphatidylglycerol phosphate in cancer than in controls, while Gao et al. [113] also observed clear separation of cancer and healthy groups by ND-EESI-MS, with phosphatidylcholines as key discriminators. Among the more mature metabolomic studies, Zheng et al. [114] developed a five-metabolite sputum panel for lung adenocarcinoma using ND-EESI-MS and reported an AUROC of 0.917, with 90% sensitivity and 80% specificity, supported by internal hold-out validation and pathway analyses implicating sphingolipid, fatty acid, and glycolytic metabolism. Other studies were less directly diagnostic: Lewis et al. [115] showed proof-of-concept separation of LC and healthy controls by FTIR spectral features, while Ardatskaya et al. [116] found increased sputum short-chain fatty acids in LC and related respiratory disease states, suggesting altered airway microbial metabolism but without formal cancer diagnostic metrics. Overall, sputum metabolomics appears biologically informative and potentially capable of discriminating LC from non-cancer states, but the literature remains limited by small discovery cohorts, case–control enrichment, mixed control populations, variable platforms, and sparse external validation.

3.4.7. Metagenomics and Microbiomics

Sputum microbiomic and metagenomic studies (total studies: 19) were generally exploratory and heterogeneous in purpose, spanning LC detection, histologic stratification, metastatic phenotyping, and treatment-response prediction (see Table 7 and Appendix A.8). In diagnostic case–control settings, several 16S-based studies showed that sputum microbial community structure differed between LC and non-cancer groups, although global diversity findings were inconsistent, and few studies reported clinically usable diagnostic metrics. Baranova et al. [117] found significant beta-diversity differences between squamous cell lung carcinoma and controls, with enrichment of Firmicutes, Streptococcus, Bacillus, Gemella, and Haemophilus in cancer sputum, and Streptococcus agalactiae emerged as the most prominent species-level signal, although no AUROC, sensitivity, or specificity estimates were provided. Earlier pilot metagenomic work by Cameron et al. [118] also suggested enrichment of Streptococcus viridans and Granulicatella adiacens in cancer-associated sputum, but interpretation was limited by the extremely small sample size. Druzhinin et al. [119] similarly reported increased Haemophilus and Bergeyella in LC sputum in one study and increased Streptococcus, Bacillus, Gemella, and Haemophilus in a larger follow-up cohort [120], alongside higher chromosomal aberration and micronucleus frequencies in peripheral lymphocytes, suggesting a possible link between airway dysbiosis and systemic genomic instability.

Some studies suggested that sputum microbiome patterns may vary by histology or disease extent rather than simply by cancer presence. Druzhinin et al. [121] reported that sputum microbiome alterations were more evident in squamous cell carcinoma than adenocarcinoma, with Streptococcus, Bacillus, Peptostreptococcus, Prevotella, Rothia, and Actinobacillus enriched in squamous cell carcinoma, whereas adenocarcinoma showed no significant differences from healthy donors. Likewise, Huang et al. [122] found that, in NSCLC, Granulicatella and Actinobacillus were enriched in early-stage disease, Actinomyces in advanced disease, Peptostreptococcus in intrathoracic metastasis, and Parvimonas in lymph node metastasis and EGFR-mutant adenocarcinoma, supporting a clinicopathologic association of sputum microbiota with tumor phenotype. Lu et al. [123] further showed that sputum microbiota was more strongly associated with NSCLC and distant metastasis than gut microbiota, with sputum-only random-forest models achieving AUROCs of 0.750 for control versus NSCLC, 0.850 for control versus stage I–III disease, and 0.720 for brain versus non-brain metastasis; Pseudomonas was particularly enriched in patients with brain metastases.

Evidence for clinically actionable sputum microbiome biomarkers was strongest when bacterial markers were integrated with other analytes or when microbiota were studied in relation to treatment outcomes. Dhilipkannah et al. [124] reported that combining sputum bacterial DNA markers with circulating plasma miRNAs produced an integrative panel with 87% sensitivity and 89% specificity for LC detection, outperforming the individual biomarker classes alone and maintaining performance in an independent validation cohort. In the immunotherapy setting, Zhang et al. [125] found that baseline sputum microbiota showed moderate ability to predict anti–PD-1 response in metastatic NSCLC, with Streptococcus yielding an AUROC of 0.77 and correlating with tumor CD8+ T-cell density, while Zapata-García et al. [126] observed that immune checkpoint inhibitor responders had higher airway alpha diversity, lower Firmicutes and Streptococcus, and greater abundance of Fusobacterium and Porphyromonas; in that cohort, Gemella predicted non-response, and Lachnoanaerobaculum predicted response. Overall, sputum microbiomics research suggests that airway dysbiosis is biologically linked to LC presence, histology, metastatic behavior, and possibly immunotherapy response, but the field remains largely hypothesis-generating, because most studies were single-center, cross-sectional, or case–control, often lacked external validation, and rarely produced standardized, transportable diagnostic models.

Table 7. Overall summary of sputum-based metagenomics and microbiomics studies.

Metagenomic Approaches	Representative Biomarkers	Representative Studies	Main Findings	Key Limitations
Exploratory diagnostic microbiome profiling	16S rRNA shotgun sequencing; microbial community composition and β-diversity analyses	Baranova et al. [127] Cameron et al. [118] Druzhinin et al. [119]	Sputum microbial composition differed between LC and non-cancer controls; airway dysbiosis accompanies LC	Small, single-center, and exploratory studies; global diversity findings were inconsistent; performance metrics missing in many studies
Histology-specific microbiome stratification	Subtype-focused sputum bacterial profiling in squamous cell carcinoma versus adenocarcinoma	Druzhinin et al. [121] Baranova et al. [117]	Microbiome alterations more pronounced in squamous cell carcinoma than in adenocarcinoma	Hypothesis-generating evidence based on cross-sectional study; external, independent validation needed
Stage, metastasis, and molecular phenotype associations	Association analyses linking sputum microbiota to early versus advanced stage, metastatic pattern, and EGFR mutation status	Huang et al. [122] Lu et al. [123]	Within NSCLC, sputum microbial patterns were associated with stage, metastasis, and molecular alterations	Biological feasibility studies; confounding bias possible; needs larger scale validation
Machine-learning diagnostic models	Random-forest models based on sputum microbial signatures	Lu et al. [123]	Machine-learning models afforded fair discriminatory performance (AUROC 0.75) for NSCLC	Limited external validation; uncertain reproducibility across populations and sequencing pipelines
Integrated multi-analyte biomarker panels	Sputum bacterial DNA markers combined with circulating plasma miRNAs	Dhilipkannah et al. [124]	Combined sputum DNA and plasma miRNA panel had 87% sensitivity and 89% specificity	Impact on patient outcomes uncertain; independent validation needed
Immunotherapy response prediction	Baseline sputum microbiota as predictors of anti-PD-1 ICI response	Zhang et al. [128] Zapata-Garcia et al. [126]	Higher airway α-diversity and enrichment of certain taxa associated with response to anti-PD-1 ICI therapy	Early-phase, retrospective, exploratory evidence; confounding bias possible; clinical utility remains to be demonstrated

Abbreviations: AUROC, area under the receiver operating characteristic curve; DNA, deoxyribonucleic acid; EGFR, epidermal growth factor receptor; ICI, immune checkpoint inhibitor; LC, lung cancer; mRNA, messenger ribonucleic acid; miRNA, micro-ribonucleic acid;;NSCLC, non–small cell lung cancer; PD-1, programmed cell death protein-1; rRNA, ribosomal ribonucleic acid.

Table 7. Overall summary of sputum-based metagenomics and microbiomics studies.

Metagenomic Approaches	Representative Biomarkers	Representative Studies	Main Findings	Key Limitations
Exploratory diagnostic microbiome profiling	16S rRNA shotgun sequencing; microbial community composition and β-diversity analyses	Baranova et al. [127] Cameron et al. [118] Druzhinin et al. [119]	Sputum microbial composition differed between LC and non-cancer controls; airway dysbiosis accompanies LC	Small, single-center, and exploratory studies; global diversity findings were inconsistent; performance metrics missing in many studies
Histology-specific microbiome stratification	Subtype-focused sputum bacterial profiling in squamous cell carcinoma versus adenocarcinoma	Druzhinin et al. [121] Baranova et al. [117]	Microbiome alterations more pronounced in squamous cell carcinoma than in adenocarcinoma	Hypothesis-generating evidence based on cross-sectional study; external, independent validation needed
Stage, metastasis, and molecular phenotype associations	Association analyses linking sputum microbiota to early versus advanced stage, metastatic pattern, and EGFR mutation status	Huang et al. [122] Lu et al. [123]	Within NSCLC, sputum microbial patterns were associated with stage, metastasis, and molecular alterations	Biological feasibility studies; confounding bias possible; needs larger scale validation
Machine-learning diagnostic models	Random-forest models based on sputum microbial signatures	Lu et al. [123]	Machine-learning models afforded fair discriminatory performance (AUROC 0.75) for NSCLC	Limited external validation; uncertain reproducibility across populations and sequencing pipelines
Integrated multi-analyte biomarker panels	Sputum bacterial DNA markers combined with circulating plasma miRNAs	Dhilipkannah et al. [124]	Combined sputum DNA and plasma miRNA panel had 87% sensitivity and 89% specificity	Impact on patient outcomes uncertain; independent validation needed
Immunotherapy response prediction	Baseline sputum microbiota as predictors of anti-PD-1 ICI response	Zhang et al. [128] Zapata-Garcia et al. [126]	Higher airway α-diversity and enrichment of certain taxa associated with response to anti-PD-1 ICI therapy	Early-phase, retrospective, exploratory evidence; confounding bias possible; clinical utility remains to be demonstrated

Abbreviations: AUROC, area under the receiver operating characteristic curve; DNA, deoxyribonucleic acid; EGFR, epidermal growth factor receptor; ICI, immune checkpoint inhibitor; LC, lung cancer; mRNA, messenger ribonucleic acid; miRNA, micro-ribonucleic acid;;NSCLC, non–small cell lung cancer; PD-1, programmed cell death protein-1; rRNA, ribosomal ribonucleic acid.

3.4.8. Integromics/Multi-Omics Approaches

Sputum integromics (multi-omics) studies consistently suggested that combining biomarker classes improves performance over single-analyte approaches, particularly for early detection and nodule triage (see Table 8 and Appendix A.7). Early integrative work combined genetic and epigenetic alterations in sputum, showing that multi-marker panels could detect tumor-related abnormalities even when cytology was negative. Hsu et al. [129] selected a seven-marker panel comprising microsatellite instability/loss of heterozygosity and promoter methylation (D9S942, D9S286, GATA49D12, D13S170, p16, RARβ) and reported 82% sensitivity and 75% specificity for identifying cancer cells in cytologically negative sputum, with one high-risk control later developing LC. In a prospective CT screening cohort of cancer-free heavy smokers, Baryshnikova et al. [130] found sputum molecular abnormalities in 6.9% of participants, most commonly p16^INK4A methylation, indicating that integrated molecular screening could detect preclinical airway field changes, although predictive value for future cancer remained limited, because few cancers developed during follow-up.

The strongest contemporary evidence came from quantitative multi-omics panels that integrated sputum biomarkers with either plasma analytes or radiologic variables. Su et al. [131] showed that combining sputum miRNAs (miR-31, miR-210) with sputum methylation markers (RASSF1A, 3OST2) produced a four-biomarker panel with 87.3% sensitivity and 90.4% specificity in training and 87.5% sensitivity and 89.5% specificity in validation for stage I NSCLC among smokers with CT-detected benign versus malignant nodules, clearly outperforming sputum cytology. Li et al. [82] extended this concept across biospecimens using microplate ddPCR, integrating sputum miRNAs and methylated DNA with plasma miRNAs; their panel achieved about 92% sensitivity and 92% specificity in both development and validation cohorts. Similarly, Liao et al. [132] combined plasma miRNAs, sputum RASSF1A methylation, and clinical-radiologic predictors such as smoking pack-years and nodule diameter into an integromic signature for LDCT-detected pulmonary nodules, yielding an AUROC of 0.97 with 90% sensitivity and 94% specificity in validation. In the screening setting, Carozzi et al. [133] integrated sputum and plasma molecular biomarkers with LDCT and showed that adding the biomarker panel improved LDCT specificity from 74% to 89% and positive predictive value from 4.3% to 10.6% while maintaining 90% sensitivity, supporting a role for multi-omics triage of screen-positive individuals. Overall, sputum integromics appears to offer the most clinically promising performance among molecular approaches, particularly when multiple analyte classes and imaging features are combined, but most studies still used enriched case–control or nodule-clinic populations and require prospective validation in true screening workflows.

Table 8. Overall summary of sputum-based integromics/multi-omics studies.

Integromics Approach	Representative Biomarkers	Representative Studies	Main Findings	Key Limitations
Multimodality sputum assessment	Combined sputum cytopathology (DNA cytometry; FISH for LOH and MSI), genomics (K-RAS and TP53 mutations), methylomics (p16INK4a and RASSF1A promoter hypermethylation), proteomics (MAGE A1-A6) and transcriptomics (miR-31, miR-210)	Kersting et al. (2000) [134] Baryshnikova et al. (2008) [130] Shin et al. (2012) [135] Su et al. (2016) [131]	Combining multiple sputum-based biomarkers improves upon the performance of conventional cytology; performance metrics suggest strong sensitivity, specificity, and overall accuracy in nested case–control studies	Although stronger than conventional approaches, assays relied on targeted biomarkers/candidates selected from earlier work; nested case–control designs with unclear applicability to real-world cohorts
Sputum LB combined with blood-based LB	Microplate ddPCR quantification of multiple sputum miRNAs, sputum DNA methylation, and plasma miRNAs	Li et al. (2021) [82]	A 96-well ddPCR workflow simultaneously quantified many candidate targets and identified an integrated biomarker panel spanning sputum and plasma that outperformed single biomarker classes for early LC diagnosis, with reproducible validation in an independent cohort	Retrospective case–control design; applicability to real-world screening cohorts remains unclear
Sputum LB in conjunction with imaging	ITALUNG biomarker panel (IBP) combined with sputum cytopathology and LDCT results	Carozzi et al. (2017) [133]	The IBP showed very high positivity among baseline screen-detected LC; when combined with LDCT, IBP improved specificity and positive predictive value relative to single-test screening	IBP alone had lower specificity when used alone; simulation-based extrapolation used for multimodal performance metrics; further validation needed in real-world screening cohorts
True multi-omics (clinical, imaging, blood, and sputum) approach	Clinical variables (e.g., smoking), imaging (e.g., SPN), sputum LB (e.g., microbial signals, methylomics) and plasma LB (e.g., ncRNA profile)	Liao et al. (2024) [132]	This mature “integromic” framework (molecular signals from plasma and sputum combined with radiologic and clinical predictors) could distinguish malignant from benign LDCT-detected nodules with sufficient specificity, outperforming other approaches	Low sensitivity for stage I disease; impact on patient outcomes needs to be validated in real-world screening cohorts

Abbreviations: ddPCR, droplet digital polymerase chain reaction; DNA, deoxyribonucleic acid; FISH, fluorescent in situ hybridization; IBP, ITALUNG biomarker panel; K-RAS, Kirsten rat sarcoma viral oncogene homolog; LB, liquid biopsy; LDCT, low-dose computed tomography; LOH, loss of heterozygosity; MAGE-A, melanoma-associated antigen A family; MSI, microsatellite instability; ncRNA, non-coding ribonucleic acid; p16^INK4a, inhibitor of cyclin-dependent kinase–4 family, 16 kDa protein; p53, tumor protein 53; RASSF1A, Ras association domain family 1 isoform A; SPN, solitary pulmonary nodule.

Table 8. Overall summary of sputum-based integromics/multi-omics studies.

Integromics Approach	Representative Biomarkers	Representative Studies	Main Findings	Key Limitations
Multimodality sputum assessment	Combined sputum cytopathology (DNA cytometry; FISH for LOH and MSI), genomics (K-RAS and TP53 mutations), methylomics (p16INK4a and RASSF1A promoter hypermethylation), proteomics (MAGE A1-A6) and transcriptomics (miR-31, miR-210)	Kersting et al. (2000) [134] Baryshnikova et al. (2008) [130] Shin et al. (2012) [135] Su et al. (2016) [131]	Combining multiple sputum-based biomarkers improves upon the performance of conventional cytology; performance metrics suggest strong sensitivity, specificity, and overall accuracy in nested case–control studies	Although stronger than conventional approaches, assays relied on targeted biomarkers/candidates selected from earlier work; nested case–control designs with unclear applicability to real-world cohorts
Sputum LB combined with blood-based LB	Microplate ddPCR quantification of multiple sputum miRNAs, sputum DNA methylation, and plasma miRNAs	Li et al. (2021) [82]	A 96-well ddPCR workflow simultaneously quantified many candidate targets and identified an integrated biomarker panel spanning sputum and plasma that outperformed single biomarker classes for early LC diagnosis, with reproducible validation in an independent cohort	Retrospective case–control design; applicability to real-world screening cohorts remains unclear
Sputum LB in conjunction with imaging	ITALUNG biomarker panel (IBP) combined with sputum cytopathology and LDCT results	Carozzi et al. (2017) [133]	The IBP showed very high positivity among baseline screen-detected LC; when combined with LDCT, IBP improved specificity and positive predictive value relative to single-test screening	IBP alone had lower specificity when used alone; simulation-based extrapolation used for multimodal performance metrics; further validation needed in real-world screening cohorts
True multi-omics (clinical, imaging, blood, and sputum) approach	Clinical variables (e.g., smoking), imaging (e.g., SPN), sputum LB (e.g., microbial signals, methylomics) and plasma LB (e.g., ncRNA profile)	Liao et al. (2024) [132]	This mature “integromic” framework (molecular signals from plasma and sputum combined with radiologic and clinical predictors) could distinguish malignant from benign LDCT-detected nodules with sufficient specificity, outperforming other approaches	Low sensitivity for stage I disease; impact on patient outcomes needs to be validated in real-world screening cohorts

Abbreviations: ddPCR, droplet digital polymerase chain reaction; DNA, deoxyribonucleic acid; FISH, fluorescent in situ hybridization; IBP, ITALUNG biomarker panel; K-RAS, Kirsten rat sarcoma viral oncogene homolog; LB, liquid biopsy; LDCT, low-dose computed tomography; LOH, loss of heterozygosity; MAGE-A, melanoma-associated antigen A family; MSI, microsatellite instability; ncRNA, non-coding ribonucleic acid; p16^INK4a, inhibitor of cyclin-dependent kinase–4 family, 16 kDa protein; p53, tumor protein 53; RASSF1A, Ras association domain family 1 isoform A; SPN, solitary pulmonary nodule.

3.4.9. Other: Laboratory Techniques and Sputum Processing

Studies in this category (see Appendix A.9) were largely methodological and focused on improving specimen adequacy, preserving analytes, and enriching diagnostically relevant material rather than directly evaluating cancer discrimination. Early proof-of-concept cell enrichment studies showed that automated or flow-based approaches could substantially increase the proportion of target respiratory or malignant cells in sputum: Frost et al. [136] reported 7.8-fold enrichment of neoplastic cells using fluorescence parameters alone and 10.5-fold enrichment when fluorescence was combined with light scatter, while Kraemer et al. [137] increased diploid respiratory epithelial cell purity from 1.1% to 42%, corresponding to approximately 38-fold enrichment, thereby facilitating downstream molecular analysis. Pre-analytic handling was also shown to materially affect sputum assay quality. Gottschall et al. [138] demonstrated that nuclear morphometric measurements were highly sensitive to fixative choice and environmental exposure, with formalin-containing BD CytoRich Red^™ (Becton Dickinson, Inc.; Franklin Lakes, NJ, USA) outperforming alcohol-based Saccomanno fixative by minimizing temperature- and sunlight-related artifacts; in contrast, sunlight-exposed Saccomanno-fixed samples often became unevaluable for computer-assisted image analysis. van der Drift et al. [139] further showed that, while cfDNA is detectable in all sputum samples, bulk cfDNA quantity is heavily confounded by neutrophilic airway inflammation and COPD, and therefore is not itself a useful discriminator of LC, underscoring the importance of separating tumor-specific signals from inflammatory background. More recent studies extended sputum processing advances to newer analyte classes: Bano et al. [140] showed that exosomes can be reproducibly isolated from sputum and that exosomes from LC patients were significantly larger and quantitatively different than those from smokers and healthy controls, supporting sputum as a technically feasible source for EV biomarker work. Likewise, Ma et al. [141] reported that optimized preservation of cytological supernatants enabled combined cfDNA/cfRNA analysis across several sample types, with sputum achieving a 100% test success rate and 95% sensitivity in the sputum subgroup, suggesting that appropriate mucolysis, fractionation, and nucleic acid stabilization can make sputum suitable for dual-analyte molecular testing. Overall, these studies indicate that sputum processing is a major determinant of downstream biomarker performance, with the most useful strategies being those that enrich epithelial/tumor-derived components, preserve nucleic acid integrity, and reduce artifactual effects from fixation and inflammatory contamination.

4. Discussion

This scoping review demonstrated that sputum LB for LC is a broad and evolving field with a substantial historical foundation and increasing molecular sophistication, but is not yet clinically mature. The included literature spans nearly five decades, beginning with cytopathologic approaches and progressing toward genomic, epigenomic, proteomic, transcriptomic, microbiomic, metabolomic, and integrative multi-omics strategies. This trajectory reflects both technological progress and a conceptual shift in how sputum is understood: it is no longer considered to be merely a cytologic specimen, but a complex and potentially informative biospecimen that may capture host, tumor, and microenvironmental biology in a noninvasive manner. However, most available data remain exploratory, retrospective, and diagnostically enriched; accordingly, encouraging biomarker signals should presently be interpreted as evidence of biologic potential rather than proof of stand-alone clinical readiness. An overall cross-domain summary of the key findings of this scoping review can be found in Supplementary Table S1.

4.1. Key Findings

A key finding of this review is that the field of sputum-based LB remains overwhelmingly centered on diagnosis and early detection. More than three-fourths of studies addressed diagnosis (78%), and more than a quarter were also relevant to screening (30%). A small proportion of studies were potentially relevant for prognostication (14%) and detection of AGAs (15%). By contrast, relatively few studies examined histologic subtyping (5%), surveillance (2%), or response monitoring (6%). This imbalance suggests that sputum LB research is still predominantly positioned in the discovery and early clinical validation phases rather than in mature implementation across the continuum of LC care. Given the low procedural burden of sputum collection and its potential suitability for repeated sampling, the limited literature on longitudinal monitoring represents an important gap.

The dominance of methylomics within the current evidence base is also notable. Among all molecular domains, methylation-based approaches appear to have gained the most traction. This likely reflects both biological and technical advantages. Epigenetic alterations may arise early in carcinogenesis, including in the setting of field cancerization, and are often detectable using relatively sensitive and scalable assays. In contrast, transcriptomic, proteomic, and metabolomic biomarkers may provide richer biological resolution but are highly susceptible to pre-analytic instability, matrix effects, and batch variation. The prominence of methylomics in sputum therefore likely reflects a combination of biological relevance and practical assay feasibility.

The overall translational readiness of sputum-based biomarkers for LC clinical care are summarized in

Table 9. Translational readiness of various biomarkers for clinical use in LC care.

Biomarker Category	Potential Clinical Use Cases	Candidate Biomarkers and/or Techniques	Translational Maturity *	Next Steps
Cytopathology	Diagnosis Screening Histologic subtyping Actionable alterations Prognostication	Flow cytometry	Phase 4 (prospective validation phase)	Evaluation in large, multicenter, prospective cohorts
		MACS	Phase 2 (assay standardization phase)	Assay standardization and further validation
		Papanicolaou smear	Phase 5 (ready for clinical use)	Combine with other approaches for clinical use
		Quantitative microscopy (LungSign®; Perceptronix Medical Inc.; Vancouver, BC, Canada), automated DNA cytometry, 3D morphologic cytometry (LuCED®; Vision Gate Inc.; Pheonix, AZ, USA), PWS microscopy	Phase 3 (retrospective validation phase)	Clinical validation and use in prospective screening cohorts
		Porphyrin labeling (CyPath®; bioAffinity Technologies, Inc.; San Antonio, TX, USA)	Phase 4–5 (prospective validation and early clinical use phase)	Follow results of NCT07168993
		FISH for MSI	Phase 3 (retrospective validation phase)	Assess performance in prospective cohorts
		CA-FISH panel	Phase 3 (retrospective validation phase)
		EGFR copy number assessment	Phase 4 (prospective validation phase)
		FAL-FISH panel	Phase 2–3 (assay development and retrospective validation phase)	Assay development and clinical validation
		TRAP	Phase 2 (assay development phase)	Assay development and clinical validation
Genomics	Diagnosis Screening Actionable alterations Monitoring response Prognostication	EGFR mutations	Phase 2–4 (assay development and clinical validation phase)	Assay standardization and clinical validation
		BRAF mutation	Phase 2 (assay development phase)	Assay development, standardization, and development
		K-RAS mutation	Phase 2–4 (assay development and clinical validation phase)	Assay standardization and clinical validation
		TP53 mutation	Phase 2–4 (assay development and clinical validation phase)	Assay standardization and clinical validation
		EML4-ALK fusion	Phase 2–3 (assay development and clinical validation phase)	Assay development, standardization, and validation
		PD-L1 status	Phase 2 (assay development phase)	Assay development and standardization; clinical validation
		HER2, ROS1, RET, MET, NTRK, and NRG gene alterations	No sputum-based data	Discovery of sputum-based methods of detection
		Multiplex ddPCR panels	Phase 2 (assay development phase)	Assay development and standardization; clinical validation
		NGS profiling
Methylomics	Diagnosis Screening Prognostication	ddMSP panels (p16INK4a, RASSF1A, SOX17, TAC1, etc.)	Phase 3–4 (clinical validation phase)	Clinical validation, evaluation in prospective cohorts, and impact on patient outcomes
		3D quantitative DNA topology imaging	Phase 2 (assay development phase)	Assay development, standardization, and calibration; clinical validation
Transcriptomics	Diagnosis Screening Subtyping Monitoring response Prognostication	Survivin mRNA hTERT mRNA	Phase 2 (assay development phase)	Assay development, standardization, and calibration
		Adenocarcinoma four-miRNA panel Squamous cell carcinoma three-miRNA panel	Phase 3 (clinical validation phase)	Clinical validation and evaluation in prospective cohorts
		snoRNA panel	Phase 2 (assay development phase)	Assay development, standardization, and calibration
Proteomics	Diagnosis Screening Histologic subtyping Actionable alterations Monitoring response Surveillance Prognostication	Cell block IHC for tumor markers	Phase 2–3 (assay development and clinical validation phase)	Clinical validation and evaluation in prospective cohorts
		Single and multiple protein panels (SELDI-TOF/MALDI-TOF and ELISA)	Phase 2–3 (assay development and clinical validation phase)	Assay development and standardization; clinical validation
		diaPASEF (MS) profiling	Phase 2 (assay development phase)	Assay development, standardization, and calibration
		EV-derived proteome profiling, similar to ExoDx™ Lung(ALK) [Exosome Diagnostics Inc.; Waltham, MA, USA]
		Portable biosensors (multichannel electrochemical transistor technology)	Phase 3–4 (assay development and clinical validation phase)	Evaluation in prospective cohorts and assessment of impact on patient outcomes
Metabolomics	Diagnosis Screening Monitoring response Surveillance	FTIR spectroscopy, Raman spectroscopy, GC-MS, FIE-MS, and ND-EESI-MS for lipid fingerprinting, and glucose and glycolytic metabolites	Phase 2 (assay development phase)	Assay development, standardization, and calibration followed by clinical validation
Microbiomics	Diagnosis Histologic subtyping Response prediction Monitoring response Prognostication	16S rRNA sequencing for specific taxa (Gemella, Firmicutes, Bacillus, Granulicatella, etc.)	Phase 2 (assay development phase)	Assay development and calibration, clinical validation, and assessment in prospective cohorts

* Translational maturity as reflected by the current phase of biomarker development, based on the National Cancer Institute Early Detection Research Network’s five-phase framework. Abbreviations: ALK, anaplastic lymphoma kinase; BRAF, v-raf murine sarcoma viral oncogene, homolog B; CA-FISH, chromosomal aneusomy–fluorescence in situ hybridization; ddPCR, droplet digital polymerase chain reaction; diaPASEF, data-independent acquisition, parallel accumulation, serial fragmentation; DNA, deoxyribonucleic acid; EGFR, epidermal growth factor receptor; EML4, echinoderm microtubule-associated protein-like 4; FAL-FISH, fractional allele loss–fluorescence in situ hybridization; FIE-MS, flow infusion electrospray ion-mass spectrometry; FISH, fluorescent in situ hybridization; FTIR, Fourier transform infrared (spectroscopy); GC-MS, gas chromatography–mass spectrometry; HER2, human epidermal growth factor receptor 2; MACS, magnetic-activated cell sorting; MALDI-TOF, matrix-assisted laser desorption/ionization time-of-flight; mRNA, messenger ribonucleic acid; miRNA, micro-ribonucleic acid; MSI, microsatellite instability; ND-EESI-MS, neutral desorption extractive electrospray ionization mass spectrometry; NRG, neuregulin; NTRK, neurotrophic tyrosine receptor kinase; PWS, pulsed wave spectrometry; RET, rearranged during transfection proto-oncogene; rRNA, ribosomal ribonucleic acid; SELDI-TOF, surface-enhanced laser desorption/ionization time-of-flight; TRAP, telomerase repeat amplification protocol.

. In line with NCI EDRN’s five-phase biomarker development framework, sputum-based cytopathologic markers (e.g., CyPath^®; bioAffinity Technologies, Inc.; San Antonio, TX, USA), ddPCR gene panels, ddMSP methylomics panels, miRNA panels, and proteomic-based approaches appear to be the most mature translationally. Nevertheless, it should be noted that, for most domains, the available evidence still supports adjunctive clinical development rather than routine stand-alone implementation of sputum LB.

4.2. Research Gaps and Barriers to Translation

Our review identified several persistent barriers to translation. The first is methodological heterogeneity; included studies varied widely in sputum collection method, whether spontaneous or induced samples were used, which specimen fractions (whole sputum, cellular pellet, or supernatant cfDNA/EV-rich fractions) were analyzed, how samples were processed and stored, what analytes were measured, and how positivity thresholds were defined. This heterogeneity makes direct comparison difficult and complicates efforts to identify which biomarker classes are genuinely the most promising. It also raises the possibility that some variation in performance across studies may reflect differences in specimen handling and assay workflow rather than true biological differences. Future studies should operationalize standardization more explicitly by prespecifying at least the following elements: collection method; time from collection to stabilization; mucolysis, centrifugation, and storage conditions; fraction selection; specimen adequacy and tumor enrichment criteria; extraction chemistry and internal controls; and locked assay thresholds with predefined quality-failure rules. Without harmonization of these pre-analytic and analytic variables, there is no conclusive way to determine if apparent differences between biomarkers are related to workflow variability or true biologic superiority.

The second major barrier is the predominance of higher-risk study designs. Most studies were judged to be at moderate, moderate-high, or high risk of bias, while only a small minority were clearly low-risk. This pattern is reflected by the preponderance of retrospective case–control studies, often using known LC cases (usually advanced stage) and healthy selected controls. Such designs may be useful for identifying candidate signals, but they tend to overestimate performance relative to real-world clinical populations such as screening cohorts, patients with indeterminate pulmonary nodules, symptomatic individuals undergoing workup, and patients with benign inflammatory or smoking-related lung disease. Clinically relevant validation should therefore move away from healthy control comparisons and instead test sputum assays in populations that mirror actual decision points in practice, such as heavy smokers with COPD or smoking-related interstitial lung disease.

A third weakness is the incomplete reporting of diagnostic performance. Sensitivity, specificity, and AUROC were not reported uniformly in all studies, and negative or positive predictive values were selectively reported. Without consistent reporting of clinically meaningful performance metrics, it is difficult to benchmark sputum assays against established standards or to determine where they might fit in the diagnostic pathway. Future validation studies should therefore report not only discrimination metrics, but also intended-use populations, threshold pre-specification, calibration, validation methods, clinically meaningful endpoints, and comparator strategies relevant to clinical practice—such as incremental value beyond LDCT and clinical risk prediction, nodule triage accuracy, reduction in unnecessary invasive procedures, and reproducibility on repeat sampling.

4.3. Implications of Available Evidence

Despite the limitations and gaps identified by this review, the literature provides several reasons for cautious optimism. First, the cumulative volume of studies across multiple domains suggests that sputum does contain biologically relevant LC signals. Second, the progression from single-marker assays to panels and, more recently, to integrative multi-omics approaches indicates increasing recognition that no single analyte class is likely to be sufficient across all clinical contexts. Third, the emergence of microbiome/metagenomic and integromic studies suggests that the field is moving toward more systems-level models of disease detection and characterization, which may ultimately prove more informative than isolated biomarkers alone. At the same time, these observations should not be misconstrued as justification for universal simultaneous multi-omics testing in all patients; rather, they support rational, staged integration within clearly defined clinical pathways.

A putative framework for a comprehensive sputum LB is provided in Figure 8. Based on the 2026 NCCN guidelines, all patients with NSCLC require comprehensive biomarker profiling, including PD-L1 status, EGFR mutations, ALK gene fusion, BRAF V600E mutation, K-RAS G12C mutation, ROS1 gene fusion, HER2 (ERBB2) mutation, NTRK gene fusions, MET exon 14 skipping mutation, RET gene fusion, and NRG1 gene fusion [142]. Sputum is a biologically proximal respiratory specimen that could potentially provide invaluable information regarding the tumor microenvironment. However, given the heavy biological noise in sputum from inflammatory cells and contamination from the upper aerodigestive tract, identifying relevant signals in this specimen is akin to finding a needle in a haystack. Such an elusive goal can only be achieved by standardization of sputum collection and processing methods, use of high-resolution molecular techniques, integration of molecular data from other specimens (if available), and incorporation of clinical and radiological data. In practical and realistic terms, a stepwise testing approach seems most plausible, rather than a universal simultaneous integration strategy. For instance, an initial clinical and radiologic risk assessment may identify a subgroup of patients where sputum LB may provide incremental information that either cannot be acquired by other means or is technically challenging to obtain. In such a stepwise approach, sputum can either serve as an initial medium for LB or be used as a last resort when blood, pleural fluid, or bronchial washings are not available.

From a clinical standpoint, the most plausible near-term role for sputum LB is likely an adjunctive decision support test rather than a stand-alone modality. A realistic implementation pathway could be a two-step or reflex model. Patients would first undergo standard clinical assessment and LDCT or diagnostic imaging; among those with indeterminate pulmonary nodules, discordant imaging–clinical features, or inadequate tissue for molecular workup, a limited sputum assay—for example, automated cytometry or a small prespecified methylation/miRNA panel performed on a defined sputum fraction—could be used to refine risk estimation and guide escalation to biopsy, short-interval imaging, or plasma/tissue genotyping. Such a strategy may be simpler and potentially more scalable than concurrent multi-omics testing across sputum, plasma, imaging, and clinical data in every patient. The inadequacy of current LDCT screening for LC was highlighted by the DELUGE (Detecting Early Lung Cancer in the Mississippi Delta Cohort) trial, which showed that a substantial proportion of patients diagnosed with early-stage LC were ineligible for LDCT screening per 2021 USPSTF criteria [11]. The Nodify Lung^® (Biodesix, Inc.; Louisville, CO, USA) nodule risk assessment strategy utilizes blood-based proteomics to reduce unnecessary testing and improve early detection of LC [21]. Additionally, the Percepta^® (Veracyte, Inc.; San Francisco, CA, USA) test uses nasal swab–based transcriptomics in conjunction with machine learning to facilitate accurate nodule triage [143]. Results of the NIGHTINGALE trial are anticipated in 2028 and will provide an objective assessment of the impact of Percepta^® (Veracyte, Inc.; San Francisco, CA, USA) testing on actual patient outcomes [144]. In screening settings, sputum-based assays may potentially complement clinical-, imaging-, and blood-based risk stratification to refine risk assessment and nodule triage and inform subsequent diagnostic steps. In diagnostic clinical pathways, sputum LB may provide additional noninvasive information when imaging is indeterminate or invasive tissue sampling is difficult, inadequate, or unfeasible. Currently, ongoing studies are assessing the role of novel sputum cytometry (such as CyPath^® [bioAffinity Technologies, Inc.; San Antonio, TX, USA]), sputum NGS (such as OncoScreen Plus^® [Burning Rock Dx; Guangzhou, China]), sputum miRNA panels, sputum DNA methylation panels, and EV-based or four-dimensional proteomic approaches [40,145,146].

Overall, clinically meaningful validation of sputum should focus on real-world populations and pragmatic endpoints such as better benign-versus-malignant discrimination in indeterminate nodules, improved confidence when tissue sampling is difficult, and reduction in unnecessary invasive procedures without sacrificing cancer detection. Complementary airway-based liquid biopsy approaches such as EBC also warrant a brief mention. Although EBC was outside the predefined scope of this scoping review, recent reviews have highlighted its promise as a noninvasive, organ-specific medium for cfDNA and miRNA-based profiling, while emphasizing the same translational barriers of low analyte yield and lack of standardization [147,148]. Pilot and proof-of-concept EBC studies, including genome-wide and NGS-based miRNA analyses, suggest that EBC may ultimately develop as a complementary airway liquid biopsy platform rather than as a direct substitute for sputum [149,150,151].

4.4. Priorities for Future Research

This review also highlights concrete priorities for future research. Prospective cohort-based study designs should be favored over retrospective case–control comparisons, and the utility of sputum LB should be assessed in clinically relevant populations, such as LDCT screening cohorts, patients with indeterminate pulmonary nodules, and patients with benign inflammatory lung disease. Healthy controls should be avoided as a comparator group, as they are not representative of real-world populations undergoing screening or testing. Moreover, cohorts of LC cancers should be representative of all stages of LC, since spectrum bias can lead to overestimation of diagnostic performance. Sputum collection and processing protocols should be standardized and reported in sufficient detail to support reproducibility. External validation across centers, populations, smoking exposures, histologic subtypes, and disease stages is essential. Studies should explicitly define their target clinical use case and benchmark performance against clinically relevant comparators rather than healthy controls alone. Greater attention is also needed for underexplored endpoints such as recurrence detection, serial disease monitoring, prediction of therapeutic response, and reduction in unnecessary invasive diagnostic procedures. Moreover, there are no sputum-based studies published to date that assess HER2 overexpression or alterations of MET, RET, ROS1, and NTRK, which signifies an area of unmet clinical need.

5. Conclusions

Based on the current scoping review, available evidence supports sputum LB as a promising but still methodologically heterogeneous field, with its strongest signal in early detection (screening) and diagnostic applications, particularly utilizing methylomic, genomic, cytopathologic, proteomic, and transcriptomic approaches. However, the literature remains limited by inconsistent reporting, variable pre-analytic methods, and predominantly moderate to high risk of bias. Sputum collection and processing protocols should be standardized and reported in sufficient detail to support reproducibility. External validation across centers, populations, smoking exposures, histologic subtypes, and disease stages is also essential. Progress toward clinical translation will require a shift from exploratory biomarker discovery to standardized, prospective, and clinically anchored validation studies. In this regard, clinical studies should focus on integromic (multi-omic) stepwise approaches that incorporate clinical and imaging data with available molecular data from tissue biopsy or other specimens and then apply sputum LB in a well-defined clinical context to provide incremental value in a simple, scalable, and realistic fashion. Isolated sputum testing is unlikely to capture tumor and patient heterogeneity, which would preclude application across the full spectrum of lung cancer care pathways. Lastly, future clinical research should also focus on currently underexplored endpoints, such as prediction of response to targeted therapy, longitudinal response monitoring, and surveillance after treatment.