Diagnostic performance of volatile organic compounds analysis and electronic noses for detecting colorectal cancer: a systematic review and meta-analysis

Introduction The detection of Volatile Organic Compounds (VOCs) could provide a potential diagnostic modality for the early detection and surveillance of colorectal cancers. However, the overall diagnostic accuracy of the proposed tests remains uncertain. Objective This systematic review is to ascertain the diagnostic accuracy of using VOC analysis techniques and electronic noses (e-noses) as noninvasive diagnostic methods for colorectal cancer within the realm of clinical practice. Methods A systematic search was undertaken on PubMed, EMBASE, Web of Science, and the Cochrane Library to scrutinize pertinent studies published from their inception to September 1, 2023. Only studies conducted on human subjects were included. Meta-analysis was performed using a bivariate model to obtain summary estimates of sensitivity, specificity, and positive and negative likelihood ratios. The Quality Assessment of Diagnostic Accuracy Studies 2 tool was deployed for quality assessment. The protocol for this systematic review was registered in PROSPERO, and PRISMA guidelines were used for the identification, screening, eligibility, and selection process. Results This review encompassed 32 studies, 22 studies for VOC analysis and 9 studies for e-nose, one for both, with a total of 4688 subjects in the analysis. The pooled sensitivity and specificity of VOC analysis for CRC detection were 0.88 (95% CI, 0.83-0.92) and 0.85 (95% CI, 0.78-0.90), respectively. In the case of e-nose, the pooled sensitivity was 0.87 (95% CI, 0.83-0.90), and the pooled specificity was 0.78 (95% CI, 0.62-0.88). The area under the receiver operating characteristic analysis (ROC) curve for VOC analysis and e-noses were 0.93 (95% CI, 0.90-0.95) and 0.90 (95% CI, 0.87-0.92), respectively. Conclusion The outcomes of this review substantiate the commendable accuracy of VOC analysis and e-nose technology in detecting CRC. VOC analysis has a higher specificity than e-nose for the diagnosis of CRC and a sensitivity comparable to that of e-nose. However, numerous limitations, including a modest sample size, absence of standardized collection methods, lack of external validation, and a notable risk of bias, were identified. Consequently, there exists an imperative need for expansive, multi-center clinical studies to elucidate the applicability and reproducibility of VOC analysis or e-nose in the noninvasive diagnosis of colorectal cancer. Systematic review registration https://www.crd.york.ac.uk/prospero/#recordDetails, identifier CRD42023398465.


Introduction
Colorectal carcinoma (CRC) stands as a substantial global public health concern, with an estimated 1.93 million new cases and 0.93 million deaths in 2020 (1).CRC is known to develop from precursor lesions, in most cases adenomas, through the adenomacarcinoma sequence (2) which can be diagnosed earlier through screening even in its early stages.Through standardized early diagnosis and treatment, the 5-year survival rate for early-stage CRC could exceed 90% (1).Fecal immunochemical test (FIT) and colonoscopy screening for colorectal cancer are pivotal tools for early diagnosis of colorectal cancer (3).However, the detection performance of FIT falls short, with a miss detection rate of 9-29% for CRC and 60-75% for advanced CRC (4).FIT-positive patients are recommended to undergo colonoscopy, but colonoscopy is painful, expensive, and invasive, with the risk of complications such as perforation and bleeding.So not all FIT-positive individuals undergo regular colonoscopy follow-up (5,6).Therefore, there is an urgent need for convenient, non-invasive, reliable, simple, and costeffective diagnostic methods to enhance early diagnosis and screening of colorectal cancer.
The analysis of Volatile organic compounds (VOCs) has been applied as a novel and promising diagnostic technique for exploration of non-invasive colorectal neoplasia biomarker.VOCs constitute the by-products of biochemical processes within the human body and typically mirror metabolic states (7,8).Pathological conditions precipitate aberrant metabolic processes, resulting in a marked increase in VOC production (9).Investigations into cancer-related VOCs have explored various matrices, including breath, blood, urine, saliva, and feces (10)(11)(12)(13).Many studies have demonstrated that the applicability of VOC analysis could be used in cancer diagnosis (14-20).
The electronic nose (e-nose) emerges as an instrument equipped with a suite of sensors endowed with specificity and an adept pattern recognition system capable of discerning both simple and complex odors (21).As a relatively recent development, the enose has become widely accepted for detecting diseases, owing to its portability, expeditious, cost-effective, and user-friendly diagnostic capabilities, rendering it particularly suited for routine clinical applications.Multiple researchers (22)(23)(24) have substantiated the commendable diagnostic accuracy of available e-nose technologies across diverse indications.Notably, van Keulen et al. (25) analyzed exhaled breath from patients with CRC and advanced adenomas (AAs), proving that the Aeonose electronic nose can distinguish CRC and AAs from controls.Additionally, de Meij et al. ( 26) reported an e-nose sensitivity of 0.85 and a specificity of 0.87 in CRC detection.
Our article aims to systematically review published studies on VOC analysis and e-nose technology concerning colorectal cancer (CRC) detection.Furthermore, we aim to compare their diagnostic performance, with the aspiration of offering a valuable reference for the application of diagnostic techniques in CRC diagnosis.

Registration
This systematic review has been registered with PROSPERO, under registration number CRD42023398465.The Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines were adhered to in both the identification and reporting phases of this review (27).

Search strategy
A comprehensive literature search encompassing PubMed, Embase, Cochrane Library, and Web of Science was conducted from inception up to September 1, 2023.This search, void of language or data publication restrictions, utilized keywords such as "Volatile Organic Compounds," "VOCs," "electronic nose," "enose," "Colorectal neoplasms," and "diagnosis" or "diagnostic" as search strategy terms.A detailed search strategy is provided in the Supplement.

Study selection
A total of 192 articles were retrieved.The eligibility of each article was assessed through a meticulous examination of titles and abstracts by two independent reviewers (Y.F. and S.Y.T.).Inclusion criteria were as follows: (1) studies conducted on adult subjects; (2) studies involving colorectal patients; and (3) studies that identified evaluating the diagnostic accuracy of using VOC analysis or e-nose technology.Exclusion criteria encompassed: (1) studies lacking information on the number of cases, controls, sensitivity, and specificity; and (2) studies published as review articles or case reports.Discrepancies between reviewers were resolved through consensus or, if necessary, with the involvement of a third investigator (Q.L.W.).A total of 32 articles met the inclusion criteria and were subsequently included in this systematic review.

Risk of bias
The quality appraisal of all incorporated literature was conducted according to the QUADAS-2 scale through Review Manager 5.4 software.The results of the risk of bias assessment are visually presented in Figures 2A, B.
In the aggregate, a few studies exhibited a high risk of bias.Concerning 'patient selection' seven studies (32, 34, 39, 42, 46, 53, 54) (21.9%) incurred a high risk of bias.The primary contributor to this high risk pertained to the absence of a detailed description of the sampling process and the implementation of a case-control study design.Regarding the 'index test' while most studies employed reference diagnostic tests to delineate the definition of a positive  ) faced a high risk of bias.The primary reason for this was that these studies do not account for the time interval between the index test and the reference test.
In evaluating clinical applicability, significant concerns in patient selection arose from the absence of matched patient groups, inadequate patient selection criteria, and applicability of the study design to the research question.Six studies exhibited a high applicability concern for patient selection criteria (26,32,42,46,49,53).No high-risk concerns were identified regarding the applicability of the index and reference tests to the research questions.
The funnel plots for publication bias are displayed in Figures 7, 8.The Deeks' regression test for funnel plot asymmetry demonstrated an absence of publication bias among the studies included, with slope coefficients P values of 0.28 and 0.62 for using VOC analysis and e-nose.

Discussion
We conducted a systematic review and meta-analysis to evaluate VOC analysis and electronic nose in detecting colorectal cancer, aiming to compare the diagnostic accuracy and clinical application value of these two methods.Pooled analysis of VOC and electronic-nose studies demonstrated high diagnostic accuracy for CRC detection, with a pooled sensitivity of 0.88 and specificity of 0.85 for VOC analysis and a sensitivity of 0.87 and specificity of 0.78 for e-nose studies.The visually assessed SROC curves indicated clinical accuracy, with VOC analysis and e-nose having SROC curves of approximately 0.93 and 0.90, respectively, both close to 1, signifying superior accuracy and diagnostic efficacy in CRC diagnosis.These findings align with prior reviews (22,59,60), but the notable heterogeneity between studies and the identified high risk of bias warrant cautious interpretation.The heterogeneity was largely due to the sample media and the analytical methods used.
Subgroup analyses revealed that breath samples in VOC analysis and urine and breath samples in e-nose studies exhibited higher sensitivity or specificity.Breath sampling is easily performed and well-received by patients, and urine samples, boasting high sensitivity and specificity, emerge as valuable alternatives.Recent  Frontiers in Oncology frontiersin.orgmeta-analysis evaluated the performance of the combined FIT and urinary.The findings revealed that the combined FIT-VOC approach could detect 33% more cases of colorectal cancers (60).Chandrapalan S et al. ( 61) showed that the combination of FIT and VOC can be a better triage tool, for CRC in patients with lower gastrointestinal symptoms than FIT alone.Due to the lack of standardization in sample collection, handling, and storage, technical barriers exist in measuring and analyzing various VOC characteristics during sampling, whether it involves alveolar air, urine, or feces.In several studies, exhaled breath was collected into a bag and subsequently analyzed (28, 29, 33, 38-40, 42, 50).The use of bag collection aligns more closely with real-world medical applications.However, this approach may be influenced by several factors, including interference from ambient VOCs, the material used for collection, and the impact of temperature, humidity, and storage time on specimens (62).For breath samples, it is essential to examine them within 6 hours of the collection's conclusion to ensure test accuracy (63).Therefore, developing methods for the collection, transmission, and handling of breath samples is crucial for the success of this approach.Some studies have indicated that the diagnostic accuracy of fecal and urine VOCs is not significantly affected by storage time (20 months for fecal and 12 months for urine VOCs) (64,65).
Urine samples are ideal detection medium because they have limited confounding factors compared to breath samples which is influenced by smoking or fecal samples influenced by diet.Further research should standardize the method of collection of such samples and investigate the effects of potential confounding factors.
Among all studies, only six reported on CRC stages, indicating limited generalizability and clinical applicability.Multi-center    Public bias analysis of all the e-nose studies.It has been demonstrated that various factors, such as age, gender, smoking, alcohol consumption, coffee intake, and the consumption of stimulating foods like leeks and garlic, as well as comorbidities and medication, may influence the composition of VOCs in exhaled breath (66).However, only a few studies considered confounding or modifying effects, limiting the validity and reliability of the results.Therefore, future studies should account for the impact of such factors on breath prints during the design phase.
Gas chromatography-mass spectrometry (GC-MS), a traditional method for VOC analysis, is a highly standardized technique providing qualitative and quantitative information on exhaled VOCs (67,68).In this study, TD-GC-MS demonstrated high sensitivity and specificity in detecting colorectal cancer, while GC-MS exhibited improved specificity but suboptimal sensitivity.The use of GC-MS and newer mass spectrometry technology devices remains the gold standard for identifying specific VOCs for analysis.However, GC-MS technology is costly and complex, with long analysis times, and it demands a high level of expertise from operators.
Based on sensors, electronic nose technology serves as a novel analytical method for disease diagnosis, offering the advantages of being cost-effective, user-friendly, portable, sensitive, and responsive.Nevertheless, there are existing shortcomings that require refinement in the application of e-nose in clinical practice.Unlike GC-MS and other techniques, e-nose lacks the precision to measure specific types and composition ratios of components in VOCs (24).It also cannot identify specific pathophysiological pathways or therapeutic targets.Furthermore, as the e-nose relies on arrays of gas sensors to distinguish and identify response spectra of mixtures composed of multiple VOCs, the diverse sensor types with distinct signal responses prevent the integration of results from one e-nose with different devices or sensor types (69).Van der Sar IG (70) recommends the establishment of a comprehensive worldwide shared database encompassing patient characteristics and other pretest probabilities.Various algorithms and methods were employed to analyze VOCs in this study, with PLA-DA and logistic regression analysis emerging as the most commonly used approaches.However, the majority of studies fail to elucidate the rationale behind selecting a specific machine learning model for analysis, only reporting the highest accuracy value, thereby impacting the reliability of the results.Additionally, studies with small sample sizes may compromise the reported accuracy.Few studies have conducted external validation to affirm the validity and reliability of these findings.Consequently, large, multi-center external validation studies should be conducted in the future to explore the applicability and reproducibility of the results in different study settings and among diverse target populations.

Limitation
This study has certain limitations.Heterogeneity was observed among studies, potentially attributed to variations in sample media and analytical methods.Some studies exhibited a high risk of bias, with seven showing concern regarding patient selection and ten having applicability concerns in one or two domains.Furthermore, the study included fewer investigations employing both VOC analysis and e-nose technology, thus impeding an accurate evaluation of the complementary effects of the two methods.In addition, VOC combined with FIT approach could increase the detection of colorectal cancer.However, there are no prospective studies evaluating the positive effect on VOC-FIT for screening prior to the onset of CRC.

Conclusion
Based on our meta-analysis, VOC analysis and e-nose technology show promise in the detection of CRC.However, several milestones must be achieved in colorectal cancer detection with these two noninvasive methods before clinical implementation.Firstly, for patients presenting with common non-specific symptoms, which may be an early indication of CRC, an exhaled breath test or a urine test or FIT +VOC could serve as screening tool.Secondly, electronic nose could be utilized in primary care units and community healthcare centers for mass screening of various intestinal diseases due to their portability, ease of use, cost-effectiveness, speed, and independence from specialized technicians.Thirdly, the identification of colorectal cancer-specific VOC biomarkers and combinations of biomarkers for colorectal cancer diagnosis is still necessary.This requires comprehensive metabolomics studies to elucidate the production of endogenous VOCs and the metabolic transformation of exogenous VOCs in colorectal cancer, aiding in the identification of VOC markers for cancer.Finally, large, multi-center external validation trials should be conducted to verify the generalizability and reproducibility of the results in different research settings and at different stages of CRC.

FIGURE 1
FIGURE 1 FIGURE 2 (A) Summary and separate outcome of risk of bias and concerns.(B) Summary and separate outcome of risk of bias and concerns regarding applicability for included studies using QUADAS-2 tool.

FIGURE 3
FIGURE 3Pooled sensitivity and specificity analyses of VOC studies.

FIGURE 4
FIGURE 4Pooled sensitivity and specificity analyses of e-noses studies.

FIGURE 5
FIGURE 5Summary receiver operating characteristic (SROC) curve Analysis of VOC studies.

FIGURE 6
FIGURE 6Summary receiver operating characteristic (SROC) curve Analysis of e-noses studies.

FIGURE 7
FIGURE 7Public bias analysis of all the VOC studies.

TABLE 1
Basic characteristics and outcomes of VOC studies in the analysis.

TABLE 2
Basic characteristics and outcomes of e-nose studies in the analysis.ANN, artificial neural network; CDA, canonical discriminant analysis; CRC, colorectal cancer; DFA, discriminant function analysis; HNSCC, head and neck squamous cell carcinoma; IBS, Irritable bowel syndrome patients; KNN, K Nearest Neighbors; LDA, linear discriminant analysis; NR, Not reported; NN, Neural Network; PNN, Probabilistic Neural Network; RF, Random Forest; SVM, support vector machine.

TABLE 4
Subgroup analysis in e-nose studies.

TABLE 3
Subgroup analysis in VOC studies.