Diagnostic accuracy of the faecal immunochemical test and volatile organic compound analysis in detecting colorectal polyps: meta-analysis

Abstract

Background

For the early detection of colorectal cancer, it is important to identify the premalignant lesions to prevent cancer development. Non-invasive testing methods such as the faecal immunochemical test are well established for the screening and triage of patients with suspected colorectal cancer but are not routinely used for polyps. Additionally, the role of volatile organic compounds has been tested for cancer detection. The aim of this review was to determine the diagnostic accuracy of the faecal immunochemical test and volatile organic compounds in detecting colorectal polyps.

Methods

Original articles with diagnostic test accuracy measures for both the faecal immunochemical test and volatile organic compounds for advanced adenomas were included. Four databases including Medical Literature Analysis and Retrieval System Online (MEDLINE), Cumulative Index to Nursing and Allied Health Literature (CINAHL), Embase, and Web of Science were searched. The quality assessment tool for diagnostic accuracy study was used to assess the risk of bias and applicability. Meta-analysis was performed using RStudio^® and the combined faecal immunochemical test-volatile organic compounds sensitivity and specificity were computed.

Results

Twenty-two faecal immunochemical tests and 12 volatile organic compound-related articles were included in the systematic review whilst 18 faecal immunochemical tests and eight volatile organic compound-related studies qualified for the meta-analysis. The estimated pooled sensitivity and specificity of the faecal immunochemical test to diagnose advanced adenoma(s) were 36% (95% c.i. 30 to 41) and 89% (95% c.i. 86 to 91) respectively, with an area under the curve of 0.65, whilst volatile organic compounds pooled sensitivity and specificity was 83% (95% c.i. 70 to 91) and 76% (95% c.i. 60 to 87) respectively, with an area under the curve of 0.84. The combined faecal immunochemical test-volatile organic compounds increased the sensitivity to 89% with a specificity of 67%.

Conclusion

Faecal immunochemical testing has a higher specificity but poor sensitivity for detecting advanced adenomas, while volatile organic compound analysis is more sensitive. The combination of both tests enhances the detection rate of advanced adenomas.

The faecal immunochemical test has poor sensitivity (36%) for advanced adenoma detection, with a significant number of patients being missed. Volatile organic compound analysis offers higher sensitivity with the flexibility of measuring it in breath, stool or urine samples. The meta-analysis also demonstrated that combining the faecal immunochemical test-volatile organic compounds enhances the polyp detection rate, with combined sensitivity up to 89%.

Introduction

Colorectal cancer (CRC) is the third most common cancer and the second leading cause of cancer-related deaths in Western countries, with 16 800 deaths every year in the UK^1–3. It is expected that by 2030, the global burden will rise by 60% with over 2.2 million new patients and 1.1 million annual deaths, with the cost exceeding €238 777 per patient to treat advanced colorectal cancer^4,5. According to the UK bowel cancer statistics, the estimated annual cost is more than 2.0 billion euros, and approximately 1:14 men and 1:18 women will develop colorectal cancer in their lifetime¹.

Colorectal cancer usually develops from polyp(s), which are one of the most common colorectal (CR) conditions and a leading factor for the development of colorectal cancer⁶. The rate of transformation of polyp to invasive cancer is thought to take 5–10 years but can be accelerated in cancer predisposition syndromes such as Lynch^7,8. The core understanding of this process carries a significant clinical impact on patient care and early detection plays a crucial role^4,9,10.

Diagnosing CR neoplasms is difficult as colorectal symptoms are poor predictors of disease detection¹¹ and pose a greater challenge to deciding which patients to reassure or investigate¹². The CR polyp prevalence in high-risk populations can reach up to 30%^13,14, however, an increasing prevalence is observed in younger populations (less than 50 years of age) with a prevalence of 33.1% for all types of polyps and 22.4% for precancerous polyps^15,16. This has resulted in an increasing rate of colonoscopies and increased burden on endoscopy services.

Colonoscopy is the ‘standard’ investigation for suspected CR neoplasms, carrying 95% sensitivity and up to 92% specificity¹⁷. However, the procedure is invasive, requires bowel preparation and carries a small but significant risk of serious complications including bowel perforation¹⁸, which limits its use as a primary triage tool.

Faecal immunochemical testing is widely used in the UK as a triage tool in both bowel cancer screening and fast-track pathways^19,20. Despite the implementation of the fast-track ‘2-week wait’ (2WW) pathway, studies have reported a 3–10% colorectal cancer diagnosis rate^21–23.

The National Institute of Clinical Excellence (NICE) recently updated guidelines (DG56) for faecal immunochemical testing of all patients regardless of high or low pretest probability for colorectal cancer^24,25. However, the faecal immunochemical test (FIT) has highly variable sensitivity (5–67%) for colorectal cancer^26,27 and for advanced adenomas (AAs), ranging from 34% to 40%²⁸, with no significant enhanced effect on sensitivity by changing the threshold values²⁸. It has a low patient adherence rate (47%), mainly due to stool-based sampling²⁹.

This represents an unmet need to identify a better triage tool for premalignant lesions with better acceptability for patients. Recently, interest in using volatile organic compound (VOC) analysis for the non-invasive detection of cancer from human waste has increased and its feasibility to detect colorectal cancer has been evaluated within the National Health Service (NHS) fast-track colorectal cancer pathways³⁰. The ability to perform VOC analysis in urine distinguishes it from FIT and other non-invasive tests (faecal DNA/RNA), making it favourable in terms of feasibility and patient compliance. Global studies support VOC analysis as a potential future triage tool for cancer^31–36, however, its role in detecting premalignant lesions is still an area for research.

Although there have been several systematic reviews and meta-analyses on the role of FIT in detecting colorectal cancer^37–42, there is little synthesis on the role of FIT or VOC in detecting polyps.

The rationale for conducting this review was to synthesize and assess the quality of available diagnostic studies in evaluating the role of FIT and VOC alone or in combination for the early detection of colorectal polyps. The main objective was to evaluate the role of combined FIT and VOC in enhancing the detection of colorectal adenomas in different population groups to bridge the gap in the literature and provide guidance for future studies.

Methods

This systematic review was conducted according to recommendations from the Preferred Reporting Items for Systematic Review and Meta-analysis of Diagnostic Test Accuracy studies (PRISMA-DTA) 2018 statement⁴³ (Fig. S1).

The patient/target condition, index test, and outcome (PIO) framework was used to search the databases. The search terms ‘colorectal polyp(s)’, ‘advanced adenoma’, ‘colorectal adenomas’, ‘faecal immunochemical test’, ‘volatile organic compound analysis’, ‘biomarkers’, ‘diagnostic accuracy’ were used with synonyms for each word for effective database search (Table S1). A literature search of articles from Medical Literature Analysis and Retrieval System Online (MEDLINE)/CINHAL (Cumulative Index to Nursing and Allied Health Literature) through Elton B. Stephens Company (EBSCO) host, Embase (using Ovid) and Web of Science was performed (Fig. S2). The last date of the database search was 15 October 2024, with no restriction to the year of publication or language. Google Scholar and previous reviews^28,44 were also searched for grey literature.

One article included in this review was in the Chinese language⁴⁵, which was translated by Google Translator to extract the required data and for the quality assessment of the article. Where information was missing, or unpublished data was required, principal investigators (PIs) for each article were contacted via e-mail.

Study selection and quality assessment

Articles from all databases were added to EndNote^® 20 and a two-phase screening process was completed according to PRISMA-DTA guidelines. During the first phase, data was screened based on the title and abstract followed by a second phase full-text article review of eligible studies. Apart from including primary studies on colorectal polyps, the articles primarily reporting colorectal cancer outcomes (for FIT or VOC analysis) were also reviewed based on inclusion criteria to extract polyp-related data. Only studies reporting both sensitivity and specificity for either test were included in the final analysis. Two independent reviewers completed the screening process and any discrepancy was resolved with the corresponding author for final inclusion in the systematic review and meta-analysis.

Quality assessment of diagnostic accuracy study-2 tool (QUADAS-2)⁴⁶ was used for risk of bias (RoB) and applicability assessment by two independent reviewers and any disagreement was resolved upon discussion with a third reviewer. The outcome for both RoB and applicability was expressed as ‘high’, ‘low’ or ‘unclear’.

Data extraction

A structured data extraction form (Table S2) was used for data collection according to inclusion and exclusion criteria (Table S3). Prospective (cross-section/observational/case-control) studies including patients who underwent the FIT or testing for volatile compounds and had colorectal polyps confirmed through reference tests, that is colonoscopy/CT colonography or flexible sigmoidoscopy, were included. All retrospective studies including patients with colorectal cancer, or lack of reporting of diagnostic accuracy measures for AAs, were excluded.

Data synthesis and statistical analysis

Only studies with data for AAs were compiled for this systematic review and further meta-analysis. Any adenoma ≥1 cm in size (or), presence of villous features (or) high-grade dysplasia was classified as AA.

Studies reporting both sensitivity (Se) and specificity (Sp) with confidence interval (c.i.) related to AAs were included in the review and any additional information on accuracy or prevalence (P) was also added. True positives, true negatives and false positives/false negatives were calculated from Se and Sp data. Data was reported as per patient, that is, the number of patients diagnosed with at least one AA (one patient can have multiple adenomas and hence categorized based on the most advanced lesions present).

Only studies reporting both sensitivity and specificity data were included for further meta-analysis (MA). For FIT studies, the threshold was limited to 10 μg/g Hg, which is currently used for patients suspected of high-risk lesions in the NHS 2WW pathway. Because FIT sensitivity is low regardless of the threshold (7–20 μg/g Hg) and the chance of missing patients with high-risk polyps remains approximately three-quarters at the threshold of 120 μg/g Hg, which is currently used in the National Bowel Screening Programme^28,47, the data for both symptomatic and asymptomatic patients was collected at the threshold of 10 μg/g Hg.

Gas chromatography and mass spectrometry (GC-MS) is the ‘standard’ for VOC analysis; only studies with GC-MS were further included for MA pooling data from all sample types. Additional subgroup analysis of VOC studies was performed based on the sample type, that is breath, stool and urine.

The RStudio^® software (v. 4.3.2) for bivariate analysis using the random-effects model (due to large interstudy variability which cannot be described by chance only as in the fixed-effect model) was used for meta-analysis to calculate pooled sensitivity and specificity for both FIT and VOC testing and these are presented as forest plots. The area under the curve (AUC) is represented by the summary receiver operating characteristic (SROC) curve. According to current PRISMA-DTA guidelines for systematic review and meta-analysis, the statistical analysis to estimate and report heterogeneity and publication bias is not very useful⁴³ and is rather misleading in the case of publication bias⁴⁸. Whilst performing MA heterogeneity is computed automatically and is mentioned in this review for descriptive purposes only and the quality of data is mainly assessed by QUADAS-2.

Combined FIT-VOC diagnostic accuracy measures were calculated in both a parallel (the result of both VOC and FIT interpreted simultaneously) and serial way (if FIT is used first to guide the need for the second test, that is VOC) and interpreted using the ‘OR’ rule⁴⁹. The ‘OR’ rule states that if both tests (FIT and VOC) are negative then the result is negative whilst if either test is positive then it is a positive diagnosis. Pre- and posttest probabilities were calculated using online MedCalc software (v. 22.023) and Fagan’s nomogram.

Results

Screening and selection of studies

A total of 1093 articles were identified from all databases, of which 22 FIT-^{45,47,50–69} and 12 VOC-related articles^{27,30,52,70–78} were included in the final analysis for the systematic review and 18 FIT^{45,50–52,54,55,57,58,60–69} and 8 VOC^{27,30,52,70–73,76} studies for the meta-analysis. The PRISMA-DTA flowsheet⁴³ (Fig. 1) shows the outcome of the screening process according to the eligibility criteria. The total number of patients with AAs was 2963 in the FIT group and 918 in the VOC group, with a mean age of 55.7^56–77 and 64.7^62–70 years respectively. Consistency in reporting features of AA, that is a size of ≥1 cm, villous histology or high-grade dysplasia, was found across all studies. Unpublished data³⁰ and missing information²⁷ for two articles were received via e-mail from PIs of the studies.

Fig. 1.

Preferred Reporting Items for Systematic Review and Meta-Analysis of Diagnostic Test Accuracy (PRISMA-DTA) flow chart of screening process

FIT, faecal immunochemical test; VOC, volatile organic compound; OVID Embase, Ovid Excerpta Medica Database; EBSCO, Elton B. Stephens Company; CINHAL, Cumulative Index to Nursing and Allied Health Literature; CRC, colorectal cancer.

Study characteristics

FIT articles summary characteristics

All data for FIT articles were collected at a threshold positivity of 10 μg/g Hg, with 18 studies using quantitative OC-Sensor^{47,50,53,56,58,59,61,63–67}, OC-Micro^55,68, HM-JACKarc Analyser^52,54 and others^45,51, whilst four studies used qualitative methods, that is OC-Light^57,62,69 and Hemosure⁶⁰ for FIT analysis. All studies were prospective except one, which was a randomized clinical trial⁵⁵, and eight studies were multicentre^{47,50,52,59,61,65,66,68}.

Most of the target population was mixed^{45,53,55,59,61,63–65,68} followed by screening^{47,56,60,62,67}, symptomatic^50,51,54,58, surveillance^52,66 and asymptomatic⁵⁷. Patients were mainly recruited from either primary^58,63 or secondary care^{45,52,56,59,61,62,64,67,68} or both^50,51,65,66 and six studies^{47,50,52,54,58,63} represented the UK population. Only seven studies^{50,52,55,58,60,62,63} followed Standards for Reporting Diagnostic Accuracy Studies (STARD) guidelines. The funding source was declared by all except three studies^55,57,67 and all studies had ethical approval (Table 1 and Table S4).

Table 1.

Summary of analytical technique and diagnostic accuracy of faecal immunochemical test (FIT) studies (all data for FIT articles were collected at a threshold positivity of 10 μg/g Hg)

Author, year	Analytical technique	Total no. of patients	No. of patients with AA	Definition of adenoma(s)	Sensitivity, % (95% c.i.)	Specificity, % (95% c.i.)	AUC (95% c.i.)
Bath et al., 202350	OC-Sensor	553	47	UK and ESGE (unspecified) LRA = 1–2 adenomas each small size <1 cm IRA = 3–4 small or 1 adenoma ≥1 cm HRA = 5 small or ≥3 adenoma at least one ≥1 cm	42 (35,48)	81 (79,82)	0.63
Hijos-Mallada et al., 202351	FOB Turbilatex®	323	53	ESGE guidelines HRA = 5 small or ≥1 cm or HGD, SL ≥ 1 cm or with dysplasia	47 (33,60)	81 (76,85)	0.71 (0.63,0.78)
Chandrapalan et al., 202252	HM-JACKarc Analyser	141	68	BSG guidelines HRA = 5 small or ≥2 adenoma at least one ≥1 cm size	54 (43,65)	79 (73,84)	0.67
Zhou et al., 202245	Hs-qFIT	910	61	AA = ≥1 cm, VA, TVA (>25% VA) or HGD	23 (13,35)	94 (92,95)	Not reported
Li et al., 202147	OC-Sensor	3423	261	BSG guidelines	77 (68,85)	Not reported	Not reported
Sekiguchi et al., 202153	OC-Sensor	476	102	AA = ≥1 cm, HGD or prominent villous component	18 (11,27)	Not reported	Not reported
Farrugia et al., 202054	HM-JACKarc Analyser	473	18	HRA = ≥1 cm (6 patients had TVA)	30 (21,41)	79 (75,83)	Not reported
Mattar et al., 202055	OC-Auto Micro 80	141	37	AA = ≥1 cm, VA, TVA (>25% VA) or HGD Early adenomas = <1 cm and LGD	24 (10,45)	86 (77,92)	Not reported
Chang et al., 201756	OC-Sensor	4516	339	WHO criteria AA = ≥1 cm or HGD or VA	32 (27,37)	Not reported	Not reported
Teixeira, 201757	OC-Light	171	123	Based on histopathology = HGD as HRA	26 (19,35)	83 (77,88)	Not reported
Mowat et al., 201658	OC-Sensor	241	40	AA = >3 adenomas or any adenoma ≥1 cm	50 (34,65)	78 (72,83)	Not reported
Symonds et al., 201659	OC-Sensor	1381	170	AA = ≥1 cm or >20% VA, or HGD, > 2 TA or stage 0 cancer	43 (36,51)	Not reported	Not reported
Wong et al., 201560	Hemosure	3643	269	AA = ≥1 cm, VA or TVA or HGD	33 (27,39)	91 (90,92)	Not reported
van Turenhout et al., 201461	OC-Sensor	2049	304	AA = ≥1 cm, VA or TVA or HGD	34 (30,41)	92 (91,93)	0.66
Chiu et al., 201362	OC-Light	1330	632	WHO criteria AA =≥1 cm or HGD or VA	28 (25,31)	93 (93,93)	Not reported
McDonald et al., 201363	OC-Sensor	194	23	BSG guidelines HRA = any adenoma ≥1 cm or ≥3 adenoma	47 (27,67)	96 (93,98)	Not reported
Ou et al., 201364	OC-Sensor	321	39	AA = ≥1 cm, VA or TVA or HGD	38 (23,55)	92 (89,94)	0.74
Terhaar sive Droste et al., 201266	OC-Sensor	935	101	AA = ≥1 cm, VA or TVA or HGD	28 (19,38)	91 (89,93)	Not reported
Terhaar sive Droste et al., 201165	OC-Sensor	236	97	AA = ≥1 cm, VA or TVA or HGD	41 (34,47)	90 (88,91)	Not reported
Khalid-de Bakker et al., 201167	OC-Sensor	243	38	AA = ≥1 cm, or any VA or HGD	15 (6,31)	96 (94,98)	Not reported
Rozen et al., 201068	OC-MICRO	1533	129	AA = ≥1 cm or >20% VA, or HGD	36 (28,44)	93 (92,94)	0.73 (0.68–0.78)
Gimeno-García et al., 200969	LA-FOBT (OC-Light)	67	12	AA = ≥1 cm, VA or HGD	83 (55,95)	91 (85,95)	Not reported

Values are n unless otherwise indicated. LRA, low-risk adenoma; IRA, intermediate-risk adenoma; HRA, high-risk adenoma; HGD, high-grade dysplasia; SL, serrated lesions; VA, villous adenoma/architecture; TVA, tubulo-villous adenoma; LGD, low-grade dysplasia; TA, tubular adenoma; AA, advanced adenoma; AUC, area under the curve; ESGE, European Society of Gastrointestinal Endoscopy; BSG, British Society of Gastroenterology; WHO, World Health Organization.

VOC articles summary characteristics

Only two studies followed STARD guidelines for reporting data^30,52. Eight studies were multicentre^{30,52,70–74,77} and all studies were prospective^{30,52,71–73,75,77,78} except four studies which were mainly cross-sectional^27,74 and case-control designs^70,76. The target population was mainly mixed^70,72,76,77 or screening^27,71,73,78 followed by symptomatic^30,74,75 population. Patients were referred or recruited from both primary and secondary care and five studies^{30,52,72,73,75} represented the UK population.

Urine was the most common type of sample used^{30,52,73,75,78} for VOC analysis followed by breath^71,72,74,76 and stool^27,70,77. Slight variability in methodology and sample collection was observed and only two studies^27,52 mentioned the cut-off threshold of positivity, that is 8.4 × 10⁸ and 0.88 (no units) respectively using the GC-MS technique for VOC analysis in AAs (Table 2). After initial analysis, quantification by GC-MS was done by eight studies^{27,30,52,70–73,76} which were included for meta-analysis.

Table 2.

Summary of sampling and diagnostic accuracies of volatile organic compound analysis studies

Author, year	VOC source	Analytical technique and threshold	Definition of adenoma(s)	No. of patients with AA (total no. of patients)	Sensitivity, % (95% c.i.)	Specificity, % (95% c.i.)	AUC (95% c.i.)
Alustiza et al., 202327	Stool	GC-MS Cut point: 8.4 × 108 (unit missing)	Not defined (results reported for >1 cm or VA taken as HRA)	12 (56)	83 (68,98)	63 (46,79)	0.70 (0.53,0.86)
Boulind et al., 202230	Urine	GC-IMS	Not defined Number of AA (HGD/ > 1 cm size) confirmed from unpublished data	31 (65) unpublished data	84 (71,96) GC-MS unpublished data	70 (62,77) unpublished data	0.82
Chandrapalan et al., 202252	Urine	GC-MS Threshold for positivity = 0.88 (unit missing)	BSG guidelines HRA = 5 small or ≥2 adenoma at least one ≥1 cm	68 (141)	VOC alone 94 (88, 98) In parallel testing VOC-FIT combined sensitivity = 97	VOC alone 69 (64,75) In parallel testing VOC-FIT combined specificity = 11	0.74
Cheng et al., 202271	Breath	GC-MS	AA = ≥1 cm, VA and/or HGD	138 (222)	79 (72,85)	70 (60,79)	0.72
Woodfield et al., 202272	Breath	GC-MS	BSG guidelines HRA = 5 small or ≥2 adenoma at least one ≥1 cm or HGD or VA or SL with dysplasia	271 (628)	66 (60,71)	58 (52,63)	0.67
Bosch et al., 202070	Stool	GC-MS	ESGE guidelines AA = ≥1 cm or villous histology or HGD	64 (291)	96 (91,99)	93 (89,96)	0.96 (0.93,0.99)
Mozdiak et al., 201973	Urine	FAIMS/GC-IMS	BSG guidelines HRA = 5 small or ≥2 adenoma at least one ≥1 cm or HGD or VA or SL with dysplasia	55 (79)	58 (44,71)	62 (41,81)	0.61 (0.47,0.75)
van Keulen et al., 202074	Breath	e-nose	AA = ≥1 cm, TVA (>25% villous) or HGD	112 (447)	73	79	0.69
Widlak et al., 201875	Urine	FAIMS	HRA = ≥3 adenoma, or ≥1 cm or HGD or VA or SL	27 (562)	93 (81,100.0)	16 (13,20)	0.56 (0.45,0.68)
Amal et al., 201676	Breath	GC-MS	AA = >1 cm, HGD or VA	10 (209)	94	94	Not available
Wang et al., 201478	Urine	Nuclear magnetic resonance spectroscopy	AA = ≥1 cm, villous histology or HGD	70 (876)	82	51	Not available
de Meij et al., 201477	Stool	Cyranose 320 e-nose	AA = ≥1 cm, TVA (>25% villous) or HGD	60 (157)	62	86	0.79

Values are n unless otherwise indicated. VA, villous architecture; HRA, high-risk adenoma; HGD, high-grade dysplasia; SL, serrated lesions; TVA, tubulo-villous adenoma; VOC, volatile organic compound; AUC, area under the curve; IMS, ion-mobility spectrometer; FIT, faecal immunochemical test; BSG, British Society of Gastroenterology; ESGE, European Society of Gastrointestinal Endoscopy; GC-MS, gas chromatography and mass spectrometry; FAIMS, field asymmetric ion mobility spectrometry; AA, advanced adenoma.

Six articles^{27,30,52,71,76,78} mentioned the type of VOCs found in adenomas. Alcohol derivatives were most commonly present as VOC markers in adenomas, acetone being most common followed by ethanol (Table S5 and Table S6).

Quality assessment

QUADAS-2 assessment of FIT-related articles

Twenty-two studies were assessed for quality; overall, studies were either unclear or at low RoB. Six studies^{54,57–59,61,67} were at high RoB mainly due to concerns with patient selection, dropouts from the study or timing of index test, sample collection method and unblinded interpretation of index test, and two studies^54,61 had applicability concerns.

The unclear risk was mainly related to lack of clarity in the description of blinding of the index^{47,52,54,55,57,59,63,66,69} and reference standard test^{47,50,52–55,57,59,61,63,68,69}, the quality of the colonoscopy, variability in describing the target condition, method of sample collection and consecutive sampling in selecting patients^{45,47,50,52–54,56–59,62,64,69} (Table S7, Fig. S3).

QUADAS-2 assessment of VOC-related articles

Twelve studies were assessed for quality; overall, more studies had an unclear RoB than applicability concerns (Table S8, Fig. S4).

Those with unclear RoB were mainly due to a lack of clarity in consecutive sampling, description of the target condition²⁷, inappropriate exclusions, blinding of the index^{27,52,70–72,76,78}, or reference standard test interpretation^{27,52,70,72,76,78}. All studies mentioned the timing for the sample collection mainly 2–4 weeks before colonoscopy except two studies^30,71. Moreover, there was variation in sample collection and storage conditions, and it remained unclear if this could impact the quality of data reported for VOCs.

Results of meta-analysis

Eighteen articles qualified for FIT analysis. The estimated pooled sensitivity was 36% (95% c.i. 30 to 41) whilst specificity was 89% (95% c.i. 86 to 91) from FIT studies with an AUC of 0.65 (Fig. 2a and 2b). Substantial heterogeneity and variance were found among studies both in sensitivity (Higgins I² = 72%, τ2 = 0.14) and specificity (Higgins I² = 93%, τ2 = 0.31) data. This variability could be due to the population, qualitative or quantitative, method of FIT analysis or blinding (see details in quality assessment of studies).

Fig. 2.

Forest plot and summary receiver operative characteristic (SROC) of faecal immunochemical test (FIT) a Forest plot using a random-effects model showing sensitivity and specificity of individual studies and corresponding 95% confidence interval (c.i.), with pooled sensitivity of 35% (95% c.i. 30 to 41) and specificity of 89% (95% c.i. 86 to 91) of FIT in detecting advanced colorectal adenomas. b SROC plot of sensitivity and specificity of FIT analysis. Each triangle represents an individual study. The small round spot on the SROC (solid purple line) represents summary sensitivity and specificity, with ellipse around showing the area of 95% confidence interval. The estimated area under the curve (AUC) is 0.65. (Continued on next page).

Eight studies qualified for VOC meta-analysis, and the pooled sensitivity was significantly higher, that is 83% (95% c.i. 70 to 91) with a specificity of 76% (95% c.i. 60 to 87) and AUC of 0.84 (Fig. 3a and 3b). Substantial heterogeneity and variance in studies both in sensitivity (Higgins I² = 85%, τ2 = 0.86) and specificity (Higgins I² = 91%, τ2 = 0.53) were observed.

Fig. 3.

Forest plot and summary receiver operative characteristic (SROC) of volatile organic compound (VOC) analysis. a Forest plot using a random-effects model showing sensitivity and specificity of individual studies and corresponding 95% confidence interval (c.i.), with a pooled sensitivity of 83% (95% c.i. 70 to 91) and specificity of 76% (95% c.i. 60 to 87) of volatile organic compounds in detecting advanced colorectal adenomas. b SROC plot of sensitivity and specificity of VOC analysis. Each triangle represents an individual study. The round spot on the SROC (solid purple line) represents summary sensitivity and specificity. The estimated area under the curve (AUC) is 0.84.

Sample-based subgroup analysis of VOC testing

Stool

Two studies including 76 patients with AA in stool-based sample analysis of VOC showed a pooled Se of 92% (95% c.i. 68 to 98) and Sp of 83% (95% c.i. 40 to 97) (Fig. 4a).

Fig. 4.

Subgroup analysis of volatile organic compound (VOC) analysis. a Subgroup analysis of studies with stool sample used for VOC analysis. b Subgroup analysis of studies with breath sample used for VOC analysis. c Subgroup analysis of studies with urine sample used for VOC analysis.

Breath

Three studies including 419 patients, using breath analysis of VOC, showed a pooled Se of 75% (95% c.i. 61 to 84) and Sp of 89% (95% c.i. 24 to 95) (Fig. 4b).

Urine

Three studies including 370 patients with AA using urine samples had a pooled sensitivity of 82% (95% c.i. 53 to 95) and specificity of 69% (95% c.i. 60 to 76), closer to the pooled Se and Sp of VOC analysis than any other subgroup (Fig. 4c).

Combined FIT-VOC analysis

When FIT was combined with the VOC test, a higher sensitivity of up to 89% was obtained but the overall specificity decreased to 67% from individual pooled values of FIT and VOC testing both in parallel and serial testing under the ‘OR’ rule (Table S9).

Pre- and posttest probability of FIT and VOC testing

The pooled prevalence of AA using FIT was approximately 15%. Using this prevalence rate, the posttest probability of FIT detecting AA was 30% while it was 10% in the case of a negative test (Tables S10 and S11). Using the posttest probability of FIT as the pretest probability for VOC (if negative) showed a further reduction of the probability of having the disease to 3% (Fig. S5a and S5b). For descriptive purposes, further analysis of the AA pooled prevalence from VOC data was done which showed a higher disease prevalence compared with FIT, however, using this prevalence rate the chance of having AA after VOC test positivity increased to 60% and reduced to 10% in the case of a negative result (Fig. S6).

Discussion

In this systematic review and meta-analysis, the evidence on diagnostic accuracy was synthesized for both FIT and VOC alone and in combination to detect advanced colorectal adenomas. The pooled estimates showed overall low sensitivity (36%, 95% c.i. 30 to 41) and high specificity (89%, 95% c.i. 86 to 91) of FIT at the threshold of 10 μg/g Hg to detect AA(s); this finding is comparable to the most recent review in 2020²⁸ on FIT performance which showed a sensitivity of AA in the range of 25–40% at a threshold of 10 μg/g Hg. The pooled estimate of VOC showed significantly higher sensitivity (83%, 95% c.i. 70 to 91) with slightly lower specificity (76%, 95% c.i. 60 to 87); this was a new finding in our review with no previous review available for comparison.

This marked difference in sensitivities of the two tests (FIT and VOC) could be explained by the different nature of these tests targeting different parts of the disease process. The VOC analysis detects alteration of the ‘volatilome’ of the individual carrying the disease, which starts at an early stage of the disease process and produces disease-specific patterns^7,27, whilst FIT is based on the detection of blood, which can be present in other conditions as well⁷, and could be a late finding, missing 65–75% of the patients harbouring high-risk precancerous colorectal lesions⁷⁹.

Along with offering higher sensitivity, another strength of the VOC test is the ability to perform VOC analysis in stool, urine or breath samples. FIT is only carried out on stool samples, which over time has shown poor acceptability among the general population with a non-adherence rate of 76%, minimizing the benefits of the screening pathway⁸⁰. Moreover, in a recent update, the uptake of FIT by the younger population was well below the target threshold⁸¹.

On the other hand, breath sampling suffers from sample instability, offers lower concentrations in samples with a 10% failure rate⁷⁴, difficulty in quality control and standardization of sample collection along with the confounding effect of VOCs in the external environment⁸². In this review, sample-based subgroup analysis showed the diagnostic accuracy for urine sampling^30,52,73 for VOC analysis was closer to the overall pooled diagnostic accuracy measures than the stool sample group^27,70, while the latter group was limited by sample size and quality of the study. The sample size for the breath analysis group was the highest, but it showed less sensitivity than pooled estimates, which could be explained by loss of VOC concentration during the collection process; however, this cannot be said with certainty and is an area of further research. Recent studies conducted on urine samples have shown promising results and feasibility^30,83, advocating its use in the 2WW pathway. Based on literature evidence and review findings, we suggest using urine as a preferable sample.

It was also demonstrated that the combination of FIT-VOC enhanced the overall sensitivity of detecting AAs. FIT has been studied previously in combination with other non-invasive modalities to increase its sensitivity for both colorectal cancer and adenomas. In a large prospective Deep-C study⁸⁴ combined mitochondrial DNA-FIT (Cologuard) had a sensitivity of 47% only for precancerous lesions, and in another review with no improvement⁸⁵. The findings suggest that the FIT-VOC combination yields maximal sensitivity (89%) compared with other non-invasive modalities, with the additional advantage of urine-based sampling with easier collection⁸⁶ and increased patient acceptability.

The significance of an index test depends on its discriminatory power to classify disease from non-diseased patients, which can be analysed by the AUC⁸⁷. According to SROC bivariate analysis in this review, the VOC test offers diagnostic accuracy with an AUC of 0.84, which suggests the significant ability of the VOC test in distinguishing patients with AAs from non-diseased patients, whilst FIT has a poor AUC (0.65), indicating its unsatisfactory diagnostic performance in clinical practice.

The VOC test can reduce the probability of missing patients with AA to 3%. This was depicted in posttest probability nomograms in which the chance of having AA doubled after FIT or VOC test positivity. However, in the case of a negative FIT test the posttest probability of having disease remained at 10% (which is still significant with the prevalence of AA at 15%), while it was reduced to 3% in the case of a negative VOC test, and these values are comparable to findings in a comparative analysis of FIT and VOC testing⁷. This could have implications in reducing false negative rates and patient safety concerns and increasing reassurance in average-risk adults.

Although diagnostic accuracy for polyps at various threshold levels was not explored in this review, previous studies suggest that a threshold of 120 μg/g Hg (used in the UK screening population) missed over half of the colorectal cancer and three-quarters of high-risk adenoma patients with a sensitivity of 25% reported at 120 μg/g Hg and for colorectal cancer 47.8%⁴⁷. A few countries are now using the FIT threshold of 20 μg/g Hg in the screening population^88,89; whether this can be applied to the UK remains an area of further research. Even though most of the population representation was mixed in this review, the pooled prevalence and FIT sensitivity were not different when compared with previous research articles^50,90 and review findings²⁸.

As colorectal cancer mainly develops from adenomas, it is important to detect precancerous lesions for cancer prevention and crucial that the test used for initial stratification will not miss a disease when it is present⁹¹. Demand for colonoscopies in the UK continues to increase, and each colonoscopy performed in the NHS outpatient department costs €665⁹². The adenoma miss rate among trained colonoscopists remains at 26–62%⁹³, which is the main cause of interval cancer⁹⁴. However, it is not clear whether increasing the sensitivity of a test to detect adenomas would increase the adenoma detection rate, and this is an area for further research.

Using FIT alone for triage could miss more than half of the advanced polyps⁵⁰, therefore, a better triage tool is required with high sensitivity and more acceptability, and considering our review findings we recommend the VOC test alone or in combination with FIT in the fast-track pathway to aid early detection of high-risk polyps. Full healthcare economic research is required to establish the cost-effectiveness of the combination of VOC and FIT as a triage test but is expected to be less expensive than an upfront colonoscopy strategy. As VOC analysis can be done on urine samples this could increase patient acceptability.

In clinical practice, despite the huge amount of work on other non-invasive methods, FIT is still used as a standard triage tool in the NHS fast-track pathway. Although it is preliminary to add VOC in clinical pathways and further studies are required to validate the findings, however, based on the review findings we propose the following pathway (Fig. 5) to introduce VOC as a triage tool alone or in combination with FIT to increase the overall sensitivity of detecting AA(s) to prevent colorectal cancer. This may be more applicable to symptomatic/surveillance⁵² or younger asymptomatic populations with a polyp prevalence of 15–30%¹⁶ or where the screening threshold for FIT is low. It is too early to draw conclusions and further validation studies are required.

Fig. 5.

Proposed algorithm for FIT and VOC testing as a triage tool for advanced colorectal adenomas

VOC, volatile organic compound; FIT, faecal immunochemical test; AA, advanced adenoma.

VOC testing is a non-invasive potential future triage tool, and its introduction in a fast-track pathway, alone or in combination with FIT can offer the following main benefits:

• Subjects with a negative FIT undergo further VOC testing which if positive mandates a referral for colonoscopy to help in the early detection of lesions.

• The combination of FIT-VOC in a fast-track pathway (especially FIT-negative symptomatic patients) will provide more reassurance, significantly reduce the false negative rate and increase patient safety.

• In the long term it will reduce the number of advanced lesions or colorectal cancer burden hence reducing morbidity rate and mortality rates.

GC-MS is recommended as the ‘standard’ and is advocated to be used for future studies in order to define thresholds for both colorectal cancer and adenomas. Urine testing for VOC analysis can be a preferable sample type to increase patient compliance in screening pathways. Due to technological considerations, sample analysis in centralized laboratory systems with networks/hubs will be required.

There were several limitations in this review. Data was limited to AAs as most of the studies were designed to primarily collect and report data on colorectal cancer and polyp data was secondarily reported with most studies reporting outcomes on AAs only. In addition, due to limited information, the ability to classify adenomas according to anatomical location (right- or left-sided) could not be reported.

Due to a lack of information, a detailed evaluation of threshold concentrations and chemical analysis for VOC for AA(s) could not be performed. The studies were mainly prospective with a lack of comparative or randomized clinical study designs of FIT and VOC tests. Another difficulty was presenting data in various population groups (that is, screening/symptomatic/surveillance) as most of the studies recruited a mixed population (with mostly symptomatic/surveillance patients) for both FIT and VOC data with a lack of reporting of diagnostic accuracies on individual groups for polyps.

The data was of moderate quality, with most of the studies showing unclear RoB. The quality of statistical reporting in most of the articles was inadequate with a lack of confidence interval and contingency tables, due to which calculation of detailed values were computed from the best available information from studies included in this review. Therefore, slight variation in numbers is possible; however, effort has been made to ensure the accuracy of data and this has been re-checked to maintain the quality of the review.

Funding

The authors have no funding to declare.

Disclosure

The authors declare no conflict of interest.

Supplementary material

Supplementary material is available at BJS Open online.

Data availability

The authors confirm that the data supporting the findings of this review are available within the article and its Supplementary material.

Author contributions

Asma Afzal (Conceptualization, Data curation, Formal analysis, Methodology, Project administration, Resources, Software, Visualization, Writing—original draft, Writing—review & editing), Yekaterina Skvarkovskaia Aranan (Data curation, Methodology, Validation), Tom Roberts (Data curation, Methodology, Validation), James Covington (Supervision, Writing—review & editing), Lorena Vidal (Writing—review & editing), Sonia Ahmed (Methodology, Supervision, Writing—review & editing), Talvinder Gill (Supervision, Writing—review & editing) and Nader K. Francis (Conceptualization, Methodology, Project administration, Supervision, Writing—review & editing)