Exhaled breath profiling for early silicosis detection in workers at risk: a pilot study Authors: Nele Van Loon, David Ruttens, Mathieu Verbrugghe, Kathleen Zwijsen, Pierre Van Damme, Thérèse Lapperre, Kevin Lamote Published: 2026 ABSTRACT BackgroundSilicosis remains a major occupational health concern in respirable crystalline silica (RCS)-exposed workers. Exhaled breath analysis using volatile organic compound (VOC) profiling has emerged as a promising noninvasive approach for early disease detection. This feasibility study evaluated multicapillary column/ion mobility spectrometry (MCC/IMS)-based VOC profiling for differentiating silicosis from other respiratory conditions and healthy controls in an occupational health setting.MethodsWe conducted a multicentre, cross-sectional observational study employing MCC/IMS to identify VOC patterns. Breath data were reduced to principal components and used in logistic lasso regression to develop predictive models. Internal and external validation assessed model performance, and subgroup analyses explored the impact of comorbidities and spirometry abnormalities.ResultsA total of 185 participants were enrolled, including 24 COPD patients, 26 with benign asbestos-related diseases (BARD) and 32 with confirmed silicosis, alongside 25 healthy controls and 78 RCS-exposed controls. Silicosis was distinguished from BARD and COPD with 100% and 87.5% accuracy (area under the receiver operating characteristic curve (AUCROC) 1.00 and 0.89), respectively. Comparison with healthy RCS-exposed controls yielded 87.8% accuracy with 98.5% sensitivity (AUCROC 0.58), improving to 82.8% accuracy and AUCROC 0.90 when considering spirometry abnormalities. Eight VOCs, notably P13 and P14, were identified as key discriminators.ConclusionMCC/IMS-based breath analysis shows promise as a rapid, portable and noninvasive technique for early detection of silicosis in RCS-exposed workers. Future research should validate these findings in larger cohorts, integrate cross-validation of candidate biomarkers across analytical platforms and establish standardised protocols to enable routine implementation in occupational health surveillance. Introduction Silicosis, the most widely recognised pneumoconiosis, results from inhalation of respirable crystalline silica (RCS), a component of rock and sand [1]. Occupational exposure occurs across multiple industries, including construction, metallurgy, mining, quarrying, glass and ceramics. In 2023, the Belgian National Social Security Office reported 211.276 construction workers, with an estimated 10% exposed to quartz. In recent decades, artificial stone-associated silicosis has been widely reported with an accelerated, 4–10 years, disease onset due to elevated exposure levels during high-intensity job tasks such as cutting, polishing and grinding, both in workshops and during in-home installations [2]. Furthermore, the presence of various comorbidities, including chronic obstructive pulmonary disease (COPD), chronic bronchitis, tuberculosis and scleroderma, are possible [3]. The International Agency for Research on Cancer (IARC) classifies silica as a group 1 human lung carcinogen [4]. A multicentre IARC study estimated an excess lifetime lung cancer risk of 1.1–1.7% for workers exposed to 0.1 mg·m−3 RCS between ages 20 and 65 years, in addition to background risks of 3–6% [5]. Although early detection and cessation of exposure can significantly slow disease progression, early-stage silicosis diagnosis in occupational health practice remains challenging [6] and underscores the need for sensitive, specific and efficient biomarkers. The quest for a simple, noninvasive diagnostic tool for early-stage silicosis has given rise to the field of “breathomics”. Exhaled breath contains volatile organic compounds (VOCs) from exogenous and endogenous sources. To date, “breathomics” has been extensively investigated in asthma [7, 8], COPD [9], lung cancer and occupational lung diseases [6, 10], although clinical implementation remains experimental; it has not been widely investigated in silicosis [11]. Exposure to asbestos or silica induces both chronic inflammation and oxidative stress, altering metabolic pathways and, consequently, the VOC profile. These changes may serve as biomarkers for conditions such as malignant pleural mesothelioma or silicosis [12–14]. Three pilot studies have investigated VOCs in exhaled breath among silicosis patients using case–control study designs [13, 15, 16]. Of these, one employed sensor technology (electronic nose), while two employed gas chromatography–mass spectrometry (GC-MS). Breath analysis resulted in a VOC signature based on the presence of specific peaks in the breath print. Despite VOC signatures identification, divergent analytical methodologies were used. GC-MS, while accurate, is both time-consuming and expensive, whereas sensor-based approaches lack compound identification capability. Multicapillary column/ion mobility spectrometry (MCC/IMS) offers a practical alternative, combining ease of use with compound identification at a lower cost than GC-MS. To date, no studies have applied MCC/IMS for silicosis detection. Therefore, this feasibility study evaluated the non-invasive assessment of silicosis through VOC profiling using MCC/IMS. Materials and methods Study design and population We conducted a multicentre, cross-sectional observational study approved by the Ethics Committee of Antwerp University Hospital/University of Antwerp (Belgian registration number B3002022000069) in compliance with the Helsinki Convention. Sample size assumed 15% silicosis prevalence and 15% dropout (80% power, H0 sensitivity 50% versus H1 80%, and p<0.05), yielding a minimum of 173 participants [17]. Confirmed silicosis cases, established by a pulmonologist through chest imaging and/or lung biopsy, were recruited from the Department of Lung Diseases at Ziekenhuis Oost-Limburg (Genk, Belgium). Chest imaging consisted of high-resolution computed tomography (HRCT) using a semi-quantitative classification system based on the International Classification of HRCT for Occupational and Environmental Respiratory Diseases (IOERD) [18]. When the occupational history was unclear, a lung biopsy was obtained for histologic confirmation of RCS exposure using standard diagnostic histological criteria. More specifically, the diagnosis of silicosis in mediastinal lymph nodes required the presence of silicotic nodules in different stages of formation, with the final stage consisting of relatively acellular sclerohyaline nodules and the presence of birefringent particles consistent with silica particles, with or without accompanying granulomatous inflammation [19, 20]. All controls were recruited from the Occupational Health Service, Mensura NV (Brussels, Belgium), while patients with COPD or benign asbestos-related diseases (BARD) were enrolled from the Department of Respiratory Medicine at Antwerp University Hospital (Antwerp, Belgium). Following written informed consent, a validated questionnaire and breath sampling was performed; control groups additionally received pulmonary function testing. All RCS-exposed controls were subject to periodic occupational health surveillance. Imaging was available in 92% of workers consisting of routine chest radiography in 60% and HRCT screening in 32%. Besides demographic characteristics (age, sex, ethnicity), smoking history, medical history and detailed information on silica exposure were recorded. Current and previous jobs with potential mineral dust exposure were documented. Occupational exposure risk was categorised based on job tasks, material characteristics and quartz content into three tiers (no, low and high) following criteria outlined in European guidance for respirable crystalline silica exposure assessment [21, 22]. Table 1 lists low-risk occupations including roles with limited or incidental mineral dust contact and high-risk occupations comprising tasks involving direct manipulation or processing of quartz-containing materials [23, 24]. Exposure duration was recorded as the total number of years exposed and further grouped into three categories (none, <15 and ≥15 years). Exclusions for VOC analysis included acute respiratory infection symptoms, moderately controlled diabetes mellitus (HbA1C >7%), uraemia and active cancer [13]. TABLE 1Risk classification of occupations with potential mineral dust exposureLow-risk occupations#High-risk occupations¶Truck driver ((un)loading raw materials)Installer/manufacturer of quartz composite productsProduction workerSilica sand extraction workerWelderSandblasterMechanicConcrete (foundation) repairer/processor/removerPainterConcrete drilling and cutting specialistCrane operatorConcrete cleanerAsbestos-removal workerMachine maintenance personnel in concrete plantsGypsum board installerManufacturer of concrete/cement/gypsum productsFinishing assemblerInstaller of interior walls/gypsum boards/ceilingsSheet metal workerFloor and/or joint grinderCarpenterTrench cutterInsulation installerDemolition/dismantling workerTiler/floor layer/bricklayer without cuttingElectrician performing grinding operationsPlastererStone or marble cutter/finisherRoad construction workerSpecialist in cutting/drilling/milling/grinding/polishing/breaking or abrasive blasting of quartz-containing materialsMiner#: Roles with limited or incidental mineral dust contact. ¶: Tasks involving direct manipulation or processing of quartz-containing materials. Risk classification of occupations with potential mineral dust exposure #: Roles with limited or incidental mineral dust contact. ¶: Tasks involving direct manipulation or processing of quartz-containing materials. Pulmonary function testing Spirometry among controls was performed by two trained occupational health nurses following American Thoracic/European Respiratory Society standards. Contraindications included recent thoracic/abdominal/ocular surgery (<1 month), diving (<3 days), aneurysm, unstable cardiovascular disease, myocardial infarction/pulmonary embolism (<3 months), pneumothorax (<6 weeks) and recurrent syncope. The largest forced vital capacity (FVC), forced expiratory volume in 1 s (FEV1) and FEV1/FVC ratio were expressed as absolute values and % predicted by the Global Lung Initiative (GLI) [25]. Normal lung function was defined as FVC and FEV1 >80% pred and FEV1/FVC >0.7 [26]. Excessive decline was defined as FEV1 decrease of >15% beyond age-related expectations or >60 mL·year–1 [27, 28]. For silicosis patients, spirometry data were retrieved from their medical record. Exhaled breath sampling Breath samples were collected using a SpiroScout breath sampler (Ganshorn Medizin Electronic, Niederlauer, Germany) connected to a BioScout MCC/IMS device (B. Braun, Dortmund, Germany). Following European technical standards [29], participants were instructed to avoid fragrances and refrain from eating/drinking/smoking for at least two hours prior to sampling, to rinse their mouth with distilled water, to wear rubber gloves and a nose clip, and sit upright. After 3 min of normal breathing through the SpiroScout mouthpiece, 10 mL of alveolar air was sampled and breath analytes were analysed: pre-separation used a nonpolar OV-5 MCC column (Multichrom Ltd, Novosibirsk, Russia), ionisation via 95MBq 63Ni β-radiation source and separation in a 12-cm drift tube under the influence of an electrical field with 99.999% nitrogen counterflow (α1-nitrogen gas, Air Liquide Medical, Belgium). VOCs were detected on a Faraday plate, evoking electrical signals proportional to VOC concentration. After breath sampling, background samples were collected identically. Disposable mouthpieces and filters were employed, and the MCC/IMS was continuously flushed with α1-nitrogen to prevent contamination. All sample lines were unheated and composed of inert Teflon (PTFE). To ensure clean IMS chromatograms, the MCC/IMS was flushed with humid air between participants and a disposable respiratory filter (Accuramed BV, Halen, Belgium) was added for COVID-19 precautions. Data processing and statistics The Shapiro–Wilk test determined the normality of the continuous variables, which were found to be non-normal and, therefore, expressed as the median with interquartile range (IQR). Kruskal–Wallis, Chi-square and Mann–Whitney U tests were used to assess differences between the groups. Chromatograms were processed using VisualNow software (B. Braun, Dortmund, Germany). Raw two dimensional chromatograms underwent baseline correction, alignment, normalisation to the reactant ion peak (RIP) and RIP-tail compensation [30]. After smoothing the data, manually selected VOCs yielded peak intensity lists. To minimise external confounding, alveolar gradients of each detected VOC were calculated (breath minus background sample peak intensity) and used as independent variables. Principal component analysis (Varimax rotation and Kaiser normalisation) supported VOC profile discovery. Receiver operating characteristic (ROC) curves determined areas under the curve (AUC), sensitivity, specificity, positive (PPV) and negative predictive value (NPV), and diagnostic accuracy with 95% confidence intervals were calculated. Goodness-of-fit assessed ROC calibration. Confounding was evaluated by stratification and multivariable analysis (R and SPSS Statistics 25). Data were pseudonymised. Modelling A multinomial logistic regression model was developed to classify participants into pre-defined outcome categories (silicosis, nonexposed and exposed controls). Data were randomly split into training (70%) and validation (30%) sets using stratified sampling to preserve class proportions. Model performance was assessed through internal validation on the hold-out set and external validation on an independent cohort. Discrimination was quantified using accuracy and AUCROC metrics, and class-specific sensitivity was calculated. Misclassification patterns were examined via confusion matrices. For each predictive model a systematic approach combining least absolute shrinkage and selection operator (lasso) regression with leave-one-out cross-validation (LOOCV) identified discriminative VOCs for silicosis and RCS exposure. VOCs selected in ≥50% LOOCV folds were considered the most relevant variables in differentiating silicosis patients from all controls (with and without RCS exposure). Backward logistic regression on the enriched VOC set determined significant contributors to group differentiation. To support classification of disease versus exposed controls, MCC/IMS findings were cross-checked with GC-MS VOC analyses [12]. Results Participant characteristics A total of 185 participants were recruited and comprised 165 adult males and 20 females aged 25–81 years (table 2). The study cohort consisted of five groups: 1) 32 patients with proven silicosis; 2) 25 unexposed healthy controls; 3) 78 RCS-exposed controls from different industries with a high exposure risk in 73% of cases; 4) 24 COPD patients; 5) 26 BARD patients. Besides demographic characteristics, smoking status and silica job exposure history, we recorded the history of lung and immune-related diseases for all controls. The silicosis and RCS-exposed groups had a higher proportion of current/former smokers (50% of the silicosis and 55% of the RCS-exposed group compared to 36% of the healthy unexposed controls). Statistical analysis comparing the exposed and unexposed controls with the silicosis group shows in addition to a wide range in age, significant differences for job exposure time (JET) (p<0.001), history of disease (p<0.01) and pulmonary function abnormalities (p<0.001). The median JET was found to be considerably higher in the silicosis group compared to the RCS-exposed group (25 years (IQR 18), versus 11 years (IQR 13.5)) with more than 80% of the silicosis group being exposed for longer than 15 years versus 35% of the RCS-exposed group. In addition, the silicosis group exhibited a higher prevalence of history of lung diseases (38% of the silicosis group, 10% of RCS-exposed controls and 0% of healthy controls, by definition) as well as autoimmune-related diseases (16% versus 3%). Spirometry evaluation for the silicosis group showed abnormal readings in 63% of all participants, suggesting that most cases were severe enough to cause significant functional impairment. TABLE 2Participant characteristicsCharacteristicBARD(N=26)COPD(N=24)Silicosis(N=32)Unexposed controls(N=25)RCS-exposed controls(N=78) Age, median (IQR)# ¶ *** 63.19 (7.17)70.27 (8.74)57.35 (14.49)53.25 (18.15)45.29 (16.95) Pack-years, median (IQR) 1.88 (27.00)50.00 (26.21)0.05 (13.05)0.00 (4.38)0.40 (9.13) RCS-JET, median (IQR) #¶ *** NANA25.00 (18.00)NA11.00 (13.50) Sex, n (%) #+ Male26 (100)16 (66.67)31 (96.88)19 (76.00)73 (93.59) Smoking, n (%) Current/former15 (57.69)24 (100)16 (50.00)9 (36.00)43 (55.13) Never11 (42.31)–16 (50.00)16 (64.00)35 (44.87) History lung disease, n (%) ¶ No26 (100)–22 (68.75)25 (100)70 (89.74) Yes–24 (100)10 (37.50)–8 (10.26) Sarcoidosis––1 (3.13)–1 (1.28) Asthma–2 (8.33)3 (9.38)–6 (7.69) COPD–24 (100)4 (15.63)–– Asthma/COPD––1 (3.13)–– Tuberculosis–2 (8.33)1 (3.13)–– Other (drug use)––––1 (1.28) History auto-immune disease, n (%) ¶ NoNA24 (100)27 (84.38)25 (100)76 (97.44) YesNA–5 (15.63)–2 (2.56) Rheumatoid arthritis––––2 (2.56) Kaplan––2 (6.25)–– Scleroderma––––– Vasculitis––1 (3.13)–– Lupus––––– Raynaud––2 (6.25)–– RCS job exposure risk, n (%) No–19 (79.17)2 (6.25)25 (100)– Low§26 (100)1 (4.17)8 (25.00)–21 (26.92) Highƒ–4 (16.67)22 (68.75)–57 (73.08) RCS JET, n (%) ¶ *** No NA19 (79.17)–25 (100)– <15 years ––6 (18.75)–51 (65.38) ≥15 years –5 (20.83)26 (81.25)–27 (34.62) Spirometry abnormalities, n(%)¶*** NoNA–12 (37.50)25 (100)62 (79.49) YesNA24 (100)20 (62.50)–16 (20.51) FCO––8 (25.00)–11 (14.10) PPM––3 (9.38)–3 (3.85) FCO+PPM––9 (28.13)–2 (2.56)BARD: patients with benign asbestos-related diseases; COPD: COPD patients; FCO: abnormal fixed cutoff values (forced expiratory volume in 1 s (FEV1) and/or forced vital capacity (FVC) <80%, FEV1/FVC <0.7); IQR: interquartile range; JET: job exposure time (total years exposed); NA: not applicable/available; PPM: abnormal percentage predicted method (FEV1 % pred decline >15%); RCS: respirable crystalline silica. #: Silicosis versus all controls, p-value <0.05. ¶: Silicosis versus RCS-exposed controls, p-value <0.05. +: Silicosis versus unexposed healthy controls, p-value <0.05. §: Low-risk occupations included roles with limited or incidental mineral dust contact, such as truck drivers ((un)loading raw materials), production workers, welders, mechanics, painters, crane operators, asbestos-removal workers, gypsum board installers, finishing assemblers, sheet metal workers, carpenters, insulation installers, tilers/floor layers/bricklayers without cutting activities, plasterers, road construction workers and miners. ƒ: High-risk occupations comprised tasks involving direct manipulation or processing of quartz-containing materials, including installers/manufacturers of quartz composite products, concrete repairers/processors, sandblasters, joint grinders, concrete drilling and cutting specialists, demolition/dismantling workers, trench cutters, concrete cleaners, electricians performing grinding operations, manufacturers of concrete/cement/gypsum products, machine maintenance personnel in concrete plants, floor and joint grinders, and skilled workers engaged in cutting/drilling/milling/grinding/polishing/breaking or abrasive blasting of quartz-containing materials. Additional high-risk activities included removal of concrete foundation heads, stone or marble cutting/finishing, and installation of interior walls/gypsum boards/ceilings. ***: p<0.001. Participant characteristics BARD: patients with benign asbestos-related diseases; COPD: COPD patients; FCO: abnormal fixed cutoff values (forced expiratory volume in 1 s (FEV1) and/or forced vital capacity (FVC) <80%, FEV1/FVC <0.7); IQR: interquartile range; JET: job exposure time (total years exposed); NA: not applicable/available; PPM: abnormal percentage predicted method (FEV1 % pred decline >15%); RCS: respirable crystalline silica. #: Silicosis versus all controls, p-value <0.05. ¶: Silicosis versus RCS-exposed controls, p-value <0.05. +: Silicosis versus unexposed healthy controls, p-value <0.05. §: Low-risk occupations included roles with limited or incidental mineral dust contact, such as truck drivers ((un)loading raw materials), production workers, welders, mechanics, painters, crane operators, asbestos-removal workers, gypsum board installers, finishing assemblers, sheet metal workers, carpenters, insulation installers, tilers/floor layers/bricklayers without cutting activities, plasterers, road construction workers and miners. ƒ: High-risk occupations comprised tasks involving direct manipulation or processing of quartz-containing materials, including installers/manufacturers of quartz composite products, concrete repairers/processors, sandblasters, joint grinders, concrete drilling and cutting specialists, demolition/dismantling workers, trench cutters, concrete cleaners, electricians performing grinding operations, manufacturers of concrete/cement/gypsum products, machine maintenance personnel in concrete plants, floor and joint grinders, and skilled workers engaged in cutting/drilling/milling/grinding/polishing/breaking or abrasive blasting of quartz-containing materials. Additional high-risk activities included removal of concrete foundation heads, stone or marble cutting/finishing, and installation of interior walls/gypsum boards/ceilings. ***: p<0.001. Breath analysis Of 185 participants, 182 provided a breath sample. Of these, four contaminated chromatograms were excluded due to elevated background noise. Across valid samples, 103 VOC peaks were detected. Based on the breath profiles, the BARD group exhibited a distinct signature (figure 1). To avoid clinical confounding, individuals with prior lung diseases, autoimmune disorders and/or illicit drug use were excluded from final VOC analysis, yielding 65 healthy RCS-exposed controls (RCS-HC) and 17 “pure” silicosis patients (SIL). Multinomial logistic regression (70:30 split) showed excellent model performance, achieving 100% sensitivity for silicosis in training. Internal and external validation accuracies were 92.1% and 90.6% (AUCROC 0.98 and 0.87), respectively, with only few misclassifications among controls (n=1). To address the study objectives, silicosis was distinguished from BARD and COPD with 100% (AUCROC 1.00) and 87.5% (AUCROC 0.89) accuracies, respectively. Model characteristics are summarised in table 3. Next, silicosis patients were compared to unexposed healthy controls, achieving 62% accuracy with 100% sensitivity and NPV (AUCROC 0.60, 95% CI 0.24–0.57). Furthermore, despite an AUCROC of 0.58 (95% CI 0.36–0.78), silicosis patients were discriminated from RCS-HC achieving 87.8% accuracy with 98.5% sensitivity and 88.9% NPV, underscoring the potential role for exclusive screening. Discriminative VOCs included P13, P14, P19, P26, P51, P53, P55 and P102 (table 4, figure 2). Final backward logistic regression yielded a rule-out model for silicosis with AUCROC 0.92. Model outputs indicated P14 and P13 as key VOC predictors (p=0.01); notably, independent of job exposure time and pulmonary function parameters, P13 most strongly separated silicosis patients from RCS-exposed workers (AUCROC 0.96 and 0.98, respectively) (table 4). Considering spirometry abnormalities, RCS-HC with abnormal spirometry (RCS-spir1) were best distinguished from silicosis patients, achieving 82.8% accuracy, 82.4% sensitivity, 83.3% specificity, 76.9% NPV and 87.5% PPV, with AUCROC 0.90 (95% CI 0.77–0.96). Principal component analysis: volatile organic compound two-dimensional plot. BARD: patients with benign asbestos-related diseases; Dim1/2: dimension 1 and 2; HC: healthy controls; RCS-C: respirable crystalline silica-exposed controls; SIL: silicosis patients. TABLE 3Model characteristics for silicosis discriminationStudied groupsSIL versus COPDSIL versus BARDSIL versus HCSIL versus RCS-HCSIL versus RCS-spir0SIL versus RCS-spir1Subjects (n)17 versus 2417 versus 2617 versus 2517 versus 6517 versus 5317 versus 12 Lasso regression LOOCV, % (95% CI) Sensitivity87.0 (68.5–96.6)100 (89.1–100)100 (83.8–100)98.5 (92.7–99.9)92.5 (82.8–97.5)82.4 (59.2–95.3) Specificity82.0 (66.3–98.0)100 (83.8–100)36.0 (19.2–55.9)47.1 (24.8–70.3)52.9 (29.8–75.2)83.3 (55.0–97.1) PPV90.9 (73.1–98.4)100 (89.1–100)51.5 (34.7–68.1)87.7 (78.6–93.8)86.0 (75.1–93.3)87.5 (64.5–97.8) NPV83.3 (61.1–95.6)100 (83.8–100)100 (71.7–100)88.9 (56.2–99.4)69.2 (41.3–89.4)76.9 (49.2–93.8) Accuracy87.5 (74.5–95.3)100 (93.3–100)61.9 (46.7–75.6)87.8 (79.4–93.6)82.9 (72.7–90.4)82.8 (65.9–93.4) AUCROC88.5 (74.2–99.2)100 (100–100)60.0 (23.5–57.4)57.5 (36.3–78.3)53.2 (31.2–74.8)89.7 (76.5–98.5) VOCs (>50% of times selected) P1P3P1P13P13P13P8P35P11P14P14P17P40P42P13P19P19P29P41P43P42P26P26P42P48P46P43P51P51P46P89P87P54P53P53P65P90P56P55P55P80P102P79P84P103AUCROC: area under the receiver operating characteristic curve; BARD: patients with benign asbestos-related diseases; COPD: COPD patients; HC: healthy controls; Lasso: least absolute shrinkage and selection operator; LOOCV: leave-one-out cross-validation; NPV: negative predictive value; P: alveolar gradient of the peak; PPV: positive predictive value; RCS: respirable crystalline silica; RCS-HC: RCS-exposed healthy controls; RCS-spir0: RCS-exposed controls with normal spirometry values; RCS-spir1: RCS-exposed controls with spirometry abnormalities; SIL: silicosis patients; VOC: volatile organic compound. Model characteristics for silicosis discrimination AUCROC: area under the receiver operating characteristic curve; BARD: patients with benign asbestos-related diseases; COPD: COPD patients; HC: healthy controls; Lasso: least absolute shrinkage and selection operator; LOOCV: leave-one-out cross-validation; NPV: negative predictive value; P: alveolar gradient of the peak; PPV: positive predictive value; RCS: respirable crystalline silica; RCS-HC: RCS-exposed healthy controls; RCS-spir0: RCS-exposed controls with normal spirometry values; RCS-spir1: RCS-exposed controls with spirometry abnormalities; SIL: silicosis patients; VOC: volatile organic compound. TABLE 4Volatile organic compounds (VOCs) of interest in the silicosis model versus healthy respirable crystalline silica (RCS)-exposed controlsSpecificationsVOCRT (s)1/K₀ (V·s·cm−2)Tentative identificationP1325.60.521PhenolP144.50.688PentanalP1961.60.6782-NonanoneP2641.00.601Y-TerpineneP512.30.513EthanolP5317.40.515O-Xylene, p-xyleneP5530.50.6021-Methyl-4-(1-methylethyl) 1,3-cyclohexadieneP1022.9−0.5923-Methyl butanolBackward logistic regressionVariableβ-coefficientStandard errorz-valuep-value Intercept −3.070.65−4.71<0.001 P26 −361.35296.53−1.340.180 P51 −53.5042.98−1.250.210 P14 77.9629.272.660.007 P13 172.0470.412.440.010Variable Intercept −20.279.19−2.210.03 Spiro abnormalities 9.654.592.100.04 P14 72.8538.891.870.06 P13 632.40273.952.310.02Forward logistic regressionVariable Intercept −3.430.82−4.18<0.001 JET±15years 2.220.852.620.009 P14 83.7433.382.510.010 P13 232.7571.973.230.0011/K: inverse reduced ion mobility expressed in V·s-1·cm-2; JET: job exposure time; P: alveolar gradient of the peak; RCS: respirable crystalline silica; RT: retention time; VOC: volatile organic compound. Volatile organic compounds (VOCs) of interest in the silicosis model versus healthy respirable crystalline silica (RCS)-exposed controls 1/K: inverse reduced ion mobility expressed in V·s-1·cm-2; JET: job exposure time; P: alveolar gradient of the peak; RCS: respirable crystalline silica; RT: retention time; VOC: volatile organic compound. Box-plot analysis of identified volatile organic compounds (VOCs) in silicosis model versus healthy respirable crystalline silica (RCS)-exposed controls. p: alveolar gradient of the peak. Mann–Whitney U tests indicate VOC significance. *: p<0.05; **: p<0.01; ***p<0.001. Discussion MCC/IMS-based VOC profiling proved feasible for differentiating silicosis from BARD and COPD. Moreover, exhaled breath analysis distinguished silicosis patients from healthy RCS-exposed workers with high accuracy and precision, minimising false positives. These findings suggest the proposed model could serve as a diagnostic tool to rule out disease among at-risk workers and enable targeted follow-up and surveillance. Our results align with previous research showing distinct “breath prints” in respiratory diseases such as COPD [9, 31] and pneumoconiosis, including silicosis [11, 13]. Most studies employed GC-MS for VOC identification, a method that, while accurate, is time-consuming and costly. In contrast, MCC/IMS offers a portable, sensitive, user-friendly alternative, enabling pseudo-identification of VOC fingerprints based on their retention time and ion mobility. Nevertheless, definitive chemical identification requires GC-MS validation [12]. To assess validity, we compared “pure” silicosis patients, those without confounding comorbidities, to healthy controls. A dedicated analysis comparing silicosis to unexposed healthy controls yielded modest discrimination. We hypothesise that this likely reflects the absence of background silica-related airway reactivity in healthy unexposed individuals, which reduces group heterogeneity, narrowing VOC variability and thereby limiting separability. By contrast, discrimination between silicosis and RCS-exposed workers, the clinically relevant comparison, is strong, which supports an “exposure-driven” hypothesis. Given that a diagnosis of silicosis mandates removal from further RCS exposure, we may assume that the exposure-induced pathophysiological responses as inflammation gradually subside; from this perspective, the VOC profile may primarily reflect the biological response to RCS exposure itself, making the distinction between unexposed healthy controls and silicosis patients inherently less pronounced. Despite low specificity (36%), LOOCV analysis using seven VOCs selected via lasso regression achieved 100% sensitivity and NPV, indicating strong potential for disease exclusion in a subset of cases. Comparatively, a 2023 study employing a larger cohort (69 silicosis patients versus 92 controls) developed a predictive model using nine VOC biomarkers via orthogonal partial least squares discriminant analysis (OPLS-DA), reporting high specificity (89.2%) but low sensitivity (60.3%) with AUCROC 0.75 [11]. Similarly, recent Australian research leveraging atmospheric pressure chemical ionisation–mass spectrometry (APCI-MS) achieved AUCROC 0.93 (±0.038) distinguishing 31 silicosis patients from 60 controls [6]. While these studies attained higher diagnostic performance, neither accounted for comorbidities, a factor addressed in our analysis. Beyond internal validation, we explored whether VOC profiling could differentiate silicosis from RCS exposure as a risk factor for disease onset. Breath-print comparisons revealed high precision in ruling out disease (sensitivity 98.5%) with minimal false positives (PPV 87.7%). Given prior evidence linking pulmonary function impairment to disease progression [32, 33], we evaluated pulmonary function parameters, confirming their relevance in distinguishing silicosis patients from RCS-exposed workers. Incorporating this variable improved model performance, with eight VOCs effectively discriminating between silicosis patients and RCS-exposed controls presenting spirometry abnormalities (RCS-spir1), achieving 82.8% accuracy and AUCROC 0.90. The metabolic alterations underlying occupational lung diseases likely explain VOC differences [6]. Inhaled dust particles reaching the alveolar space trigger macrophage-mediated inflammation, generating reactive oxygen species and lipid peroxidation [34], a key mechanism in silicosis pathogenesis. Animal studies have linked alkanes, such as pentane, to lipid peroxidation [35], while case–control investigations in pneumoconiosis patients identified alkanes and methylated alkanes as characteristic breath print components [11, 13]. Complementary GC-MS analyses reported elevated acetaldehyde, hexanal, nonanal and other compounds in silica-exposed individuals and silicosis patients [16]. However, due to the small number of silicosis cases (n=4), these findings were inconclusive regarding diagnostic performance. More recently, the 2023 OPLS-DA study proposed nine candidate biomarkers for silicosis, including 2,3-butanedione, ethyl acetate, chlorobenzene and others, achieving 89.6% accuracy [11], while APCI-MS profiling suggested leukotriene-E3 as a potential marker, though validation remains pending [6]. Despite heterogeneity across studies, the present analysis identified two VOCs, P13 and P14, via lasso regression as significant discriminators between silicosis patients and at-risk workers. Tentative compound identification based on a reference library suggests P13 may potentially correspond to phenol and P14 to pentanal. Phenol has been associated with bacterial catabolism of aromatic amino acids and linked to gastrointestinal malignancies [31], whereas pentanal, an aldehyde and lipid peroxidation product, is recognised as a biomarker of oxidative stress-mediated lung injury [36]. Although precise chemical identification was beyond the scope of this study, table 4 shows that several identified discriminating VOCs, although tentative, overlap with compound classes reported in all three aforementioned VOC pilot studies for silicosis [13, 15, 16]. Particularly aldehydes and alkanes, as well as aromatic hydrocarbons. These convergent findings highlight biologically plausible pathways, namely oxidative stress and altered aromatic metabolism, underlying silicosis and warrant further validation in larger cohorts. The primary strength of this study lies in its exploration of VOC profiling within occupational health surveillance for respiratory health prevention. Despite encouraging findings, several limitations warrant consideration. First, although smoking status and pack-years were matched between silicosis patients and controls, significant age (p<0.001) and sex (p=0.01) differences were observed. While no correlation was found between the most discriminative VOCs and age or gender, residual confounding cannot be entirely excluded, as previous research shows age-specific [37] and gender-related [38] VOC patterns. Additional confounders, such as medication use, cigarette smoke exposure and ambient air contaminants, may influence the VOC pharmacokinetics [39–41]. To mitigate these effects, we adhered to European technical standards for breath sampling [29], excluded low-threshold/background VOCs and focused analyses on group-level differences rather than individual variations. Second, the relatively small number of silicosis patients included in the final VOC analysis may have reduced statistical power; findings should therefore be regarded as hypothesis-generating and require validation in larger, multicentre cohorts. Third, limitations inherent to the observational design must be acknowledged. The absence of HRCT imaging for most RCS-exposed workers limits definitive exclusion of incipient silicosis and misclassification cannot be ruled out. Although HRCT is considered the diagnostic gold standard, its use in occupational surveillance is currently not recommended, as evidence for improving individual outcomes in the absence of clinical indications remains insufficient [42]. Finally, limitations related to analytical technology must be considered. MCC/IMS is limited to tentative VOC identification without definitive chemical characterisation. Compared with reference analytical methods, GC-MS offers unambiguous compound identification but requires substantial time, infrastructure and financial resources, whereas APCI-MS provides high sensitivity with rapid analysis but requires specialised costly instrumentation. In contrast MCC/IMS is rapid, noninvasive and portable, enabling real-time breath analysis at the point of care, features well-suited to occupational health settings. Our results therefore complement GC-MS and APCI-MS data and support MCC/IMS as a pragmatic frontline screening platform that can be paired with targeted MS-based validation when compound-level resolution is required. The clinical context for deploying exhaled breath monitoring as a screening tool for occupational respiratory diseases in RCS-exposed populations requires careful consideration. The model's high sensitivity and NPV support its potential use for ruling out silicosis in periodic surveillance, combined with spirometry, to enable possible early detection of exposure-driven airway reactivity and timely referral. This is particularly relevant for silicosis, given its irreversible and incurable nature. Despite limitations, our findings underscore the potential of “breathomics” using MCC/IMS as a practical approach for early detection of RCS-related preclinical biological changes before radiological disease is routinely detected and for optimising risk management strategies in exposed populations. Conclusion Occupational lung diseases alter exhaled VOC profiles, creating disease-specific “breath prints”. These findings support the concept that targeted breath tests can be developed for diagnostic and surveillance purposes. To determine whether VOC profiling adds value to job-exposure matrices and enhance risk assessment strategies for RCS-exposed workers, rigorous cross-validation studies are essential. These studies should compare candidate biomarkers identified by different research groups using diverse analytical techniques. This approach will strengthen the evidence base for VOC-based monitoring and enable focused investigation into underlying biological pathways modulating these biomarkers. Furthermore, it will contribute to developing standardised analytical protocols, ultimately facilitating integration of “breathomics” into occupational health surveillance and preventive strategies.