Table 2 Literature overview eNose technology in lung disease

Asthma

Dragonieri, 2007 [18]

n = 20 asthma

• n = 10 mild

• n = 10 severe

n = 20 HC

• n = 10 old

• n = 10 young

Diagnostic accuracy

Mild vs young HC

CVV 100%

Severe vs old HC

CVV 90%

Mild vs severe

CVV 65%

Cyranose 320

PCA; CDA

Fens 2009 [19]

n = 20 asthma

n = 30 COPD

n = 20 non-smoking HC

n = 20 smoking HC

Diagnostic accuracy

COPD vs asthma

CVA 96%

COPD vs smoking HC

CVA 66%

Non-smoking vs smoking HC

Not significant

Cyranose 320

PCA

Lazar 2010 [20]

n = 10 asthma

• induction of bronchoconstriction with methacholine or saline

n = 10 controls

Disease course

Bronchoconstriction causes no significant change in breathprint

Cyranose 320

PCA; mixed model analysis

Montuschi 2010 [21]

n = 27 asthma

n = 24 HC

Diagnostic accuracy

eNose only

Acc 87.5%

eNose + FeNO

Acc 95.8%

Tor Vergata

PCA; feed-forward neural network

Fens 2011 [26]

Training: [19]

n = 20 asthma

n = 20 COPD

Validation:

n = 60 asthma

• n = 21 fixed obstruction

• n = 39 classic

n = 40 COPD

Diagnostic accuracy

Validation: Classic asthma vs COPD

Sens 85%

Spec 90%

AUC 0.93 (0.84–1.00)

Acc 83%

Validation: Fixed asthma vs COPD

Sens 91%

Spec 90%

AUC 0.95 (0.87–1.00)

Acc 88%

Validation: Fixed vs classic asthma

No significant difference

Cyranose 320

PCA; CDA

Van der Schee 2013 [22]

n = 25 asthma

n = 20 HC

Diagnostic accuracy

Before OCS

Sens 80.0%

Spec 65.0%

AUC 0.766 ± 0.14

After OCS

Sens 84.0%

Spec 80%

AUC 0.862 ± 0.12

Before OCS (FeNO only)

AUC 0.738 ± 0.15

Cyranose 320

PCA; CDA

n = 18 asthma

• maintenance ICS, stop ICS (4 weeks) and OCS (2 weeks)

Therapeutic effect

OCS responsive vs not

Sens 90.9%

Spec 71.4%

AUC 0.883 (± 0.16)

n = 25 asthma

• maintenance ICS, stop ICS (4 weeks) and OCS (2 weeks)

• n = 13 Loss of control (LOC)

Disease course

LOC vs no LOC

Sens 90.9%

Spec 71.4%

AUC 0.814 ± 0.17

Correlation sputum eos—breathprint

R = 0.601

Plaza 2015 [30]

n = 24 eosinophilic asthma

n = 10 neutrophilic asthma

n = 18 paucigranulocytic asthma

Diagnostic accuracy

Neutro vs pauci

Sens 94%

Spec 80%

AUC 0.88

CVA 89%

EoS vs neutro

Sens 60%

Spec 79%

AUC 0.92

CVA 73%

EoS vs pauci

Sens 55%

Spec 87%

AUC 0.79

CVA 74%

Cyranose 320

PCA; CDA

Brinkman 2017 [32]

n = 22 asthma, induced LOC

• maintenance ICS, stop ICS (8 weeks) and restart ICS

Disease course

Baseline vs LOC

Acc 95%

LOC vs recovery

Acc 86%

Correlation sputum eos—breathprint

Not significant

Cyranose 320

PCA

Bannier 2019 [23]

n = 20 asthma (age > 6 years)

n = 22 HC

Diagnostic accuracy

Sens 74%

Spec 74%

AUC 0.79

Aeonose

ANN

Brinkman 2019 [31]

n = 78 severe asthma

• n = 51 longitudinal follow-up

Clustering

3 clusters (baseline), acc 93%

Differences: chronic OCS use, percent serum eosinophil and neutrophil count

Follow-up (18 months)

n = 21 cluster stable

n = 30 migrated

Cyranose 320, Tor Vergata, Comon Invent

PCA; Ward clustering; Non-hierarchical K-means clustering; PLS-DA; PAM; Topological data analysis

Cavaleiro Rufo 2019 [34]

n = 64 suspected asthma (age 6–18 years)

• n = 45 asthma

• n = 29 persistent

• n = 16 intermittent

• n = 19 no asthma

Diagnostic accuracy

Asthma vs no asthma

Sens 77.8%

Spec 84.2%

AUC 0.81 (0.69–0.93)

Acc 79.7%

Persistent vs no asthma

Sens 79.7%

Spec 68.6%

AUC 0.81 (0.70–0.92)

Acc 79.7%

Intermittent vs no asthma

Not significant

Cyranose 320

PCA; Hierarchical clustering

Dragonieri 2019 [24]

Training:

n = 14 AAR

n = 14 rhinitis

n = 14 HC

Validation:

n = 7 AAR

n = 7 rhinitis

n = 7 HC

Diagnostic accuracy

Training:

AAR vs HC

AUC 0.87 (0.70–0.97)

CVA 75.0%

Validation:

AAR vs HC

AUC 0.77 (0.62–0.93)

CVA 67.4%

Validation:

AAR vs rhinitis

AUC 0.92 (0.84–1.00)

CVA 83.1%

Cyranose 320

PCA; CDA

Abdel-Aziz 2020 [118]

Training:

n = 486 atopic asthma (age > 4 years)

Validation:

n = 169 atopic asthma (age > 4 years)

Diagnostic accuracy

Training:

AUC 0.837–0.990

Sens, spec and acc only visually available

Validation:

AUC 0.18–0.926

Sens, spec and acc only visually available

Cyranose 320, Tor Vergata, Comon Invent, SpiroNose

PLS-DA; adaptive least absolute shrinkage and selection operator; gradient boosting machine

Farraia 2020 [28]

Training:

n = 121 asthma suspected (age > 6 years)

Validation:

n = 78 asthma suspected

(age > 6 years)

Clustering

Training: 3 clusters (hierarchic), differences:

food/drink intake 2 h prior to sampling, percentage of asthma diagnosis in group, PEF%, age < 12 y

Validation: 3 clusters (hierarchic), differences: food/drink intake 2 h prior to sampling

Cyranose 320

Unsupervised hierarchic clustering; Non-hierarchical K-means clustering; PAM

Tenero 2020 [25]

n = 28 asthma (age 6–16 years)

• n = 9 controlled

• n = 7 partially controlled

• n = 12 uncontrolled

n = 10 HC

Diagnostic accuracy

HC + controlled vs. partially + uncontrolled

Sens 79%

Spec 84%

AUC 0.85 (0.72–0.98)

Cyranose 320

Penalized logistic regression

PCA

Chronic obstructive pulmonary disease (COPD)

Fens 2011 [45]

n = 28 GOLD I + II

• airway inflammation (sputum eosinophil cationic protein and myeloperoxidase)

Disease course

Correlation eosinophil cationic protein and breathprint

r = 0.37

Correlation myeloperoxidase and breathprint

Not significant

Airway inflammation vs no

Sens 50–73%

Spec 77–91%

AUC 0.66–0.86

Cyranose 320

PCA

Hattesohl 2011 [37]

n = 23 COPD (pure exhaled breath, PEB)

n = 10 COPD (exhaled breath condensate, EBC)

n = 10 HC (EBC, PEB)

n = 10 AATd (EBC, PEB)

Diagnostic accuracy

COPD vs HC

Sens 100%

Spec 100%

CVV PEB 67.6%

CVV EBC 80.5%

COPD vs AATd

Sens 100%

Spec 100%

CVV PEB 58.3%

CVV EBC 82.0%

HC vs AATd

Sens 100%

Spec 100%

CVV PEB 62.0%

CVV EBC 59.5%

Cyranose 320

LDA

n = 11 AATd COPD (PEB)

• augmentation therapy

Therapeutic effect

Before vs 6 d after therapy

Sens 100%

Spec 100%

CVV 53.3%

Fens 2013 [42]

n = 157 COPD

Clustering

4 clusters (acc 97.4%)

Differences: airflow limitation, health related QoL, sputum production, dyspnoea, smoking history, co-morbidity, radiologic density, gender

Cyranose 320

Hierarchical cluster analysis

Non-hierarchical K-means clustering

Sibila 2014 [41]

n = 10 COPD bacterial colonised

n = 27 COPD non-colonised

n = 13 HC

Diagnostic accuracy

Colonised vs non-colonised

Sens 82%

Spec 96%

AUC 0.922

CVA 89%

HC vs non-colonised

Sens 81%

Spec 86%

AUC 0.937

CVA 83%

HC vs colonised

Sens 80%

Spec 93%

AUC 0.986

CVA 87%

Cyranose 320

PCA; CDA

Cazzola 2015 [38]

n = 27 COPD

• n = 8 AECOPD ≥ 2 per year

• n = 19 AECOPD < 2 per year

n  = 7 HC

Diagnostic accuracy

COPD vs HC

Sens 96%

Spec 71%

CVA 91%

AECOPD ≥ 2 vs < 2 per y

Not significant

Prototype (6 QMB sensors)

PLS-DA

Shafiek 2015 [39]

n = 50 COPD

• n = 17 sputum PPM growth

n = 93 AECOPD

• n = 42 sputum PPM growth

n = 30 HC

Diagnostic accuracy

COPD vs HC

Sens 70–72%

Spec 70–73%

COPD vs AECOPD no PPM

Sens 89%

Spec 48%

(with PPM not significant)

AECOPD PPM vs AECOPD no PPM

Sens 88%

Spec 60%

Cyranose 320

LDA; SLR

n = 61 AECOPD

• during and 2 months after recovery

Disease course

During vs recovery

Sens 74%

Spec 67%

Van Geffen 2016 [46]

n = 43 AECOPD

• n = 18 with viral infection

• n = 22 with bacterial infection

Diagnostic accuracy

With vs without viral infection

Sens 83%

Spec 72%

AUC 0.74

With vs without bacterial infection

Sens 73%

Spec 76%

AUC 0.72

Aeonose

ANN

De Vries 2018 [43]

Training:

n = 321 asthma/COPD

Validation:

n = 114 asthma/COPD

Clustering

5 clusters

Differences: ethnicity, systemic eosinophilia/ neutrophilia, FeNO, BMI, atopy, exacerbation rate

SpiroNose

PCA; Unsupervised Hierarchical clustering

Finamore 2018 [49]

n = 63 COPD

• n = 32 n6MWD worsened 1 year

• n = 31 n6MWD stable or improved 1 year

Disease course

n6MWD change predicted by eNose

Sens 84%

Spec 88%

CVA 86%

n6MWD change predicted by eNose + GOLD

Sens 81%

Spec 78%

CVA 79%

BIONOTE

PLS-DA

Montuschi 2018 [50]

n = 14 COPD

• maintenance ICS, stop ICS (4 weeks) and restart ICS

Therapeutic effect

Maintenance vs restart ICS

Change in 15 of 32 Cyranose sensors; 3 of 8 Tor Vergata sensors

Maintenance vs restart ICS

Spirometry + breathprint prediction model

AUC 0.857

Cyranose 320, Tor Vergata

Multilevel PLS; KNN

Scarlata 2018 [44]

n = 50 COPD

• standard inhalation therapy (12 weeks)

Therapeutic effect

Baseline vs after 12 w

Significant decline in VOCs

BIONOTE

PLS-DA

n = 50 COPD

Clustering

3 clusters

Differences: BODE index, number of comorbidities, MEF75, KCO, pH/pCO2 arterial blood

Unsupervised K-means clustering

Van Velzen 2019 [47]

n = 16 AECOPD

• before, during and after recovery

Disease course

Before vs during

Sens 79%

Spec 71%

CVA 75%

During vs after

Sens 79%

Spec 71%

CVA 75%

Before vs after

Sens 57%

Spec 64%

CVA 61%

Cyranose 320, Tor Vergata, Comon Invent

PCA

Rodríguez-Aguilar 2020 [40]

n = 116 COPD

• n = 88 smoking, n = 28 household air pollution associated

• n = 64 GOLD I-II, n = 52 GOLD III-IV

n = 178 HC

Diagnostic accuracy

COPD vs HC

Sens 100%

Spec 97.8%

AUC 0.989

Acc 97.8% (CDA), 100% (SVM)

Smoking vs air pollution associated

Not significant

GOLD I–II vs GOLD III–IV

Not significant

Cyranose 320

PCA; CDA; SVM

Cystic fibrosis (CF)

Paff 2013 [52]

n = 25 CF

n = 25 primary ciliary dyskinesia (PCD)

n = 23 HC

Diagnostic accuracy

CF vs HC

Sens 84%

Spec 65%

AUC 0.76

CF vs PCD

Sens 84%

Spec 60%

AUC 0.77

Exacerbation CF

Sens 89%

Spec 56%

AUC 0.76

Cyranose 320

PCA

Joensen 2014 [53]

n = 64 CF

• n = 14 pseudomonas infection

n = 21 PCD

n = 21 HC

Diagnostic accuracy

CF vs HC

Sens 50%

Spec 95%

AUC 0.75

CF vs PCD

Not significant

Pseudomonas vs. non-infected CF

Sens 71.4%

Spec 63.3%

AUC 0.69 (0.52–0.86)

Cyranose 320

PCA

De Heer 2016 [54]

n = 9 CF colonised A. fumigatus

n = 18 CF not colonised

Diagnostic accuracy

Sens 78%

Spec 94%

AUC 0.80–0.89

CVA 88.9%

Cyranose 320

PCA; CDA

Bannier 2019 [23]

n = 13 CF (age > 6 years)

n = 22 HC

Diagnostic accuracy

Sens 85%

Spec 77%

AUC 0.87

Aeonose

ANN

Interstitial lung disease (ILD)

Dragonieri 2013 [58]

n = 31 sarcoidosis

• n = 11 untreated

• n = 20 treated

n = 25 HC

Diagnostic accuracy

Untreated vs HC

AUC 0.825

CVA 83.3%

Untreated vs treated

CVA 74.2%

Treated vs HC

Not significant

Cyranose 320

PCA; CDA

Yang 2018 [59]

Training: 80% of

n = 34 pneumo-coniosis

n = 64 HC

Validation: 20% of

n = 34 pneumo-coniosis

n = 64 HC

Diagnostic accuracy

Training:

Sens 64.3–67.9%

Spec 88.0–92.0%

AUC 0.89–0.91

Acc 80.8–82.1%

Validation:

Sens 33.3–66.7%

Spec 71.4–78.6%

AUC 0.61–0.86

Acc 65.0–70.0%

Cyranose 320

LDA; SVM

Krauss 2019 [60]

n = 174 ILD

• n = 51 IPF

• n = 25 CTD-ILD

n = 33 HC

n = 23 COPD

Diagnostic accuracy

IPF vs HC

Sens 88%

Spec 85%

AUC 0.95

CTD-ILD vs HC

Sens 84%

Spec 85%

AUC 0.90

IPF vs CTD-ILD

Sens 86%

Spec 64%

AUC 0.84

Aeonose

ANN

Dragonieri 2020 [61]

n = 32 IPF

n = 36 HC

n = 33 COPD

Diagnostic accuracy

IPF vs HC

AUC 1.00 (1.00–1.00)

CVA 98.5%

IPF vs COPD

AUC 0.85 (0.75–0.95)

CVA 80.0%

IPF vs COPD + HC

AUC 0.84

CVA 96.1%

Cyranose 320

PCA; CDA; LDA

Moor 2020 [57]

Training:

n = 215 ILD

• n = 57 IPF

• n = 158 non-IPF

n = 32 HC

Validation:

n = 107 ILD

• n = 28 IPF

• n = 79 non-IPF

n = 15 HC

Diagnostic accuracy

Training + validation:

ILD vs HC

Sens 100%

Spec 100%

AUC 1.00

Acc 100%

Training:

IPF vs non-IPF ILD

Sens 92%

Spec 88%

AUC 0.91 (0.85–0.96)

Acc 91%

Validation:

IPF vs non-IPF ILD

Sens 95%

Spec 79%

AUC 0.87 (0.77–0.96)

Acc 91%

SpiroNose

PLS-DA

Lung cancer (LC)

Machado 2005 [75]

Training:

n = 14 LC

n = 20 HC

n = 27 other lung disease

Validation:

n = 14 LC

n = 30 HC

n = 32 other lung disease

Diagnostic accuracy

Training: LC vs HC + other

CVA 71.6% (CDA)

Validation: LC vs HC + other

Sens 71.4%

Spec 91.9%

Acc 85% (SVM)

Cyranose 320

SVM

PCA

CDA

Hubers 2014 [71]

Training:

n = 20 LC

n = 31 HC

Validation:

n = 18 LC

n = 8 HC

Diagnostic accuracy

Training:

Sens 80%

Spec 48%

Validation:

Sens 94%

Spec 13%

Cyranose 320

PCA

Schmekel, 2014 [88]

n = 22 LC

• n = 10 survival > 1 year

• n = 12 survival < 1 year

n = 10 HC

Disease course

 < 1 y vs HC

R = 0.95–0.98

 < 1 y vs > 1 y

R = 0.86–0.97

Prediction model survival days

R = 0.99

Applied Sensor AB model 2010

PCA; PLS; ANN

McWilliams 2015 [68]

n = 25 LC

n = 166 smoking HC

Diagnostic accuracy

Sens 84–96%

Spec 63.3–81.3%

AUC 0.84

Cyranose 320

Classification and regression tree; DFA

Gasparri 2016 [76]

Training:

n = 51 LC

n = 54 HC

Validation:

n = 21 LC

n = 20 HC

Diagnostic accuracy

Training + validation:

Sens 81%

Spec 91%

AUC 0.874

Training:

Sens 90%

Spec 100%

Validation:

Sens 81%

Spec 100%

Prototype (8 QMB sensors)

PLS-DA

Rocco 2016 [16]

n = 100 (former) smokers

• n = 23 LC

Diagnostic accuracy

Detection LC

Sens 86%

Spec 95%

AUC 0.87

BIONOTE

PLS-Toolbox; PLS-DA

Van Hooren 2016 [81]

n = 32 LC

n = 52 head-neck SCC

Diagnostic accuracy

Sens 84–96%

Spec 85–88%

AUC 0.88–0.98

Acc 85–93%

Aeonose

ANN

Shlomi 2017 [67]

n = 30 benign nodule

n = 89 LC

• n = 16 early stage LC

• n = 53 EGFR tested (n = 19 mutation)

Diagnostic accuracy

Early stage LC vs benign

Sens 75%

Spec 93.3%

Acc 87.0

EGFR mutation vs wild type

Sens 79.0%

Spec 85.3%

Acc 83.0%

Prototype (40 nanomaterial-sensors)

DFA

Tirzite 2017 [83]

n = 165 LC

n = 79 HC

n = 91 other lung disease

Diagnostic accuracy

LC vs HC + other

Sens 87.3–88.9%

Spec 66.7–71.2%

CVV 72.8%

LC vs HC

Sens 97.8–98.8%

Spec 68.8–81.0%

CVV 69.7%

LC stages

Not significant

Cyranose 320

SVM

Huang 2018 [70]

Training: 80% of

n = 56 LC

n = 188 HC

Validation: 20% of

n = 56 LC

n = 188 HC

External:

n = 12 LC

n = 29 HC

Diagnostic accuracy

Validation:

LC vs HC

Sens 100, 92.3%

Spec 88.6, 92.9%

AUC 0.96, 0.95

Acc 90.2, 92.7%

External validation:

LC vs HC

Sens 75, 83.3%

Spec 96.6, 86.2%

AUC 0.91, 0.90

Acc 85.4, 85.4%

Cyranose 320

LDA; SVM

Van de Goor 2018 [73]

Training:

n = 52 LC

n = 93 HC

Validation:

n = 8 LC n = 14 HC

Diagnostic accuracy

Training:

Sens 83%

Spec 84%

AUC 0.84

Acc 83%

Validation:

Sens 88%

Spec 86%

Acc 86%

Aeonose

ANN

Tirzite 2019 [77]

n = 119 LC smoker

n = 133 LC non-smoker

n = 223 HC + other lung disease

• n = 91 smoking

Diagnostic accuracy

LC non-smoker vs HC + other

Sens 96.2%

Spec 90.6%

LC smoker vs HC + other

Sens 95.8%

Spec 92.3%

Cyranose 320

LRA

Kononov 2020 [78]

n = 65 LC

n = 53 HC

Diagnostic accuracy

Sens 85.0–95.0%

Spec 81.2–100%

CVA 88.9–97.2%

AUC 0.95–0.98

Prototype (6 MOS)

PCA; Logistic regression; KNN; Random forest; LDA; SVM

Krauss 2020 [79]

n = 91 LC active disease

• n = 51 incident LC

n = 29 LC complete response

n = 33 HC

n = 23 COPD

Diagnostic accuracy

LC active vs HC

Sens 84%

Spec 97%

AUC 0.92

Incident LC vs HC

Sens 88%

Spec 79%

AUC 89%

Aeonose

ANN

Lung cancer—(non-)small cell lung cancer ((N)SCLC)

 Dragonieri 2009 [69]

n = 10 NSCLC

n = 10 COPD

n = 10 HC

Diagnostic accuracy

NSCLC vs HC

CVV 90%

NSCLC vs COPD

CVV 85%

Cyranose 320

PCA; CDA

 Kort 2018 [72]

n = 144 NSCLC

n = 18 SCLC

n = 85 HC

n = 61 suspected, LC excluded

Diagnostic accuracy

NSCLC vs HC

Sens 92.2%

Spec 51.2%

AUC 0.85

NSCLC vs HC + LC excluded

Sens 94.4%

Spec 32.9%

AUC 0.76

SCLC vs HC

Sens 90.5%

Spec 51.2%

AUC 0.86

Aeonose

ANN

 De Vries 2019 [87]

Training:

n = 92 NSCLC

• n = 42 response

• n = 50 no response

Validation:

n = 51 NSCLC

• n = 23 response

• n = 28 no response

Therapeutic effect

(anti-PD-1 therapy)

Training:

CVV 82%

AUC 0.89 (0.82–0.96)

Validation:

AUC 0.85 (0.7–0.96)

Sens 43%

Spec 100%

SpiroNose

LDA

 Mohamed 2019 [80]

n = 50 NSCLC

n = 50 HC

Diagnostic accuracy

Sens 92.9%

Spec 90%

Acc 97.7%

PEN3

PCA; ANN

 Kort 2020 [74]

n = 138 NSCLC

n = 143 controls

• n = 59 suspected, LC excluded

• n = 84 HC

Diagnostic accuracy

NSCLC vs controls

(eNose data only)

Sens 94.2%

Spec 44.1%

AUC 0.75

NSCLC vs controls

(multivariate)

Sens 94.2–95.7%

Spec 49.0–59.7%

AUC 0.84–0.86

Aeonose

ANN; Multivariate logistic regression

 Fielding 2020 [82]

n = 20 bronchial SCC

• n = 10 in situ

• n = 10 advanced stage

n = 22 laryngeal SCC

• n = 12 in situ

• n = 10 advanced stage

n = 13 HC

Diagnostic accuracy

BSCC in situ vs HC

Sens 77%

Spec 80%

Misclassification rate 28%

BSCC vs LSCC adv

Sens 100%

Spec 80%

Misclassification rate 10%

Cyranose 320

Bootstrap forest

Lung cancer—Malignant Pleural Mesothelioma (MPM)

 Chapman 2012 [86]

Training:

n = 10 MPM

n = 10 HC

Validation:

n = 10 MPM

n = 32 HC

n = 18 benign ARD

Diagnostic accuracy

MPM vs HC

Training: CVA 95%

Validation: Sens 90%

Spec 91%

MPM vs ARD

Validation: Sens 90%

Spec 83.3%

MPM vs ARD vs HC

Validation: Sens 90%

Spec 88%

Cyranose 320

PCA

 Dragonieri 2012 [85]

n = 13 MPM

• internal validation with training set: n = 8, validation set: n = 5

n = 13 HC

n = 13 AEx

Diagnostic accuracy

MPM vs HC

Sens 92.3%

Spec 69.2%

AUC 0.893

CVA 84.6%

Validation: AUC 0.83

CVA 85.0%

MPM vs AEx

Sens 92.3%

Spec 85.7%

AUC 0.917

CVA 80.8%

Validation: AUC 0.88

CVA 85.9%

MPM vs AEx vs HC

AUC 0.885

CVA 79.5%

Cyranose 320

PCA; CDA

 Lamote 2017 [84]

n = 11 MPM

n = 12 HC

n = 15 AEx

n = 12 benign ARD

Diagnostic accuracy

MPM vs HC

Sens 66.7% (37.7–88.4)

Spec 63.6% (33.7–87.2)

AUC 0.667 (0.434–0.900)

Acc 65.2% (44.5–82.3)

MPM vs benign ARD

Sens 75.0% (45.9–93.2)

Spec 64% (33.7–87.2)

AUC 0.758 (0.548–0.967)

Acc 48.9–85.6% (48.9–85.6)

MPM vs benign ARD + AEx

Sens 81.5% (63.7–92.9)

Spec 54.5% (26.0–81.0)

AUC 0.747 (0.582–0.913)

Acc 73.7% (58.1–85.8)

Cyranose 320

PCA

Pulmonary infections

De Heer 2016 [100]

n = 168 bottles with strain

• n = 135 bacteria + yeast

• n = 30 medium only

• n = 62 mould (A. fumigatus and R. oryzae)

Diagnostic accuracy

(in vitro)

Mould vs other

Sens 91.9%

Spec 95.2%

AUC 0.970 (0.949–0.991)

Acc 92.9%

Cyranose 320

PCA; CDA

Suarez-Cuartin 2018 [101]

n = 73 bronchiectasis

• n = 41 colonised (n = 27 pseudomonas)

• n = 32 non-colonised

Diagnostic accuracy

Colonised vs non-colonised

AUC 0.75

CVA 72.1%

Pseudomonas vs other PPM

AUC 0.96

CVA 89.2%

Pseudomonas vs non-colonised

AUC 0.82

CVA 72.7%

Cyranose 320

PCA

Pulmonary infections—Ventilator-associated pneumonia (VAP)

 Hanson 2005 [104]

n = 19 VAP (clinical pneumonia score, CPIS ≥ 6)

n = 19 controls (CPIS < 6)

Diagnostic accuracy

Correlation CPIS -breathprint

R² = 0.81

Cyranose 320

PLS

 Hockstein 2005 [105]

n = 15 VAP (pneumonia score ≥ 7)

n = 29 HC (ventilated)

Diagnostic accuracy

Acc 66–70%

Cyranose 320

KNN

 Humphreys 2011 [99]

n = 44 VAP suspected

• 98 BAL samples

• Groups: gram-positive, gram-negative, fungi, no growth

n = 6 HC (ventilated)

Diagnostic accuracy

(in vitro)

Differentiation groups (LDA)

Sens 74–95%

Spec 77–100%

Acc 83%

Differentiation groups (cross-validation)

Sens 56–84%

Spec 81–97%

Acc 70%

Prototype (24 MOS)

PCA; LDA

 Schnabel 2015 [106]

n = 72 VAP suspected

• n = 33 BAL + 

• n = 39 BAL−

n = 53 HC (ventilated)

Diagnostic accuracy

BAL + VAP vs HC

Sens 88%

Spec 66%

AUC 0.82 (0.73–0.91)

BAL + vs BAL− VAP

Sens 76%

Spec 56%

AUC 0.69 (0.57–0.81)

DiagNose

Random Forest; PCA

 Chen 2020 [15]

Training: 80% of

n = 33 VAP

n = 26 HC (ventilated)

Validation: 20% of

n = 33 VAP

n = 26 HC (ventilated)

Diagnostic accuracy

Training:

AUC 0.823 (0.70–0.94)

Validation:

Sens 79% (± 8)

Spec 83% (± 0)

AUC 0.833 (0.70–0.94)

Acc 0.81 (± 0.04)

Cyranose 320

KNN; Naive Bayes; decision tree; neural network; SVM; random forest

Pulmonary infections—Tuberculosis (TB)

 Fend 2006 [109]

n = 188 TB

n = 142 TB excluded

Diagnostic accuracy

(in vitro)

Sens 89% (80–97)

Spec 88% (85–97)

Bloodhound BH-114

PSA; DFA; ANN

 Bruins 2013 [107]

Training:

n = 15 TB

n = 15 HC

Validation:

n = 34 TB

n = 114 TB excluded

n = 46 HC

Diagnostic accuracy

Training:

Sens 95.9% (92.9–97.7)

Spec 98.5% (96.2–99.4)

Validation: TB vs HC

Sens 93.5% (91.1–95.4)

Spec 85.3% (82.7–87.5)

Validation: TB vs TB excl

Sens 76.5% (57.98–88.5)

Spec 74.8% (64.5–82.9)

DiagNose

ANN

 Coronel Teixeira 2017 [108]

Training:

n = 23 TB

n = 46 HC

Validation:

n = 47 TB

n = 63 HC + asthma + COPD

Diagnostic accuracy

Training:

Sens 91%

Spec 93%

Validation:

Sens 88%

Spec 92%

Aeonose

Tucker 3–like algorithm; ANN

 Mohamed 2017 [110]

n = 67 TB

n = 56 HC

Diagnostic accuracy

Sens 98.5% (92.1–100)

Spec 100% (93.5–100)

Accuracy 99.2%

PEN3

PCA; ANN

 Saktiawati 2019 [111]

Training:

n = 85 TB

n = 97 HC + TB excluded

Validation:

n = 128 TB

n = 159 TB

excluded

Diagnostic accuracy

Training:

Sens 85% (75–92)

Spec 55% (44–65)

AUC 0.82 (0.72–0.88)

Validation:

Sens 78% (70–85)

Spec 42% (34–50)

AUC 0.72 (0.66–0.78)

Aeonose

ANN

 Zetola 2017 [112]

n = 51 TB

n = 20 HC

Diagnostic accuracy

Sens 94.1% (83.8–98.8)

Spec 90.0% (68.3–98.8)

Prototype (QMB sensors)

PCA; KNN

Pulmonary infections—Aspergillosis

 De Heer 2013 [102]

n = 11 neutropenia

• n = 5 probable/proven aspergillosis

• n = 6 no aspergillus

Diagnostic accuracy

Sens 100% (48–100)

Spec 83.3% (36–100)

AUC 0.933

CVA 90.9% (59–100)

Cyranose 320

PCA; CDA

 De Heer 2016 [54]

n = 9 CF colonised A. fumigatus

n = 18 CF not colonised

Diagnostic accuracy

Sens 78%

Spec 94%

AUC 0.80–0.89

CVA 88.9%

Cyranose 320

PCA; CDA

Pulmonary infections—Corona Virus Disease (COVID-19)

 Wintjens 2020 [114]

n = 219 screened

• n = 57 COVID-19 positive

Diagnostic accuracy

Sens 86% (74–93)

Spec 54% (46–62)

AUC 0.74

CVA 62%

Aeonose

ANN

Obstructive sleep apnoea (OSA)

Greulich 2013 [89]

n = 40 OSA

n = 20 HC

Diagnostic accuracy

OSA vs HC

Sens 93%

Spec 70%

AUC 0.85

Cyranose 320

PCA

N = 40 OSA

• 3 months CPAP ventilation

Therapeutic effect

Before vs after CPAP

Sens 80%

Spec 65%

AUC 0.82

Incalzi 2014 [95]

n = 50 OSA

• 1 night CPAP ventilation

Therapeutic effect

Change in breathprint (visually different, no statistical analysis)

BIONOTE

PCA; PLS-DA

Dragonieri 2015 [90]

n = 19 OSA

n = 14 obese

n = 20 HC

Diagnostic accuracy

Obese OSA vs HC

CVA% 97.4

AUC 1.00

Obese OSA vs obese

CVA% 67.6

AUC 0.77

Obese vs HC

CVA% 94.1

AUC 0.94

Cyranose 320

PCA; CDA; KNN

Kunos 2015 [96]

n = 17 OSA

n = 9 non-OSA sleep disorder

n = 10 HC

• 7AM and 7PM sample

n = 26 HC

–7AM sample

Diagnostic accuracy

OSA 7AM vs 7PM

Significantly different

Non-OSA or HC 7AM vs 7PM

Not significantly different

(Non-)OSA 7AM vs HC 7AM

Significantly different

Acc 77–81%

Cyranose 320

PCA

Dragonieri 2016 [92]

Training:

n = 13 OSA

n = 15 COPD

n = 13 overlap

Validation:

n = 6 OSA

n = 6 COPD

n = 6 overlap

Diagnostic accuracy

Training:

OSA vs overlap

CVA 96.2%

AUC 0.98

Validation:

OSA vs overlap

CVA 91.7%

AUC 1.00

Validation:

OSA vs COPD

CVA 75%

AUC 0.83

Cyranose 320

PCA; CDA

Scarlata 2017 [91]

n = 40 OSA

• n = 20 hypoxic

n = 20 obese

n = 20 COPD

n = 56 HC

Diagnostic accuracy

OSA vs HC

Acc 98–100%

Non-hypoxic vs hypoxic OSA

Acc 60–80%

HC vs COPD

Acc 100%

BIONOTE

PLS-DA

Other—Acute respiratory distress syndrome (ARDS)

Bos 2014 [115]

Training:

n = 40 ARDS

n = 66 HC

Validation:

n = 18 ARDS

n = 26 HC

Diagnostic accuracy

Training:

Sens 95%

Spec 42%

AUC 0.72

Validation:

Sens 89%

Spec 50%

AUC 0.71

Cyranose 320

Sparse-partial least square logistic regression

Other—Lung transplantation (LTx)

Kovacs 2013 [117]

n = 16 LTx recipients

n = 33 HC

Diagnostic accuracy

LTx recipients vs HC

Sens 63%

Spec 75%

AUC 0.825

Cyranose 320

PCA; Linear regression

Therapeutic effect

Correlation breathprint—tacrolimus levels

R = -0.63

Cyranose 320

PCA; Linear regression

Other—Pulmonary embolism (PE)

Fens 2010 [116]

n = 20 PE

• n = 7 comorbidity

n = 20 PE excluded

• n = 13 comorbidity

Diagnostic accuracy

Comorbidity: PE vs excluded

Acc 65%

AUC 0.55

No comorbidity: PE vs excluded

Acc 85%

AUC 0.81

No comorbidity: PE vs excluded (breathprint + Wells)

AUC 0.90

Cyranose 320

PCA