Daily Respiratory Research Analysis
Three high-impact studies advance respiratory medicine: a multicentre prospective external validation shows exhaled-breath eNose can accurately detect lung cancer; multimodal AI integrating CT and histopathology markedly improves UIP diagnosis and agreement among pathologists; and a large Thai cohort links third-trimester maternal RSV illness to increased preterm birth risk, informing maternal immunization strategies.
Summary
Three high-impact studies advance respiratory medicine: a multicentre prospective external validation shows exhaled-breath eNose can accurately detect lung cancer; multimodal AI integrating CT and histopathology markedly improves UIP diagnosis and agreement among pathologists; and a large Thai cohort links third-trimester maternal RSV illness to increased preterm birth risk, informing maternal immunization strategies.
Research Themes
- Noninvasive diagnostics (breathomics) for lung cancer
- Maternal respiratory virus infection and perinatal outcomes
- AI-driven integration of radiology and pathology for ILD
Selected Articles
1. Lung cancer detection by electronic nose analysis of exhaled breath: a multicentre prospective external validation study.
In 364 adults with suspected lung cancer, the original eNose model achieved an AUC of 0.92 (95% CI 0.85–0.99) in COPD patients and 0.80 (0.75–0.85) overall. At 95% sensitivity, specificity/PPV/NPV were 72%/95%/72% in COPD and 51%/74%/88% overall. A newly trained model yielded AUC 0.83, sensitivity 94%, specificity 63%, PPV 79%, NPV 89% in validation, performing consistently across tumor characteristics and stages.
Impact: Provides rigorous external validation of a noninvasive breathomics test for lung cancer triage with high negative predictive value, potentially reducing invasive procedures and expediting workup.
Clinical Implications: eNose could be adopted as a front-line triage tool in thoracic oncology clinics to prioritize diagnostic pathways, potentially reducing unnecessary biopsies and accelerating diagnosis, especially in COPD patients.
Key Findings
- Original eNose model AUC: 0.92 in COPD and 0.80 overall.
- At 95% sensitivity, specificity/PPV/NPV were 72%/95%/72% (COPD) and 51%/74%/88% (overall).
- New model validation: AUC 0.83, sensitivity 94%, specificity 63%, PPV 79%, NPV 89%.
- Performance was consistent across tumor characteristics, disease stage, centers, and clinical features.
Methodological Strengths
- Prospective multicentre external validation with predefined sensitivity target
- Model performance assessed in clinically relevant subgroups (e.g., COPD) and with a new tailored model
Limitations
- Study conducted in specialized thoracic oncology clinics in two centers, potentially limiting generalizability
- Not yet tested as a population screening tool or in primary care pathways
Future Directions: Prospective impact studies in broader care pathways (primary care to oncology), cost-effectiveness analyses, device standardization, and integration with imaging/risk models.
BACKGROUND: Electronic nose (eNose) analysis of exhaled breath shows potential for accurate and timely lung cancer diagnosis, yet prospective external validation studies are lacking. Our study primarily aimed to prospectively and externally validate a published eNose model for lung cancer detection in chronic obstructive pulmonary disease (COPD) patients and assess its diagnostic performance alongside a new eNose model, specifically tailored to the target population, in a more general outpatient population. PATIENTS AND METHODS: This multicentre prospective external validation study included adults with clinical and/or radiological suspicion of lung cancer who were recruited from thoracic oncology outpatient clinics of two sites in the Netherlands. Breath profiles were collected using a cloud-connected eNose (SpiroNose®). The diagnostic performance of the original and new eNose models was assessed in various population subsets based on receiver operating characteristic-area under the curve (ROC-AUC), specificity, positive predictive value (PPV), and negative predictive value (NPV), targeting 95% sensitivity. For the new eNose model, a training cohort and a validation cohort were used.
2. Antenatal Respiratory Syncytial Virus and Human Metapneumovirus Illness Rates Among Pregnant Women in Thailand and the Association Between Antenatal Respiratory Syncytial Virus and Perinatal Outcomes: A Prospective Cohort Study.
Among 2764 pregnant women (median enrollment GA 10 weeks), RSV and hMPV illness incidences were 57 and 23 per 10,000 pregnant woman-months, respectively; 42% and 34% sought medical care. Third-trimester antenatal RSV illness was associated with increased preterm birth risk (adjusted HR 2.50; 95% CI 1.04–6.00), but not SGA (aHR 0.79; 95% CI 0.29–2.16).
Impact: Quantifies antenatal RSV/hMPV burden and links third-trimester RSV illness to preterm birth, directly informing maternal RSV immunization policy and obstetric risk counseling.
Clinical Implications: Supports prioritizing maternal RSV vaccination and heightened surveillance in late pregnancy; obstetricians should consider recent RSV illness when assessing preterm birth risk.
Key Findings
- Incidence: RSV 57 and hMPV 23 per 10,000 pregnant woman-months.
- Medical care sought by 42% (RSV) and 34% (hMPV) illnesses.
- Third-trimester RSV illness associated with preterm birth (aHR 2.50; 95% CI 1.04–6.00).
- No association between antenatal RSV illness and SGA (aHR 0.79; 95% CI 0.29–2.16).
Methodological Strengths
- Prospective active surveillance with twice-weekly contacts and RT-PCR confirmation
- Time-to-event analysis with trimester-specific exposure assessment and adjusted Cox models
Limitations
- Single-country cohort may limit global generalizability
- Illness capture based on symptom-triggered sampling could miss asymptomatic infections
Future Directions: Evaluate maternal RSV vaccine effectiveness against preterm birth endpoints, assess mechanisms linking maternal infection to parturition timing, and extend surveillance across diverse settings.
BACKGROUND: We estimated respiratory syncytial virus (RSV) and human metapneumovirus (hMPV) illness incidences among pregnant women and examined the association between antenatal RSV illness and preterm birth and small-for-gestational-age (SGA) infant. METHODS: Pregnant women aged ≥18 years were contacted twice weekly until the end of pregnancy to identify illness episodes with ≥1 of the following: myalgia, cough, runny nose/nasal congestion, sore throat, or difficulty breathing. Midturbinate nasal swabs were collected and tested for RSV and hMPV by real-time reverse transcription polymerase chain reaction. Incidences were calculated. Cox proportional hazards regression was used to estimate hazard ratios (HRs) comparing participants with and without RSV illnesses for preterm birth (live birth before 37 weeks' gestation) and SGA infant.
3. Enhancing Interstitial Lung Disease Diagnoses Through Multimodal AI Integration of Histopathological and CT Image Data.
The multimodal AI achieved an AUC of 0.92 for distinguishing UIP from non-UIP. Among general pathologists, diagnostic agreement improved markedly (κ 0.737 post-model vs 0.273 pre-model), and consensus with expert pulmonary pathologists increased (κ from 0.278–0.53 to 0.474–0.602). The model also increased diagnostic confidence.
Impact: Demonstrates that integrating radiology and histopathology via AI can elevate UIP diagnostic accuracy and interobserver agreement, addressing a critical bottleneck in ILD multidisciplinary diagnosis.
Clinical Implications: AI-assisted multimodal review may standardize UIP diagnosis, reduce variability in multidisciplinary discussions, and guide more consistent treatment decisions (e.g., antifibrotics, transplant referral).
Key Findings
- Multimodal AI achieved AUC 0.92 for UIP vs non-UIP.
- General pathologist agreement improved (κ 0.737 post-model vs 0.273 pre-model).
- Consensus with expert pulmonary pathologists increased (κ 0.474–0.602 vs 0.278–0.53).
- AI increased diagnostic confidence among general pathologists.
Methodological Strengths
- Integration of CT features (28 radiologic traits) with prior histopathology model in a unified framework
- Direct comparison against both expert and general pathologists with kappa statistics
Limitations
- Single-test cohort (n=114) limits external generalizability; external validation across institutions is needed
- Potential spectrum bias given retrospective image selection
Future Directions: Prospective multicentre validation, integration into multidisciplinary team workflows, and assessment of clinical impact on treatment decisions and outcomes.
BACKGROUND AND OBJECTIVE: The diagnosis of interstitial lung diseases (ILDs) often relies on the integration of various clinical, radiological, and histopathological data. Achieving high diagnostic accuracy in ILDs, particularly for distinguishing usual interstitial pneumonia (UIP), is challenging and requires a multidisciplinary approach. Therefore, this study aimed to develop a multimodal artificial intelligence (AI) algorithm that combines computed tomography (CT) and histopathological images to improve the accuracy and consistency of UIP diagnosis. METHODS: A dataset of CT and pathological images from 324 patients with ILD between 2009 and 2021 was collected. The CT component of the model was trained to identify 28 different radiological features. The pathological counterpart was developed in our previous study. A total of 114 samples were selected and used for testing the multimodal AI model. The performance of the multimodal AI was assessed through comparisons with expert pathologists and general pathologists.