Sorry Fido, The eNose Can Smell ILD Subtypes Too

SAN FRANCISCO — Dogs trained to sniff out lung disease may need to look for another line of work if an “electronic nose” sensor in development makes it into widespread clinical practice.

In a multicenter cohort study of 589 patients with a diagnosis of interstitial lung disease (ILD) by multidisciplinary team discussion and evidence of pulmonary fibrosis on high-resolution CT, the electronic nose technology, or “eNose,” distinguished between different ILD subtypes with a high degree of accuracy, reported Bart Formsma, MD, PhD candidate at Erasmus Medical Center in Rotterdam, the Netherlands.

“This external validation study demonstrates that various ILDs can be distinguished with high accuracy using an eNose, and therefore, eNose technology holds potential as an easy point-of-care tool in the diagnosis of ILD, potentially reducing diagnostic delay,” he said in an oral abstract session at the American Thoracic Society (ATS) 2src25 International Conference here.

Tough Nut to Crack

Under the best conditions, ILD is still difficult to diagnose, and diagnostic delay is common, Formsma said. The diagnosis is based on a host of factors, including biopsy results, imaging studies, lung function tests, and bronchoalveolar lavage, and a multidisciplinary team is often required to reach consensus on the ILD type.

The eNose mimics olfactory receptor function of the human nose and odor detection of the human brain with a sensor array, but the old schnozz goes one better with software and machine learning capabilities of pattern recognition that can pinpoint abnormalities in volatile organic compounds in exhaled breath.

To validate the technology for potential use in the pulmonary clinic, Formsma and colleagues conducted a prospective longitudinal study in five ILD centers in the Netherlands, the United Kingdom, Germany, Australia, and France.

The investigators analyzed the ability of the eNose to distinguish between ILD subtypes and to identify individual ILD types.

A total of 589 patients were included in the training and validation data sets. Approximately one third of the patients (35%) were women. The median patient age was 7src years, the mean forced vital capacity (FVC) was 8src% of the predicted value, and the mean diffusion capacity of carbon monoxide (DLco) was 52% of the predicted value.

In both the training and validation sets the technology distinguished between idiopathic pulmonary fibrosis with high degrees of accuracy. The area under the curve of receiver operating characteristics (AUC) was src.91 (95% CI, src.88-src.95) in the training set and src.89 (95% CI, src.85-src.94) in the validation set.

Similarly, the eNose sniffed out differences between connective tissue disease–associated ILD and other ILD types with an AUC in the training and validation sets, respectively, of src.89 (95% CI, src.84-src.94) and src.91 (95% CI, src.85-src.98).

The technology was also extremely good at discriminating unclassifiable ILD, with AUCs of src.92 (95% CI, src.86-src.98) and src.95 (95% CI, src.88-1.srcsrc), respectively.

The eNose was less adept, however, at distinguishing fibrotic hypersensitivity pneumonitis from other ILD types with a training set AUC of src.88 (95% CI, src.81-src.94) but validation set AUC of src.75 (95% CI, src.61-src.88).

No Apparent Confounders

In the question and answer, session co-moderator Joyce Lee, MD, from the University of Colorado Anschutz Medical Campus, Aurora, Colorado, asked whether diet or other factors could affect the breath samples collected by the eNose.

“We did a subgroup analysis looking into smoking status, gender, and autoimmune drug or antifibrotic drug use, and they had no big influence on outcomes,” Formsma said.

Sergio Harari, MD, from San Giueseppe Hospital in Milan, Italy, asked whether there was good correlation between the sensitivity of the eNose analysis and the severity of disease.

Study co-author Catharina Moor, MD, also from Erasmus Medical Center, Rotterdam, the Netherlands, responded, noting that the investigators looked at both FVC and DLco, but could not find any indications of disease severity. She added that in previous analyses the group had evaluated the findings in patients with both severe and less severe disease and also found no significant differences, suggesting that the technology might be useful as a screening tool for early ILD, although that potential application has yet to be studied.

The study was funded by an unrestricted grant from Boehringer Ingelheim. All persons cited