Molecular Classifier Helps Spot Idiopathic Pulmonary Fibrosis

By David Douglas

April 11, 2019

NEW YORK (Reuters Health) - A molecular classifier based on RNA-sequencing data from transbronchial lung biopsies can recognize an expression signature associated with usual interstitial pneumonia (UIP), the defining pathology of idiopathic pulmonary fibrosis (IPF).

"Diagnosis of interstitial lung disease is frequently characterised by delay, misdiagnosis, use of costly and invasive procedures, and high use of health-care resources," Dr. Ganesh Raghu of the University of Washington, in Seattle, and colleagues write in The Lancet Respiratory Medicine, online April 1.

The team notes that the diagnosis of IPF "requires a pattern of usual interstitial pneumonia to be present on high-resolution chest CT (HRCT) or surgical lung biopsy."

In a statement, Dr. Raghu added, "IPF is often challenging to distinguish from other interstitial lung diseases, but timely and accurate diagnosis is critical so that patients with IPF can access therapies that may slow progression of the disease, while avoiding potentially harmful treatments."

To develop a machine-learning algorithm using less-invasive transbronchial-lung-biopsy samples, the researchers recruited 237 patients undergoing evaluation for interstitial lung disease. The patients already had available samples obtained by clinically indicated surgical or transbronchial biopsy or cryobiopsy.

Three to five samples were collected from all patients and extracted for transcriptomic sequencing. The researchers used diagnostic histopathology and RNA sequence data from 90 patients to train a machine-learning algorithm to recognize a usual-interstitial-pneumonia pattern.

The algorithm they developed - the Envisia Genomic Classifier (Veracyte) - was then tested on transbronchial-lung-biopsy samples from 49 patients and compared with diagnoses based on clinical and imaging data and histopathological results.

In 42 patients who had possible or inconsistent usual interstitial pneumonia on HRCT, the classifier showed a positive predictive value of 81% for underlying biopsy-proven usual interstitial pneumonia.

In clinical-utility analysis in 94 patients, the classifier results significantly increased diagnostic confidence in the 18 patients diagnosed as having IPF and in all 48 patients with non-diagnostic pathology or non-classifiable fibrosis histopathology.

Multidisciplinary diagnoses based on clinical and imaging data and either molecular classifier results or histopathology also showed good agreement (86%).

Dr. Raghu said in the statement, "along with clinical information and radiological features in high-resolution CT imaging, physicians through multidisciplinary discussions, may be able to utilize the molecular classification as a diagnostic tool to make a more informed and confident diagnoses."

Dr. Simon Hart, author of an accompanying editorial, told Reuters Health by email that "it is important that the physician can distinguish lung inflammation that is likely to respond to treatment with steroids or similar drugs, from fibrosis (scarring) which needs a different treatment approach. Whilst a surgical lung biopsy provides this important information, surgery carries a risk of complications or death. A chest CT scan can help, in many cases it is not reliable enough and we can only guess about the underlying pathology."

Dr. Hart, of the University of Hull, in Cottingham, U.K., pointed out that in the current study, the team took "small lung biopsies safely through a bronchoscope and used the gene signatures in these samples to train a computer to recognise the common pattern of lung scarring. The computer's diagnosis proved to be reliable and clinically useful in distinguishing lung scarring from other pathology."

"The machine-learning approach," he concluded, "needs to be tested further in a range of similar lung diseases, but it may prove to be particularly useful in those patients where there is uncertainty about the CT scan, or when a surgical lung biopsy is too risky."

Dr. Martin Kolb, director of the Division of Respirology at McMaster University, in Hamilton, Canada, who was not involved in the study, noted in an email to Reuters Health, "This is an interesting approach with novel findings."

The new approach "seems of particular relevance to centers that do not have a highly experienced pathologist available, especially in cases in which the UIP pattern on HRCT falls into the 'probable' and 'indeterminate' category," he said.

The study was funded by Veracyte. Dr. Raghu and other authors have relationships with the company, and seven are employees.


Lancet Respir Med 2019.