August 02, 2012
2 min read
Save

Automated detection method identified patients with incident lung nodules

You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact customerservice@slackinc.com.

A combination of ICD-9 codes, CPT codes and a natural language processing algorithm for searching radiology reports accurately and efficiently identified patients with incident lung nodules, according to study findings published in the Journal of Thoracic Oncology.

Although pulmonary nodules are frequently encountered in clinical practice, their management in community settings has not been well studied. Alternatives for the clinical management of patients with screening-detected or incidentally identified lung nodules include imaging tests, nonsurgical biopsy, surgical resection and CT surveillance.

“The inability to readily identify patients with lung nodules in community settings has been a key factor limiting research on how best to manage these patients,” the researchers wrote. “Most prior studies of lung nodules were performed in tertiary care settings, and many enrolled nonrepresentative samples in uncontrolled studies of diagnostic accuracy, screening or surgical outcomes.”

To remove significant obstacles to research and facilitate future studies of comparative clinical management effectiveness, the researchers developed a method to identify patients with incident lung nodules by using electronic administrative records and automatic text processing of diagnostic radiology reports within a large, integrated health system from 2006 to 2010.

The researchers combined five nonmutually exclusive subcohorts with possible lung nodules by using five ICD-9 codes, including 793.1 (nonspecific findings on radiological and other examinations of lung field), 786.6 (swelling, mass or lump in chest), 518.89 (other diseases of lung not classified elsewhere), 519.8 (other diseases of respiratory system not classified elsewhere) and 519.9 (unspecified disease of the respiratory system).

An experienced pulmonologist also reviewed a random sample of 116 radiology transcripts, providing a reference standard for the natural language processing algorithm.

The researchers also acknowledged relevant CT scans with the following CPT codes: 71250 (CT thorax without contrast), 71260 (CT thorax with contrast), 71270 (CT thorax without and with contrast) and 71275 (CT angiography chest).

Exclusion criteria for this study included a possible lung nodule diagnosis before Jan. 1, 2006, and a diagnosis of lung cancer on or before the dates of the relevant diagnosis or CT scan.

Of the 12,389 patients subjected to the automated method, 7,112 were identified as having one or more incident lung nodules, according to study findings. Fourteen percent of these patients were later diagnosed with lung cancer.

When compared with clinician review, the sensitivity of the natural language processing algorithm for identifying the presence of lung nodules was 96%, and the specificity was 86%.

“By using a combination of ICD-9 codes, CPT codes and [natural language processing] of full-text radiology reports, we developed an automated method that seemed to have excellent sensitivity and good specificity for identifying incident lung nodules,” the researchers wrote. “With further refinement and subsequent validation, such a method will help to identify patients with lung nodules in large populations more efficiently, thereby facilitating research that compares alternative practices for lung nodule management in community-based settings.”

Disclosure: One of the researchers involved with the study received a grant from NCI and has a pending grant from the Patient-Centered Outcomes Research Institute.