HealthDay News — A deep learning model based on chest radiograph (CXR) images and data from the electronic medical record (EMR) has better discrimination for smokers at high risk for incident lung cancer than Centers for Medicare & Medicaid Services (CMS) eligibility, according to a study published online in the Annals of Internal Medicine.

Michael T. Lu, MD, MPH, from Harvard Medical School in Boston, and colleagues developed a convolutional neural network (CXR-LC) that predicts long-term incident lung cancer using data commonly available in the EMR. The CXR-LC model was developed in the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial and was validated in additional PLCO smokers and National Lung Screening Trial heavy smokers.

The researchers found that for incident lung cancer, the CXR-LC model had better discrimination than CMS eligibility (area under the receiver operating characteristic curve [AUC], 0.755 vs 0.634). The CXR-LC model performance was similar to that of the PLCO Model 2012 risk score with 11 inputs in the PLCO dataset (AUC, 0.755 vs 0.761) and the NLST data set (AUC, 0.659 vs 0.650). The CXR-LC was more sensitive than CMS eligibility in the PLCO dataset when compared in equal-sized screening populations (74.9% vs 63.8%); 30.7% fewer incident lung cancers were missed with the CXR-LC.

“The use of patient EHR [electronic health record] data to assess disease risk will likely continue to grow in the near future,” write the authors of an accompanying editorial. “Along with the potential benefit, there are concerns among patients, physicians, and health care organizations about how to responsibly manage this use. This will be an important field of research going forward.”


Continue Reading

Abstract/Full Text (subscription or payment may be required)

Editorial (subscription or payment may be required)