Deep natural language processing may be able to estimate the presence of active cancer, cancer worsening or progression, and cancer improvement or response from radiologic reports, according to a study published online in JAMA Oncology.

Kenneth L. Kehl, M.D., from the Dana-Farber Cancer Institute in Boston, and colleagues evaluated whether deep natural language processing can extract relevant cancer outcomes from radiologic reports in electronic health records. A retrospective cohort of 1,112 patients with a lung cancer diagnosis who underwent tumor genotyping were divided into the curation set, with an additional 109 in the test subset.

The researchers found that in the test subset, deep learning models identified the presence of cancer, improvement/response, and worsening/progression with accurate discrimination (area under the curve >0.90). Machine and human analysis resulted in similar measurements of disease-free survival (hazard ratio [HR], 1.18; 95 percent confidence interval [CI], 0.71 to 1.95), progression-free survival (HR, 1.11; 95 percent CI, 0.71 to 1.71), and time to improvement/response (HR, 1.03; 95 percent CI, 0.65 to 1.64). Using a reserve set of 1,294 patients (15,000 additional radiologic reports), algorithm-detected cancer worsening/progression was associated with decreased overall survival (HR for mortality, 4.04; 95 percent CI, 2.78 to 5.85) and improvement/response was associated with increased overall survival (HR, 0.41; 95 percent CI, 0.22 to 0.77).

"Deep natural language processing appears to speed curation of relevant cancer outcomes and facilitate rapid learning from electronic health record data," the authors write.

Several authors disclosed financial ties to medical systems and the pharmaceutical industry.

-----------------------------------------------------------------------
Editor's Note:

If you have any suggestion to the content, 

please email: Julia.Zhang@ubmsinoexpo.com