AT ACG 2016

LAS VEGAS (FRONTLINE MEDICAL NEWS) – A new scoring system for military veterans assigns colorectal cancer risk based on a list of lifestyle and clinical factors, and could help triage patients for colonoscopy or other follow-up. The study used natural-language processing to analyze patient medical records from the VA health care system.

“Those who were are at very low risk had no colon cancer, and their risk of advanced polyps was pretty low, while those in the high-risk group should probably give thought to having a colonoscopy,” Thomas F. Imperiale, MD , a gastroenterologist and professor of medicine at the Indiana University School of Medicine, Indianapolis, said in an interview. Dr. Imperiale presented the results at the annual meeting of the American College of Gastroenterology.

If the results are confirmed, Dr. Imperiale suggested that veterans in the highest-risk group could be prioritized for colonoscopy, while those in the low-risk group could consider a fecal immunochemical test.

When asked during the Q&A session if he could identify a subset of patients who could skip colonoscopies altogether, Dr. Imperiale said that he could not. “Prediction is possible, and if those predictions stand up in validation, they may be used to at least provide veterans with a choice, and perhaps provide the health care system a choice about how they screen their patients,” he said.

The researchers identified 66,725 veterans who underwent a diagnostic or screening colonoscopy between 2002 and 2009 at one of 14 VA medical centers. They used natural language processing to identify the most advanced finding and the location within the colorectum.

The rate of advanced neoplasia was 8.8%, and the rate of colorectal cancer (CRC) was 1.2%. Independent risk factors included age, sex, tobacco use, and exposure to COX-1 and COX-2 nonsteroidal anti-inflammatory drugs. The researchers defined four risk categories, low to high, with advanced neoplasia risks of 2.9% (CRC risk, 0%), 5.3% (CRC risk, 0.5%), 8.5% (CRC risk, 0.9%), and 11.4% (CRC risk, 2.1%). For advanced neoplasia, the goodness-of-fit P value was 1.00 and the c statistic was 0.58. For CRC, these values were 1.00 and 0.64, respectively.

Dr. Imperiale noted that the results remain relatively crude, and that the field of natural-language processing continues to evolve. In fact, software available today outperforms that used in the study. “It’s a field that’s moving quickly,” said Dr. Imperiale.

The results could help prioritize patients for colonoscopies, according to Victor Levy, MD, a senior attending physician at Mary Imogene Bassett Healthcare, Cooperstown, N.Y., who attended the presentation. And the study represents good news for the field overall. “Big data analytics has been accepted in other industries, and it’s now finally beginning to exert its effects in health care,” Dr. Levy said in an interview.

The results don’t generalize to other groups, because veterans have unique covariables and confounding factors. A new model would have to be developed for other populations.

Dr. Imperiale and Dr. Levy have no disclosures.