SAN DIEGO — Researchers used analysis of saliva, a cross-section of data from a patient questionnaire and an artificial intelligence analysis tool to develop a tool to predict risk in esophageal cancer, according to findings presented at Digestive Disease Week 2016.
David G. Graham, MD, of the University College London, and colleagues, aimed to develop an accurate risk prediction tool for assessing esophageal cancer risk. They used salivary transcriptomic data along with factors related to patient demographics, symptoms and risk factors. “A novel, non-invasive tool is required,” he said.
He noted that extracellular RNAs in body fluids are emerging as potential biomarkers for detecting this disease.
The study included 80 patients from whom RNA was extracted from the saliva. Twenty patients of this cohort were healthy controls, while 20 had non-dysplastic Barrett’s esophagus, 20 had high-grade dysplasia and 20 had esophageal cancer.
The researchers collected saliva samples and 204 data points from the questionnaire. The researchers conducted transcriptomic screening using 24 amplicons with known links to the development esophageal cancer from inflammatory or oncogenic pathways, according to Graham. They used quantitative real-time PCR with 18s rRNA as the housekeeping gene to conduct targeted RNA expression analysis. A novel Infogain Measure with standard feature selection artificial intelligence techniques was used to analyze transcriptomic, demographic, symptom and risk factor data to create the predictive tool.
“The idea was to collect huge amounts of data,” Graham said.
Ten aberrant expressions of the 24 amplicons of the metaplasia-dysplasia-carcinoma sequence were found in analysis of RNA in the saliva. Those included TP53, CDKN2a, SMAD7, EGFR, AMY2, TLR6, KIT9, KRAS, PTEN and PIK3CA (P < .05), according to Graham.
“Once we had this expression data, we did statistical testing between the groups,” he said. “We saw increasing expression in 10 amplicons, the most striking of which was CDKN2a.”
The researchers observed 11 key attributes that may be useful in developing the predictive tool. These included six of the amplicons mentioned above and five features from the questionnaire data, including waist/hip ratio, proton pump inhibitor use and previous cancer.
“We have shown that the questionnaire data provides some ability to identify those at risk,” Graham said. “Saliva provides potentially key biomarkers for prevention and early cancer detection.”
Another key element to the study is accurate predication using artificial intelligence analysis, Graham added. “These are very early data, but we think they are quite exciting,” he said.
Graham D, et al. Abstract #305. Presented at: Digestive Disease Week; May 21-24, 2016; San Diego.
Disclosures: Graham reports no relevant financial disclosures.
Editor's Note: This item has been updated to reflect Graham's correct institution.