Clinical OMICS

MAR-APR 2018

Healthcare magazine for research scientists, labs, pathologists, hospitals, cancer centers, physicians and biopharma companies providing news articles, expert interviews and videos about molecular diagnostics in precision medicine

Issue link: https://clinicalomics.epubxp.com/i/958440

Contents of this Issue

Navigation

Page 18 of 51

www.clinicalomics.com March/April 2018 Clinical OMICs 17 cerSEEK demonstrated a specificity of >99% with only seven out of 812 sam- ples from healthy individuals giving false-positives. According to co-au- thor Anne Marie Lennon, M.D., Ph.D., associate professor of medicine, "Hav- ing a small, robust panel allows us to minimize false positive results, which, in screening, is extremely important, because you really need to have a specificity of 99%. In addition, having a smaller panel allows us to keep the test affordable." The researchers applied the law of diminishing returns to optimize their panel of genetic mutations by identi- fying the minimum number of short DNA sequences, or amplicons, that would allow them to detect at least one driver mutation in each of the eight tumor types. They observed a plateau in detection sensitivity around 60 amplicons, and exceeding this number, the authors claimed, would increase the probability of false posi- tives without a substantial increase in sensitivity. Armed with an optimized 61-amplicon panel, the team used an immunoassay platform to whittle their list of 41 protein biomarker can- didates down to 8 that proved partic- ularly useful for distinguishing cancer patients from healthy controls. However, integrating data from multiple analytes presented a unique challenge. "This is the first time that an algorithm has been built for this specific type of data," said co-author Cristian Tomasetti, Ph.D., associ- ate professor of oncology. Tomasetti led the team of biostatisticians that developed the algorithms supporting CancerSEEK, which uses a logistical regression to classify samples as "pos- itive" or "negative" based on muta- tion frequency and protein levels. If scored positive, a supervised, machine-learning algorithm uses the information on mutation frequency and protein levels to determine the most likely location of the tumor based on tissue-specific patterns, and CancerSEEK successfully localized cancers to a small number of anatomic sites in 83% of patients. Since different tumors often share the same driver mutations, the ability to localize the cancer after detection is one of the major impediments to the successful implementation of a blood-based test that screens for multiple tumor types. Despite these promising results, the study has received several critiques. For instance, in an interview with NPR, Joshua Schiffman, M.D., pediat- ric oncologist at the Huntsman Cancer Institute at the University of Utah, pointed out that the test only identi- fied 43% of Stage I cancers. "If we have a test result that is negative we don't want to give false reassurance to the patient," he remarked. Lennon expressed a different per- spective on these numbers: "If you had a drug that improved cancer sur- vival by 40% people would say that it is brilliant… If we can identify 43% of the patients with Stage I cancer who are asymptomatic—that is a huge thing for those patients." In a true screening setting, how- ever, patients will have less advanced disease than the cohort used for the study, which already had diagnosed (continued on next page) The DETECT study at Geisinger Health system will enroll 10,000 women between the ages of 65 and 75 in the first large scale research to determine CancerSEEK's viability as a clinical diagnostic tool. TEK IMAGE/SCIENCE PHOTO LIBRARY / Getty Images

Articles in this issue

Links on this page

Archives of this issue

view archives of Clinical OMICS - MAR-APR 2018