Using machine-learning techniques on data from a large gastric cancer trial, researchers have identified a potential biomarker of survival benefit from treatment with paclitaxel. "Applying machine-learning/artificial intelligence tools on complex cancer genomic datasets may identify clinically-relevant molecular patterns undetectable by direct human interpretation," Drs. Patrick Tan and Raghav Sundar of Duke-NUS Medical School in Singapore told Reuters Health by email. "These patterns may constitute novel biomarkers or potential therapeutic targets." As reported in Gut, using machine learning modeling, the authors analyzed 499 samples from the Stomach cancer Adjuvant Multi-Institutional group Trial (SAMIT), a phase 3 study in which gastric cancer patients were randomized to Pac-S-1 - i.e., paclitaxel +S-1 (an oral combination of tegafur, gimeracil and oteracil); Pac-UFT (paclitaxel +UFT, an oral combination of uracil and tegafur); S-1 alone; or UFT alone after curative surgery. An independent cohort of patients with metastatic gastric cancer treated with paclitaxel and ramucirumab (Pac-Ram) served as an external validation cohort. The analysis of the Pac-S-1 training cohort yielded a 19-gene signature that was different in those who had delayed progression after treatment with paclitaxel (Pac-sensitive) and those who did not (Pac-resistant). The team tested the signature on 375 samples from the The Cancer Genome Atlas stomach adenocarcinoma cohort, and 76% were classified as Pac-sensitive. In the internal Pac-UFT validation cohort, Pac-sensitive patients showed a significant improvement in three-year disease-free survival: 66% versus 40% (HR, 0.44). No between-group survival difference was seen between Pac-sensitive and Pac-resistant patients in the UFT or S-1 alone arms. In the external Pac-Ram validation cohort, the signature predicted benefit for Pac-sensitive patients: median progression-free survival, 147 days versus 112 days (HR, 0.48). Drs. Tan and Sundar said, "Further validation work is required to improve the sensitivity and specificity of the genomic signature biomarker, as well as prospective validation in a clinical trial." Currently, there is no time frame for when the test might be available for the clinic. Dr. Paul Oberstein, Director, Gastrointestinal Medical Oncology Program, at NYU Langone's Perlmutter Cancer Center in New York City, commented on the study in an email to Reuters Health. "This is promising, as it may help determine who should receive paclitaxel therapy and who may be able to avoid therapy when predicted to be unhelpful." "However, there are limitations," he said. "The main one is that the tool does not identify if a different chemotherapy will be more effective, so it is difficult for this to guide practice." "To further validate this finding, it would be useful to conduct a prospective trial where one could assess whether using this tool to guide therapy led to better outcomes," he said. "It is important to be aware that the authors looked at this tool in another group of samples (the TCGA database) and found that 76% were predicted to be sensitive to paclitaxel, so therapy still seems to be potentially beneficial for most patients." "This study is a step on the path to predicting response to therapy, but requires refinement before it will change practice," Dr. Oberstein concluded. —Marilynn Larkin Source
