Individualized Pancreatic Cancer Treatment: New Technology, Old Questions
Three articles in the current Journal of Clinical Oncology describe sophisticated methods designed to match pancreatic cancer patients to available drug combinations.
The first applied an AI-based approach in 377 pancreatic cancer patients to define the microscopic features of each patient’s tumor in paraffin-fixed blocks to select between 5-FU and Gemcitabine-based therapies. The results show a small improvement in time to disease progression at 8.6 months versus 7.5 months in one group and 9.6 versus 7.2 in the second group. While interesting as a proof of concept, changing outcomes in this almost uniformly lethal disease by between 1.1 and 2.4 months seems underwhelming when one examines the intensive, multiplexed, algorithm-driven, computational muscle required for even these modest benefits.
The second study also applied a microscopic tissue-based AI platform to compare 5-FU versus Gemcitabine based therapy in the in the post-operative (adjuvant) setting using results from the PRODIGE-24 clinical trial. Again, this computer driven refinement of histological appearance (borrowing from over 2 centuries of pathologist’s expertise) showed improved discrimination between 5FU and Gemcitabine based therapies.
The third study used a very different platform. In this analysis a DNA-based Gene profile was used to discriminate outcomes for FOLFIRINOX versus Gemcitabine, again using the Prodige-24 clinical trial patient outcomes in pancreatic cancer. Results did distinguish the “high-risk” basal-like tumors from the standard-risk groups, but this is already easily accomplished by any trained, qualified pathologist.
The accompanying editorial by Gabriel Brooks, MD, MPH and Raghav Sundar, MD, PhD is informative.
There are several interesting takeaways from these studies.
The first and most obvious is that phenotype (measured by microscopic appearance) trumps genotype (measured by DNA), as it is the biological features of each tumor (examined by these sophisticated microscopic techniques) and not the tumor’s genetic makeup (genomics) that predict patient outcomes.
The second and most important is that all cancer physicians, despite their protestations to the contrary, actually want to use tissue-driven patient drug selection.
Chemosenstivity testing remains the holy grail of oncology, it’s just that one needs to use their chemosensitivity test to play.
While the authors are to be commended for making the effort, the editorial points out several caveats. As they note, the AI system is a bit of a “black box”, opaque to those who might try to reproduce the results. Furthermore, physician selection of pancreatic cancer patients for these treatments one over the other is known to be influenced by age, infirmity, kidney function and other factors that can confound interpretation of the results. They conclude that the findings are hypothesis generating and not definitive.
More important, we find that the scientific community is now applying highly sophisticated AI and molecular biology tools to answer comparatively basic questions i.e. 5-Flourouracil versus Gemcitabine. Questions that our group successfully answered decades ago.
Critical questions remain: Do these investigators have the ability to examine the newest classes of drugs, such as the recently FDA-approved KRAS inhibitors or do they need to start from scratch with a new AI-model for each choice? More important, can they examine drug interactions like drug synergy (more than additive effects i.e. 1 + 1 = 3) or drug antagonism (where drugs work against each other, i.e. 1 + 1 = 0)? I think not.
The most concerning question is: How quickly can these platforms realistically develop actionable databases for new drugs and combinations? Will it be years or decades before patients in need can access these predictive models for drugs like Daraxorasib (KRAS inhibitor) and the newer more effective drug combinations.
Our platform (EVA/PCD) has already been applied in hundreds of pancreatic cancer patients and shown to provide everything that these investigators so sorely lack: 1) Rapid turnaround (5-7 days) 2) Broad applicability to new (i.e. KRAS inhibitors, etc.) as well as established drugs and combinations (i.e. FOLFIRINOX vs Nab-Paclitaxel plus Gemcitabine) 3) Synergy analysis and 4) Published response and survival outcomes.
We are in an era in cancer therapy, not unlike the 1980’s when we confronted the AIDS epidemic. Our poor understanding of the HIV virus and lack of effective therapies led to the discovery of AZT.
Samuel Broder, MD, PhD, the director of the National Cancer Institute from 1989 to 1995 promoted the use of AZT as the first effective therapy for HIV. While the scientific community actively pursued new drugs, Dr Broder explained to his detractors “The search for a perfect AIDS drug should not prevent the use of a good AIDS drug”.
My take: “The search for a perfect predictor of response to cancer therapy should not prevent the use of a good predictor of response chemotherapy response.” Patients can no longer wait for the academic community to catch up. They deserve the best treatments available for their cancers right now.
