NEW YORK – A new classifier may be able to guide pancreatic cancer patients toward treatments that are more likely to work against their disease and potentially improve their outcomes.
"It's currently difficult to determine what is the optimal therapy to place patients on upfront," Naim Rashid, a biostatistician at the University of North Carolina at Chapel Hill's Lineberger Comprehensive Cancer Center, said in an interview. In pancreatic cancer, where overall survival is relatively poor compared to other cancers, he believes this classifier may help improve patients' outcomes.
Nearly 57,000 people are diagnosed with pancreatic cancer each year in the US, but according to the National Cancer Institute, only 9.3 percent of pancreatic patients survive five years after their diagnosis. This, Rashid said, underscores the need to improve clinical outcomes for patients by placing them on therapies that are more likely to be effective against their particular disease.
Standard treatments for pancreatic cancer include surgery and radiation or chemotherapy with drugs like FOLFIRINOX — a combination of folinic acid, fluorouracil, irinotecan, and oxaliplatin — or gemcitabine with or without other drugs, depending on the extent of the cancer. But since there is no consensus on how to best group pancreatic cancer patients based on their prognosis, clinicians currently are challenged to figure out which treatments to start them on.
Rashid and his colleagues have recently compared a handful of subtyping schemes to determine which is the most robust and best explains overall survival and treatment response. Through their analysis, which was published recently in Clinical Cancer Research they found that a two-subtype classification could best gauge treatment response and overall survival, and they developed a classification algorithm that could be more easily implemented clinically, an avenue they are currently pursuing.
Rashid and his colleagues focused on three recently reported classifiers for pancreatic cancer: one they previously developed that divides patients' cancers into basal-like or classical tumors, which they called Moffitt; one they call Collisson separates tumors into three groups of quasi-mesenchymal, classical, and exocrine-like tumors; and a third they dubbed Bailey groups cancers into four subtypes of squamous, pancreatic progenitor, immunogenic, and aberrantly differentiated endocrine exocrine tumors.
The researchers applied these three classification schemes to publicly available datasets that include clinical outcomes and gene expression information and overlaid that on data from two clinical trials that include treatment response information.
The Moffitt scheme, they found, could best explain differences in both treatment response and overall survival. Basal-like tumors under this classification scheme exhibited no response to treatment with FOLFIRINOX or to FOLFIRINOX plus the CCR2 inhibitor PF-04136309 in one clinical trial. Similarly, basal-like tumors did not respond to modified FOLFIRINOX or gemcitabine plus nab-paclitaxel in another clinical trial.
Classical tumors under the Moffitt scheme, meanwhile, exhibited a much stronger treatment response in both those trials.
When they grouped the tumors according to the Collisson or Bailey classification approaches, however, there was no difference between subtype and treatment response in the first trial. For the second trial, there were some differences in treatment response by Collisson or Bailey subtypes, but the researchers said they provided no extra benefit over the Moffitt scheme.
The Moffitt subtyping scheme could similarly differentiate patient survival, as basal-like tumors under the Moffitt subgrouping approach had worse prognosis, as compared to classical tumors.
But the way the two Moffitt subtypes are differentiated is based on clustering, an approach Rashid noted is not conducive to determining the subgroup to which a sample from a single patient belongs. "[When] we have one individual patient coming into the clinic, it's very difficult to have to take that patient and try to cluster it with some existing set of patients," Rashid said.
To address this challenge, he and his colleagues developed a single-sample classifier they dubbed Purity Independent Subtyping of Tumors, or PurIST, which relies on the expression of eight gene pairs to classify tumors into either basal-like or classical subtypes. The relative expression of the genes in these pairs is used to generate a score that predicts tumors' probability of being basal-like. This sort of gene-pair approach, Rashid said, makes the model more robust to the type of platform used and less sensitive to technical factors that can bias the data.
This approach is highly accurate, according to Rashid and colleagues. Within the validation datasets, the reported accuracy of PurIST's predictions was about 90 percent or higher, and it was largely highly sensitive and specific.
PurIST can also be applied to various sample types and is a platform-independent approach. Rashid and his colleagues found that its predictions were replicable across the types of samples that are commonly collected in clinical practice, including formalin-fixed paraffin-embedded and fine-needle aspiration samples.
They also tested PurIST using gene expression data generated not only via microarray and RNA-seq, but also using the Nanostring nCounter platform, which is already widely used in determining molecular cancer subtypes.
"It makes [PurIST] more applicable for use by [Clinical Laboratory Improvement Amendments (CLIA)-certified] labs, which increases the clinical applicability for decision-making for treatment," Rashid said.
That's something he and his colleagues are currently pursuing for PurIST. About two or three months ago, they began working on establishing PurIST so that it can be run in a CLIA-certified environment and hope to accomplish this by the end of the year or early next year, he said.
Once that is in place, Rashid said he and his colleagues have lined up several clinical trials into which they want to incorporate PurIST. He added that those trials are going to validate the performance of the PurIST classifier.
Ultimately, he said he hopes that PurIST can help improve patients' clinical outcomes by putting patients on the best therapies for them, based on that subtyping information and the knowledge that patients with basal-like tumors have poorer responses to common first-line therapies, while patients with classical tumors have better responses.
"That would be a perfect example of how you may want to tailor therapy," Rashid said. "The patient shows up to the clinic and we find that they are basal-like subtype. We would want to put them on something else than standard FOLFIRINOX."