Skip to main content
Premium Trial:

Request an Annual Quote

Caris Life Sciences' MI FOLFOXai Predictor Brings Precision to Colorectal Cancer Chemo Decisions

Premium

NEW YORK – To date, most of the predictive tools for precision oncology have been focused on identifying patients likely to respond to targeted therapies. The field has made great strides when it comes to identifying and inhibiting molecular features driving patients' cancer cells.

The same has not been true, however, for identifying patients likely to respond to many chemotherapy agents, which are commonly their first-line options.

Through its newly launched advanced artificial intelligence platform, MI FOLFOXai, the molecular diagnostic company Caris Life Sciences is seeking to do just that — identify patients with colorectal cancer who are likely to benefit from first-line treatment with the FOLFOX chemotherapy regimen, as opposed to other chemotherapy regimens available in this setting, such as FOLFIRI or the combination of both, FOLFIRINOX.

"So far, we haven't seen that similar [predictive] advancement in chemotherapy," said Mohamed Salem of Atrium Health's Levine Cancer Center, an institution that has partnered with Caris Life Sciences through the company's Precision Oncology Alliance. "This [predictive capability] is very important in patients with colorectal cancer, because most of the benefits we are achieving in treating these patients are driven by the chemotherapy backbone."

MI FOLFOXai, which Caris launched officially just over a month ago, uses tumor profiling results from its Molecular Intelligence AI platform to gauge the likelihood that colorectal cancer patients will benefit from FOLFOX as a first-line regimen in combination with the anti-VEGF agent bevacizumab (Genentech's Avastin).

Salem, who treats patients with colorectal cancer, said that, to-date, his decision-making process surrounding which chemotherapy regimen to use in the first-line setting — be it FOLFOX, FOLFIRI, or FOLFIRINOX — has been based primarily on side effect profiles and patient priorities, as opposed to molecularly informed information.

"This [MI FOLFOXai] will be very helpful, not only for the patient, but also for the physician," said Salem. "It will give us a tool to select the right treatment for the right patient."

The overall efficacy of FOLFOX versus FOLFIRI in the first-line setting for patients with advanced colorectal cancer has been established as more or less equivalent, Salem said, and there hasn't been a simple way to assess which patients will respond best to which regimen. While there is a known progression-free survival benefit with the combination of the two, FOLFIRINOX, treating patients with both agents can cause significant toxicities that many can't tolerate.

In terms of choosing one or the other, Salem said that the slightly different side effect profiles have often guided his treatment decisions. For instance, Salem has treated patients who are professional piano players, and chosen accordingly, after conversations with these patients, to prescribe FOLFIRI over FOLFOX, since the oxaliplatin element of the latter tends to damage nerves and cause weakness, numbness, and pain in the hands and feet.

"I cannot give them something that would cause neuropathy, because that would very much kill their career," Salem said, adding that the opposite would be true for a patient with a very public profile for whom the more severe hair loss associated with the irinotecan in FOLFIRI would be detrimental to their careers. For these patients, the decision might be to opt for FOLFOX first.

While conversations about side effect profiles and patient priorities should remain an important part of the treatment decision process even when using Caris' new tool, Salem said that MI FOLFOXai will provide additional predictive information that he can use to develop the right treatment plan for his patients.

Caris Chief Medical Officer Michael Korn acknowledged that when patients present with advanced colorectal cancer most oncologists do already perform molecular analysis as part of standard workup. But this analysis, he said, only assesses "the bare minimum of molecular information" needed for starting first-line treatment.

These initial analyses look for single markers to indicate whether a patient might be a candidate for targeted therapies like EGFR inhibitors. According to Korn, the MI FOLFOXai tool provides physicians and their patients with information regarding these specific markers and many others following full comprehensive molecular profiling — based on a 592-gene DNA sequencing panel — to predict FOLFOX benefit.

"In a way, oncologists are getting everything that they need to know to make their standard decisions on how to treat the patient," for example, KRAS, EGFR, NRAS, HRAS, and PRAS mutation status, "and then they're also getting this extra information," said Caris CSO David Spetzler.

The way MI FOLFOXai works in practice for a patient presenting with metastatic colorectal cancer is that the oncologist would send a tumor sample to Caris' labs, and after genomically profiling that sample and running the data gleaned from the 592-gene sequencing panel through the AI platform, Caris would send back a report within eight to 14 days. The front page of the report would indicate clearly whether the patient is likely to have "increased benefit" with FOLFOX or "decreased benefit" with FOLFOX. The subsequent pages of the report would list details from the molecular analysis, including mutations and informative biomarkers.

According to Spetzler, Caris originally built out the algorithm using 26,035 cases collected throughout its Precision Oncology Alliance over four years. The company then validated the MI FOLFOXai algorithm using two independent datasets. The first dataset consisted of 296 real-world evidence cases, the data for which was acquired through insurance claims records, electronic medical records, and death registries. In this dataset, there was an 11.2-month median overall survival difference between the group of patients predicted to have an increased benefit on FOLFOX by the AI algorithm and the patients predicted to have a decreased benefit on FOLFOX.

The second dataset — which both Spetzler and Korn highlighted as having lent the strongest validity to the tool so far — was a cohort of 149 cases from the Phase III TRIBE 2 study. Spetzler and Korn referred to the TRIBE 2 validation as a "retrospective prospective" analysis — meaning that researchers performed new, prospectively defined analysis on samples collected within a previously conducted study. The prospective aspect hinged on the fact that Caris' researchers were completely blinded to the TRIBE 2 data when they received it. They ran the MI FOLFOXai predictor on the samples and accompanying clinical data, and only after determining the results were they unblinded and able to discern which patients benefitted from upfront FOLFOX treatment.

Through the TRIBE 2 analysis, the AI tool was able to demonstrate a median overall survival difference of six months between those it predicted to benefit and those it predicted would have decreased benefit. Based on the results of the two validation cohorts, Caris claims that MI FOLFOXai has demonstrated approximately 50 percent improvement in overall survival.

"This [validation] led us to launch this test," said Spetzler. "Because the survival [difference] between those patients is really extraordinary."

Salem, who recently sent his first patients' sample for analysis via MI FOLFOXai, said he expects to use the tool consistently going forward. In his view, the TRIBE 2 analysis was "the best level of evidence you can get in terms of retrospective studies."

It's not uncommon to launch diagnostics based on retrospective analysis. Salem noted that the widely used breast cancer recurrence test, Oncotype DX (developed by Genomic Health, which is now part of Exact Sciences) was launched based on retrospective data and then evaluated prospectively over more than a decade in the large-scale TAILORx study.

In Salem's opinion, prospective validation is the gold standard and said this should be done for MI FOLFOXai. However, as with the TAILORx trial, he also pointed out that such validation will take time to complete.

While Korn is of the view that the two retrospective datasets provide sufficient confidence in the AI tool's validation, the company does plan to continue to collect data using a prospective registry established within its Precision Oncology Alliance, which includes leading cancer centers that have partnered with Caris.

"We're running this registry to continue to accumulate data to see how we can improve and refine this," said Spetzler. The launched AI tool "is the very first version of this, and as we add more data we're going to be able to improve the predictions that we're making and better advise physicians on how to treat their patients."

For example, Korn highlighted that one of the analyses fueled by the registry data has shown that patients who received FOLFIRI as first-line treatment rather than FOLFOX, and were predicted to perform poorly on FOLFOX using the AI tool, actually had a significantly higher chance of benefitting from FOLFIRI.

"So, this is an inversion of the prediction, which could be used to predict patients who would perform better on FOLFIRI first," Korn said, acknowledging that more research is needed to validate and quantify the tool's inverse predictive capability.