Flatiron, Foundation Med Link Records of 20,000 Cancer Patients

[Updated, 11/3/16, 11:43 a.m. See below.] Some of the world’s largest non-profit cancer treatment centers are pooling patient information to gain new insights into the disease and potentially find new treatments. Now two for-profit companies that trade in cancer data are trying the same thing.

Flatiron Health and Foundation Medicine have begun combining the health information of patients who are in both of their databases—about 20,000 people so far, the firms announced Thursday. Each company has what the other one lacks. Flatiron, based in New York, is in the medical records business and says it has signed up more than 250 cancer clinics around the country. It is privately held, has 350 employees, and has raised more than $300 million in venture capital.

Cambridge, MA-based Foundation (NASDAQ: [[ticker:FMI]]) has a tumor sequencing business. Oncologists send patients’ tumor samples to Foundation to decode their genetic signatures, which can help doctors make treatment decisions. The company has grown aggressively, with 417 employees as of January, and sustained large losses since it was founded in 2009.

Both firms also sell the data they gather, with patient identities stripped out, to drug companies and others who use it in the search for new products or when planning clinical trials. Those pharma ties are crucial for Foundation. It receives little revenue from its tests, because insurers have been slow to cover them. Only $8.7 million, or 30 percent, of its $29.4 million in revenue for the third quarter came from its testing business. The other $20.7 million came from pharma customers—and $13 million of that from its majority owner Roche, which bought a controlling stake in 2015. (Foundation executives declined to comment, citing “quiet period” regulations.)

Flatiron officials by press time did not disclose revenues or how much comes from selling data to pharmaceutical companies. The firm has built a medical records system that it says can make sense of important cancer patient data in documents even if those data aren’t neatly divided into boxes. In everyday practice, doctors will often dictate their daily notes into an electronic form. But there’s no guarantee the notes will provide easy understanding of a patient’s condition or treatment. It’s even hard to tell sometimes when a patient has died.

“I’ve been treating breast cancer for 20 years, and looking through the records of my colleagues, sometimes it’s still not exactly clear what transpired,” says Philippe Bedard, who practices at the Princess Margaret Cancer Centre in Toronto, Canada’s largest. (He is not involved with Flatiron.)

Bedard is also Princess Margaret’s point person on a data-sharing project, dubbed GENIE, which links the center with six other major centers, including Memorial Sloan Kettering Cancer Center in New York and Dana-Farber Cancer Institute in Boston.

GENIE launched a year ago and has so far pooled the genetic data from about 20,000 patients’ tumors. It is in some ways similar to the Flatiron-Foundation project. But GENIE is not trying to extract health-record information and combine it with the tumor data. Instead, if researchers tapping into the GENIE database find something interesting in the genomic pool—say, 300 patients who share a rare mutation—they have to ask the participating centers to do a manual search through those patients’ medical records to answer specific questions.

“What you might consider a clunky manual method is also a very efficient way to ask and answer questions, rather than extracting everything from medical records and analyzing it later,” says Charles Sawyers, a leading cancer researcher at Memorial Sloan Kettering who is leading GENIE. “I’d love a medical-record system to reveal the answers by writing a little code and searching, but that’s not the reality today.”

[Updated to clarify the description of Flatiron.] Flatiron uses both artificial intelligence and human curators to create its database. Sawyers says the machine-learning part—natural language processing—is a “Google-like” strategy. Both Foundation and Flatiron have in fact received venture funding from Google. If the “big data approach works,” says Sawyers, “it would be disruptive.” He thinks it eventually will, but says “the jury is still out.” (IBM is also using its Watson system to read and analyze doctor’s notes, medical publications, and more.)

By the end of the year, GENIE will release publicly the pooled, anonymized genomic data from its first 20,000 patients, and will release new data every three months, Sawyers says. The consortium aims to discuss results from its first two projects next spring.

Flatiron chief medical officer Amy Abernethy gives one example that shows both the promise and the difficulty of turning