Deal struck to mine cancer patient database for new treatment insights

The plan took shape over a decade: A prominent society of cancer doctors would create a massive patient database and use it to discover the most effective treatments for specific cancers.

The first part of the effort worked — the American Society of Clinical Oncology secured data from more than 100 oncology practices nationwide. But it has since struggled to reconcile myriad conflicts in language and formatting, leaving crucial insights buried in the records.

On Thursday, the 40,000-member organization unveiled an unusual arrangement to solve that problem by partnering with two private companies that will pay for the rights to analyze the database, known as CancerLinQ, and deliver clinical information to doctors nationwide.

The licensing deal with Tempus and Precision HealthAI is not just a new line of attack in the effort to apply big data in cancer care. It is also a novel way of financing that effort by professional societies, which are uniquely capable of amassing patient information, but lack the money and technical expertise to unlock its secrets.

“From a practical point of view, this means that our subscribers, and the world at large, will get better and faster insights from our data than we could deliver on our own,” said Dr. Clifford Hudis, the organization’s chief executive and chairman of the CancerLinQ board of governors.

He said CancerLinQ, a nonprofit subsidiary of the society of clinical oncology, will received fixed annual payments from the companies. Tempus, based in Chicago, specializes in analyzing molecular data and is led by Eric Lefkofsky, the co-founder and chairman of Groupon. Precision HealthAI uses artificial intelligence to guide precision treatments for cancer patients and helps pharmaceutical companies develop new cancer drugs.

The companies will mine CancerLinQ data to organize it around subsets of patients, making it easier for oncologists to identify the most effective treatments and improve the speed and precision of clinical trials. CancerLinQ includes de-identified information on more than 1 million patients, including their disease characteristics, treatments, and outcomes. The analysis of its data could eventually lead to new products to help doctors diagnose cancer and match individual patients with specific therapies.

Though promising, it remains in its early stages of development. Even with the help of well-financed private companies, it is far from certain that the effort will result in discoveries that will significantly improve the quality of cancer care.

And like any enterprise that relies on artificial intelligence, its effectiveness depends on the ability of algorithms to properly sort and understand vast amounts of complicated data. It will remain incumbent on doctors to understand how those algorithms work, and whether any treatment recommendations they deliver are appropriate for their patients.

“This information is out there for people to use, but does it mean I’m going to use it based on their analyses? Absolutely not,” said Dr. Michael Keng, chief quality officer of hematology and oncology at the University of Virginia. “I would want to see how they are analyzing it and evaluate it for myself.”

The University of Virginia has signed up to use CancerLinQ, but has not yet implemented it. Keng said its immediate potential lies in its ability to give doctors more data on the types of patients they see in their practices.

Instead of relying on personal experience, doctors could tap a nationwide pool of information to learn how certain treatments are affecting patients — whether they are prolonging their lives or resulting in harmful side effects. That task has become much more complicated in recent years with the availability of genomic data and development of new treatment options.

“We have more markers of prognosis and new subtypes [of cancer]. We have so many things going on,” Keng said. “So if we’re able to pool the numbers, we’re really able to see what patients are responding to and design trials” to develop personalized treatments.

Hudis said the new licensing arrangement will ensure that the effort remains financially sustainable. He declined to disclose specific numbers, but said it costs CancerLinQ tens of millions of dollars annually to build and maintain the database and establish partnerships with new providers. Even with the additional licensing revenue, he said, CancerLinQ will remain a money-losing operation.

But teaming up with private companies allows the organization to stanch the losses, maintain its nonprofit status, and continue to help doctors deliver the most effective treatments to every patient.

“It should bother us that we have built up a system where we tolerate the notion that which cancer center you walk into maybe makes a difference in your outcome,” Hudis said. “If there are best standards, we should be demanding them at the societal level for everybody.”