Time is of the essence when it comes to treating cancer, the second leading cause of death in the U.S. according to the Centers for Disease Control and Prevention. Between diagnosis and the first day of treatment, days and even weeks may tick by as doctors convene to discuss treatment plans and order testing to gather as much information as possible. But as a new decade dawns, artificial intelligence may buy more time for those who need it most.
Both President Donald Trump and former Vice President Joe Biden have promised to prioritize curing cancer should they win the 2020 election. But because of its complex biology, cancer has been historically difficult to cure with a pill or injection. As new treatments like immunotherapy undergo further research, health systems are starting to harness data-sharing and artificial intelligence to better predict a patient’s prognosis, and determine the most effective treatment plan for their cancer based on other patients with similar medical histories.
The Knight Cancer Institute at Oregon Health & Science University, in partnership with Intel, created a “collaborative cancer cloud” with this vision. Once doctors have met a patient, within hours, they can diagnose them and begin treatment. Thanks to AI’s ability to analyze massive amounts of real patient data, doctors can better predict the best possible treatment for their patient, based on the responses of others with genetically similar cancers. OHSU says the project uses “machine learning techniques against a collective set of molecular and imaging data in order to support big data analytics in a federated, aligned environment,” and hopes to shorten the diagnosis-to-treatment time period to 24 hours in 2020. Similar programs include CancerLinq, ORIEN and the Cancer Moonshot Research Initiative.
“In the past, vast quantities of information were lost to file cabinets and unconnected servers. The big data era is allowing CancerLinQ and other initiatives to access this large and growing amount of de-identified data to help physicians provide high-quality care,” said Dr. Robert Miller, medical director of CancerLinq, which gathers anonymized patient data from electronic health records nationwide. “Instead of learning only from the 3% of patients who join clinical trials, CancerLinQ aims to enable practices to learn from everyone with cancer.”
“Using informatic techniques to mine data in this way is useful, and we’re in the very beginning stages of understanding it. The promise is there,” said Dr. John Vu, a board-certified medical oncologist and director of clinical informatics at Baptist MD Anderson Cancer Center in Jacksonville, Florida. “With advanced analytics, we might be able to get more specific information, like prognosis in a particular situation, or have a better idea of their genetic makeup and the turnaround time for successful treatment.”
Vu believes that same-day diagnosis and treatment would help patients beat cancer because every day counts with a notoriously fast-progressing disease. But for this to become a reality, the rest of medical technology would need to catch up. For example, pathology labs may not have same-day biopsy results. If other steps in the diagnosis process speed up, Vu said, he would feel equipped to make treatment decisions more quickly than is currently possible.
“If someone needs treatment right away, and I had results that day, that patient could be treated with targeted therapies that day. We don’t have the technology to turn it around in one day, but when we get there, yes, that would help that patient,” he said.
The NCI’s Cancer Moonshot initiative aims to improve cancer care and accelerate research, and one arm of that mission focuses on data sharing. Cancer Moonshot-funded projects require investigators to submit a plan of how their conclusions, and data throughout the trial, will be shared with the public. Their answer should end with the data being FAIR: findable, accessible, interoperable and reusable. While data sharing has always happened between colleagues and between experts who know of each other’s work, data sharing on an international level has only gained traction in recent years.
“One of our overarching goals is to enhance data sharing, making sure that data that’s coming from any of the projects funded through Moonshot funds are getting into the hands of people who can use that information well,” Dr. Jaime Guidry-Auvil, director of the Office of Data Sharing in the National Cancer Institute (NCI), told The Daily Beast. “It hasn’t always been appreciated how useful data sharing is. Once a hypothesis is published, it can be thought that there has been a sufficient scientific process. What’s coming about as more data is being generated, is we can take the data and use it in different ways: both to validate a hypothesis, and also to generate a hypothesis.”
Vu agrees that, while AI is not a replacement for a medicinal cure that may one day end cancer, advances in data sharing and analysis in health care could lead to what does. By finding associations within the data, AI can point researchers in new directions they may not have considered otherwise.
“You can see an association that, if you give this drug to this patient, in 1,000 patients, it seems to help, but it doesn’t show a cause-and-effect. It can drive hypothesis,” he explained. “Looking at these large data sets, you start thinking of clinical questions about the data that need to be answered, and generating clinical trials to answer them, and we may find answers that otherwise wouldn’t have been found. In that way, AI can shorten the life cycle of the clinical trial. It can generate hypotheses we may not think of otherwise.”
For example, Vu said, most lung cancer patients today receive genome sequencing testing, which looks for specific mutations within their cancer that doctors know respond well to certain treatments. Hypothetically, AI analysis of patient data may reveal a trend that a specific mutation predicts longer survival regardless of treatment, or show a new link between a less-understood mutation and a developing treatment.
AI can also help customize treatment plans for patients with complex cancers, which some cancer data-sharing platforms have begun offering.
“There are more than 100 types of cancers with different genetic variations,” said Miller. “Data sharing and AI have the potential to help us further personalize care for each individual patient. For example, if a physician is treating a patient with a rare cancer, he or she can examine the outcomes of patients across the country with the same cancer and similar characteristics to help choose the right therapy for the right patient at the right time.”
Vu cited the Eastern Cooperative Oncology Group (ECOG) performance scale, which physicians use to determine how fit a patient is to undergo cancer treatment. The system is subjective, and adding objective data based on AI’s analysis of objective data to the process could result in more appropriate plans for treatment.
“Analytics could help us parse that scale into further detail. Do they have a history of smoking? Diabetes? Obesity? Then when we have a patient in front of us, we could say, ‘Based on your history and this algorithm, you’ll tolerate treatment really well,’ or ‘No, we don’t recommend it for you.’ Currently ECOG is very vague. With analytics, we could get much more specific, but no one has done that yet.”
In 2020, the National Institutes of Health (NIH) will formalize their new data sharing policy, which will require data gathered from any NIH-funded project to be made available to the scientific community. This policy has not been updated since 2003, and the changes reflect a growing attitude that the more scientific data is available, the faster innovations in care can be discovered.
“Much of it centers on the idea that data sharing should be considered throughout the life cycle of research and not just at the end of research. If we promote better data sharing, then we can accelerate our field further,” said Guidry-Auvil.
Ultimately, AI is not the cure the world has been waiting for. But oncologists like Vu are beginning to use it as a way to improve treatment until the cure is discovered. And, if researchers take the opportunity, AI may be the tool that helps science discover a cure sooner.
“All of this AI stuff is really new. It’s been out about five years and hasn’t taken hold in many places. It has a lot of promise: I think the way we treat cancer in 10 years will be a lot different than we do today, and some of that will be due to advanced informatics,” said Vu. “You can use AI to help accelerate discovery, to find new processes and situations to suppress cancer’s growth. I think AI can help us get there faster than traditional means, like standard clinical trials. Definitely it’s moving us in the right direction.”
As Guidry-Auvil says, “having very quick access to a data set that is much larger, that you could not easily obtain on your own, will hopefully lead to an improvement in treatments. Maybe it’s not a direct cure, but without that information, it will be harder to get to one. Adding AI to that, if the AI can analyze the data well, would add to that progress.”