- Of all the areas of medicine that could benefit from the number-crunching power big data analytics, cancer care might be at the top of the list. With endless variations in the development, progression, and aggressiveness of this broad category of diseases, and surprisingly little concrete data to guide physicians towards standardized treatment decisions, the American Society of Clinical Oncology (ASCO) has decided to take on the hefty challenge of bringing better data to cancer care.
The CancerLinQ platform, a learning health IT system that collects and synthesizes clinical data from the EHRs of participating oncologists, started as a small pilot project in 2013. After vastly exceeding expectations for participation and industry interest, the database is set for general launch by the end of the year with a new co-innovation partner on board.
Dr. Clifford A. Hudis, MD, FACP, a medical oncologist at Memorial Sloan Kettering Cancer Center and the immediate past president of ASCO, sat down with HealthITAnalytics to discuss how CancerLinQ will change the big data analytics landscape for cancer specialists by bringing better information to the point of care.
How does CancerLinQ fill a need for oncologists and patients?
In oncology, we’ve made most of our treatment decisions on the basis of prospective, randomized, clinical research studies. Historically, this was needed because many of the treatments for cancer are a bit more toxic than we’re used to in other areas of medicine. And frankly, their benefits are sometimes less than people think.
So in order to determine whether a treatment is better than doing nothing or better than an established standard, we would do randomized trials, watch them for a long time and measure the outcomes and then interpret the results. That’s why often see clinical trial results that are sort of hard to figure out. Do people really live longer? How much longer? How much is it worth?
Because these studies are done rigorously, they’re often done in relatively healthy groups of patients. They have cancer of one sort of another, but apart from that they’re healthy. They don’t have diabetes, high blood pressure; they’re not very old; they’re not very young, and so on. Only about 3 percent of people with cancer actually participate in research. Ninety-seven percent of the experiences we get while treating cancer are outside the clinical research setting, and up until now we couldn’t mine any of it. With rare exception, nothing is truly learned from it.
If I ask my patients what they think happens with their experience, they all think they’re contributing to our learning. Even patients having nothing to do with a clinical trial think their doctors are learning from the experience of treating them. Sadly, that’s never been particularly true. So we hope to change that.
So faced with all of this, we thought that we need a way to start to learn from all of these millions of treatment decisions that are made every single year in America. Because we do know what was done and what actually happened to those patients. And while there’s bias and limits to what you can truly learn from this data, it would certainly contribute to a faster pace of advance in drug development than what we do now.
For example, right now the FDA has a tremendous burden and balancing act when they see new data and a new drug. Do they let it into the market? Do they take the risk that it’s toxic? How much data do they need in order to conclude that the drug is useful?
What if we had a system that didn’t rely only on a couple of hundred patients and one or two randomized research studies? Where the FDA would then know that they were going to get every experience with that drug from virtually all the patients treated in the next three months, six months, or a year? They might have the courage to approve more drugs more quickly while knowing that they could withdraw approval if they learned that they weren’t safe or didn’t work. CanerLinQ is a first step – but big step – to try to address this opportunity.
How will CancerLinQ approach the challenge of EHR interoperability?
It works simply by drawing data in from many EHR systems. It doesn’t matter which one. It associates the data and begins to interpret it by trying to identify and recognize patterns that might not be noticed by individual clinicians. It can provide measurement against standard guidelines, so that doctors know how they’re doing on a day-to-day basis compared to the established standard, and it can provide outcomes data into the future to support the ongoing clinical research that we all do.
We piloted this approach with mostly all open source software, and we launched a pilot in 2013 aiming to gather data from multiple clinical practices around the country using different EHRs. Our goal was to amass data from 30,000 patients with breast cancer. In fact, the thing was so popular we ended up closing the project with 177,000 patients from multiple EHRs and it worked at the demonstration project. It’s a technical challenge, and it requires technical skills in data science that CancerLinQ and the American Society of Clinical Oncology as a whole don’t really have. It’s not our area of expertise, so we needed someone who had that experience with big data analytics to match our skills in medicine.
So we’ve chosen SAP and their HANA technology to be able to very quickly, in real time, process huge amounts of data to get analytic results in terms of associations, trends over time, and the like. So this data set is going to be larger and more transformative than what comes out of a single, closed healthcare system. We’re going to get data from multiple healthcare systems largely outside of the clinical research arena, and this is an opportunity to bring that output into the real world.
How will this data be presented to the clinician at the point of care?
There are many different outputs that would be important. For example, for a rare disease, you might see an aggregated summary of how other doctors are treating it and what their outcomes are. You might see the standard, agreed upon guidelines for a given clinical situation and your compliance on average with that guideline, as well as the specific recommendation for patients like the one in front of you.
You might see a report that patients in that circumstance are experiencing a higher than expected rate of some complication that you didn’t know about, or an interaction with a statin or blood pressure medicine that you wouldn’t have remembered on your own.
Those are some simple examples of how this could be used to drive up the quality of care. That’s the overarching goal and the reason we’re doing this. If you’re a clinician treating breast cancer, and you’re supposed to give a specific kind of targeted therapy based upon the presence of a specific genetic abnormality or a more classic phenotypic variation like estrogen receptor, wouldn’t it be helpful to you to know that 95 percent of the time you were on target but 5 percent you weren’t? Wouldn’t it be great to go back and look at why you were off target on those 5 percent? Maybe there’s a good reason, but maybe there’s a bad reason. This is about improving quality day in and day out.
I’ll stop short of calling this clinical decision support, because there are certain regulatory hurdles and things to calling it that. But ultimately, I care less about why planes crash than I do about avoiding plane crashes, right? So you do want CDS right up front someday, guiding you to do the right thing, allowing you to deviate when it’s for a good reason. But it’s a pretty big challenge to get to that right now, because we don’t actually have the highest level of evidence for the majority of decisions in cancer care, believe it or not. So we wouldn’t want to necessarily provide CDS at the beginning. What we want to do is start to gather the data on how people treat.
How will CancerLinQ evolve in the future?
Version 1.0 is promised for the end of 2015, and we’re working furiously towards that goal. And in terms of the ambitions for the future…ultimately you start to brush up against artificial boundaries. When does it stop being just common cancers? When is it all cancers? When is it all medicine? What’s the difference between doing this with a patient who also has diabetes and ultimately helping manage diabetes outright?
I think these are all going to be evolving boundaries. We did the pilot with breast cancer because it’s a common disease. It has many and many instances a higher level of evidence than other diseases, because there’s been a large amount of prospective randomized studying over the years and the outcomes are measurable relatively easily. So for all those reasons we went with breast cancer in the pilot. But let’s be clear. When version 1.0 CancerLinQ is used, it’s not restricted to any particular cancer. It’s all cancer. And then in the future, we’ll see where it takes us.