You may have heard of Watson, IBM's friendly-sounding artificial intelligence system, a robot that can win at Jeopardy. In the future, it might solve the world's most challenging medical cases.
Already, Watson Oncology can read your doctors' notes, review your labs, all images and new kinds of test results including gene sequences, take in all the pertinent literature, and scan a growing, virtual universe of cancer patients' information to identify other cases like yours. Soon, Watson might inform an oncologist which cancer treatments, precisely, offer the best chances of curing someone like you.
If there's one thing that's become clear about precision medicine, and the future delivery of optimal care based in science – and not just on what you or your doctor happen to have read or seen or recall – it will require super technology. This need is especially pressing, urgent in oncology.
The pace of advances in cancer care – new drugs, new diagnostic tests and new research findings – are too much for any human physician to keep track. Combine all that progress with the rising numbers of cancer cases and burdensome costs of care, which need come down somehow – preferably by reason rather than by rationing – and informatics becomes an essential prescription.
The field might be called Oncology IT. The goal is to harness information from millions of cancer patients' records, now and in years ahead (anonymized, of course), connect the dots and – as new medications arrive from the pipeline – nudge doctors and patients toward optimal and what might soon be curative and better combinations of drugs.
And it's feasible: think Facebook (it knows everything about you), plus Google (it pulls relevant information, even images, from a virtual universe of information) and Amazon (it delivers).
Now Watson is just a well-known, popular example of an information technology (IT) system that's already entered clinical practice, so to speak. Already it's being tried and is, essentially, learning from cancer patients' records at over a dozen cancer centers around the country.
Meanwhile, whether or not cancer patients are aware that this has been happening, a mix of biotech and IT companies, some linked, have been dealing with top U.S. cancer facilities, hospital networks and individual institutions. They've been setting up relationships to variously test cancer patients' (physical) specimens, electronically store and analyze clinical and laboratory records and, by machine-learning processes, devise algorithms to suggest best treatments.
Oncology IT is an expanding and, from what I can tell, competitive biotech area, rich by investment of patients' (samples and trust); doctors, scientists and engineers (careers); and money (lots).
At this point, it's hard to compare the in-flux methods of precision oncology firms like Caris, Flatiron Health, Foundation Medicine, Illumina and Lifecode, to name a few, besides ASCO's CancerLinQ platform and the National Cancer Institute's programs in this realm. But having explored this emerging field for most of a year, some questions seem clear.
Here are five key issues for cancer patients and physicians, concerning now and future Oncology IT:
1. Which Information Goes In?
There are zettabytes of cancer patients' data, subjective and objective, raw and tallied, from which future investigators might cull patterns. While I tend to think it's worth entering all of each patient's information, not everyone would agree. Some might include only "hard" facts like pathology, labs (including standard genetics), drug doses and survival.
I think it's worth entering everything possible for each case, with each patient's consent (see below). Because in creating cancer IT systems, private and public, and merging "islands" of patient-data from single hospitals, what gets omitted may be impossible to retrieve.
If we're going to shoot the moon, we shouldn't skip the soft (and cheap) stuff: how cancer patients feel, what they've tended to eat and drink, where they've lived and worked.
Consider, for instance, the potential value of knowing a million cancer patients' anonymized occupational histories to reveal exposures to, say, benzene while working on an oil rig for 5 vs. 10 years. That sort of information may yield important clues – data – for epidemiologists working 20 years from now, or for future-epigeneticists trying to establish a correlation of certain DNA changes in a gene yet to be identified, linked to leukemia, upon exposure to a chemical.
It's impossible to know which bits of information will turn out to be helpful to individual cancer patients, for diagnosis and treatment decisions, and also for society. The abbreviated list of information to be uploaded in an Oncology IT cloud somewhere includes these components:
- each patient's history and clinical exam, from the time of diagnosis until death – regardless of whether they change insurance or hospitals
- PROMs (patient-reported outcomes)
- standard and newer lab studies (genomics and beyond)
- images (pathology and radiology)
- a clear timeline of all treatments (including brands, doses, and any complementary or alternative meds, such as herbal tea or vitamins, blueberry extract or whatever a patient might be trying)
- outcomes including documentation of each tumor's size, starting at the time of diagnosis, and, if it's spread, which organs are affected (not just a stage); if a patient dies or enters a prolonged remission, all of this should be in the database.
2. Interoperability
We shouldn't make the same mistake twice. One lesson from the fiasco of current electronic health records (EHRs), which were promising and remain necessary, and should get better, is that the information should be interoperable. Too often, a patient's note, lab report or image from hospital or office can't be reviewed at another medical center. It's really bad for patients, who need access to their lab tests if they change insurance and switch to a new hospital network, or go for a second opinion at a facility that "partners" with a separate IT firm.
This problem can only be circumvented by having uniform data standards. Clinical information and lab results, however encrypted, need to be (readily) accessible to all cancer patients and their physicians.
3. Human Input Needed
Best cancer treatments will, eventually, be determined by informatics. With detailed clinical and pathology information on millions of cancer cases available, patterns will emerge. It could be, after a person has a tiny biopsy, it'll be clear what's the best combination of two or three or four cancer meds for that patient, how often and which labs should be monitored, etc. We'll be managing cancer as we treat many (formally incurable) infectious diseases.
What's interesting to consider – and I think this a real possibility – is that Oncology IT could obviate the need for clinical trials. Because each patient's decision would be informed by information extracted from the cloud, on the treatments and outcomes of the most-similar cases.
Despite all this I do think, however far forward Oncology IT does go, that it's best for patients if real, human doctors, physician-oncologists, are involved in their care, including treatment choices – which should include a personalized component – taking into account each person's preferences, fears, values, etc.
4. Who Owns the Samples, Test Results and Information?
Today, a cancer patient might have her inherited genome, and her cancer genome, sequenced by a biotech company affiliated with a hospital where she receives care. Imagine, now, that she survives for five years and, upon a recurrence, wants to retrieve her sequence information. There is no doubt she should have access to all of her test results. If there's a bit of sample that went unused, like part of a biopsy that was frozen for possible future studies, she should have access to that, and be able to take it to another place for analysis, if she chooses.
5. Consent, Required
In the context of a new cancer diagnosis, people are vulnerable to coercion: "Our hospital is participating in a program where all the samples and test results are being collected to be analyzed and help future cancer patients. Will you sign here?"
An ethical problem, rarely discussed, is that patients may feel, or indeed be, pressured to follow the doctor's suggestion, especially if they have cancer and the oncologist appears to be knowledgeable. I have known more than a few individuals who've signed up for experimental trials, so that their medications and co-pays for visits would be covered. And I've had educated friends, upon meeting "the best" oncologist in a region, sign anything. The fine print is neither easy to read, nor comprehend.
At the minimum, patients need to "opt in." Ideally, and somehow, this needs to happen in the context of a meaningful conversation with providers, and not under pressure.
Solving this problem, of informed consent for all of the above, won't be easy.
