Tons of data is generated, but we can't possible analyze and mine it all = what are we missing? #patientjourneys— Alicia C. Staley (@stales) June 12, 2017
To begin, I responded that I'm sure we can analyse it all... that's what machine learning is for. Of course, machine learning has hit some bumps with @ibmwatson lately. In my humble opinion, the issue with Watson hasn't been with the learning, but with the predicting. Machine learning is fairly standardised at this point.
A quick tutorial from a novice... feel free to skip this or correct me...
Machine learning takes samples known cases. So, let's say I want to analyse a journey of a breast cancer patient. I may want to pick data from @stales and review her:
- search terms and search agents
- social activity in twitter and facebook
- blog posts
- medical billing records
- electronic health records (EHR)
- prescription purchases
before, during, and after her diagnosis. Then I'll also pick about 50-100 or or so other breast cancer patients' data and feed it all into the program. The machine learning tool develops patterns of search terms, social posts, physician visits, prescriptions... the whole ball of wax. Then, and this is where the data gets to be useful, we take other patients who might not yet have been diagnosed, and pop their search data, their posts, their medical data into the program. As the machine learning tool identifies similarities to the baseline, the system will flag them for review. "Here's a potential match..." That sort of thing. Where this all gets dicey is if the system tries to predict what will happen next... "Well, you've done a, b, and c... so d-z may be logical next steps."
If this all sounds like I'm some fiend from insert your favourite evil technology movie here... I'm sorry. The point of data is to analyse it. WHAT WE DO WITH THE ANALYSIS IS WHAT MARKS US OUT AS HUMAN/ EVIL FIEND.
One of my other points in my conversation with @stales was that we need to have trust from patients and honesty from Big Pharma. You see, I'm assuming that Big Pharma will use the data to develop therapies which can help stem the progression of disorders, prevent disorders from occurring, or help manage the disorder until a cure can be discovered. This kind of agreement, this tit for tat, assumes that patients give up an immense amount of privacy and Big Pharma gains a metric buttload of data that helps these patients and those coming after them.
The potential legal conundra emanating from this are legion. If a bulk of unblinded health data exists, what happens when insurance get their hands on it? If predictive analysis becomes the standard, what happens when decisions are taken for continuing coverage based on the analytics? When Big Pharma can microtarget patients, what does that advertising look like? HIPAA? Will governmental agencies exclude immigrants based on health history?
Of course, I mention those things because they're already happening. Insurance companies spend a lot of money to have the most robust actuarial tables for inclusion/ exclusion criteria. Big Pharma already combs through blinded EHR data to identify markets for clinical trials or drug development. HIPAA sometimes feels like a cruel joke intended to make it harder for my data to be used by multiple physicians. I'll leave the last one alone... but google your own examples.
So, there's no clear reason to assume that Big Pharma or the insurance industry can be trusted. Unless you have a monopoly on the research, development, approval, manufacture, distribution, and payment for treatments though, there's no other game in town.
An honest, intermediate broker though... that has some legs.
If we look at the example of the Cystic Fibrosis Foundation (CFF), and their development of Kalydeco, we may have the best example of what happens when an honest broker gets involved in the patient journey.
By establishing a "Venture Philanthropy" model, the CFF was able to solicit funding which then went to targeted research clinicians who developed a drug approved by the FDA... which then went on to be marketed by a company for 300,000USD per year...
But the model exists. It's there. It got the job done... right up until that last part. So, we can modify the working model. It's what you do. You don't say, "Well, that didn't work... looks like we have to find some completely other new way to do it." You say, "Well, that *almost* worked... what failed? Let's fix that."
To that end, the trusted broker is probably going to have to come from each disease constituency. CFF showed that a dedicated group can get shit done. So, the National Multiple Sclerosis Society, American Diabetes Association, Short Bowel Syndrome Foundation, American Cancer Society, Sjogrens Syndrome Foundation, American Heart Association, Crohn's & Colitis Foundation, National Organization for Rare Disorders, and the multitude others can act as data repositories for the journey of their patients. You can guarantee that each group is staffed with a Chief Data Officer or Technology Officer or other job. Those staffers needs to lead the charge to get the data resources for their analyses.
Of course, data analysis is a back-office task. It's not pretty. It's viewed as a bunch of data-loving nerds crunching numbers, building statistical models, producing charts... like patient journeys... for other people to never read in white papers.
If we want the data for patient journeys, we have to gather and analyse the data for patient journeys. And we have to find people who love healthcare data specifically to do these analyses. You *do* fall into a flow that may make you seem like Dr. Evil while you're doing your analysis... but those of us who love healthcare data love it because we know if we find a new data point that helps identify patients early, or finds a new influencer with a new point of view, or tease out that one lab test that seems to be helping identify patients more commonly than others, or help develop appropriate pricing models... we want health care for people not just for feeding shareholder value.
No comments:
Post a Comment