On the Ramifications of High-Tech, Big-Data Medical Care
Data & Society / saved 2014-03-22
Summary:
As mentioned this morning , in our new issue I have an interview with Dr. David Blumenthal about the paradox of modernization in the American health care system. We all know that everything about medicine is becoming technologized, in ways good and bad. On the good, see previous interview with Eric Lander about the genomic-knowledge revolution. On the bad, see Jonathan Rauch on the industrialized process of dying . But we also know that nearly every visit to a medical facility begins with the tedium of filling out forms by hand.
David Blumenthal was in charge of the Obama administration's effort to speed the adoption of electronic medical records, and in the interview he explains why that has been hard but will be worthwhile.
Now, responses from readers in the tech and medical worlds. First, from David Handelsman, of a health-related data company in North Carolina:
One of the things that Dr. Blumenthal didn’t include in his response was that the health care industry needs to continue to create a culture of evidence-based medicine, beyond the activities at those organizations that are further along the maturity curve regarding electronic health records and healthcare technology.
The reality is that much of healthcare is administered to patients based upon the practitioner’s experience (patients he or she has seen with similar conditions), the practitioner’s ability to accurately recall the appropriate care for the patient being seen and, where time has allowed, the opportunity to stay current on healthcare research and able to then apply that research correctly to the patient at hand.
While I have the utmost respect for health care practitioners, there’s an awful lot of room here for what I’ll call non-optimized care. The practitioner’s experience may be incomplete – he or she may not have seen a patient “like this patient”. They’re ability to recall the best care options for “this type of patient” may be unreliable given the vast numbers of patients in their care. They may be lagging behind in current research and recommendations because there isn’t enough time in the day, and if they are current, they may still have the same issue of recall regarding complex health decisions.
Evidence-based medicine aims to provide optimal recommendations regarding patient care. When electronic data is available, the patient’s current situation can be electronically compared to other similar patients AND their respective healthcare outcomes. At that point, a recommendation for care for “this patient”, whose profile is aligned with “all of those similar patients”, can be made based upon the recorded outcomes. Please note that this should be considered a recommendation – the responsibility for providing care ultimately falls on the practitioner, and not an algorithm, but that practitioner should always have the best information at hand to determine the best course of care.
From a retired MD:
There is a rub [among many, I suspect] as seen in our personal records from digitized offices.
To respond to insurance company demands for documentation of visits, physicians can simply cut and paste the previous visit data onto the current visit, making such changes as are necessary. Every visit looks remarkably complete!
The volume of material viewed makes finding anything new difficult. The record becomes a document for the insurance company, barely useful for physician's own use or physician to physician communication.
From the academy:
I'm a PhD student in statistics working on prediction and causal inference using health data. I'd like to comment on a quote from your (great) interview with David Blumenthal about the promise of Electronic Medical Records:
Dr. Blumenthal says: "This will move us into a field that is taking shape right now, that of analytics. It will help us take these data and turn them into diagnostic information—into recommendations a physician can give a patient or that patients can get directly, online."
He seemingly conflates 'diagnostic information' and '[treatment] recommendations'. But they actually pose fundamentally different problems from a statistical perspective, and I think EMRs will play a much more transformative role in diagnosis than treatment. This is because diagnoses live in the realm of pure prediction, while treatment decisions live in the realm of causal inference. EMR data will be observational. Using observational data for pure prediction is completely valid, but using it for causal inference is only valid under strong assumptions of no unobserved confounding.
A common fallacy of Big Data Hype is the assumption that gathering boatloads of observational data will enable us to solve problems that are fundamentally causal in nature. There will certainly be special situations where EMR data can reliably drive treatment decisions (and this will be a big deal!), but such cases will be the minority.
By contrast, statistical algorithms should be able to almost always make excellent, reliable predictions about what conditions a new patient is likely to have or acquire given her own health