HealthLawProf Blog

Chilling Thoughts from Chilmark about Data Analytics and Patients

Chilmark Research produces evidence-based reports of health IT and market trends in the health IT industry.

A recently issued Chilmark report, 2013 Clinical Analytics for Population Health Market Trends Report, which I have not read because it costs $4500, details the conflicting interests of clinicians and payers with respect to insights gleaned from data analytics.  The hope of EHRs in combination with data analytics is better patient health, for example through alerts about needed preventive measures or care management strategies.  But different payment may reimburse categories of care differently--so a diabetic covered by one type of payment structure might get reminders when her counterpart with different coverage might not.  Even worse, patients whose prognosis is seen as "hopeless" through the predictive lens of analytics might get very different treatment recommendations under cost-conscious reimbursement structures.

Cora Sharma's post on the Chilmark blog details these likely conflicts with chilling precision.

[LPF]

Is Big Data Overhyped?

For some in Silicon Valley, the rise of new data and communication networks creates unprecedented opportunities to solve problems like obesity, traffic, and flu pandemics. For example, an app like FitBit or LoseIt can keep track of calories and buzz a dieter once he goes over his daily limit. Futuristic early warning systems can warn drivers away from bottlenecks, and detect emerging influenza outbreaks.

Evgeny Morozov’s illuminating book To Save Everything, Click Here challenges both “internet centrism” and “solutionism.” The internet may, for instance, make traffic worse. Moreover, solutionism tends to “reach for the answer before the questions have been fully asked.” Is the problem really traffic, or something deeper in the way cities and opportunities are arranged? Solutionism tends to prioritize issues that widely accessible tech can address: small, algorithmically decomposable bits of wicked problems.

While a solutionist might think of gamified calorie counting as a wonderful new way to fight obesity, a more sober analysis of the problem will lead us to doubt the smartphone will make us svelte. Similarly, calorie counts may be a great disclosure tactic, but disclosure is only the first step on the road to changing behavior. And our food problem, like our traffic problem, may entail reconsideration of privilege, taste, and inequality as far deeper problems than individual struggles for self-control.

Big data has been linchpin of solutionist narratives about the future of tech in health care. However, there are still major challenges in data quality. Even if the data were perfect, causal inference still may be a challenge, as Hoffman & Podgurski explain:

EHR [electronic health record] vendors are making slow progress towards achieving interoperability, the ability of two or more systems to exchange information and to operate in a coordinated fashion. In 2010 only 19% of hospitals exchanged patient data with providers outside their own system. Vendors may have little incentive to produce interoperable systems because interoperability might make it harder to market products as distinctive and easier for clinicians to switch to different EHR products if they are dissatisfied with the ones they purchased. . . .

Even if the EHR data themselves are flawless, analysts seeking to answer causal questions, such as whether particular public health interventions have had a positive impact, will face significant challenges relating to causal inference. These include selection bias, confounding bias, and measurement bias.

Paul Ohm adds to the data skepticism in a recent essay:

[A]s medical research follows the lead of Google Flu Trends and begins to slip outside these traditional institutions and their concomitant safeguards, we should be concerned about the relative lack of controls. Particularly as more medical research is conducted by profit-driven companies—–whether large corporations or small startups—–we should worry about forcing the public to accept new risks to privacy with little countervailing benefit and none of the controls. The worst of all worlds would occur if medical researchers at non-profit institutions began to clamor for relaxed human subjects review in a race to the bottom to compete with their forprofit counterparts.

Ohm’s point about maintaining a baseline of standards is prescient: I have heard at least one behavioral scientist argue that research will migrate out of universities and into private companies if the universities don’t relax IRB standards. Ohm also questions whether something as celebrated as Google Flu Trends has led to actionable data:

Who has created an app, therapy, or epidemiological study based on the colors on [Google's flu maps]? Has a traveler ever avoided boarding a plane to a city on a distant coast because of the relative difference in the shading of the oranges between home and destination? The answer, I suspect, is that none of these positive results has occurred. Instead, the project’s primary mission is to market Google: we are reminded by a colorful map that Google is not evil.

As I’m sure the Washington Post will be soon be reminding us of cloud services generally.

[FP]

X-Posted at Concurring Opinions.