Sunday, January 11, 2015

Uncertain Health in an Insecure World – 18

“Big Data, Analytics and The Emperor’s Clothes”

The brave new digital world makes measuring everything, yes everything, possible.

My Fitbit tells me if I slept soundly, after walking >10,000 steps during the day. The iCloud tells me that I’m out of new photo storage capacity, encouraging me to purchase more. I’ve already forgotten my login password to the electronic medical record (EMR) that I used in 2014, but faxes of my patients’ test results follow me everywhere.

This is progress, right?

Whether you’re a digital native, migrant or alien, what most of the world experiences of big data can be classified as “the tail wagging the dog.” The password protected entry of our vital signs (age, height, weight, etc.), PIN’s and patient information into secure databases serves a system that is supposed to benefit the users of services (i.e., us).

But does this really serve us?
The real value of big data resides in the capacity of information systems to predict a failure before it actually happens. If big data could foretell a bad night’s sleep, prevent incipient disease, or avert an airplane crash, then that would indeed be powerful.

But for big data to reach the level of useful predictive power, sophisticated analytics are needed.

Pattern recognition that predicts failure (or success) is the forte of high-speed computing. While IBM’s Watson can win TV’s ‘Jeopardy’ on the basis of massive random access memory, it is the analysis of complex chess move scenarios (i.e., modeling) that predisposes Watson to beat chess masters >95% of the time.
Software is necessary to detect problematic conditions early, and to predict bad outcomes that system operators could then use to mitigate harm, to their own platforms or to us. Familiarity is comfortable to humans. Software programs also recognize familiar patterns, giving them a beneficent “pass”. But until data patterns suggest that a predefined negative threshold or boundary has been crossed, it concludes that nothing is wrong.

Failure, like icebergs, loves to lurk below the surface of such familiarity.

Variability in data quality is the enemy of big data usefulness, and the nemesis of analytics. When U.K. patients fill in EQ-5D forms or record outcome measures (PROM’s) after inpatient hip or hernia surgery, the related cost-effectiveness and cost-utility of these procedures varies greatly across NHS hospitals, even after adjusting for case-mix index (CMI).

The “garbage in, garbage out” adage is applicable.

 Health care systems in the developed world collect gobs of data for reasons usually related to rational health care service delivery (administrative databases), and to the process & outcomes of health care (patient databases). Both types of big data sets should ideally provide health systems with insights into achieving the triple aims of patient access, sustainability (cost control) and care quality. Increasingly, health care payers (purchasers) require timely system performance data capture & submission before making payments to hospitals and caregivers (providers).

The purchaser-provider split remains a gulf, and costs are still rising in most health care jurisdictions.

For years, such central health care databases and the diverse EMR’s used in the field have underpinned triple aim aspirations of health care systems where insurance coverage is legislated and access is ±assured. But the successes (and failures) of these big data systems have mostly related to protecting patient privacy, assuring insurance portability, and measuring compliance with care guidelines & pathways.

The hoped-for benefits of big data to triple aim success (i.e., to us) remain largely unfulfilled.

A 2012 survey by MIT Sloan Management Review and SAS Institute Inc. of 2,500 big data users in 24 diverse industries, including Kaiser Permanente, revealed that the analytics revolution was incomplete. Only 11% of respondents were analytical innovators, while 29% were analytically challenged. Unlike in the data mining companies of Silicon Valley, with an estimated market cap of US$41.5 billion by 2018, the emerging field of data science has yet to produce tangible ROI in the health care sector.

To us, the possible is always alluring.

But in reality, the analytics revolution is not now… not yet.

The Emperor of Big Data was just spotted walking through the Square… sadly, without any clothes.


No comments:

Post a Comment