Part 1 of 2
If Big Data needs a big industry to prove itself, there’s no better candidate than American healthcare. “Big” only begins to describe an industry in which the National Healthcare Expenditure, one measure of the sector’s size, exceeds $3 trillion.
Admittedly, that number is in some dispute, but the dispute centers on the proper accounting treatment of some $300 billion in spending. When an industry can view a $300 billion discrepancy as one of those pesky accounting errors, the size of that industry speaks for itself.
It’s also an industry that generates vast amounts of data. We all interact with the industry throughout our lives. Practitioners create extensive records on their patients. Everything is charted.
At Elephant Ventures, we're used to dealing with data sets that are enormously complex.
Insurers, including Medicare, insist that they be provided with the details. Medical billing and reimbursement depend entirely on information derived from coded systems that are applied everywhere, and electronic record-keeping has taken the place of paper.
There’s no shortage of data now, and, if anything, the amount of information in the system will only increase.
We have, then, a big industry that, by its very nature, generates big data, but it’s an industry with problems.
American healthcare delivers quality care, but it does so at great cost, and there’s a general consensus that those spectacular costs are not yielding similarly spectacular results. Life expectancy in the United States, for example, trails life expectancy in countries that spend less than half as much per capita on healthcare.
Clearly, we want a healthcare system that delivers value, and that’s always a two-sided proposition involving both costs and outcomes. If costs are out of control, we need to rein them in. If results are short of optimal, we need to know how and where to improve.
To date, most value-directed efforts have aimed at controlling costs, doing things like bargaining for lower drug prices and eliminating duplicative testing and treatment.
Cost control may be the easier side of the equation, however, and improving results may well be the more difficult task. That task is one in which big data can play a vital role, not least by giving us evidence-based insights into outcomes and results. Those insights are objective, they’re testable and they allow us to make apples-to-apples comparisons of different options.
The “apples-to-apples” factor is particularly important. It’s all too easy to look at knee surgeries, for example, and conclude that some set of best practices applies across the board. That generalized approach isn’t likely to yield the kinds of improvements you’re hoping for.
Healthcare is one venue in which we can all justifiably be seen as special snowflakes. Indications for surgery may be very different for a fit 20-something than they are for an older patient whose joints are showing their age, and that distinction may only be the beginning.
Does it matter if one surgical candidate is mildly obese? What about a history of diabetes? What if one is a smoker and the other suffers from a number of allergies?
The right application of big data principles can tell us what variables actually matter. In the right hands, big data can do its work at a truly granular level, and that’s what’s needed if it’s going to contribute to good clinical decision-making and to the realignment of today’s provider incentives, given that those incentives are often at odds with a value-driven approach.
Big data makes those goals attainable. Here at Elephant Ventures, for example, we’re used to dealing with data sets that are enormously complex, and we have the tools to drill down into the data to distinguish between the variables that matter and the ones that qualify as noise.
That’s part of what it will take to tackle the results part of the healthcare problem, but it’s not the complete picture. It’s an obstacle we can surmount, but there are other issues that need attention before big data can fully deliver on its promise. For more on some of those issues, see Part 2.