Predictive analytics has become a mission-critical tool in healthcare, but managing its responsible use is no small feat.
The impact of analytics on the healthcare industry has been so substantial that articles, books, training courses and entire websites have been dedicated to its application. Few arenas of human service and commerce offer such unbounded opportunity—or such formidable challenges.
There are reasons it's a double-edged sword. Forget that healthcare is an essential service and big business; it's an arena of rapid and constant change that's persistently buffeted by economic and political forces, which make its path uncertain—if not precarious—at any given moment.
It's one thing to analyze and optimize the shifting supply and demand in the domain of actual care, where change is already tough to anticipate and cope with; it's another to try to track those shifts while also struggling to follow a fluctuating public-policy landscape.
Put simply, healthcare analytics can be oppressively complicated. Unlike retail and logistics, where the task is to tease out and capitalize on subtleties in well-established patterns in events and behavior, healthcare—much like the stock market—presents the challenge of figuring out what the patterns are, when they will suddenly change and how much.
This pattern-finding results in time savings and cost reduction in industries across the board. In healthcare, however, analytics delivers another significant reduction: risk. And getting to that can be oppressively complex.
Don't get me wrong: Conventional analytics are a plus in healthcare, from administration to patient care and beyond. The descriptive analytics tasks alone can make the difference between success and failure. Population segmentation is an essential first step: knowing which patients are most likely to respond, who will respond to marketing outreach and who likely won't respond at all.
But in healthcare, those numbers aren't just about selling insurance and care plans; they're about who responds best to managed care, which patients are most likely to engage in proactive behavior and where caregiver efforts are most likely to make a difference.
Consider the role of outcomes. In retail, an outcome is an event—a product purchased or a vote cast. In healthcare, outcomes are about much more than events: Buying a product or casting a vote seldom changes a person, but healthcare outcomes can result in lifelong changes—for better or worse. It's not enough to simply describe an outcome; it has to be placed in the context of lives in progress, with many moving parts, and that's much more complex than mere demographics. Healthcare and the people receiving it create an ongoing dynamic.
Strong descriptive analytics, while important, aren't the true competitive edge; it's fine-tuning the understanding of outcomes and following the data toward the best ones—however counterintuitive.
Predictive analytics can improve healthcare outcomes by analyzing patterns in current behavior to help predict future outcomes. Certainly, the bottom line is a consideration, and the metrics of institutional performance are factors. But in the end, good predictive analytics in healthcare creates better lives for those in need.
How does it work? Beyond the aforementioned advantages, predictive analytics enables earlier detection of health issues, enabling clinicians to proactively address them before they become severe, or even stop them before their onset.
In the case of a mother-to-be, descriptive analytics can detect patterns in her health claims data or clinical care management records that foresee the emergence of a high-risk condition—gestational hypertension, for instance. When the cofactors commonly seen in those pregnancies are identified, descriptive analytics can provide a pattern that predictive analytics can use to red-flag an upcoming high-risk condition, and the clinician can then help the mother-to-be avoid it altogether with intervening treatment.
The same holds true for any number of other healthcare issues, including cancer. Though all cancer emerges from the same pathology—misreplicating cells—different cancers in different parts of the body affect the afflicted person in different ways. Here, as in preterm birth, early detection is key. Subtle changes in appetite, weight, mobility and other factors can be descriptive analytics inputs, yielding patterns that trigger positive cancer predictions. And then an MRI and biopsy can occur far sooner in the progress of the cancer, yielding a much better outcome.
Not even climate models present the complexity and challenges to be found in healthcare analytics. It's certainly daunting, but it's also exciting. The puzzles to be solved in healthcare are so varied and intricate that even articulating the range of the puzzles could fill a book or seminar.
There are some downsides to predictive analytics in healthcare. Given the essential nature of healthcare as a service, it's important that data not be used to segment patients into high- and low-risk groups for healthcare services.
One downside that will be quick to emerge is embedded in the inevitable self-service risk-scoring apps and services that have already begun to appear on healthcare websites. These are intended to empower individuals who feel they might be at risk of having a dangerous medical condition to get a sense of how serious it might be, and to visit their physician sooner—sooner is always better, for the individual and for the healthcare system overall.
But this creates its own risk: When anecdotal or misconceived information is entered into such a system, there will be plenty of false positives in the mix, which will burden the waiting room. And there's just as great a danger of true negatives, when someone feeds incomplete or incorrect information into the scoring app, creating a false sense of security and delaying actual diagnosis.
Moreover, expanding the data-gathering process beyond the doctor's office, with freely available analytics spurring individuals on to gather more and more information about themselves to explain what they are experiencing—self-service DNA testing and record-keeping personal fitness apps, for instance -- complicates healthcare coverage and cost, if that data becomes available to healthcare insurance companies.
Where are we headed? Clearly, healthcare analytics is already a mission-critical discipline, becoming deeply entrenched at all levels. And if that's the case, what will healthcare look like in five or 10 years?
First, risk assessment models will rapidly improve, as more data pours into the systems they support. Government healthcare programs like Medicaid have turned to analytics for efficiency gains and improved outcomes. They are making patient-agnostic data available to third-party analytics providers. As the models improve, and those efficiency and outcome gains proliferate, analytics becomes an increasingly desirable investment.
Population analysis will also rapidly improve. More data and higher-quality analysis results in more accurate modeling of at-risk populations, which makes both insurance and preventative care more effective; the best time to address a health issue is before it becomes an issue. When the portrait of a high-risk population is finely detailed and constantly refined, it becomes easier to spot the increase in risk to individuals.
Putting it all together, we can see analytics as much more than just a growing tool set gaining popularity in one of our biggest industries; it's reshaping the industry entirely, changing the very nature of how we approach the management of health. That's certainly challenging, even intimidating -- and it will be bumpy. But it's also exciting, even inspiring -- and foreshadowing of a vast expanse of emerging opportunity.
Scott Robinson is director of business intelligence at Lucina Health in Louisville, Ky.