Anyone who works in healthcare—or who has pondered, even fleetingly, how it differs from other sectors of the American economy—likely won’t be surprised to hear it’s the final frontier for predictive analytics.
For indeed, while predictive analytics has long since reshaped retail, shipping & logistics, and even national security, healthcare—to draw an imprecise but nonetheless illuminating comparison—is still in the throes of transitioning from paper to electronic records.
Given this disparity, it’s natural to wonder how established forms of data science are reshaping healthcare, particularly for hospitals and other large healthcare organizations.
Dig a level deeper and it becomes more interesting still. The volume and velocity of health data continue to increase due to government-mandated adoption of electronic health records (EHR) and similarly mandated sharing of data among healthcare providers. The latter might not sound so challenging, but in fact it’s a hot-button issue—the most recent government efforts are squarely focused on countering obstacles to interoperability among health IT vendors.
Let’s get down to the nitty-gritty.
On the financial front, hospitals are using predictive analytics to manage the almost comically byzantine world of healthcare finance and reimbursement.
A little context: In the hospital world there’s a living document known as the charge master, also known as the charge description master, and commonly referred to as the CDM. (Let’s not get bogged down in a mix of acronyms—we can all agree that’s hardly the point.)
Picking back up: A typical hospital CDM has around 30,000 to 40,000 individual line items. In simple terms, each one is the amount that the hospital charges for a specific device, procedure, medication, and so forth. Yet nothing in healthcare is ever “simple.” New line items are continuously added to the CDM, sometimes replacing existing ones, other times not. Then, during the billing process, each line item is subjected to various non-uniform adjustments based on negotiated reimbursement agreements with individual health-insurance plans.
Imagine trying to manually keep track of the adjusted charge amounts for each line item for each insurance plan—to say nothing of detecting missing or incorrect charges across hundreds of thousands of individual insurance claims. Now you begin to see why machine learning and data mining are so critical to managing healthcare financial data—even something as ostensibly straightforward as a price-list has tens of thousands of data points and dozens if not hundreds of relevant variables, far beyond what humans can manage without the aid of AI.
The bigger picture: tying financial reimbursement to clinical outcomes.
Predictive analytics is also helping hospitals understand the true drivers of unplanned readmissions. These are patients for whom we can find a better way. These are the patients for whom the current US health system isn’t working, not because healthcare organizations are at fault, but because managing the health of at-risk or chronically ill patient populations is a complex undertaking. Improving outcomes for these patients is a moral imperative, and requires analyzing CDM-level data in combination with patient-level data points such as diet, exercise, and engagement with their preventative and recuperative care plans.
Yes, hospitals are financially penalized for unplanned readmissions. And yes, these readmissions drive up healthcare costs overall. But the most important truth is more granular: unplanned readmissions translate to lower quality of life for ill and at-risk patient populations—the sick, the elderly—and elevate their risk of acquiring antibiotic-resistant infections such as staph.
That encapsulates why it’s rewarding to work on this final frontier. The challenges in healthcare are significant—because overcoming them matters to all of us.
Click here for the full article.
About the Author
Paul Bradley is ZirMed’s Chief Data Scientist.