Marrying Math and Science in Healthcare
In the coming years, healthcare organizations’ success managing population health will hinge on capabilities that other sectors of the American economy have relied on for over a decade. These same capabilities will be the key to containing costs and optimizing system utilization, goals that are only becoming more important as providers enter into new value-based contracts with commercial and public payers and strive to meet and report on new quality metrics.
I’m talking about algorithms … and since I’m a data scientist, that might not surprise you. But what these algorithms uncover will surprise you. And that’s one reason they’re worth talking about, especially as you map out your goals for 2015 and beyond.
So What’s the Real Deal?
Algorithms power predictive analytics; predictive analytics puts the right information in front of the right healthcare professional at the right time so they can make a quick decision. A good quick decision. Just like the ones they make every day, based on clinical expertise and evidence-based clinical best practices.
Without predictive analytics, the number of potential decision points is unmanageable. You can’t manually review every claim, double-check every bit of clinical documentation, or comb through every patient’s records to determine which gaps in care need to be addressed most immediately. Large healthcare organizations can’t even do that for a significant subset of their patients or records. Healthcare data is simply too big, and it’s only going to get bigger.
Predictive analytics uncovers the cases, patients, and care instances that paint the true picture of how gaps in care, over- or under-utilization, and cost are connected. They outline the comprehensive end-to-end picture of the patient, the continuum of care, and highlight meaningful ways your organization can improve when you’re already running a world-class facility.
Can even the best-performing hospitals and health systems further strengthen or streamline their operations? No. Absolutely not.
And on a related note, Hell just froze over.
Of course they can! The caveat is that these gains in efficiency and efficacy can’t be realized through the indiscriminate piling on of additional decision points. In some cases this even reduces efficacy and efficiency across the board, because it forces staff to spend more time on more tasks that have negligible impact on patient outcomes and clinical performance (if the impact is even measurable..which it often isn’t until sophisticated Big Data analytics are introduced).
What’s needed instead is a refined set of decision points, a curated list or auto-populated work queue realized through data-driven solutions that automate the prioritization process (i.e., make smart baseline decisions) and that cluster claims, accounts, or patient records in meaningful ways so that root causes can be addressed strategically.
What’s needed, in other words, are advanced statistical and predictive algorithms: the proven effective way to extract timely, targeted, and actionable insights from Big Data.
For example, after applying predictive modeling algorithms to longitudinal data on patients diagnosed with morbid obesity and type 2 diabetes, sub-populations of patients were statistically identified. Some sub-populations were low-risk and relatively healthy (type 2 diabetics who are managing their condition through regular outpatient visits with their doctor, etc.) and others are high-risk and less healthy (erratic visits, most through the ER, etc.). From these sub-populations, the algorithms extract the key data markers that then fire early warnings if a currently obese patient’s data suggest they are migrating toward uncontrolled type 2 diabetes. These markers can be presented to the care team, suggesting changes that make the patient’s transition to a healthier, low-risk group of type 2 diabetics more likely. The patient ends up healthier and the provider ends up with less financial risk.
Data Mining and the Triple Aim: Quality, Utilization, and Cost
The triple aim of value-based care models is meant to align incentives with outcomes, and what you need when it comes to mining your data isn’t so different.
Data mining uses statistical algorithms to identify and analyze meaningful trends and anomalies in patient cost and clinical data. Management can then leverage these insights to improve efficiencies and the efficacy of their operations and initiatives. Insights aren’t the same as “findings,” yet the latter are often passed off as the former by vendors who package sub-surface data scraping as data mining. To manage population health and advance patient outcomes, you have to go deep. The challenge isn’t just identifying at-risk populations. It’s understanding the correlations and commonalities that exist within and across those populations, accurately modeling what their care needs are today and what they’ll be in the future, and gaining insight into how you can enhance your organization’s ability to meet those needs efficiently and cost-effectively.
Your patients’ past behavior patterns help determine their future behavior and future care needs; data-driven predictive modeling enables you to determine who will end up costing your practice the most or least amount of time and money. Leveraging provider data and utilizing advanced statistical and predictive algorithms also enables you to automate the process for flagging illnesses and conditions that if untreated will result in other comorbid conditions that would worsen the health of your patients and your business.
Predictive models have been used in several other sectors for many years. Now it’s time to marry math and science in healthcare, time to truly leverage healthcare data to strengthen the continuum of care while containing costs. This isn’t a pipe-dream; it’s the cold hard truth that lies just beneath the surface of the challenges and frustrations we are all aware of and that we see impacting our organizations, our loved ones, and ourselves.
We have all wished for healthcare in America to be different, to be better. And today, across all players in healthcare, the strongest consensus I see is that part of the answer lies in tapping into the wealth of information and insight in healthcare datasets with predictive modeling to direct action, advance clinical outcomes, and improve clinical operations.
And that’s another reason (the most important reason, in fact) that these algorithms are well, well worth talking about.
Click here for the full article.
About the Author
Paul Bradley is ZirMed’s Chief Data Scientist.