Years ago, you’d ask your doctor during his house call, “What’s in your black bag?” and you’d get to peek inside.
Old Faithful: The Medical “Black Bag”
Now that MD has been replaced by a telehealth practitioner who arrives by video. And the “black bag” is virtual, attracting many providers and executives who are beginning to use it. It’s harder to peek in, but a more relevant question is, “How will the new technologies inside drive value-based care for providers and health plan payers alike?”
It should be said that headline-grabbing “new tech” like machine learning and NLP tend to leave some understandably puzzled — but these aren’t mere novelties that will fade over time.
Make no mistake, the many new technologies help tackle tough issues like interoperability and are operations-ready, able to create new insights for care and value for providers. How all this developed so quickly will interest both healthcare payers and practitioners. It’s a tale with lessons for the future, as we explain below.
The New Black Bag: What’s Inside, What’s Worth Knowing
Consider the good news — useful to hear as healthcare’s challenges remain: unstoppable big data growth, the expanding load of medical documents (growing at 2 billion a year) and stubborn interoperability issues.
Quietly over the course of 10 years, an array of new data analytics technologies has emerged. Back in 2006, Google planted a stake on this timeline with its “TRANSLATE” project, focused on machine learning (ML) and AI. That expanded through work at Facebook, Amazon’s new cloud service (AWS). Some innovation came from academia, but key advances came from smaller, nimbler groups, as others evolved machine learning, deep learning and natural language processing (NLP), plus semantic search tools. All depending on scalable cloud platforms that use cheaper storage to support analytics at a far greater scale.
This new Black Bag now drives innovation in major HIT divisions. It reflects advances in healthcare analytics through tools like NLP – a driver of workflow efficiencies for providers, because it can extract insights from EHRs’ unstructured clinical notes.
INNOVATIONS IN HIT TECHNOLOGY SINCE 2006
“THE QUIET REVOLUTION”
Google+Facebook Amazon cloud NLP EMR Uptake PaaS+AI
As mature technologies, these tools attract HIT units needing to tackle data in new ways. They’re used singly or more often, integrated on platforms. So now we’re seeing strong results with NLP-driven platforms offering enhanced precision through radiology insights and maturing decision support.
But can these new-pedigree tools, arrayed on platforms, drive better care more widely at lower cost?
Related: How NLP Is Unlocking Data Driven Insights for Radiologists
A Test Case: The ICU Patient
Consider a woman rushed from the Emergency Department to the ICU. Her new physicians and nurses will make over 120 decisions daily about her care. Some decisions may trigger new issues.
What’s important: In the ICU, the patient’s outcomes depend largely on the data available about her. Electronic and other data are one part — charts, imaging, labs, EHR care records. But that’s just the start.
The other patient data story is her longitudinal record, or lack of it. Ideally, that should inform the interventions needed but it’s often unavailable. And that lack of a “single-lens” patient picture can impact the care she gets. Despite advances in EHRs, providers find it hard to tap the unstructured, often badly formatted clinical data, but they’re vital to every ICU patient.
Robert Wachter (University of California–San Francisco’s Department of Medicine Chair) reminds us why this is hard work — developing smart, sharable data. Medical applications simply do not “read” and understand physicians’ EHR clinical notes. These are unstructured — neither searchable nor “insight-ready” when accessed. They need to be broken down with the right technology from the new Black Bag of analytics tools.
What if the ICU had that longitudinal care record and it was usable (no interoperability issues)? That’s what NLP-driven platforms bring: They extract unstructured data from varied sources, make them interoperable and draw out smart data insights for clinicians.
Technologies like NLP have matured at a timely point. “Big data” is key to healthcare’s vocabulary, and big data sets grow unstoppably. By 2020, we expect 25,000 petabytes of healthcare data, 80% of it unstructured. (Medical documents, meanwhile, grow at 5.4 million sets per day.) The opportunity to tackle this healthcare “data tsunami” grows bigger each day.
Related: Why Unstructured Data Holds the Key to Intelligent Healthcare Systems
NLP Platforms and “Next-Gen” Analytics
Robert Wachter has a question for us: Is NLP-enabled-data mining the “digital breakthrough we’ve been waiting for?” 1 SyTrue CEO and Founder Kyle Silvestro has his answer: “We’ve helped teams transform into data-driven units. But with an NLP driven platform (SyTrue’s NLP OS ™), the outcomes improve dramatically.”
Wachter adds that NLP “turns a medical record into a vast opportunity for learning.” In cardiac care, today’s systems are unprepared to sort thousands of patient records for diagnoses, measures taken and outcomes created – and spot the complications (pre- or post-discharge). That has major financial impact, and NLP helps tackle these issues.
NLP-driven platforms, in short, address the meaning in clinical notes. And, adds Silvestro, our platform “extracts over 60% more usable data from EHRs and data warehouses than otherwise possible.”
NLP Platform Focus: The Longitudinal View of Patient Data
For Silvestro, his key principle as he helps partners through their data transformation: Analytics stands or falls on the completeness and accuracy of data. To get the patient’s longitudinal record, they start at point of care. “If you get accurate data at the point of care, everything else becomes much easier.”
But many analytics efforts start with billing data, the Achilles Heel in the process. As Silvestro puts it: “By doing this you miss 70% of data, and you impact downstream outcomes. You create a ‘fun-house’ mirror view of data that’s distorted.” Errors and omissions add up, so a wrong start also punctures revenues. Only an end-to-end analytics process produces an accurate, “single lens” view of patients’ data.
Related: Inside The Secret Life of Your Healthcare Data
Why Black Bag Technologies are Different
One thing is clear in piloting these technologies: Many groups have yet to commit confidently to these tools. In a telling insight, Gartner research forecasts that –
“Through 2017, 60 percent of big data projects will fail to go beyond piloting and experimentation, and will be abandoned.”
Many executives see this first hand. As Yogi Berra famously said, “You learn a lot just by watching.” With some effort, we can find HIT units that routinely create work-arounds for old problems, instead of tapping the new technologies through APIs — like NLP. In a recent executive survey, 86% report seeing only minimal progress toward their data transformation goals.
Why? Digital transformation is culture changing work, with high-level decisions needed to start and keep it going. Some may see pilot projects underperform at first, but that first step is inevitable in value-based care.
Best Rationale: Gaining The Interoperability Dividend
A core reason to test the new Black Bag is the interoperability dividend. Interoperability is the “ability to share information across multiple technologies,” according to the Center for Medical Interoperability. ONC expects to achieve it after 2020. For now, it’s a $30 billion loss in annual revenues in much of healthcare.
When it’s lacking, that impacts how multiple healthcare systems work together — accountable care groups and population health efforts, both depend on accurate costing to keep them financially sound. Beyond that are ICD-9 versus ICD-10 coding issues, issues in lab data and radiology findings – all challenging, but manageable with the new technologies.
With the new Black Bag, interoperability is at the core of what these technologies enable, so their strategic value is large.
Related: 5 Key Roadblocks to Data-Driven Healthcare
New NLP Insights For Healthcare Plans
Healthcare payers, meanwhile, see new options in these technologies as they sharply focus on value-based care. Example: the use of HEDIS (Healthcare Effectiveness Data and Information Set), tapped by most U.S. health plans to gauge care and services. With its 71 measures, it helps lower costs and improves revenues. When supported by an NLP platform, some health plans are developing robust new models to improve Medicare risk adjustments and enhance claims adjudication, both critical to the business.
One U.S. health payer using SyTrue’s NLP OS ™ platform has identified new cost savings, and new ways for pricing risk for its partners. “Whether it’s new or old issues, our NLP-platform payer clients find entire HIT divisions discovering new value in their data,” states Silvestro.
A striking example: document splitting using machine learning on the NLP platform. Healthcare data come from many sources (structured and unstructured), all tough to normalize for analytics. A 5-year patient record may be 500 pages long and arrive as a PDF (or a low-resolution PNG file via fax). It must then be read or undergo OCR to render it usable. Those 500 pages may then show over 700 encounters and reflect over 70% of that patient’s data. So the document has to be split and classified as multiple documents showing each of the 700 encounters. This is crucial in complex care cases.
It’s equally important as MACRA rolls out this year. As HIMSS 17 discussions showed, MACRA is a key focus for many in 2017, and all the new technologies will support the organizations that target it seriously.
Testing The New “Black Bag”
Executives facing digital transformation should take Confucius’s advice: “A man who does not plan well ahead finds trouble at his door.” Planning needs clear long-term data transformation goals – so they generate new insights for patient care and new revenue for operations. Now is the right moment for peering inside the new “Black Bag” and putting its technologies to work.