Q
You compare SyTrue’s work to what happens in an oil refinery. You’re extracting data, but also cleaning and refining it for use. How does someone looking into NLP technology find a quality NLP solution — which indicators help?
Recognize the role of “dirty data” and healthcare language usage. Free text data is dirty and tough to clean. But HIT companies, providers and payers are also challenged by free text because there are no healthcare usage standards. “Heart attacks” can be labeled in 5 ways. Misspellings are infinite in number and organizations can classify the Medications section within documents 146 different ways. Documents can show carriage returns removed, or unique characters inserted. Text documents (like a Discharge Summary) might not be classified at all, and there are often puzzling headers and footers. In this complexity, remember one number: 60%. Data scientists spend 60% of their time cleaning data. Too often, this is still a one-off job, done client by client, time consuming and expensive.
It’s important to look for comprehensive or platform solutions. If your vendor’s platform isn’t scalable with an integrated workflow, you’re probably considering the wrong platform.
Our partners tell us that we’ve anticipated these issues with a framework that addresses them. They have tested our platform head-to-head against others and seen results that are outstanding. We’ve constantly outperformed the competition in accuracy, throughput, and functionality, reducing the time to market significantly. In addition the NLP OS ™ components have reduced document curation and clean up time by half compared to other vendors, and improved productivity by double-digit factors. And we typically roll out our solution in a matter of weeks, not months.
Q
Your partners see healthcare reforms, reduced staffs and budgets, moves toward precision medicine, and need population health management. These related elements offer great opportunities for NLP technology. What drives this push? Which factors help predict how NLP technology will evolve?
The main push is for accurate data that supports better outcomes. Organizations now face financial risks linked to new patients. And they see an old issue. Claims data can’t offer the insights into their populations to really manage risk.
But there’s more. We see 2 billion documents created annually. It’s a human resource problem with no easy solution. No one can review all the documents created daily. And it’s an exponentially growing issue. Medicare Advantage, one example, audits only 10% of all claims submitted, but 90% (worth billions of dollars) goes unaudited. The same with state based insurance plans — costing tax payers billions of dollars each year thru fraud. Organizations know they need accurate data at the point of care. What they have now is incomplete data — often billing data – that won’t yield the clinical and operational outcomes in today’s outcomes-driven healthcare. But NLP can address the issue of accuracy and exploding growth in documents being created.
How will NLP evolve? The strongest predictor is algorithms, which take time to develop. With robust algorithms, we’ll see two things: improved health decisions and ways to monitor patients’ longitudinal records – we’ll turn suggestions into care models. But before that happens, providers critically need to produce accurate data.
Q
What should your industry colleagues do as they consider embracing NLP?
NLP is a major tool to create smart data for downstream use, and organizations must be ready to use their data so it meaningfully impacts patients and stakeholders.
Remember that NLP provides desperately needed interoperable data. Organizations need interoperable clinical information to develop data models that impact patients across multiple systems.
Q
What do you tell those who are waiting until NLP matures further?
Remember Elon Musk’s Tesla Motors. Car companies fought the idea of electric cars, until Tesla produced cars with a new vision, and created a new experience for the market. Now other car manufactures are running to catch up but probably never will.
Why? Today’s technology advances at an exponential clip, reflecting Moore’s Law — changing every 24 months. So every 2 years on average, today’s HIT companies are doubling their product output and driving down cost exponentially. That outstrips the overall performance of any company.
So today’s HIT laggards may find themselves too far behind to catch up. Take an orthopedic practice: It starts using data smartly, and gains an advantage in negotiating contracts and marketing. Their data can show that their patients get back to work faster than other patients. This smart use of data transforms referrals — their physicians get paid on outcomes, so a data-rich practice gets higher reimbursement rates. All this is impossible without good clinical data. For that, an NLP operating system is the answer.
Q
As competitors emerge, who will be left standing as the market for NLP grows? What must NLP solution providers do to survive?
Currently, we see 3 “flavors” of NLP technology:
–An NLP engine working like a black box that produces some data.
–NLP workflow-embedded technology that solves Computer Assisted Coding (CAC) challenges.
–An NLP operating system that solves enterprise issues.
We’ve discussed how an NLP Engine is an incomplete solution for the growing need for accuracy. The same is true for workflow-embedded NLP systems. We’ve got great confidence saying that only an enterprise-strength solution can solve the broader issues.
Q
If you’d like to leave readers one key perspective on NLP, what is it?
As you consider NLP, move beyond the technology to the daily reality of clinical documents and how NLP manages them. Ignore technical details on NLP – does it use statistical, rule-based, or hybrid models? These are just paths toward one big goal: improving our healthcare across the nation.
What’s vital here is the usability, scalability, quality and accuracy of the data you produce. To ensure your team can achieve this, start with an operating system designed to address the challenges of refining 2 billion unstructured documents from hundreds of thousands of physicians annually. These documents differ in key ways: in structure, format, misspellings, abbreviations, diagnostic labels, synonyms. So your tools must account for these complex variations as you create accurate data for your health records. Be sure your operations, workflow and data are all integrated.
Can you do all that with your tools? What’s missing? Can you iterate your data extraction as needs change – since tomorrow’s needs will differ from today’s? A solid NLP operating system answers these questions.
