Medical coding and clinical documentation requirements have become more complex in recent years. This added complexity can create significant administrative burden for physicians and contribute to operational inefficiencies. Hospitals and health systems are increasingly leveraging technology solutions, such as natural language processing, to help alleviate physician burden and improve clinical documentation accuracy.
Becker’s Hospital Review recently spoke with Jason Mark, manager of Data Science and Natural Language Understanding teams for 3M Health Information Systems, about how natural language understanding (NLU) — the latest evolution in natural language processing (NLP) technology — is streamlining workflows for clinicians, clinical documentation improvement specialists and coders. He shared how speech recognition, more comprehensive patient information and better pattern detection can improve efficiency and empower clinicians with improved patient insights. The current COVID-19 pandemic offers a unique backdrop to illustrate the power of these technologies.
Natural language understanding takes natural language processing to the next level
NLP is a fairly mature technology. With NLP, machines can read language, understand parts of speech like nouns and verbs, handle entity recognition such as the names of physicians or medications and understand message intent. Natural language understanding (NLU) enhances NLP, augmenting it with deep content sources, such as clinical information and deep coding knowledge. This generates a robust medical information model.
“As we apply NLU to a patient’s electronic health record, we can develop a more complete picture of the person’s conditions,” Mr. Mark said. “We aren’t just picking out a few symptoms and saying, ‘We see this plus that, so we think the patient has diabetes.'”
In addition to delivering richer insights into specific health conditions, NLU offers information on how different health conditions interact with one another. NLU also can distinguish between clinical language and the administrative and financial language used for revenue cycle and coding.
Another advantage of NLU is the ability to analyze the full patient encounter, not just a single document. This is important, especially when working with hierarchical condition categories that require a longitudinal record.
“We have a more comprehensive model which helps us understand the patient’s clinical picture across all of the information available to us,” Mr. Mark said. “For example, we know that three visits ago, you were checked for a condition that requires follow-up and that hasn’t happened yet.”
Natural language understanding improves patient care by reducing provider overload
NLU-based solutions eliminate unnecessary tasks for healthcare professionals. By adding layers of deep understanding to electronic health records (EHRs), these systems recognize what data is available and limit queries to providers as they dictate information.
A very real limiting factor in adoption by providers is “alert fatigue”. Every time a clinician says the word ‘diabetes,’ for instance, some systems prompt the provider to ask the patient five questions. In contrast, NLU systems know whether these questions have already been answered elsewhere in the patient record.
“We prevent provider overload by only asking about what is relevant and what can’t be answered with existing information,” Mr. Mark said. “When we get documentation right from the beginning, it reduces query rework loops which saves time and effort for everyone downstream.”
When questions arise in patient records, NLU-based systems address them based on different user roles. For example, if a patient record has a documentation gap or if an inconsistency exists between lab results and a medication, the system might nudge the provider and remind them to document the issue up front. If the user is a clinical documentation improvement specialist, the system would help them craft a query. Taking a smart approach to user tasks lowers distraction levels for everyone.
The benefits of NLU extend beyond the revenue cycle. Thanks to higher levels of computing power and access to larger volumes of data, NLU-based solutions conduct more advanced pattern recognition related to care practices and patient demographics.
“We can now be more intelligent about social determinants of health,” Mr. Mark said. “When a doctor prescribes a medication, systems will increasingly be able to determine whether the patient will be able to fill it based on their insurance, income, access to transportation and more.”
NLU enables health systems to apply the right resources to the right patients
The unique combination of speech recognition and NLU has the potential to eliminate considerable waste in healthcare.
During the COVID-19 pandemic, for example, there is a heightened awareness to save every minute of physician time possible while maximizing the value of the information that they capture as they treat patients. Even today, months into the pandemic, there are many things still unknown about co-morbidities, treatment efficacies and care practices that could lead to better outcomes. Unlike common conditions, where the medical field maintains historical data spread over a significant time span, there is little to draw from for COVID-19, other than the volume of documentation being generated right now as patients are treated. It’s possible that NLU can help to unlock the insights and patterns in the physician narrative to help expedite consolidation of knowledge on how to best combat COVID-19.
“If doctors can document something once, up front, and all the other downstream processes such as revenue cycle and quality have the information they need, it frees time for doctors and other staff,” Mr. Mark said. “This means clinicians and employees can focus on more complex cases or be redeployed in different ways.”