Imagine if your dictated visit notes could update patient EMRs. Clinical language understanding can make that happen.
More than 40 years ago, the introduction of problem lists in clinical care was aimed at controlling the complexity of the rapidly expanding universe of specialized medicine. It was a way to provide physicians with a complete, concise and clinically consistent view or “entry point” into a clinical case. While their value in patient care has been demonstrated in countless studies, physicians have historically adopted them with much less enthusiasm than one would expect.
The advent of electronic medical records (EMR) fostered hopes that the problem list could become the primary reference for care providers approaching a patient. And yet, according to multiple studies,1 creating and maintaining computerized problem lists has proven to be an elusive goal, even for the most diligent practitioners equipped with the most advanced EMR systems.
Cumbersome data entry, I presume
There are many reasons for the lack of adoption of electronic problem lists, but the most common is that they need to be entered and updated manually, a tedious and time-consuming task for physicians. The result? While patients' diseases, symptoms and risk factors evolve and change, the corresponding items on the electronic problem list tend to age rapidly and may soon become irrelevant or even inaccurate. For example, a certain symptom may have disappeared, or an initial diagnosis may have been further defined, making the initial description too generic to guide actual care. Additionally, as multiple specialists engage with a patient, they focus on problems that are both different and overlapping. While each provider contributes to the problem lists (from different perspectives), patient data rapidly becomes repetitive or redundant, rendering the electronic problem list less useful. This could help explain why, in Massachusetts where the adoption of EMRs is about four times higher than the national average, problem lists are routinely used by less than 50 percent of doctors using EMRs.2
The gap between clinically active problems and the data in the EMR affects not just problem lists. Clinical applications that support proactive care protocols, such as case management and real-time decision support, are severely hindered when patient data is missing or out of date. In order to function correctly, these tools require the most current and up-to-date problem lists and other clinical information, some of which is continuously being generated during the episode of care that could still be unfolding during ongoing hospitalization.
Most modern EMRs offer integrated and customizable tools and pick lists for physicians to create and manage patient problems in the EMR using keyboard and mouse. Unfortunately, physicians find these data-entry modalities inefficient and hard to use. Not only do they take away valuable time from direct patient care, but many also complain that the resulting data sets do not fully observe the natural flow of clinical thinking. Conversely, narrative-based documentation methods are viewed as able to preserve detailed and expressive descriptions of patients and their stories and are commonly accepted as the best way to capture and arrange the informational background on which effective diagnostic reasoning is based.3
For these reasons, direct dictation of clinical narratives is still the preferred method to document a visit. For many years, speech recognition and computer-aided transcription systems have seamlessly supported physician dictation work flow. The final output of such systems is a textual clinical note.
Clinical language understanding bridges the gap
Clinical language understanding (CLU) is a developing technology that automatically captures problems and other key patient clinical data from dictated visit notes. This data is standardized and saved into the EMR and other clinical systems directly from dictation, saving physicians the time and tedium of entering data manually. As CLU becomes available for routine clinical use in hospitals and practices, the solution to the problem of creating and maintaining the problem lists will be more feasible.
The goal of clinical language understanding is to directly integrate and synchronize through language technologies constantly evolving clinical scenarios with their representation inside the EMR. For problem lists, as the CLU software extracts data from the note, these can be matched in real time with what's in, or missing from, the list. A number of scenarios can be supported by this paradigm. For instance, problems documented in the note but absent from the problem list can be automatically added, completing the documentation. Furthermore, by utilizing ICD-9 codes to normalize different terminologies, the system can decide if a more detailed description of a problem becomes available and update the problem list accordingly.
Consider this sentence: “The otitis media for which the patient was seen last month appears to be fully resolved.” CLU automatically and reliably assesses that the “otitis media” is “resolved” and thus should be removed from the list of current problems. Today, this action would require manual editing of the data. However, with CLU this happens automatically, with the physician confirming the deletion.
Documenting patient visits is complex and does not lend itself to discrete and structured data entry. A problem name or its status, how well the condition is controlled, the onset or remission of complications, or the history of recurrences and other attributes are better described in dictated notes. The increasing prevalence of chronic conditions that require complex clinical management makes only more urgent the need for information-rich, flexible documentation tools that can directly and intelligently interact with EMR applications. The purpose of CLU is to bridge the gap between dictated narratives and structured data in the EMR. As CLU applications find their way into commercial EMR systems, the bridge will finally be established to align the ways in which both physicians and EMRs document, represent and interpret clinical data.
1. DesRoches, CM et al. “Electronic Health Records in Ambulatory Care — A National Survey of Physicians,” New England Journal of Medicine 2008; 359, no. 1: 50—60.
2. Blue Cross Blue Shield of Massachusetts press release (October 28, 2008).
3. Gordon D. Schiff, David W. Bates. “Can Electronic Clinical Documentation Help Prevent Diagnostic Errors? New England Journal of Medicine 2010; 362:1066-1069.
How does CLU work?
As physicians dictate their clincal notes into the EHR using voice-recognition technology, CLU analyzes the narrative and automatically extracts pertinent clinical facts, such as problems, symptoms, severity, allergies and medications. This data is automatically normalized and coded into standards, such as ICD-9 and SNOMED.
The final result is akin to the physician taking the time to enter the data into the EHR using menus and pick lists. However, with CLU, physicians can continue to dictate efficiently, while preserving the patient clinical story and the clinical decision-making process that is so valuable in communicating the uniqueness of the patient conditions to other members of the clinical team. At the same time, the discrete and coded data is automatically available in the EHR to support and streamline other applications, such as clinical decision support tools, quality and regulatory reporting, trending and billing.
Davide Zaccagnini, M.D., is director of medical informatics, Nuance Communications.
For more information on Nuance Communications: