The meaningful use-compliant electronic health record (EHR) has quickly become very adept at capturing and sharing standardized, structured clinical content that can be communicated, stored, and to some extent consumed by other systems. Unfortunately, this strength is also the EHR's greatest limitation. Amid the structured templates and required fields of the EHR, the essential critical knowledge a provider needs to know is often so obscured that the EHR becomes more of an obstacle or annoyance than a truly useful source of clinical information.
No Place for Clinician's Thought-Process?
The critical clinical insights that providers most need from an EHR are simply not available to allow for informed decision-making. The required fields may all be populated, but the patient's story remains frustratingly incomplete.
The reason for this is simple: by its very nature, the EHR paradigm of capturing clinical information by way of mouse-and-keyboard input into structured forms limits the expressiveness of content. Because there is no place for non-standard information or for the clinician's thought process in reaching certain diagnoses in the templates, we not only miss out on the details of a patient's clinical history, but also on the critical information that reflects the way doctors think.
Documentation of the rationale for conclusions, relevant temporal and sequential facts, causal information, etc. is either lost or obscured beyond efficient retrieval. Some EHRs have incorporated options to allow providers to capture unstructured narrative information, but the resulting text usually has limited utility since it remains unstructured data buried inside various notes fields.
This dilemma is significant. It will take more than incremental feature improvements to realize the promise of the EHR: to support everything from disease management to clinical decision support to major operational efficiencies. To deliver on the expectations for eHealth, we need the EHR not only to capture and effectively use structured data, but also to capture the full patient story and support clinical collaboration based on that story.
What is needed is collaborative intelligence, a solution that enables and supplements the kind of complete and focused clinical picture physicians convey via face-to-face collaboration. Providing such intelligence requires an understanding of clinical workflows, and an ecosystem of people, process and technology to provide the clinical insights that permit clinicians to zoom in on the most critical information quickly and effectively.
All of the pieces required for such collaborative intelligence are in place today: Recognition and understanding of spoken content, semantic coding and analysis to drive actions and learning algorithms that continuously improve the performance of automated systems based on human feedback. Four key technologies provide the backbone:
Speech Understanding: Speech is the most natural way for humans to convey complex information, and it is the preferred mode of clinical documentation for most physicians today. Speech-based documentation is fast and interferes with the provider-patient interaction least. Converting speech into structured clinical notes using computers reduces costs and time lag associated with human transcription.
The availability of next-generation speech understanding technology now provides significantly higher accuracies across medical disciplines and documentation types than what has previously been available through speech recognition systems. Integration with various clinical systems further optimizes the efficiency of the technology.
Natural Language Understanding (NLU): Sophisticated technology to "read" and understand unstructured clinical narrative is a critical ingredient for collaborative intelligence. We can now produce meaningful structured information from narrative content, merging the benefits of dictation and structured documentation.
Irrespective of whether clinical narrative is captured through dictation or directly in textual form, the synergistic combination of speech and natural language processing (NLP) technologies now yields highly accurate, context-aware clinical content that is codified to standardized medical ontologies such as SNOMED-CT. This in turn drives actionable information and together with structured EHR data enables clinical decision support and improves the quality of care.
Semantic Clinical Reasoning: Once meaningfully structured narrative information is available, it must be made accessible in workflow-friendly, flexible modes. Newly available tools allow physicians to gain access and insights into clinical data that were impossible to get a few years ago. Also, these tools make physicians more productive because they are capable of abstracting and summarizing the relevant clinical information for each provider. They can reason across millions of documents or drill down on the relevant information about one patient in a given context.
Information mined from narrative content can be combined with structured data from EHRs to obtain holistic insights into the patient's story. From retrospective analyses to real-time feedback for physicians at the time of documentation that enables more timely clinical documentation improvement (CDI) to the ability to share clinical insights among caregivers in a collaborative system, the fruits of this reasoning are game-changing.
Machine Learning: To realize the full scope of its benefits, a collaborative intelligence system must be both highly scalable and responsive to the incessant changes in medical knowledge. The only way to achieve these objectives is through "machine learning" - intelligent systems that improve their predictions as they process more information.
Many NLP systems lack a robust capability to do this or rely on hand-crafted rules for knowledge updates, an inherently non-scalable approach. Learning from human feedback is crucial as it provides a constant opportunity to adapt to the changing environment as well as to improve the results and insights gained from collaborative intelligence.
Taken together and combined in the right manner, these technologies and workflows offer the best path to fulfill the goals of eHealth. The EHR remains an essential tool for advancing the quality and efficiency of care, but all stakeholders in healthcare have to remember that it is far from a panacea. To reach the goals of complete, accurate and seamlessly interoperable clinical information, we need to take into account that the most complete, accurate and interoperable way of communicating clinical information is via the spoken word. It also happens to be the most efficient way of capturing such information.
Juergen Fritsch is the chief scientist of MedQuist. He was previously chief scientist and co-founder of M*Modal, and before that, he was one of the founders of Interactive Services (ISI), where he served as principal research scientist.