By Steve Oden, Senior Vice President of Products Operations, medCPU
Elevating overall quality in healthcare requires closing care gaps across all site-specific services. This undertaking requires IT’s key role of ensuring that information needed to close gaps is delivered uniformly at all points of care. With valuebased models placing greater emphasis on care quality — ultimately determining a care network’s financial performance — IT is leading the charge in clinical decision support (CDS) systems.
Some CDS systems combine evidencebased best practices with a patient profile derived from EMR discrete data elements alone, as capturing data from narrative free-text documentation is technically challenging. However, partial patient information is, in part, responsible for delivering too many inaccurate care alerts leading to alert fatigue. Improving alerts through better information acquisition is vital to improved patient care, workflow, and uniform quality expected from a CDS system.
So what is any given CDS system really doing if its technology fails to capture and accurately read vital information? For a system to excel, it must receive and precisely comprehend all relevant information from each EMR involved in delivering patient care, and create a complete patient profile capable of generating precise alerts.
Please log in or register below to read the full article.
By Steve Oden, Senior Vice President of Products Operations, medCPU
Elevating overall quality in healthcare requires closing care gaps across all site-specific services. This undertaking requires IT’s key role of ensuring that information needed to close gaps is delivered uniformly at all points of care. With valuebased models placing greater emphasis on care quality — ultimately determining a care network’s financial performance — IT is leading the charge in clinical decision support (CDS) systems.
Some CDS systems combine evidencebased best practices with a patient profile derived from EMR discrete data elements alone, as capturing data from narrative free-text documentation is technically challenging. However, partial patient information is, in part, responsible for delivering too many inaccurate care alerts leading to alert fatigue. Improving alerts through better information acquisition is vital to improved patient care, workflow, and uniform quality expected from a CDS system.
So what is any given CDS system really doing if its technology fails to capture and accurately read vital information? For a system to excel, it must receive and precisely comprehend all relevant information from each EMR involved in delivering patient care, and create a complete patient profile capable of generating precise alerts.
Challenges To Generating A Comprehensive And Accurate Patient Profile
CDS systems face two major challenges in generating an accurate patient picture:
- The inability to read and precisely gather and comprehend free-text or unstructured patient information entered by clinicians into EMR systems; and
- the absence of a unified and comprehensive source containing all relevant patient information across all departments and sites, including ER, labs, radiology, etc.
For a CDS system to truly do its job — including acquiring all relevant clinical data, accurately evaluating the clinical data in real-time, and producing highly accurate and meaningful alerts that truly affect the care delivered to the patient — the system’s technology must fully address these two challenges.
Finding The Hidden Data Within Narrative Documentation
The inaccuracies and lack of precision historically seen with CDS systems has been the limiting factor in CDS adoption, as operating with only a portion of insight into the patient care process is a serious shortcoming. Today’s advanced CDS systems address this by looking for the rest of the data within the narrative documentation that providers effectively use to document and describe conditions, diagnoses, and treatments. Attempting to shift these providers to more structure documentation techniques within the EMR have created provider dissatisfaction by slowing providers down, and actually can lead to a loss of critical information being documented.
Therefore, CDS systems must find a way to discover the hidden data within narrative documentation. Beginning with NLP technologies is the logical place to start, yet the accuracy and precision of this technology is at best around 70 percent or less. Advanced machine learning can improve on this technology to improve the understanding of what is actually being said within the documentation, enabling a refinement of NLP extractions. However, what ultimately is needed is a way to place what has been extracted from the free text into the clinical context of what we already know about the patient’s clinical care up to this point. The clinical context is needed to move the accuracy and precision of free-text data extraction into the range of 90 percent or better. Only then can the information hidden within the free text be reliable enough to inform effective CDS alerts.
How CDS Assembles A Complete Patient Picture
CDS systems should function as effective and efficient information aggregators, capturing data from throughout a healthcare network. Therefore, the CDS system must complete the patient data picture by combining the newly discovered data from the narrative free-text with the volumes of discrete data and information available.
The most common source of discrete and patient care data is the HL7 interfaces available from virtually all EMR systems. However, this data source can create a burden on the implementation of the CDS system. Therefore, a method of consuming and processing large volumes of interface data without labor-intensive and time-consuming interface integration efforts is needed. To achieve this important yet atypical outcome, the CDS system must reinvent the way systems collect data from HL7 interfaces. One such innovation would be to flip the HL7 data paradigm around and view this data as free-text rather than discrete data. Assuming the CDS system can extract data from narrative free-text as describe above, utilizing this advanced technology could create a method of acquiring data from traditional HL7 messages while bypassing the majority of the time–consuming, interface data-mapping efforts.
However, even with such a significant innovation, the CDS system will be missing a significant portion of the full clinical picture of a patient’s record. As traditional HL7 interfaces represent only 30 percent to 60 percent of the data needed for a complete clinical record, additional innovation is required to fill this data gap. CDS systems could attempt to interoperate directly with the EMR databases or via new interoperability methods such as Fast Healthcare Interoperability Resources (FHIR). FHIR is picking up momentum as a replacement for traditional HL7 interfaces; however, even with FHIR restful services, there will be limits on the data that is available to the CDS system. Attempting to partner with all of the EMR suppliers to design and develop direct database queries would be challenging and excessively time-consuming from both a business and technical perspective.
Therefore, the CDS system must provide an alternative method of filling the data gap. The alternative must provide real-time access to all relevant data elements that are not available from interface or direct-query solutions. This method must be flexible enough to work from the wide range of EMRs in use today. The CDS system must consider methods of extracting data in real-time as the healthcare professional is interacting with the EMR. This approach is dependent upon identifying a generally available technology that is compatible with a wide range of EMRs, and can deliver data as the end user enters it into the EMR without impacting the stability or performance of the EMR itself. An example of this type of game-changing technology is Microsoft Active Accessibility (MSAA), which provides a common application-programming interface at the Windows operating-system level that could creatively become the foundation of filling the CDS data gap.
The ability to combine real-time, direct data acquisition from the EMR as the clinician is entering data with real-time HL7 interface data, and to extract actionable data from narrative free-text, creates a data-acquisition architecture that can collect the complete clinical record needed for accurate and meaningful CDS alerts.
Bringing Together Information And Overcoming Interoperability Issues
A CDS system that can aggregate structured and unstructured data from systems throughout a care network is invaluable, but standardizing quality across an entire care network requires more. CDS systems must assimilate large quantities of accurate and up-to-date information in order to provide suggestions and alerts specific to a patient’s complete health picture based on all clinical data, current best practices and evidence-based literature without being cumbersome or interrupting clinical workflow. A CDS that addresses these challenges can contribute significantly to healthcare outcomes — and to the financial outcomes of healthcare organizations and networks.
