problematic. The “terminology problem” can be characterized by several interrelated issues including: 1) the development of multiple standardized clinical terminologies, 2) inadequacies related to computers not understanding complex human language, and 3) minimal policy initiatives or federal incentives to drive adoption.
21 2.4.1 Multiple Standardized Clinical Terminologies
The first problem related to operationalizing standardized terminologies is associated with the simultaneous development of multiple, overlapping terminologies, created by numerous organizations and groups (Hardiker et al., 2011). For example, the American Nursing
Association recognizes ten different standardized terminologies, seven of which are nursing specific (Kim & Matney, 2014). However, the paradox of quantity is that the proliferation of these terminology systems has not yet produced one singular terminology that is comprehensive enough to cover the extensive health care needs of its multidisciplinary users (Cimino, 1998;
Hammond et al., 2014; Kim & Matney, 2014). Another paradox is that the quantity and overlap of these terminologies perpetuates the problem of coded data which are not standardized.
As an effort to mediate this aspect of the “terminology problem”, international
organizations, researchers, health IT professionals and clinicians are working to laterally cross-map and harmonize standardized terminology systems. For example, SNOMED-CT has had the ICD standardized terminology and a number of nursing terminology systems (Clinical Care Classification; ICNP; North American Nursing Diagnosis Association International (NANDA-I);
Nursing Interventions Classification; Nursing Outcomes Classification; Omaha System;
Perioperative Nursing Data Set) cross-mapped to its comprehensive clinical terminology sets (Kim & Matney, 2014). This is an essential practice to ensure the meaning of exchanged information is preserved and the precise meaning is not lost when interpreted by different health care professionals (Kim & Matney, 2014).
22 2.4.2 Human-Computer Concept Interpretation
A second contributor to the “terminology problem” is the juxtaposition of human concept interpretation and computer concept interpretation within an EHR (Hardiker et al., 2011). The problem being that humans create concepts based on natural language, full of diverse, semantic, contextual, and equivalent meanings; computers cannot readily understand these especially when clinical terms are intended to be re-used for interoperability, decision support and data extraction (Hardiker et al., 2011; Kim & Matney, 2014).
But herein lies the connection to the first problem: not enough concepts are available in one standardized terminology to describe everything for everyone in health care (Rosenbloom et al., 2006). An idea to mediate this could be to keep adding concepts and recognize them as unique and distinct; this is the exercise of “pre-coordination” or “enumerative terminology”
(Coiera, 2015; Rosenbloom et al., 2006). Further, to aid functionality, a concept could be tagged with several synonyms and employed when developing the user interface for an EHR (Kim &
Matney, 2014). From the discussion above, a pre-coordinated concept could be “postural hypotension” and as such, would be given all the necessary attributes required in a standardized terminology. However, if organizations continually add to their existing terminology system without oversight and knowledge of pre-existing concepts, the result can lead to an assortment of terms poorly organized with troublesome granularity, clarity, and redundancy (Cimino, 1998;
Coiera, 2015). When Kim, Hardiker, & Coenen (2014) mapped ICNP to SNOMED-CT, lack of clarity was cited between the two systems. In one instance, they found that the SNOMED-CT concepts, “abnormal behaviour (finding)” and “problem behaviour (finding)” both could be mapped to the one ICNP concept of “negative behaviour”.
23 As an alternate idea, one could use existing concepts and combine them to produce a new clinical concept that is not available in a pre-coordinated format; this is the exercise of “post-coordination” or “compositional terminology” (Coiera, 2015; Rosenbloom et al., 2006). Post-coordination occurs when complex concepts with varying levels of detail are described using more fundamental ones (also known as atomic concepts) (Rosenbloom et al., 2006). Again from the discussion above, “postural hypotension” could be dynamically coded within an EHR using the separate concepts of “posture” and “hypotension.” There is an inherent risk however, with this practice; Consequences of post-coordination include: i) difficulty restricting an EHR to medically meaningful concepts, ii) the potential for unrecognized duplicate entries, iii) inefficiency composing complex concepts from simpler ones, and iv) the burden to look up several unrelated concepts from distinct multiple sub-class lists (Hardiker et al., 2002; Lee, Cornet, Lau, & de Keizer, 2013; Rosenbloom et al., 2006).
Keeping in mind the risks of pre- and post-coordination, both practices are fundamental when translating clinical need to computer need. In his seminal paper, Cimino (1998)
recognized the importance of terminology systems having the capacity to increase the number of concepts it included, but stressed the significance of formal, reproducible, and systematic
methods of adding content. These early realizations have become today’s standards, where the quality of health care terminology structure, content, mapping, and its life cycle have been addressed by the International Organization for Standardization (ISO) (Kim & Matney, 2014).
As one example of an ISO standard related to standardized terminology, ISO 18104:2014 defines the categorical structure methodology for nursing diagnosis and nursing action, which developers and clinical infomaticians use when adding and developing content for standardized nursing terminologies (International Standards Organization, 2014).
24 2.4.3 Limited Usage of Standardized Clinical Terminologies
A third challenge contributing to the “terminology problem” is the limited uptake in knowledge, clinical priority, health policy, and research available on the implementation of standardized clinical terminologies into existing EHRs (Westra et al., 2008). Historically, items that have been systematically coded are related to concepts associated with billing and
reimbursement and not specific to clinical usage (Kim & Matney, 2014). Despite 40 years of defined, coded, and available nursing terminology systems, the wide integration of such standards into EHR systems available for purchase (i.e., vendor systems) has yet to be realized (Westra et al., 2015). Associated with this challenge is the reality that the language nurses use to describe practice continues to be inconsistent across and within health care settings (Technology Informatics Guiding Education Reform, 2009). This leaves many nursing-sensitive patient care indicators non-standardized, limiting the ability to capture nurses’ work and to share and
compare information as patients move through the health care system (Chow et al., 2015).
Though there has been work to increase the availability of better data to create knowledge from aggregated EHRs, clinicians still do not fully realize how it can be used to support their practice and their organizations (Canadian Health Outcomes for Better Information and Care, 2015). Often, clinicians view the incorporation of standards in documentation as a bureaucratic process, suggesting that on-going education and engagement are necessary to ensure the
information produced is of value and utilized to inform practice (Canadian Health Outcomes for Better Information and Care, 2015). In an integrative review of the literature, Stallinga et al.
(2015) found that health care professionals were not concerned when ambiguous language was used to document patient care. They suggest this could be the reason why the implementation of
25 standardized terminologies have failed and that these systems will only be successful when data can be meaningfully reused in areas such as decision-making and research.
To add another dimension, even for those organizations who have an agreed upon set of standardized nursing clinical charting parameters, and intend to operationalize a standardized clinical terminology to codify nursing care in their EHR, the methods which to map these existing parameters is diverse with no clear agreed-upon gold-standard (Monsen et al., 2016;
Harris et al., 2015; Kim et al., 2014; Richesson, Andrews, & Krischer, 2006). For example, manual mapping is often cited as a method for searching through a standardized clinical terminology; however, in one study, researchers have shown graduate prepared clinical nurses only had a 57.1% success rate when manually mapping a nursing clinical concept to SNOMED-CT (when compared to a pre-existing inter-terminology mapping list) (Monsen et al, 2016).
Further, in another study, when two researchers were mapping nursing clinical concepts from ICNP to SNOMED-CT, each individually manually mapped 436 concepts, and between the two had a kappa value of 0.45 (moderate agreement) when compared. These few examples suggest that manual mapping may be inconsistent; however, necessary when exploring concept
representation, not otherwise known. Other studies have attempted to use automated or semi-automated measures to map existing clinical concepts to standardized clinical terminologies (also known as candidate mapping). For example, inter-terminology mapping exercises might start with UMLS as a tool to investigate previous laterally mapped concepts as potential matches (Kim, 2016; Kim et al., 2014). Yet, the generated result of these automated or semi-automated methods is often insufficient, requiring another method (e.g., manual mapping) to fill in the missing matches (Saitwal et al, 2012; Lau, Simkus, & Lee, 2008; Richesson et al., 2006).
26 Though these nuanced differences in mapping methods may seem trivial to some (after all, they do result in coded data elements matched to the desired standardized clinical terminology), the outcome may have a significant patient safety and quality care impact which needs to be addressed and clearly understood before the large scale implementation of standardized clinical terminologies are pushed out in an EHR. Specifically, this relates to the actualized benefits of the codified standardization process itself; semantic interoperability, data aggregation and decision support. These are threatened if dissociated and disparate mapping methods are employed without oversight by organizations tasked with integrating these standardized terminologies within existing EHRs. To illustrate this, imagine if two different organizations decided to include the concept “dizziness when standing up” in their EHR. One mapped this to the SNOMED-CT concept Orthostatic Hypotension (disorder) 28651003, the other mapped it to Dizzy Spells (finding) 315018008. What would happen if the same patient was treated at
different times for this condition, in each organization? Would their documentation be coded in SNOMED-CT as Orthostatic Hypotension (disorder) or Dizzy Spells (finding)? What would happen if we tried to employ semantic interoperability, data aggregation, or decision support networks on what we thought was the same concept, but coded differently? The scope of this paper does not expand on the outcome of dissociated cross-organizational mapping; however, it would be an important consideration for future research.
Another aspect to the uptake of standardized clinical terminologies relates to the
engagement of vendors whose proprietary applications are (or are not) built with the capacity and standards adoption to support sematic interoperability (Harris et al., 2015). For example, when free-text, natural language is the main form of clinical documentation, the content becomes largely inaccessible to manipulate and use with decision support tools or statistical research
27 (Hardiker et al., 2002). Clinical information systems need to be designed in such a way that data are placed in a meaningful context and ready to be reused without the need for manual
transformation and manipulation (Lenz, Beyer, & Kuhn, 2007). Some of these challenges may be solved by the use of natural language processing in which free text is analyzed and mapped to a codified clinically equivalent concept (Topaz et al, 2016); however, this is yet to readily
available through commercial clinical EHR vendors. Another reflection on this limitation requires a view of current marketplace demands; currently, there are no Canadian federal governmental policies requiring health care organizations to choose EHRs with the built in standards and structure to facilitate sematic interoperability. In BC, the Ministry of Health (2014) released a cross sector policy discussion paper, Enabling Effective, Quality Population and Patient-Centered Care: A Provincial Strategy for Health Information Management and Technology, recognizing the need to improve data sharing throughout the health care system.
Though provisions and plans are underway to drive foundational information standards, the scope and scale of this work is still in the early phases of development in BC.