Representation of Pain Concepts and Terms in Existing Ontologies and Taxonomies

Dear Editor,

Pain is linguistically and conceptually imprecise. For example, low back pain is semantically categorized as both a disease and a symptom [1]. Pain is the leading cause of adult outpatient [2] and emergency department visits [3], impacting more than 100 million Americans at a cost of more than $600 billion annually [4]. The National Institutes of Health (NIH) has begun the process of organizing common data elements for pain research [5], but clinical data about pain lacks objective metrics, findings, and testability, and pain presents a greater challenge to electronic health record text mining than do many other disease states [6].

Pain is a branch of numerous research vocabularies, scattered throughout as an accessory to other diseases. Without standardized research and clinical language, pain data lack the structure to build toward effective clinical data mining and research data organization. In other specialties, studies of these factors have worked toward controlled vocabularies, developed via semi-supervised machine learning methods [7], and behavioral phenotyping [8], developed via electronic health record–embedded dimensional behavioral documentation.

Although expert consensus has developed taxonomic frameworks to dimensionally categorize pain conditions [9], these frameworks do not necessarily align with documentation generated in clinical encounters. The NIH’s Patient-Reported Outcomes Measurement Information System (PROMIS) measures pain domains [10]. These first steps work toward conceptual, not linguistic, interoperability. Building on this framework would allow a common language to analyze existing data, link with clinical trials, and develop clinical data acquisition templates.

Review of Existing Ontologies

A search in the BioPortal of the National Center for Biotechnology Information (NCBI) (http://bioportal.bioontology.org/) for the word “pain” returned results from 52 ontologies. Of these, 17 were selected for more extensive review on the basis of their use in clinical applications or likelihood of relevancy to pain research and treatment (Figure 1).

Figure 1.

Pain-specific terms in existing vocabularies.

The selected ontologies were reviewed with respect to commonly used metrics, including the size, maximum depth, and average and maximum number of data branches for each ontology, as well as the number and proportion of terms in the ontology related to the parent term “pain,” the organization of these concepts, and the availability of pain-related terms in categories such as anatomic descriptions of pain location. The size of the reviewed ontologies varied widely, from more than 300,000 terms in the Systematized Nomenclature of Medicine—Clinical Terms (SNOMED-CT) to only 124 in the Ontology for General Medical Science (OGMS). As expected, the larger vocabularies had more terms related to pain in general. However, this was not universally true. The Medical Subject Headings (MeSH) vocabularies, though second in size, had only 39 pain-related terms (0.01% of the vocabulary). A review of the 17 ontologies for terms relating to specific types of pain—i.e., terms related to the anatomic location of the experienced pain, the severity of the pain, or the type or description of the pain—revealed that all but one of the ontologies covered anatomic pain terms, such as back pain or knee pain. However, few covered pain severity or pain type (Figure 1).

Both the National Cancer Institute Thesaurus (NCIT) and SNOMED-CT have relatively large numbers of terms dealing with the anatomy, severity, and types of pain. However, the organization of these terms in the two vocabularies is not the same. The NCIT groups these terms together as direct outputs of the parent term “pain,” whereas in SNOMED-CT, these terms are divided among several subcategories under the overarching term “pain.” Anatomic terms are largely grouped under “pain finding at anatomic site.” Pain type terms are split between the “pain by sensation quality” (aching pain; burning pain) and “finding of pattern of pain” (acute pain; chronic pain) sub-branches. Terms that describe the severity of the pain, on the other hand, can be found both as direct outputs of “pain” (“mild pain”; “severe pain”) and in the “finding of present pain intensity” sub-branch (“mild present pain”; “excruciating present pain”).

Discussion

Our review found no existing primary vocabulary or classification comprehensively reflecting or integrating the complex biopsychosocial and functional components of pain conditions. Additionally, the existing vocabularies lack uniform coverage of concepts that reflect the documentation generated in pain clinical encounters. The integration of pain classification to include functional status metrics and pain interference dimensions will allow for more accurate assessment and categorization.

Genetics

To better question and understand the genetic basis of the pain experience and response, data need to be organized to allow for a more specific identification of gaps. Despite enormous efforts to understand the development and amplification of pain, variation among responses makes pain a candidate for genetic studies.

Data Standardization

To mine and identify pain information in electronic health records, data standardization is a critical foundational step. Allowing ongoing clinical data collection to align with research standards supports larger interoperable databases. By standardizing pain language, we create the infrastructure to build the connection between the clinical spheres and research domains.

Data Integration

One of the challenges of creating a multidimensional assessment of pain is the diverse, text-rich data sources. Pain is a subjective experience, and thus the plethora of descriptive terms used minimizes the ability to cross-link with numerical data. Common and standardized terminology allows for common language across reporting from multiple sources across clinical spheres. Integration of research and clinical data allows for a deeper understanding of the pragmatic experience of patients.

Ongoing Surveillance

Tracking treatment responses, such as opioid responses, and linking them to genetic data also are a critical part of evaluating interventions. This information can be incorporated into prediction models, clinical decision support programs, and genomic approaches to understand pain conditions.

Pain Condition Cross-Linking

Building this infrastructure and vocabulary for pain conditions will allow for cross-linking with numerous other clinical specialties in which pain is a secondary diagnosis impacting clinical care and quality of life. Ultimately, using this pain organization and treatment response data to inform and educate providers across specialties would be an important goal.

Conclusions

No current data acquisition and organization reflecting or integrating the complex biopsychosocial and functional components of pain conditions exists. By standardizing pain language, we could advance organizational, research, and clinical goals. Future applications include data standardization and integration, ongoing integration, genomic and pain condition cross-linking, and software and tool development.

 

Funding sources: Research reported in this publication was supported by the NIH National Institute of Biomedical Imaging and Bioengineering under grant number K08EB022631, by the National Institute of Drug Abuse under grant number R24DA055306, and by the NIH National Heart, Lung, and Blood Institute under grant number R01HL064541.

Conflicts of interest: The authors declare that they have no competing interests.

Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Authors’ contributions: MA conceived the project with MS and carried out the study with assistance of JRS, SJW, and MS.