Assessing the Uniformity of Uveitis Clinical Concepts and Associated ICD-10 Codes Across Health Care Systems Sharing the Same Electronic Health Records System

Key Points

Question  Is the mapping of International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes to uveitis clinical concepts uniform across systems using the same electronic health records (EHR) system?

Findings  This survey study found that ICD-10 code mapping to uveitis terminology is highly uniform among centers with the Epic EHR system. However, longitudinal changes were observed, as were inadequacies of ICD-10 codes for the description of uveitic disease.

Meaning  This study suggests that rare discrepancies in ICD-10 code mapping among centers with the same EHR system exist; these differences may affect the accurate identification of patients with major anatomical categories of uveitis and complicate the design of big data studies in uveitis.

Importance  Big data studies may allow for the aggregation of patients with rare diseases such as uveitis to answer important clinical questions. Standardization of uveitis-related variables will be necessary, including the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes used to identify patients of interest. There are currently limited data on the uniformity of diagnosis mapping to ICD-10 codes for uveitis diagnoses among different health systems.

Objective  To assess the degree of uniformity in mapping of uveitis clinical concepts to ICD-10 codes across health care systems using the same electronic health record (EHR) system.

Design, Setting, and Participants  This multicenter survey study was conducted between September 14 and October 9, 2020, at 5 academic health care systems that use the Epic EHR. Researchers from the University of Washington, Harvard University, Stanford University, Yale University, and the University of California, San Francisco queried 54 uveitis-related diagnostic terms and recorded the associated ICD-10 codes.

Main Outcomes and Measures  The degree of uniformity for uveitis clinical concepts and associated ICD-10 codes.

Results  Fifty-four uveitis-related diagnostic terms were queried within the Epic EHR at 5 different health care systems. There was perfect agreement among all 5 centers for 52 of the 54 diagnostic terms. Two diagnostic terms had differences in ICD-10 coding: juvenile idiopathic arthritis associated chronic uveitis and intermediate uveitis. Intermediate uveitis was associated with codes H20.1x (ICD-10 description: chronic iridocyclitis) or H20.9 (ICD-10 description: unspecified iridocyclitis) in 3 centers while being associated with code H30.2x (ICD-10 description: posterior cyclitis) at the 2 remaining centers. The discrepancies appear to be related to a recent update in diagnostic mapping in the Epic EHR.

Conclusions and Relevance  This study suggests that ICD-10 code mapping to uveitis diagnostic terminology appears to be highly uniform at different centers with the Epic EHR. However, temporal changes in diagnosis mapping to ICD-10 codes and a lack of 1-to-1 mapping of diagnosis to ICD-10 code add additional sources of complexity to the interpretation of big data studies in uveitis.

Uveitis encompasses a diverse group of rare clinical entities.¹ Its rarity poses challenges for performing traditional clinical research and reaching consensus on characteristic disease features.² Increased availability of large clinical databases from electronic health records (EHRs) may allow for the aggregation of patients with such rare diseases to answer important clinical questions.³ However, the reliability of these databases may be compromised by sources of error, including missing or inaccurate data,^4,5 inconsistencies in physician coding⁶ or documentation, and limitations inherent to observational data.⁵ Accurately identifying patients with uveitis from health care databases and validating the accuracy of clinical data rely on standardizing uveitis-specific variables,³ including International Classification of Diseases (ICD) codes, which are frequently used for patient selection.^7-9

The process by which ICD codes are assigned to patients relies on the mapping of medical terminology provided by clinicians to ICD codes via a mapping dictionary embedded in the EHR. To meet EHR meaningful use requirements set forth by the Centers for Medicare & Medicaid Services, clinical problems must be described using Systematized Nomenclature of Medicine—Clinical Terms (SNOMED CT),¹⁰ which provides comprehensive clinical terminology for EHR systems. As of October 2015, EHR diagnoses must use International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes for billing purposes. Mapping of SNOMED CT clinical terminology to ICD-10 codes is provided by the National Library of Medicine¹¹ and used by EHRs to generate ICD-10 codes from medical problems. Electronic health records systems may also use third-party vendors such as Intelligent Medical Objects (Intelligent Medical Objects Inc) to provide additional terminology and mapping.¹² Owing to differences in granularity between these systems, a 1-to-1 map is not always possible.¹¹ Multiple SNOMED CT clinical concepts may converge onto a single ICD-10 code, and multiple ICD-10 codes may be required to capture a single clinical concept. Therefore, the ICD-10 code assigned to a patient’s medical problem may be consistent with, but not specific for, the clinician’s intention.

The intricacies of clinical concept mapping to ICD-10 codes may give rise to several sources of error resulting in limited reliability of patient characterization based on ICD codes. Previous studies using the International Classification of Diseases, Ninth Revision (ICD-9) showed that many uveitis ICD codes had low positive predictive value for accurate diagnosis.¹³ With the ICD-10, Palestine et al¹⁴ demonstrated issues of etiologic, prognostic, and anatomic nonspecificity, as well as variability in ICD-10 code mapping for uveitis diagnoses between different EHR systems. This issue highlights the potential for misclassification when ICD-10 codes are used for research studies. Ultimately, machine learning algorithms for structured and unstructured data extraction may allow for automated interpretation of medical records, including patient diagnosis. Until then, it will be important to standardize uveitis-specific clinical variables, including ICD codes.

Currently, it is unknown whether uveitis ICD-10 code mapping varies within the same EHR, which may be important for understanding the limitations of database research in uveitis. We sought to assess the degree of uniformity in mapping of uveitis diagnoses to ICD-10 codes among institutions using the same EHR (Epic EHR; Epic Systems Corporation).

Researchers from ophthalmology departments of 5 institutions that use the Epic EHR system were invited to participate in the study: the University of Washington (UW); Harvard University Massachusetts Eye and Ear; Stanford University; Yale University; and the University of California, San Francisco. Study investigators recorded ICD-10 codes associated with 54 uveitis-related diagnostic terms (eTable 1 in the Supplement). A total of 54 uveitis diagnostic terms were adapted from previously published studies^2,14 and expanded to include synonyms and anatomic subtypes of those entities. These terms encompass a diverse anatomic and etiologic set of uveitic diseases described by uveitis health care professionals. One investigator (K.M.M.) recorded all ICD-10 codes associated with the diagnostic terms at each center via video conference. ICD-9 codes were not assessed in our study because the Epic diagnostic dictionary no longer reports ICD-9 code associations. For each center, data were collected within a single day except for the longitudinal analysis conducted at 2 centers (Yale University and UW) to explore observed mapping discrepancies for intermediate uveitis. At UW, intermediate uveitis was queried 1 additional time after the period of bulk data collection. At Yale University, intermediate uveitis was queried 1 additional time prior to the date of bulk data collection. For each ICD code, where indicated, the character representing laterality was replaced with “x” for ease of analysis. For example, H20.10, H20.11, H20.12, and H20.13 were recorded as H20.1x. Search terms associated with combination codes such as B25.9 and H30.9x for cytomegalovirus retinitis were recorded as (B25.9, H30.9x), as a paired result. Examples of data collection are shown in eFigure 1, 2, 3, and 4 in the Supplement. This study was reviewed by the institutional review board at the University of Washington and determined not to involve human participants; it therefore did not require exemption or approval. The tenets of the Declaration of Helsinki¹⁵ were followed.

The main outcome measure was uniformity of uveitis diagnostic mapping among surveyed centers. Uniformity was calculated for each pair of centers as the proportion of diagnoses with identical associations. Exact binomial 95% CIs for the proportion of codes with exact ICD-10 code agreement were calculated. If a diagnosis was associated with multiple ICD-10 codes and any were different between centers, that diagnosis was considered to be nonuniform. If a diagnosis was not associated with any ICD-10 code, a result of none was recorded but still included in the uniformity calculations. SNOMED CT identification numbers were obtained from the SNOMED CT browser (release: United States Edition 2020-09-01).¹⁶ Proportion calculations were performed in Excel (Microsoft Corp); exact binomial 95% CIs were computed in R, version 4.0.3 (R Group for Statistical Computing).¹⁷

Between September 14 and October 9, 2020, 54 uveitis-related diagnostic terms were queried within the Epic EHR at 5 different health care systems and the corresponding ICD-10 codes were evaluated (Table 1). The proportion of diagnoses with identical ICD-10 code mapping on pairwise comparison among all centers was greater than 96% (52 of the 54 diagnostic terms) (Table 2). Two pairs of centers had identical mapping for all searched terms. One center was not a perfect match to any of the other 4 centers.

Two diagnostic terms were responsible for the differences among centers: juvenile idiopathic arthritis associated chronic uveitis and intermediate uveitis. At each center, juvenile idiopathic arthritis–associated chronic uveitis was associated with a combination of 1 uveitis-related ICD-10 code and 1 juvenile idiopathic arthritis–related ICD-10 code. The uveitis-associated ICD-10 code was H20.1x (ICD-10 description: chronic iridocyclitis) at all the centers. However, the juvenile idiopathic arthritis code was M08.80 at 3 centers and M08.90 at 2 centers; both of these codes are consistent with juvenile arthritis. For intermediate uveitis, more variability was observed. At 3 centers, ICD-10 codes associated with intermediate uveitis were H20.1x (ICD-10 description: chronic iridocyclitis) and H20.9 (ICD-10 description: unspecified iridocyclitis). In contrast, H30.2x (ICD-10 description: posterior cyclitis) was associated with intermediate uveitis at the remaining 2 centers. Intermediate uveitis combination codes (a pair of ICD-10 codes associated with a single clinical concept) included an additional G35 code (ICD-10 description: multiple sclerosis) at all centers and B99.9 (ICD-10 description: unspecified infectious disease) at 2 of the 5 centers. At 1 center, B99.9 was used in isolation for intermediate uveitis.

A limited longitudinal assessment for the term intermediate uveitis was performed at 2 centers. In June 2020, prior to bulk data collection, intermediate uveitis was queried at Yale University and found to be associated with codes H20.9 and H20.1x, but it had been updated to H30.2x by the time of official data collection. At UW, intermediate uveitis was again queried after bulk data collection, and the ICD-10 association had been updated from codes H20.9 and H20.1x to code H30.2x.

Fifteen of 54 terms (28%) had no ICD-10 association at any surveyed center (eTable 2 in the Supplement), including anatomic subtypes and synonyms of other mapped terms. Four anatomic subtypes of sarcoidosis uveitis had no associated ICD-10 code, whereas both sarcoid uveitis and sarcoid iridocyclitis were associated with code D86.83 (ICD-10 description: sarcoid iridocyclitis), despite the difference in anatomic specificity of these terms. Furthermore, even though the 3 terms herpes zoster iridocyclitis, herpes zoster anterior uveitis, and herpes zoster uveitis were assigned to code B02.32 (ICD-10 description: zoster iridocyclitis), only herpes simplex iridocyclitis was associated with code B00.51 (ICD-10 description: herpesviral iridocyclitis). Other synonyms of herpesvirus anterior uveitis had no code association, highlighting an example of mapping asymmetry between 2 similar entities. Another commonly used diagnostic term with no corresponding ICD-10 code was punctate inner choroiditis, despite punctate inner choroidopathy being associated with code H31.8 (ICD-10 description: other specified disorders of choroid). Behçet uveitis, posterior placoid chorioretinitis, and serpiginous-like choroiditis had no associated ICD-10 code at any center. Of the 15 diagnostic terms with no associated ICD-10 codes, 4 (punctate inner choroiditis, Behçet uveitis, syphilitic posterior uveitis, and syphilitic panuveitis) were encoded within SNOMED CT, whereas 11 were not, demonstrating incomplete ICD-10 code mapping of existing SNOMED CT concepts in the Epic EHR.

Of 39 terms with at least 1 associated ICD-10 code, 18 (46%) were associated with a code shared by at least 1 other term. In 8 instances, overlapping diagnoses onto the same ICD-10 codes pertained to differing anatomic and etiologic entities (Table 3). Code H20.1x was associated with both anterior uveitis and intermediate uveitis at 3 centers, an example of anatomic overlap. Posterior uveitis, birdshot chorioretinopathy, and multiple evanescent white dot syndrome were associated with code H30.9x, an example of etiologic overlap. Serpiginous choroiditis and progressive outer retinal necrosis were associated with code H30.89x at all centers, an example of infectious and noninfectious overlap.

We surveyed 5 academic health care systems with the Epic EHR to assess the uniformity in mapping of uveitis diagnoses to ICD-10 codes. Our results suggest a high degree of uniformity in mapping across centers that use the Epic EHR. However, a critical difference in the ICD-10 code association for intermediate uveitis was observed between centers, which may affect the accurate identification of these patients in database studies. Furthermore, a substantial number of commonly used diagnostic terms were not mapped to specific ICD-10 codes in the Epic diagnostic dictionary. Our results highlight the potential limitations in identifying patients with uveitis accurately using EHRs for research.

Despite standardized mapping from clinical terminology (SNOMED CT) to ICD-10 codes provided by the National Library of Medicine, differences in mapping from uveitis diagnoses to ICD-10 codes have been demonstrated between different EHRs.¹⁴ Potential explanations include third-party vendors that provide additional terminology and ICD-10 associations to EHRs.¹² Thus, the same clinical diagnosis may result in different ICD-10 codes depending on the type and version of the EHR used. For example, a previous study demonstrated that intermediate uveitis was assigned to codes H20.1x or H20.9 in the Epic EHR while the term was assigned to code H30.2x in the MDIntellisys EHR.¹⁴ Furthermore, when querying intermediate uveitis using the Interactive Map-Assisted Generation of ICD Codes (I-Magic),¹⁸ a search engine that uses National Library of Medicine mapping of SNOMED CT to ICD-10 codes, the code H20.9 is generated, which is described as unspecified iridocyclitis rather than intermediate uveitis. This result illustrates how critical differences can exist not only between different EHRs but also between EHRs and official SNOMED CT to ICD-10 code mapping.

Database studies have increasingly used ICD codes to identify patients of interest.^7,9,19 With the rise of big data, studies involving rare diseases have become more accessible with the use of ICD codes. However, understanding the limitations of this approach is critical. Often, it is presumed that the clinical diagnosis and the linked ICD-10 code at the time of visit are reliable and uniform between all centers with the same EHR. Although the agreement among 5 surveyed institutions in our study was high, discrepancies were observed for 2 clinical entities. Differences pertaining to juvenile idiopathic arthritis chronic uveitis were due to a difference in the juvenile arthritis code rather than the eye code itself. On the other hand, important differences were observed for intermediate uveitis. At 3 of the centers, intermediate uveitis was associated with anterior uveitis codes (H20.1x or H20.9). At the 2 remaining centers, intermediate uveitis was associated with code H30.2x, known as posterior cyclitis, which is never used by uveitis specialists, as discussed by Palestine et al.¹⁴ This discrepancy has implications for the accurate identification of patients with intermediate uveitis in health care databases. Assigning either code H20.1x or code H20.9 to a patient with intermediate uveitis would render them indistinguishable from patients with other anterior uveitis diagnoses. Assigning code H30.2x to a patient with intermediate uveitis would overlap with posterior cyclitis and pars planitis, diagnoses with more etiologic similarity to intermediate uveitis, and likely a more appropriate association. Furthermore, posterior cyclitis is an antiquated term not used by uveitis specialists¹⁴ and is unlikely to be of interest in database studies or to generate significant bias. Similar nonspecificity for intermediate uveitis appears to exist for the ICD-9, potentially impacting conclusions that can be drawn from previous studies focused on anatomically specific uveitic disease.^7,19 Patients with intermediate uveitis assigned an anterior uveitis ICD-9 or ICD-10 code would be indistinguishable from patients with anterior uveitis. Code H30.2x (and its corollary in the ICD-9, 363.21) may have a higher specificity for intermediate uveitis owing to its association with pars planitis, but sensitivity may be low.

A July 2020 update to the Epic and Intelligent Medical Objects diagnostic dictionary likely accounts for the observed mapping discrepancies between centers and was consistent with the observed longitudinal changes at the 2 surveyed centers (Intelligent Medical Objects, written communication, September 2020). In this update, intermediate uveitis transitioned from codes H20.1x and H20.9 to code H30.2x, but we found mapping discrepancies of this term owing to the update not occurring at the same time between institutions who share Epic. Even though the mapping of uveitis diagnoses within Epic can change over time to improve limitations in previous iterations, these updates may affect major anatomical categories of uveitic disease,²⁰ often a critical component of clinical care and study questions. Efforts to improve the reliability of ophthalmic studies on administrative databases²¹ may need to track temporal changes in mapping. Specific to uveitis, identifying intermediate uveitis cases by ICD code may require a different strategy depending on when data were encoded.

Both ICD-10 and SNOMED CT are medical ontologies, or systems for codification of medical terminology. An ideal ontology for encoding uveitic disease would allow for the classification by disease cause, course, laterality, anatomy, morphology, infection, and systemic disease association.¹ Our study demonstrates an inadequacy in these systems to encode uveitis-related diagnoses. First, we observed incomplete mapping of clinical concepts to ICD-10 in Epic, including some diagnostic terms that are encoded in SNOMED CT. More than one-fourth of the diagnoses included in this survey had no associated ICD-10 codes. Although most of these diagnoses were also not encoded in SNOMED CT, some diagnoses, such as syphilitic posterior uveitis and syphilitic panuveitis, were an exception. In contrast, 5 clinical terms were associated with ICD-10 codes but were missing in SNOMED CT, demonstrating that even this ontology, inclusive of hundreds of thousands of clinical concepts, has incomplete coverage for description of uveitic disease. Finally, similar to previous results reported by Palestine et al,¹⁴ our results demonstrate that the same ICD-10 code may be applied to multiple diseases, which may lead to etiologic, anatomic, and infectious nonspecificity when interpreting ICD-10 codes to derive uveitic disease entities.

Quantifying disease misclassification associated with nonspecificity in code mapping is challenging. Although the cause of the discrepancy cannot be confirmed, major epidemiologic studies that have confirmed uveitis diagnosis by medical record review found that 3.4% to 6.3% of cases are associated with intermediate uveitis,^22-24 whereas 1 database study relying only on the pars planitis ICD-9 code estimated intermediate uveitis to account for less than 1% of cases.⁷ The coding inadequacies identified in this study suggest risk for nonrandom misclassifications that may be specific to a particular group of patients or diagnoses. For example, owing to the observed discrepancy in intermediate uveitis mapping, a patient with intermediate uveitis may be assigned code H30.2x at one center but code H20.1x at another, leading to a particularly high rate of misclassification for this entity. On the other hand, a diagnosis of Vogt-Koyanagi-Harada syndrome would be associated with code H20.82x at all centers; thus, the misclassification rate for this entity would likely be lower. Medical record review, the criterion standard for ensuring the accuracy of clinical data, is not possible with many large data sets. Therefore, knowing which diagnoses are prone to frequent misclassifications or errors in the EHR mapping and updating this knowledge periodically would be important for researchers.

Many sources of error beyond ICD mapping may lead to inaccurate identification of disease entities when using ICD codes in isolation, including inaccurate patient diagnosis and physician coding errors. Error analysis focuses on identifying and quantifying different sources of uncertainty in measurements and the propagation of this uncertainty.²⁵ Minimizing diagnostic misclassification and other sources of uncertainty is especially important for rare diseases such as uveitis, for which big data studies may be the only source of adequate patient numbers to draw meaningful conclusions. Increasing numbers of uveitis-related studies are using large databases, but few focus on identifying sources of uncertainty and how conclusions may be affected. Machine learning algorithms, such as those allowing for natural language processing for interpretation of unstructured data, may significantly improve our ability to extract reliable clinical data in lieu of medical record review or an overhaul of ICD coding. However, standardizations in uveitis diagnostic criteria and feature description will be required to train future algorithms.

This study has some limitations, including the small number of surveyed centers. Inclusion of other centers may reveal additional coding inconsistencies. However, we included a larger number of uveitic disease entities than previous studies, with additional anatomic subtypes and synonyms of those previously examined. In addition, we surveyed multiple centers for ICD code mapping, unlike an earlier, similar study that surveyed a single center with each EHR.¹¹ Our longitudinal data collection was limited in scope. A more extensive longitudinal review of disease associations may be both informative and necessary to understand the limitations of many health record–based studies. Finally, our analysis was limited to the ICD-10 because ICD-9 code associations are no longer reported in the Epic diagnostic dictionary.

Our study’s findings support rare differences in the mapping of uveitis diagnostic terminology to ICD-10 codes, which may affect the ICD-10 codes associated with intermediate uveitis. In addition, mapping of clinical concepts to ICD codes appears to change with time. Although the ICD coding system was not designed for capturing clinical concepts or documenting clinical care,¹² it is increasingly being used for the identification of specific diseases in research using large clinical databases. These findings suggest that efforts should be considered to standardize mapping, track and disseminate information regarding impactful changes with time, and improve precision and coverage for uveitic conditions. Mapping of ICD-10 codes to uveitis diagnostic terminology appears to be highly uniform at different centers using the Epic EHR. However, temporal changes in diagnosis mapping to ICD-10 codes and a lack of 1-to-1 mapping of diagnosis to ICD-10 codes add additional sources of complexity to the interpretation of EHR-based studies of patients with uveitis.

Accepted for Publication: April 19, 2021.

Published Online: July 1, 2021. doi:10.1001/jamaophthalmol.2021.2045

Corresponding Author: Cecilia S. Lee, MD, MS, Department of Ophthalmology, University of Washington, 325 Ninth Ave, Box 359608, Seattle, WA 98122 (leecs2@uw.edu).

Author Contributions: Dr C. S. Lee had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: McKay, Pepple, C. S. Lee.

Acquisition, analysis, or interpretation of data: McKay, Apostolopoulos, Dahrouj, Nguyen, Reddy, Blazes, Lacy, A. Y. Lee, C. S. Lee.

Drafting of the manuscript: McKay.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: McKay, Dahrouj, Lacy, C. S. Lee.

Obtained funding: Pepple.

Administrative, technical, or material support: McKay, Nguyen, Reddy, Blazes.

Supervision: Pepple, A. Y. Lee, C. S. Lee.

Conflict of Interest Disclosures: Dr A. Y. Lee reported receiving grants from Santen and Carl Zeiss Meditec; personal fees from the US Food and Drug Administration, Genentech, Gyroscope, and Verana Health; and nonfinancial support from Topcon and Microsoft outside the submitted work. Dr C. S. Lee reported receiving grants from the National Institute on Aging outside the submitted work. No other disclosures were reported.

Funding/Support: This study was supported by the Cynthia and Joseph Gensheimer Fellowship, Latham Vision Research Innovation Award, grant K23EY029246 from the National Eye Institute, and grant R01AG060942 from the National Institute on Aging.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: This article does not reflect the opinions of the US Food and Drug Administration.