Finding buried genetic test results in the electronic health record is inefficient and variable across institutions

Abstract

Background:

The absence of standardized approaches for handling genetic test results in electronic health records (EHRs), combined with a lack of diagnostic codes for most rare disorders, hinders accurate and timely identification of patients with rare genetic variants. This impedes access to research opportunities and genomic-driven care. To reduce the diagnostic odyssey, identify research-eligible subjects, and ultimately enhance patient care, it is critical to optimize approaches to retrieve genetic results.

Objectives:

To characterize resource requirements, yield, and biases among methods for identifying and retrieving genetic test results across 11 Intellectual and Developmental Disability Research Centers (IDDRC).

Design:

A survey was used to collect details from the authors on approaches to identify EHRs from patients who had genetic testing and variants of interest were reported; surveys were completed in 2022.

Methods:

Strengths and limitations in approaches to identify and retrieve genetic test results conducted from the implementation of EHR systems were evaluated. A standard template was used to collect genetic testing storage formats, methods to identify patients with rare disease variants, estimates of time/cost, nature of accessed data, method-specific bias in types of American College of Medical Genetics and Genomics classified variants identified. When possible, precision when performing gene name searches in the EHR was calculated.

Results:

Four approaches were used: (1) manual searches, reviews, and extractions, (2) natural language processing software-aided manual reviews and extractions, (3) custom databases via testing lab collaborations, and (4) testing EHR vendor-designed genomics modules. The fully manual approach required minimal infrastructure and allowed access to clinical notes but missed variants of unknown clinical significance. Precision for gene name matches based on searches of 59 genes was 0.16. Natural language processing software minimized effort but required considerable informatics support. Custom databases and EHR vendor modules necessitated substantial computational support; however, genetic testing results retrieval was efficient.

Conclusion:

Leveraging the IDDRC network, we found that methods to store, search and extract genetic testing results vary widely, especially regarding older test results, and have distinct benefits and limitations. Limitations are best addressed through practice guidelines that standardize storage and retrieval of genetic test results to facilitate efficient identification of research eligible subjects and genomic-informed patient care.

Plain language summary

Differences in finding genetic test results in the EHR between institutions

This study addresses challenges in managing genetic test results in electronic health records (EHRs), which can hinder research and access to personalized care for patients with rare genetic variants. Eleven centers specializing in intellectual and developmental disabilities reviewed different methods of extracting genetic data from EHRs, focusing on tests conducted over several years. Four approaches were used: 1) manual searches, 2) using natural language processing with manual reviews, 3) custom databases in collaboration with testing labs, and 4) EHR modules designed for genomics. Each method had its advantages and challenges. Manual searches were simple but missed important data; natural language processing required technical support but reduced manual work; custom databases and EHR modules were more efficient but needed more resources. The study highlights the need for standardized guidelines to improve how genetic test results are stored and accessed, ensuring better research opportunities and care for patients with genetic conditions.

Keywords

biomedical informatics electronic health records genetics

Introduction

Genomic medicine is rapidly becoming integrated into clinical care across medical specialties, as sequencing becomes more accessible and associations between genetic variation and human disease are elucidated.^1–5 The identification of clinically relevant variants may accelerate diagnosis, inform treatment, guide genetic counseling, and confer clinical trial eligibility. For individuals touched by rare genetic disorders, rapid recruitment into registries and natural history studies can provide the foundations for therapeutic development. While the field of genetic medicine has made enormous progress over the last two decades, the historic lack of systematic integration of genetic testing results into electronic health record (EHR) systems means that the full potential of these advances is not being realized.⁶

Several factors hinder the efficient identification and retrieval of genetic testing results integrated into EHR systems, limiting the utility of a genetic diagnosis for subsequent clinical care and research opportunities.⁷ Genetic test results were historically—and are often still—stored in the EHR as scanned images or PDFs. Results in these formats are not structured or mapped to standard clinical vocabularies, reducing their usefulness in searches for research-eligible patients, clinical decision support trials and timely notification of variant reclassification or result actionability as the knowledgebase grows. The recognition of these obstacles to delivery of precision medicine has inspired a number of initiatives aimed at converting data related to genetic variants from standardized nomenclature which requires human readability (e.g., Human Genome Variation Society (HGVS)),⁸ to computational terminology (e.g., Global Alliance for Genomics and Health (GA4GH) Variation Representation Specification (VRS),⁹ and Fast Healthcare Interoperability Resource (FHIR)¹⁰) to allow for direct communication among various electronic platforms. Pilot projects to launch frameworks such as these in both institutional and cross-institutional settings have highlighted the need for significant resources, leadership, and engagement with participating laboratories in order for such initiatives to succeed.^11,12 Some institutions are beginning to implement systems that structure entry of genetic test results into the EHR; this has the potential to reduce future barriers to retrieving these results.^13,14 Notably, important considerations include ensuring genetic test vendors standardize result output, providing technical training for physicians and researchers, and careful attention to the sensitivity, privacy and need for accompanying clinical interpretation of genetic variants as recommended by the American College of Medical Genetics and Genomics (ACMG) guidelines.⁶

While these substantial efforts will inform the storage and retrieval of future results, considerable barriers remain impeding the ability of researchers and healthcare providers to efficiently retrieve retrospective (i.e., legacy) data from genetic testing whose results have been stored in unstructured formats since the advent of EHR systems. This limits the ability to identify patients for whom clinically relevant variants have already been identified who could benefit research participant recruitment. Further, identifying ways to reduce the effort of retrieving legacy data will allow investigators to track changes in variant classifications over time to better understand how often reanalysis or further testing should be conducted for patients with uncertain findings reported as having variants of uncertain significance—especially as variant classifications are informed by larger and more diverse samples of individuals undergoing testing.

In this effort, we leveraged an existing research network with a diverse range of genetic testing laboratory providers, EHR structure, and informatic capabilities to characterize inter-institutional variability in accessing genetic test results. Funded by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), the Intellectual and Developmental Disability Research Centers (IDDRCs) are a network of research institutions that collaborate in translational research efforts to understand the causes of Intellectual and Developmental Disabilities (IDDs) and to develop effective therapeutics to improve the quality of life for individuals affected by IDDs.¹⁵ Issues related to using EHR-derived data to efficiently identify potential research participants drove this network to establish the Genetic Reports: Enhancing Access to Test results in the Electronic Health Record (GREATest EHR) working group. The mission of this group was to characterize the ways in which genetic test results are stored and subsequently retrieved in the EHR across institutions, and in doing so, to establish a foundation for collectively working toward improvement and harmonization. While the focus of this study was rare genetic IDDs, the learnings are relevant across all disciplines in which genomic medicine is being integrated.

Methods

Eleven IDDRC institutions (Supplemental Figure 1) systematically compared the variability in methods and yield in identifying individuals eligible for research based on the presence of a clinically identified rare genetic variant. Of these 11 institutions, nine were recruiting sites for the National Brain Gene Registry (BGR; https://braingeneregistry.wustl.edu/),¹⁶ an IDDRC network initiative for which eligibility is determined by the presence of a clinical variant classified as a variant of uncertain significance (VUS) or pathogenic/likely pathogenic according to ACMG criteria on clinical genetic testing.¹⁷ The challenges of identifying eligible participants for the BGR informed this qualitative study. Each site assembled an investigative team that included clinical genetics and biomedical informatics experts. A structured template (Supplemental Figure 2) was used and completed by the authors of this study, to collect (1) details regarding the format of genomic testing results stored within each institution’s EHR system (e.g., clinical notes, scanned PDFs); (2) how patients with rare variants are identified with estimates of effort (e.g., person hours, cost, and level of informatic service needs); (3) the nature of accessed results (e.g., identified with protected health information (PHI) vs de-identified with PHI removed; cloud-based vs local), and (4) which class(es) of ACMG classified variants (e.g., variants of unknown clinical significance (VUS), pathogenic, likely pathogenic)¹⁷ were more readily identified. Surveys were completed between March 2022 and April 2022 and included current information for each participating site at this time.

In addition, the primary site for the BGR—which utilized the fully manual method—collected data regarding the number of true positives and false positives when searching for gene name matches in EHRs for 59 genes relevant to IDDs. The precision for identifying records of individuals with a genetic testing result relevant to genes of interest using this approach was calculated by determining the mean for true positives and false positives across all genes.

Results

Review of structured responses provided by participating sites revealed that approaches for identifying individuals with specific genomic testing results fell into four main categories (Figure 1). The most frequent method (“Method 1”) used by 8 of the 11 sites was fully manual chart reviews and information retrieval. In Method 1, clinicians and/or researchers manually searched for text within the records to identify character strings that matched a gene name of interest. For any record with a potential gene name match, the investigator manually reviewed all records. From records with a confirmed gene name match the investigator manually retrieved genetic testing results of interest and transposed this information in a separate database. An illustrative example of Method 1 is provided in Figure 2. All sites using Method 1 accessed structured (e.g., diagnostic codes, procedure codes.), semistructured (e.g., clinical laboratory values), and unstructured (e.g., free text clinical notes) EHR data. Most sites using this approach searched within identified EHRs stored locally on internal servers (7/8). Some sites (2/8) accessed EHR-derived clinical information via de-identified databases which had been modified to protect privacy of information (e.g., no full names, skewed birthdates, limited access to many scanned documents). Notably, most sites (6/8) indicated that retrieved results were biased toward capturing ACMG-classified pathogenic or likely pathogenic genetic variants (i.e., non-VUS).¹⁷ While all sites using the manual approach reported minimal computational infrastructure requirements, this was offset by the enormous amount of personnel effort, time, and skill required to manually review records and ensure accurate collection of genetic testing results. For the primary BGR site using this approach, the average number of true positives when searching for 59 different IDD-related gene names was 2.56 ± 13.88. The average number of false positives was 4.84 ± 73.14, resulting in a precision of 0.16.

Figure 1.

Proportion of sites reporting information regarding approaches used to retrieve genetic testing results.

Figure 2.

Example approach to initially query a de-identified version of data extracted from an EHR system using structured data elements, followed by manual reviews and extraction of genetic testing results from the identified EHR to identify potentially eligible participants for a research study.

The second approach (“Method 2”) currently used by 1 of the 11 sites was to first scan records using commercially available natural language processing (NLP) software prior to manual review. This software, Linguamatics I2E (https://www.linguamatics.com/), was previously trained to accurately (F1 scores = 0.75–0.84) identify phenotypic concepts within strings of text and map these identified concepts to Human Phenotype Ontology terms, increasing the likelihood of positive gene matches (see Figure 3 for illustrative example of approach). For more details on training and performance of Linguamatics I2E software please see Trivedi et al., 2020.¹⁸ The site using this approach accessed both structured and unstructured data stored in an identified local database. In contrast to Method 1, no perceived bias was reported in the types of ACMG classifications retrieved by this method. Although substantial computational infrastructure was required for Method 2—including purchasing of software licenses—using NLP software to evaluate medical records was noted to reduce the effort required for subsequent manual reviews as investigators only evaluated a subset of parsed records that were more likely to contain Human Phenotype Ontology mapped terms relevant to IDDs.

Figure 3.

Example approach to extract genetic testing results as aided by natural language processing from the identified EHR at the University of Iowa.

The third approach (“Method 3”) was used by four of the 11 sites. Method 3 involved searching for patients with specific genetic testing results stored in a custom database external to the EHR. See Figure 4 for illustrative example and Rockowitz et al., 2020¹⁹ for more details on this approach. These sites established agreements with commercial genetic testing laboratories to have test results returned to the ordering institution in a structured format that could be quickly and efficiently uploaded into the custom database. This approach allowed sites direct access to organized structured data elements, but not free-text clinical notes. All sites using this approach (4/4) reported no evidence of bias in the ACMG variant classifications identified. Although substantial computational infrastructure was required to design and maintain the custom databases, users of this method reported considerable benefits from the relative ease of result retrieval. Further, encrypted files sent by the genetic testing laboratory that mapped custom database IDs to patient names and dates of birth allowed users to access records for specific patients with variants of interest.

Figure 4.

Example approach to store genetic testing results that have been stored in a custom database that is not directly linked to the EHR at BCH.

The newest and least tested approach (“Method 4”) which is in the process of being implemented at 4 of the 11 sites involved testing the use of custom modules designed by EHR vendors to structure entry of genetic test results into the EHR. Although these modules are still being tested and have yet to be officially deployed at each of the testing sites, this approach by design helped ensure that results from genetic testing ordered after module implementation were readily retrieved in a structured format. None of the sites testing this approach indicated any bias in the types of variants (i.e., ACMG classified or VUS) that could be retrieved to identify patients of interest for clinical and research purposes.

Additional benefits and limitations described for each of these four overarching methods are detailed in Table 1. A benefit of Methods 1 and 2 over Methods 3 and 4 was the ability to more readily extract retrospective, or “legacy,” variant information collected dating back to the establishment of institutional EHR systems. Method 3 had the benefit of easy de-identification of data useful for research-related purposes—especially for large inter-institutional collaborative projects like those conducted by the IDDRC network. Disadvantages of Methods 3 and 4 included the extensive informatics support required to design and establish the necessary infrastructure. Finally, a limitation of these two approaches compared to Methods 1 and 2 was the need for close integration with genetic testing laboratories since results were either transmitted in a format to allow efficient uploads into the custom database or mapped to internationally accepted standards for integration with institutional EHR systems.

Table 1.

Benefits and limitations of methods used to retrieve genetic testing results in the GREATest EHR network.

Method (n_{Sites Using Method}/n_{All Sites})	Method 1Fully manual(8/11)	Method 2NLP + manual(1/11)	Method 3Custom database(4/11)
Benefits	• Retrospective genetic testing results accessible		• Structured entry allowing more accurate retrieval of genetic variants
	• Comprehensive data extraction	• Reduces manual review and identifies patterns not apparent via fully manual reviews	• PHI easily removed
Limitations	• Difficult to access scanned documents (in the case of deidentified databases with no PHI)		• Requires IT and informatics support
	• Increased risk for error		• Limited retrospective data access
	• Time and effort for manual review		• Arduous and costly setup
		• Requires NLP software license, development, and training of models	• Requires testing lab collaboration

Results of survey indicated there were four main methods used to extract genetic testing results previously collected for clinical purposes. For each method, the number of sites using the approach over the total number of sites is provided (some sites utilize more than one approach) along with reported benefits and limitations.

EHR, electronic health record; NLP, natural language processing; PHI, protected health information.

Discussion

Dramatic expansions in the availability and applications of genetic testing have the potential to bring enormous benefits to individuals affected by rare disease. However, individuals with specific variants of interest will only benefit from advances in genomic knowledge—or have the opportunity to contribute to the generation of new knowledge—if results can be ascertained from a collection of EHR systems. Finding ways to leverage EHRs for research purposes holds enormous promise for mitigating disparities that arise from research recruitment strategies that often skew opportunities away from marginalized or socially disadvantaged communities.^20–23 In this light, the imperative for implementing efficient and harmonized systems for integrating structured genetic variant entry into EHRs takes on added urgency. These efforts are critical to bridging the divide between rapid advances in genomic knowledge and access to research opportunities and cutting-edge genomic-informed clinical care.

In this study, we identified four approaches to identifying patients who had clinical genetic test results indicating variants of interest in specific genes. The fully manual approach, in which EHRs were searched to find records with text strings matching gene names of interest, was by far the most common strategy. Precision estimates suggest this approach has high rates of false positives and—although we were unable to document false negatives with this approach—it also has the potential for high false negative rates. There were some notable benefits to this approach, including comprehensive retrieval of relevant phenotypic and clinical management information, as well as more flexibility in cases of changes in EHR platforms or data sources. However, it was also the most tedious of all approaches and prone to mismatches. For example, manual searches for the gene name SPAST matched several terms commonly used for clinical care purposes such as “spasticity.” Another frequent source of false positive results in manual searches were genes included in a multigene panel; if a result or clinical note listed all the genes included in the panel, that patient’s record would be captured in search results, regardless of the result for the specific gene of interest. In these cases, substantial staff effort and expertise was required to discriminate between false and true positives. Notably, many sites indicated that this approach was also biased toward capturing non-VUS reports. Although no clear pattern emerged regarding the reason for this bias, since individuals performing manual chart reviews were trained to identify VUS, it is likely that research personnel were often only documenting variants with pathogenic/likely pathogenic classifications. Furthermore, opportunities for human error were noted to increase with each act of manual information retrieval or manipulation.^24,25

The addition of NLP software to preselect records with text mapped to IDD-relevant Human Phenotype Ontology terms reduced the frequency of gene name mismatches and alleviated the manual review burden. Refinement of gene name queries using negation constructs and proximity, for example, could eliminate false positive results by excluding results found within the context of other commonly used medical terms matching the same order of letters as a gene name (e.g., SPAST and “spasticity”). Reversal of the query through the addition or removal of a negation concept allowed the user to easily examine both included and excluded results. In addition, use of NLP software to detect phenotype information helped prefilter records with both: (1) gene names of interest, and (2) relevant symptoms. However, the development and application of these queries required time, effort, and multiple rounds of optimization via manual review. For example, NLP required input to be either discrete fields or in sentence format. This requires the infrastructure to first extract EHR data and accurately parse content into sentence format. Therefore, Method 2 required infrastructure and resources to establish NLP queries, optimize them, and actively monitor for bias.

While NLP is often stand-alone and not integrated within the EHR, there are on-going efforts to integrate these technologies in ways which may substantially improve the efficiency of extracting important information related to genetic test results.¹¹ NLP can also identify variants in multiple formats, thus helping to standardize final variant nomenclature.^26–29 Recent advances in artificial intelligence (AI) and Large Language Models (LLMs) have dramatically improved natural language understanding.^30,31 There is hope that LLMs which have already proven to be efficient and accurate at retrieving some content of interest from clinical notes with minimal training and development time (see Meyer et al., 2023 for descriptions of strengths and biases associated with LLMs³²), can be further fine-tuned with domain specific data, providing wide application in genetics and genomics across medical specialties (e.g., automating clinical phenotype extraction from notes/images and deciphering new insights from extracted data). However, when using AI tools, it is important to consider the potential bias based on training data and the possibility of leaking PHI.

An additional limitation of Methods 1 and 2 is the under-ascertainment of variants of interest when gene names are not included in the unstructured notes. One potential approach to circumvent this issue is to evaluate predictive models that include structured data elements (e.g., ICD codes, CPT codes, RxNorm CUIs) to identify which of these features can accurately predict if an individual has text string matching a gene name of interest included in the clinical documentation.

Efforts to directly establish partnerships with genetic testing laboratories alleviated issues related to manual chart reviews and NLP.^11,19 These collaborations allowed for the incorporation of genetic test results directly into custom databases in discrete and computer readable format. This ensured genetic variant information was mapped to common data models. These collaborative efforts forced genetic test results into a structured format that was more readily analyzed. Unfortunately, as sites using Method 3 stored discrete genetic results outside of the EHR, it is not tightly coupled with other patient data. In addition, a bottleneck to the deployment of Method 3 was the need for genetic testing laboratories to be directly involved throughout the process so data were submitted in a format which allows efficient uploads into the custom database or that are mapped to internationally accepted standards, such as Health Level 7 (HL7).¹⁰ Furthermore, the ability to identify which genetic testing laboratories should be contacted to implement this approach requires knowledge of labs that are currently (or have previously) been used to perform genetic testing—this information may not be readily available to many providers.

Despite these limitations, having genetic testing results mapped to HL7 standards in custom databases will be beneficial for integration with institutional EHR systems, triggering automated computational tasks such as literature searches of variants and biomarkers, related known therapies, pharmacogenomic and pharmacokinetic findings, ongoing clinical trials, and triggering clinical decision support algorithms.

The last method being tested among sites in our sample—genomic modules designed by EHR vendors—allowed ready access to genetic testing results in structured formats that are obtained after module implementation. The four IDDRC network sites who are involved in testing Method 4 are actively collaborating with EHR vendors to streamline workflows for data capture and integration; however, this required a substantial amount of effort from providers and coders. Although the use of structured data capture for genetic testing results did not appear to have bias in the types of ACMG-classified variants that were ascertained, these EHR vendor-designed modules are still in their infancy and still require additional technical analyses as well as investigations into the ease or difficulty in implementation across a diverse portfolio of clinical practice. As such, it is possible that biases may become evident once these modules are used for longer durations of time across clinical specialties with limited expertise in the use of genetic testing for clinical care; active surveillance for such emergent biases will be critical to minimize their occurrence and impact.

While this study focused on approaches to identify subjects for research purposes, integration of genetic test results into the EHR is critical for optimal implementation of the most up-to-date advances in precision care. Challenges in locating results of genetic tests can lead to unnecessary further testing, prolong the diagnostic odyssey, and failures to recognize the need for genetic re-evaluation. Identifying more efficient ways to retrieve genetic testing results dating back decades will reduce the burden on providers who are considering genetic testing for a particular patient and allow researchers to determine the frequency of variant reclassifications which will add insight into how often genetic test results should be revisited. Structured entry of genetic test results also opens the opportunity for future clinical decision support applications to take into account real-time classification updates from genomic databases such as ClinVar, in which submitter-driven variants and classifications are made publicly available.^33–35 Future work that layers EHR-derived information (e.g., family history, clinical context, and phenotypes) on top of genetic laboratory classifications offers the best prospects of leveraging genomic advances and medical informatics to benefit individual patients.⁶

Furthermore, better integration of test results into EHR systems will improve our understanding of which clinical departments order genetic tests and how often genetic test results improve care—or, conversely, are misinterpreted by providers due to lack of genomic literacy.⁶ This will inform approaches for educating more providers in genetic medicine, as well as informing the design and implementation of decision support related to ordering and interpreting genetic tests. Based on the current study, the best approach seems to be to force entry of genetic test results into the EHR in a codified manner so they are readily searchable for clinical decision support; however, this will require collaborations among key stakeholders (e.g., providers, genetic testing lab, EHR vendors, and experts in genetic medicine). A major hurdle will then be to liberate legacy data which may require more advanced technological approaches like incorporation of AI.

Limitations and future directions

Although limitations have been noted throughout, one overarching limitation of the current study is that there was no formal assessment of recall and precision of each method. Notably, many institutions involved in this study required access to PHI to prospectively retrieve genetic testing results. The inter-institutional resource and regulatory requirements that would have been necessary for prospective retrieval, recall, and precision were beyond the scope of this project. We therefore opted to leverage the experience of an existing Institutional Review Board approved project, the BGR, in which 9 of our 11 sites were involved,¹⁶ although we recognize the methodological limitation of this strategy.

The ultimate goals of the IDDRC GREATest EHR working group are to accelerate translational IDD research by optimizing methods to rapidly identify individuals who are eligible for clinical trial opportunities in real-time, based on the presence of a genetic variant. Future directions for this working group include focusing on more precisely defining which methods offer the most accurate and rapid way to access genetic test results across various EHR systems. A pilot project to identify the structured data elements that best predict that a patient at an IDDRC center has a genetic test result of interest is also being planned. An advocacy subgroup is also being proposed, to ensure lessons learned from this working group will inform the next iteration of EHR-based genomic modules through partnerships with EHR vendors and genetic testing companies.

Conclusion

Despite these limitations, this study successfully leveraged a broad national network to identify and underscore the considerable methodological heterogeneity employed by U.S. institutions in the storage, retrieval, and interrogation of genetic testing data—particularly with respect to legacy test results. The diverse strategies utilized each confer specific operational strengths and limitations. Addressing these challenges requires the establishment of standardized practice guidelines to unify data management workflows, thereby improving the identification of research-eligible populations and facilitating the delivery of genomically informed clinical care. Ultimately, the development of an EHR infrastructure that effectively serves the needs of clinicians, researchers, patients, and genetic testing laboratories will be critical to advancing and scaling genomic medicine.

Supplemental Material

sj-jpg-1-trd-10.1177_26330040251356521 – Supplemental material for Finding buried genetic test results in the electronic health record is inefficient and variable across institutions

Supplemental material, sj-jpg-1-trd-10.1177_26330040251356521 for Finding buried genetic test results in the electronic health record is inefficient and variable across institutions by Olivia J. Veatch, Jomol Mathew, Shira Rockowitz, Dustin Baldridge, Alyssa Wetzel, Maria Niarchou, Megan Clarke, Prabhu Shankar, Suma Shankar, Julie S. Cohen, Kendell German, Seth Berger, Angela Sellitto, Inez Y. Oh, Rashi Raizada, Piotr Sliz, Selvin Soby, Mihailo Kaplarevic, Dan Doherty, Andrea Gropman, Constance Smith-Hicks, Jeffrey L. Neul, Virginia Lanzotti, Benjamin Darbro, Qiang Chang, Mustafa Sahin and Maya Chopra in Therapeutic Advances in Rare Disease

Supplemental Material

sj-jpg-2-trd-10.1177_26330040251356521 – Supplemental material for Finding buried genetic test results in the electronic health record is inefficient and variable across institutions

Supplemental material, sj-jpg-2-trd-10.1177_26330040251356521 for Finding buried genetic test results in the electronic health record is inefficient and variable across institutions by Olivia J. Veatch, Jomol Mathew, Shira Rockowitz, Dustin Baldridge, Alyssa Wetzel, Maria Niarchou, Megan Clarke, Prabhu Shankar, Suma Shankar, Julie S. Cohen, Kendell German, Seth Berger, Angela Sellitto, Inez Y. Oh, Rashi Raizada, Piotr Sliz, Selvin Soby, Mihailo Kaplarevic, Dan Doherty, Andrea Gropman, Constance Smith-Hicks, Jeffrey L. Neul, Virginia Lanzotti, Benjamin Darbro, Qiang Chang, Mustafa Sahin and Maya Chopra in Therapeutic Advances in Rare Disease

Footnotes

Acknowledgements

We would like to acknowledge Dr. Tracy King at NIH National Institute of Child Health and Human Development (NICHD) for her facilitation of the Intellectual and Developmental Disabilities Research Centers (IDDRC) GREATest EHR Workgroup. We would also like to acknowledge the support provided by the excellent research coordinators at each site.

Declarations

ORCID iDs

Olivia J. Veatch

Jeffrey L. Neul

Mustafa Sahin

Maya Chopra

Supplemental material

Supplemental material for this article is available online.

References

Bowling

Thompson

Amaral

, et al. Genomic diagnosis for children with intellectual disability and/or developmental delay. Genome Med 2017; 9(1): 43.

Franks

Melen

Friedman

, et al. Technological readiness and implementation of genomic-driven precision medicine for complex diseases. J Intern Med 2021; 290(3): 602–620.

Jiang

Wang

Personalized medicine in oncology: tailoring the right drug to the right patient. Biomark Med 2010; 4(4): 523–533.

Kessler

Genomics and precision medicine: implications for critical care. AACN Adv Crit Care 2018; 29(1): 28–35.

Kotze

Luckhoff

Peeters

, et al. Genomic medicine and risk prediction across the disease spectrum. Crit Rev Clin Lab Sci 2015; 52(3): 120–137.

Grebe

Khushf

Chen

, et al. The interface of genomic information with the electronic health record: a points to consider statement of the American College of Medical Genetics and Genomics (ACMG). Genet Med 2020; 22(9): 1431–1436.

Ayatollahi

Hosseini

Hemmat

Integrating genetic data into electronic health records: medical geneticists’ perspectives. Healthcare Inform Res 2019; 25(4): 289–296.

den Dunnen

Dalgleish

Maglott

, et al. HGVS recommendations for the description of sequence variants: 2016 update. Hum Mutat 2016; 37(6): 564–569.

Wagner

Babb

Alterovitz

, et al. The GA4GH variation representation specification: a computational framework for variation representation and federated identification. Cell Genom 2021; 1(2): 100027.

10.

Alterovitz

Heale

Jones

, et al. FHIR genomics: enabling standardization for precision medicine use cases. NPJ Genom Med 2020; 5: 13.

11.

Lau-Min

McKenna

Asher

, et al. Impact of integrating genomic data into the electronic health record on genetics care delivery. Genet Med 2022; 24(11): 2338–2350.

12.

Tenenbaum

Avillach

Benham-Hutchins

, et al. An informatics research agenda to support precision medicine: seven key areas. J Am Med Inform Assoc 2016; 23(4): 791–795.

13.

Lau-Min

Asher

Chen

, et al. Real-world integration of genomic data into the electronic health record: the PennChart genomics initiative. Genet Med 2021; 23(4): 603–605.

14.

Vento

JO.

BIO23-019: Precision Oncology: Integrating Structured Genomic Data Into the Electronic Health Record via the EPIC® Genomics Module. J National Comprehensive Cancer Network 2023; 21(3.5): 7165.

15.

Walkley

Abbeduto

Batshaw

, et al. Intellectual and developmental disabilities research centers: fifty years of scientific accomplishments. Ann Neurol 2019; 86(3): 332–343.

16.

Chopra

Savatt

Bingaman

, et al. Clinical variants paired with phenotype: a rich resource for brain gene curation. Genet Med 2023; 26(3): 101035.

17.

Richards

Aziz

Bale

, et al. Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Genet Med 2015; 17(5): 405–424.

18.

Trivedi

Gildersleeve

Franco

, et al. Evaluation of a concept mapping task using named entity recognition and normalization in unstructured clinical text. J Healthcare Inform Res 2020; 4(4): 395–410.

19.

Rockowitz

LeCompte

Carmack

, et al. Children’s rare disease cohorts: an integrative research and clinical genomics initiative. NPJ Genom Med 2020; 5: 29.

20.

Frazier

Brown

Rockowitz

, et al. Toward representative genomic research: the children’s rare disease cohorts experience. Ther Adv Rare Dis 2023; 4: 26330040231181406.

21.

Cole

Sellitto

Baratta

, et al. Social determinants of genetics referral and completion rates among child neurology patients. medRxiv 2023. DOI: 10.1101/2023.09.12.23295450.

22.

Muller

Lee

Barge

, et al. Low referral rate for genetic testing in racially and ethnically diverse patients despite universal colorectal cancer screening. Clin Gastroenterol Hepatol 2018; 16(12): 1911.

23.

Delikurt

Williamson

Anastasiadou

, et al. A systematic review of factors that act as barriers to patient referral to genetic services. Eur J Hum Genet 2015; 23(6): 739–745.

24.

Feng

Anoushiravani

Tesoriero

, et al. Transcription error rates in retrospective chart reviews. Orthopedics 2020; 43(5): e404–e408.

25.

Mathes

Klassen

Pieper

Frequency of data extraction errors and methods to increase data extraction quality: a methodological review. BMC Med Res Methodol 2017; 17(1): 152.

26.

Parikh

Genetti

Aykanat

, et al. A data-driven architecture using natural language processing to improve phenotyping efficiency and accelerate genetic diagnoses of rare disorders. HGG Adv 2021; 2(3): 100035.

27.

Deisseroth

Birgmeier

Bodle

, et al. ClinPhen extracts and prioritizes patient phenotypes directly from medical records to expedite genetic disease diagnosis. Genet Med 2019; 21(7): 1585–1593.

28.

Sevim Bayrak

Stein

Jain

, et al. Identification of discriminative gene-level and protein-level features associated with pathogenic gain-of-function and loss-of-function variants. Am J Hum Genet 2021; 108(12): 2301–2318.

29.

Koleck

Dreisbach

Bourne

, et al. Natural language processing of symptoms documented in free-text narratives of electronic health records: a systematic review. J Am Med Inform Assoc 2019; 26(4): 364–379.

30.

Krittanawong

Artificial intelligence in clinical practice: how AI technologies impact medical research and clinics. 1st ed. Langford Lane, England: Elsevier Inc., 2024.

31.

Yang

Liu

Deng

, et al. Enhancing phenotype recognition in clinical notes using large language models: PhenoBCBERT and PhenoGPT. Patterns (N Y) 2024; 5(1): 100887.

32.

Meyer

Urbanowicz

Martin

PCN

, et al. ChatGPT and large language models in academia: opportunities and challenges. BioData Min 2023; 16(1): 20.

33.

Landrum

Kattman

. ClinVar at five years: delivering on the promise. Hum Mutat 2018; 39: 1623–1630. DOI: 10.1002/humu.23641.

34.

Landrum

Lee

Benson

, et al. ClinVar: public archive of interpretations of clinically relevant variants. Nucleic Acids Res 2016; 44: D862–868. 20151117. DOI: 10.1093/nar/gkv1222.

35.

Landrum

Lee

Benson

, et al. ClinVar: improving access to variant interpretations and supporting evidence. Nucleic Acids Res 2018; 46: D1062–D1067. DOI: 10.1093/nar/gkx1153.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.15 MB