Deep learning approach to detection of colonoscopic information from unstructured reports.

BACKGROUND: Colorectal cancer is a leading cause of cancer deaths. Several screening tests, such as colonoscopy, can be used to find polyps or colorectal cancer. Colonoscopy reports are often written in unstructured narrative text. The information embedded in the reports can be used for various purposes, including colorectal cancer risk prediction, follow-up recommendation, and quality measurement. However, the availability and accessibility of unstructured text data are still insufficient despite the large amounts of accumulated data. We aimed to develop and apply deep learning-based natural language processing (NLP) methods to detect colonoscopic information. METHODS: This study applied several deep learning-based NLP models to colonoscopy reports. Approximately 280,668 colonoscopy reports were extracted from the clinical data warehouse of Samsung Medical Center. For 5,000 reports, procedural information and colonoscopic findings were manually annotated with 17 labels. We compared the long short-term memory (LSTM) and BioBERT model to select the one with the best performance for colonoscopy reports, which was the bidirectional LSTM with conditional random fields. Then, we applied pre-trained word embedding using large unlabeled data (280,668 reports) to the selected model. RESULTS: The NLP model with pre-trained word embedding performed better for most labels than the model with one-hot encoding. The F1 scores for colonoscopic findings were: 0.9564 for lesions, 0.9722 for locations, 0.9809 for shapes, 0.9720 for colors, 0.9862 for sizes, and 0.9717 for numbers. CONCLUSIONS: This study applied deep learning-based clinical NLP models to extract meaningful information from colonoscopy reports. The method in this study achieved promising results that demonstrate it can be applied to various practical purposes.

Shared Interoperable Clinical Decision Support Service for Drug-Allergy Interaction Checks: Implementation Study.

BACKGROUND: Clinical decision support (CDS) can improve health care with respect to the quality of care, patient safety, efficiency, and effectiveness. Establishing a CDS system in a health care setting remains a challenge. A few hospitals have used self-developed in-house CDS systems or commercial CDS solutions. Since these in-house CDS systems tend to be tightly coupled with a specific electronic health record system, the functionality and knowledge base are not easily shareable. A shared interoperable CDS system facilitates the sharing of the knowledge base and extension of CDS services. OBJECTIVE: The study focuses on developing and deploying the national CDS service for the drug-allergy interaction (DAI) check for health care providers in Korea that need to introduce the service but lack the budget and expertise. METHODS: To provide the shared interoperable CDS service, we designed and implemented the system based on the CDS Hooks specification and Health Level Seven (HL7) Fast Healthcare Interoperability Resources (FHIR) standard. The study describes the CDS development process. The system development went through requirement analysis, design, implementation, and deployment. In particular, the concept architecture was designed based on the CDS Hooks structure. The MedicationRequest and AllergyIntolerance resources were profiled to exchange data using the FHIR standard. The discovery and DAI check application programming interfaces and rule engine were developed. RESULTS: The CDS service was deployed on G-Cloud, a government cloud service. In March 2021, the CDS service was launched, and 67 health care providers participated in the CDS service. The health care providers participated in the service with 1,008,357 DAI checks for 114,694 patients, of which 33,054 (3.32%) cases resulted in a "warning." CONCLUSIONS: Korea's Ministry of Health and Welfare has been trying to build an HL7 FHIR-based ecosystem in Korea. As one of these efforts, the CDS service initiative has been conducted. To promote the rapid adoption of the HL7 FHIR standard, it is necessary to accelerate practical service development and to appeal to policy makers regarding the benefits of FHIR standardization. With the development of various case-specific implementation guides using the Korea Core implementation guide, the FHIR standards will be distributed nationwide, and more shared interoperable health care services will be introduced in Korea.

Standard Document Development for Health Information Exchange in Korea.

BACKGROUND: Health information exchange (HIE) allows healthcare providers to access a patient's medical information to improve patient care continuity. The standardized data realize the HIE values. Since the Health Level 7 Clinical Document Architecture (CDA) is flexible, implementation guides (IG) are needed for use cases. Although many CDA IGs have been developed, they did not describe how these CDA IGs were developed. A national CDA IG that meets the local requirements is demanded since the data differs according to the digital divide and social-cultural background of the country that wants to establish HIE. Due to their localized contents, other countries cannot directly adopt the published CDA IGs. OBJECTIVES: We developed the national CDA IG, namely, Korean (K)-CDA IG that meets the local requirement, including reusable structured templates, value sets, and object identifiers (OIDs). We present a detailed description of the development process and the technical methods of the national CDA IG in the Korean context. METHODS: The K-CDA IG was developed in the following stages: analysis, development, and evaluation. First, we investigated the health information environment and electronic health record (EHR) systems and conducted a gap analysis with published CDA IGs. Second, a templated CDA approach was taken for designing modular. Lastly, we consulted a technical advisory group for comments on the validity of the K-CDA IG. RESULTS: A total of 35 CDA templates were developed. We improved 28 value sets of which 13 were Korea specific and 15 were based on the ones used in other IGs, and made a set of rules to establish the OID structure. CONCLUSION: We presented the development process and the technical specifications of K-CDA IG. We explored how the results can be used as interoperability criteria in the national EHR systems certification program. Finally, we provided recommendations that could guide other entities planning their HIE programs.

Fast Healthcare Interoperability Resources (FHIR)-Based Quality Information Exchange for Clinical Next-Generation Sequencing Genomic Testing: Implementation Study.

BACKGROUND: Next-generation sequencing (NGS) technology has been rapidly adopted in clinical practice, with the scope extended to early diagnosis, disease classification, and treatment planning. As the number of requests for NGS genomic testing increases, substantial efforts have been made to deliver the testing results clearly and unambiguously. For the legitimacy of clinical NGS genomic testing, quality information from the process of producing genomic data should be included within the results. However, most reports provide insufficient quality information to confirm the reliability of genomic testing owing to the complexity of the NGS process. OBJECTIVE: The goal of this study was to develop a Fast Healthcare Interoperability Resources (FHIR)-based web app, NGS Quality Reporting (NGS-QR), to report and manage the quality of the information obtained from clinical NGS genomic tests. METHODS: We defined data elements for the exchange of quality information from clinical NGS genomic tests, and profiled a FHIR genomic resource to enable information exchange in a standardized format. We then developed the FHIR-based web app and FHIR server to exchange quality information, along with statistical analysis tools implemented with the R Shiny server. RESULTS: Approximately 1000 experimental data entries collected from the targeted sequencing pipeline CancerSCAN designed by Samsung Medical Center were used to validate implementation of the NGS-QR app using real-world data. The user can share the quality information of NGS genomic testing and verify the quality status of individual samples in the overall distribution. CONCLUSIONS: This study successfully demonstrated how quality information of clinical NGS genomic testing can be exchanged in a standardized format. As the demand for NGS genomic testing in clinical settings increases and genomic data accumulate, quality information can be used as reference material to improve the quality of testing. This app could also motivate laboratories to perform diagnostic tests to provide high-quality genomic data.

Non-alcoholic fatty liver disease and progression of coronary artery calcium score: a retrospective cohort study.

BACKGROUND AND AIM: Non-alcoholic fatty liver disease (NAFLD), a hepatic manifestation of the metabolic syndrome, was associated with subclinical atherosclerosis in many cross-sectional studies, but the prospective association between NAFLD and the progression of atherosclerosis has not been evaluated. This study was conducted to evaluate the association between NAFLD and the progression of coronary atherosclerosis. METHODS: This retrospective cohort study included 4731 adult men and women with no history of cardiovascular disease (CVD), liver disease or cancer at baseline who participated in a repeated regular health screening examination between 2004 and 2013. Fatty liver was diagnosed by ultrasound based on standard criteria, including parenchymal brightness, liver-to-kidney contrast, deep beam attenuation and bright vessel walls. Progression of coronary artery calcium (CAC) scores was measured using multidetector CT scanners. RESULTS: The average duration of follow-up was 3.9 years. During follow-up, the annual rate of CAC progression in participants with and without NAFLD were 22% (95% CI 20% to 23%) and 17% (16% to 18%), respectively (p<0.001). The multivariable ratio of progression rates comparing participants with NAFLD with those without NAFLD was 1.04 (1.02 to 1.05; p<0.001). The association between NAFLD and CAC progression was similar in most subgroups analysed, including in participants with CAC 0 and in those with CAC >0 at baseline. CONCLUSIONS: In this large cohort study of adult men and women with no history of CVD, NAFLD was significantly associated with the development of CAC independent of cardiovascular and metabolic risk factors. NAFLD may play a pathophysiological role in atherosclerosis development and may be useful to identify subjects with a higher risk of subclinical disease progression.

Persistent Nonalcoholic Fatty Liver Disease Increases Risk for Carotid Atherosclerosis.

BACKGROUND & AIMS: Nonalcoholic fatty liver disease (NAFLD) has been associated with subclinical atherosclerosis in cross-sectional studies. We investigated the longitudinal association of NAFLD with the development of subclinical carotid atherosclerosis. METHODS: We performed a retrospective cohort study of 8020 adult men (average age, 49.2 y) without carotid atherosclerosis at baseline who underwent repeated health check-up examinations from January 1, 2005, through December 31, 2013. NAFLD status was diagnosed by ultrasonography and classified into 4 groups based on baseline and follow-up findings: none, developed, regressed, or persistent NAFLD. Subclinical carotid atherosclerosis was measured by ultrasound. RESULTS: The age-adjusted hazard ratio for subclinical carotid atherosclerosis development comparing participants with persistent NAFLD with those without NAFLD was 1.23 (95% confidence interval [CI], 1.13-1.35; P < .001). The association persisted after adjustment for smoking, alcohol, body mass index, and weight change (hazard ratio, 1.13; 95% CI, 1.03-1.25; P = .014), but disappeared after adjustment for metabolic variables. The hazard ratio, comparing subjects with regression of NAFLD vs those with persistent NAFLD, was 0.82 (95% CI, 0.69-0.96; P = .013). The risk of subclinical carotid atherosclerosis development also was higher among participants with a high NAFLD fibrosis score, fibrosis-4 scores, or levels of γ-glutamyl transferase at baseline. CONCLUSIONS: In a large cohort study, persistent NAFLD was associated with an increased risk of subclinical carotid atherosclerosis development. This association was explained by metabolic factors that could be potential mediators of the effect of NAFLD. Markers of liver fibrosis also were associated with subclinical carotid atherosclerosis development. Prospective studies are needed to determine whether treatment of NAFLD can reduce this risk.

Nonalcoholic Fatty Liver Disease for Identification of Preclinical Carotid Atherosclerosis.

Nonalcoholic fatty liver disease (NAFLD) is associated with cardiovascular disease, yet whether identification of NAFLD in asymptomatic individuals is helpful over established risk factors remains unknown. A total of 37,799 asymptomatic adults aged 20 years or older who underwent comprehensive health check-up examination, including abdominal and carotid artery duplex ultrasonography (US) were included in the analysis. Nonalcoholic fatty liver disease was diagnosed with US and exclusion of secondary causes of fat accumulation or other causes of chronic liver disease, and graded as mild or moderate to severe fatty liver. Individuals with carotid plaque identified on carotid artery US were considered at risk for cardiovascular disease. Metabolic syndrome (MetS) was defined by the adult treatment panel III criteria. Nonalcoholic fatty liver disease was an independent factor associated with carotid plaque in a dose-dependent manner (odds ratio [OR]; 95% confidence interval [CI]: 1.09 [1.03-1.16] and 1.13 [1.06-1.21] for mild and ≥ moderate degree of NAFLD). Among clinically-relevant subgroups, NAFLD was more closely associated with carotid plaque in young adults (aged < 60 years) without MetS (OR [95% CI]: 1.13 [1.03-1.19] and 1.16 [1.06-1.27] for mild and ≥ moderate degree of NAFLD) than old adults (aged ≥ 60 years) or with MetS (OR [95% CI]: 1.06 [0.97-1.17] and 1.07 [0.97-1.19] for mild and ≥ moderate degree NAFLD). In young adults without MetS, the prevalence of carotid plaques was 32.8% and the sensitivity and specificity of NAFLD for carotid plaque was 0.38 and 0.67, respectively. In conclusion, NAFLD is associated with carotid plaque independent of traditional risk factors, especially in young adults without MetS. Nonalcoholic fatty liver disease could help identify additional individuals with preclinical atherosclerosis in asymptomatic young adults without MetS, yet, showed suboptimal performance as a screening tool.