Citations

AbstractSummaryBuilding a high-quality annotation corpus requires expenditure of considerable time and expertise, particularly for biomedical and clinical research applications. Most existing annotation tools provide many advanced features to cover a variety of needs where the installation, integration and difficulty of use present a significant burden for actual annotation tasks. Here, we present MedTator, a serverless annotation tool, aiming to provide an intuitive and interactive user interface that focuses on the core steps related to corpus annotation, such as document annotation, corpus summarization, annotation export and annotation adjudication.Availability and implementationMedTator and its tutorial are freely available from https://ohnlp.github.io/MedTator. MedTator source code is available under the Apache 2.0 license: https://github.com/OHNLP/MedTator.Supplementary informationSupplementary data are available at Bioinformatics online.

AbstractObjectiveWith COVID-19, there was a need for a rapidly scalable annotation system that facilitated real-time integration with clinical decision support systems (CDS). Current annotation systems suffer from a high-resource utilization and poor scalability limiting real-world integration with CDS. A potential solution to mitigate these issues is to use the rule-based gazetteer developed at our institution.Materials and MethodsPerformance, resource utilization, and runtime of the rule-based gazetteer were compared with five annotation systems: BioMedICUS, cTAKES, MetaMap, CLAMP, and MedTagger.ResultsThis rule-based gazetteer was the fastest, had a low resource footprint, and similar performance for weighted microaverage and macroaverage measures of precision, recall, and f1-score compared to other annotation systems.DiscussionOpportunities to increase its performance include fine-tuning lexical rules for symptom identification. Additionally, it could run on multiple compute nodes for faster runtime.ConclusionThis rule-based gazetteer overcame key technical limitations facilitating real-time symptomatology identification for COVID-19 and integration of unstructured data elements into our CDS. It is ideal for large-scale deployment across a wide variety of healthcare settings for surveillance of acute COVID-19 symptoms for integration into prognostic modeling. Such a system is currently being leveraged for monitoring of postacute sequelae of COVID-19 (PASC) progression in COVID-19 survivors. This study conducted the first in-depth analysis and developed a rule-based gazetteer for COVID-19 symptom extraction with the following key features: low processor and memory utilization, faster runtime, and similar weighted microaverage and macroaverage measures for precision, recall, and f1-score compared to industry-standard annotation systems.

AbstractIntroductionThe efficacy of cardiac resynchronization therapy (CRT) has been widely studied in the medical literature; however, about 30% of candidates fail to respond to this treatment strategy. Smart computational approaches based on clinical data can help expose hidden patterns useful for identifying CRT responders.MethodsWe retrospectively analyzed the electronic health records of 1664 patients who underwent CRT procedures from January 1, 2002 to December 31, 2017. An ensemble of ensemble (EoE) machine learning (ML) system composed of a supervised and an unsupervised ML layers was developed to generate a prediction model for CRT response.ResultsWe compared the performance of EoE against traditional ML methods and the state‐of‐the‐art convolutional neural network (CNN) model trained on raw electrocardiographic (ECG) waveforms. We observed that the models exhibited improvement in performance as more features were incrementally used for training. Using the most comprehensive set of predictors, the performance of the EoE model in terms of the area under the receiver operating characteristic curve and F1‐score were 0.76 and 0.73, respectively. Direct application of the CNN model on the raw ECG waveforms did not generate promising results.ConclusionThe proposed CRT risk calculator effectively discriminates which heart failure (HF) patient is likely to respond to CRT significantly better than using clinical guidelines and traditional ML methods, thus suggesting that the tool can enhanced care management of HF patients by helping to identify high‐risk patients.

Background: Familial hypercholesterolemia (FH) is an important public health burden as it is a relatively common Mendelian genetic disorder which is associated with dramatically increased lifetime risk for premature atherosclerotic cardiovascular disease (ASCVD) due to elevated plasma low-density lipoprotein cholesterol (LDL-C) levels. Little is known about prevalence and control of FH in the US. Objective: To develop an electronic phenotyping algorithm for rapid identification of FH in electronic health records (EHRs) and thereby address knowledge gaps in prevalence and control of FH. Methods: We identified patients in the Mayo Employee and Community Health (ECH) system who met the following inclusion criteria: 1) gave research authorization, 2) any LDL-C in EHR ≥190 mg/dL; 2) triglycerides <400 mg/dL; 3) absence of secondary causes of hyperlipidemia (hypothyroidism, cholestatic liver diseases, nephrotic syndrome, renal failure, and pregnancy). Next we applied the Dutch Lipid Clinic Network (DLCN) criteria to ascertain FH cases. ASCVD-related diagnosis codes, laboratory data, and medications were obtained from structured EHR datasets. We used the Mayo natural language processing (NLP) system to ascertain presence of family and personal history of premature ASCVD, xanthomas and corneal arcus from clinical text notes. Results: Of a total of 131,000 patients seen in ECH clinics between July 1993 and December 2014, 6018 met the inclusion criteria. Implementing the electronic phenotyping algorithm for FH in these patients identified 178 definite and 369 probable cases (DLCN score >8 and 6-8 points, respectively) with an overall FH prevalence of 0.4% (1:240). Blinded expert review of 160 randomly chosen patients showed positive and negative predictive values for electronic phenotyping algorithm at 85% and 90%, respectively. Only 40% of these patients achieved LDL-C ≤100 mg/dL on treatment. Conclusions: 1) An EHR-based phenotyping algorithm that included NLP had reasonable accuracy in ascertaining FH cases. 2) Implementing this algorithm revealed the prevalence of FH in the study cohort to be nearly twice the current estimate of 1:500 in the general population. 3) Less than half of the FH patients had optimal LDL-C levels on treatment.

Introduction: Proton pump inhibitors (PPIs) are used by an estimated 29 million Americans. PPIs increase the levels of asymmetrical dimethylarginine, a known risk factor for cardiovascular disease (CVD). Data from a select population of patients with CVD suggest that PPI use is associated with an increased risk of stroke, heart failure, and coronary heart disease. The impact of PPI use on incident CVD is largely unknown in the general population. Hypothesis: We hypothesized that PPI users have a higher risk of incident total CVD, coronary heart disease, stroke, and heart failure compared to nonusers. To demonstrate specificity of association, we additionally hypothesized that there is not an association between use of H 2 -blockers - another commonly used class of medications with similar indications as PPIs - and CVD. Methods: We used the Rochester Epidemiology Project’s medical records-linkage system to identify all residents of Olmsted County, MN on our baseline date of January 1, 2004 (N=140217). We excluded persons who did not grant permission for their records to be used for research, were <18 years old, had a history of CVD, had missing data for any variable included in our model, or had evidence of PPI use within the previous year.We followed our final cohort (N=58175) for up to 12 years. The administrative censoring date for CVD was 1/20/2014, for coronary heart disease was 8/3/2016, for stroke was 9/9/2016, and for heart failure was 1/20/2014. Time-varying PPI ever-use was ascertained using 1) natural language processing to capture unstructured text from the electronic health record, and 2) outpatient prescriptions. An incident CVD event was defined as the first occurrence of 1) validated heart failure, 2) validated coronary heart disease, or 3) stroke, defined using diagnostic codes only. As a secondary analysis, we calculated the association between time-varying H 2 -blocker ever-use and CVD among persons not using H 2 -blockers at baseline. Results: After adjustment for age, sex, race, education, hypertension, hyperlipidemia, diabetes, and body-mass-index, PPI use was associated with an approximately 50% higher risk of CVD (hazard ratio [95% CI]: 1.51 [1.37-1.67]; 2187 CVD events), stroke (hazard ratio [95% CI]: 1.49 [1.35-1.65]; 1928 stroke events), and heart failure (hazard ratio [95% CI]: 1.56 [1.23-1.97]; 353 heart failure events) compared to nonusers. Users of PPIs had a 35% greater risk of coronary heart disease than nonusers (95% CI: 1.13-1.61; 626 coronary heart disease events). Use of H 2 -blockers was also associated with a higher risk of CVD (adjusted hazard ratio [95% CI]: 1.23 [1.08-1.41]; 2331 CVD events). Conclusions: PPI use is associated with a higher risk of CVD, coronary heart disease, stroke and heart failure. Use of a drug with no known cardiac toxicity - H 2 -blockers - was also associated with a greater risk of CVD, warranting further study.

Background: Real world studies of silent brain infarction (SBI) and white matter disease (WMD) are impeded by challenges in cohort identification. Natural language processing (NLP) from imaging reports may facilitate future studies. However, electronic health records can be heterogeneous and the process of interpreting neuroimages and generating reports can vary. Understanding knowledge representation and relationships between neuroimages and imaging reports is crucial for using NLP to facilitate disease management and cohort identification. Methods: A balanced sample of head neuroimages (CT, MRI) of patients >50 years without clinical histories of symptomatic stroke, TIA, or dementia were obtained at Mayo Clinic and Tufts Medical Center. A team of 4 radiology residents performed report interpretation (RI) on 1000 reports according to a standardized protocol for the presence of SBIs, the presence of WMD, and WMD grade. A random subsample of 400 was doubly read for interrater reliability. For benchmarking, a team of 4 neuroradiologists directly reviewed and described findings on a subsample of 182 images, each doubly read. We assessed interrater reliability for direct review (DR) and RI, and agreement between these 2 information sources. An NLP algorithm was developed to review and extract findings from 1000 imaging reports. Results: For DR, interrater reliability was moderate for SBIs and WMD (k = 0.53, 95% CI 0.43-0.64 and k = 0.47, 95% CI 0.33-0.61) and good for WMD grade (Spearman 0.71, p<0.001). For RI, interrater reliability for SBIs, WMD and WMD grade was good (k = 0.88, 95% CI 0.80-0.97; k = 0.98, 95% CI 0.97-1.00; and Spearman = 0.985, p<0.001, respectively). Agreement between DR and RI was good for SBIs (k = 0.77, 95% CI 0.67-0.86) and WMD (k = 0.65, 95% CI 0.54-0.77). Spearman rank correlations comparing WMD grade interpretation DR to RI was 0.60 (p<0.001). In identifying the presence of SBIs and WMDs, the accuracy of the NLP algorithm was 0.991 and 0.994, respectively. Conclusion: For the presence of SBI and WMD, and WMD grade, agreement between RI and DR was similar to agreement between 2 neuroradiologists directly reviewing neuroimages. It is highly feasible to use NLP to identify patients with SBIs and WMDs for clinical studies.

Background . Patients increasingly use asynchronous communication platforms to converse with care teams. Natural language processing (NLP) to classify content and automate triage of these messages has great potential to enhance clinical efficiency. We characterize the contents of a corpus of portal messages generated by patients using NLP methods. We aim to demonstrate descriptive analyses of patient text that can contribute to the development of future sophisticated NLP applications. Methods . We collected approximately 3,000 portal messages from the cardiology, dermatology, and gastroenterology departments at Mayo Clinic. After labeling these messages as either Active Symptom, Logistical, Prescription, or Update, we used NER (named entity recognition) to identify medical concepts based on the UMLS library. We hierarchically analyzed the distribution of these messages in terms of departments, message types, medical concepts, and keywords therewithin. Results . Active Symptom and Logistical content types comprised approximately 67% of the message cohort. The “Findings” medical concept had the largest number of keywords across all groupings of content types and departments. “Anatomical Sites” and “Disorders” keywords were more prevalent in Active Symptom messages, while “Drugs” keywords were most prevalent in Prescription messages. Logistical messages tended to have the lower proportions of “Anatomical Sites,”, “Disorders,”, “Drugs,”, and “Findings” keywords when compared to other message content types. Conclusions . This descriptive corpus analysis sheds light on the content and foci of portal messages. The insight into the content and differences among message themes can inform the development of more robust NLP models.

Abstract Objective : Developing clinical natural language processing systems often requires access to many clinical documents, which are not widely available to the public due to privacy and security concerns. To address this challenge, we propose to develop methods to generate synthetic clinical notes and evaluate their utility in real clinical natural language processing tasks. Materials and Methods : We implemented 4 state-of-the-art text generation models, namely CharRNN, SegGAN, GPT-2, and CTRL, to generate clinical text for the History and Present Illness section. We then manually annotated clinical entities for randomly selected 500 History and Present Illness notes generated from the best-performing algorithm. To compare the utility of natural and synthetic corpora, we trained named entity recognition (NER) models from all 3 corpora and evaluated their performance on 2 independent natural corpora. Results : Our evaluation shows GPT-2 achieved the best BLEU (bilingual evaluation understudy) score (with a BLEU-2 of 0.92). NER models trained on synthetic corpus generated by GPT-2 showed slightly better performance on 2 independent corpora: strict F1 scores of 0.709 and 0.748, respectively, when compared with the NER models trained on natural corpus (F1 scores of 0.706 and 0.737, respectively), indicating the good utility of synthetic corpora in clinical NER model development. In addition, we also demonstrated that an augmented method that combines both natural and synthetic corpora achieved better performance than that uses the natural corpus only. Conclusions : Recent advances in text generation have made it possible to generate synthetic clinical notes that could be useful for training NER models for information extraction from natural clinical notes, thus lowering the privacy concern and increasing data availability. Further investigation is needed to apply this technology to practice.