Search

Introduction: Human and veterinary medicine are practiced separately, but literature databases such as Pubmed include articles from both fields. This impedes supporting clinical decisions with automated information retrieval, because treatment considerations would not ignore the discipline of mixed sources. Here we investigate data-driven methods from computational linguistics for automatically distinguishing between human and veterinary medical texts. Methods: For our experiments, we selected language models after a literature review of benchmark datasets and reported performances. We generated a dataset of around 48,000 samples for binary text classification, specifically designed to differentiate between human medical and veterinary subjects. Using this dataset, we trained and fine-tuned classifiers based on selected transformer-based models as well as support vector machines (SVM). Results: All trained classifiers achieved more than 99% accuracy, even though the transformer-based classifiers moderately outperformed the SVMbased one. Discussion: Such classifiers could be applicable in clinical decision support functions that build on automated information retrieval. [ABSTRACT FROM AUTHOR]

Finding relevant information in the biomedical literature increasingly depends on efficient information retrieval (IR) algorithms. Cross-Encoders, SentenceBERT, and ColBERT are algorithms based on pre-trained language models that use nuanced but computable vector representations of search queries and documents for IR applications. Here we investigate how well these vectorization algorithms estimate relevance labels of biomedical documents for search queries using the OHSUMED dataset. For our evaluation, we compared computed scores to provided labels by using boxplots and Spearman's rank correlations. According to these metrics, we found that Sentence-BERT moderately outperformed the alternative vectorization algorithms and that additional fine-tuning based on a subset of OHSUMED labels yielded little additional benefit. Future research might aim to develop a larger dedicated dataset in order to optimize such methods more systematically, and to evaluate the corresponding functions in IR tools with endusers. [ABSTRACT FROM AUTHOR]

Digital individual participant data (IPD) from clinical trials are increasingly distributed for potential scientific reuse. The identification of available IPD, however, requires interpretations of textual data-sharing statements (DSS) in large databases. Recent advancements in computational linguistics include pretrained language models that promise to simplify the implementation of effective classifiers based on textual inputs. In a subset of 5,000 textual DSS from ClinicalTrials.gov, we evaluate how well classifiers based on domain-specific pretrained language models reproduce original availability categories as well as manually annotated labels. Typical metrics indicate that classifiers that predicted manual annotations outperformed those that learned to output the original availability categories. This suggests that the textual DSS descriptions contain applicable information that the availability categories do not, and that such classifiers could thus aid the automatic identification of available IPD in large trial databases. [ABSTRACT FROM AUTHOR]

Abstract Background In 2004 the adoption of a modular curriculum at the medical faculty in Muenster led to the introduction of centralized examinations based on multiple-choice questions (MCQs). We report on how organizational challenges of realizing faculty-wide personalized tests were addressed by implementation of a specialized software module to automatically generate test sheets from individual test registrations and MCQ contents. Methods Key steps of the presented method for preparing personalized test sheets are (1) the compilation of relevant item contents and graphical media from a relational database with database queries, (2) the creation of Extensible Markup Language (XML) intermediates, and (3) the transformation into paginated documents. Results The software module by use of an open source print formatter consistently produced high-quality test sheets, while the blending of vectorized textual contents and pixel graphics resulted in efficient output file sizes. Concomitantly the module permitted an individual randomization of item sequences to prevent illicit collusion. Conclusions The automatic generation of personalized MCQ test sheets is feasible using freely available open source software libraries, and can be efficiently deployed on a faculty-wide scale.

While clinical information systems usually store patient records in database tables, human interpretations as well as information transfer between institutions often require that clinical data can be represented as documents. To automate document generation from patient data in conjunction with the rich computational facilities of clinical decision support, we propose a template-based extension of the Arden Syntax, and discuss the benefits and limitations observed during a pilot application for patient recruitment. While the original Arden Syntax supports string concatenation as well as the substitution of unnamed placeholders, we integrated an additional method based on embedding expressions into strings. A dedicated parser identifies the expressions and automatically substitutes them at runtime, which can for example be harnessed to display the most recent value from a time series. The resulting mechanism supports the generation of extensive clinical documents without the need to apply specific operators. To evaluate the proposed extension, we implemented an Arden module that identifies an intensive care patient cohort that conforms to the eligibility criteria of a clinical trial and outputs a concise patient overview in different document formats. While string interpolation in the original Arden standard has been tailored to clinical event monitoring, we interpret that our accessible approach usefully extends Arden's data-to-text capabilities. Future research might target the development of an interactive template editor that would hide the complexity of formatting directives and conditional expressions behind a graphical user interface, and explore how computer-linguistic formalisms might facilitate advanced features such as automatic inflections of verbs and nouns. [ABSTRACT FROM AUTHOR]

Preparations for anesthesiological management of patients build on preoperative patient self-reports concerning risk factors and comorbidities. In this setting, electronic documentation could facilitate innovative computerized functions, although patient-facing digital questionnaires require appropriate tools that patients can access effectively. To explore the feasibility of an electronic application for preoperative data acquisition directly from patients, a digital, tablet-based prototypical application has been developed within a user-centered design process in order to replace a previously used paper-based anamnesis sheet for perioperative risk evaluation. The implemented prototype has been extensively tested and iteratively improved to progressively provide an easy-to-use data entry function. To assess the suitability of this tool for everyday data acquisition by patients and physicians and to identify usability problems, the stepwise development process was accompanied by a heuristic evaluation as well as a thinkaloud evaluation, while another 56 participating patients completed a feedback sheet according to ISO 9241/10. The latter method detected additional usability problems that occurred during the use of the application, which contributed to iterative improvements of the prototype. Throughout the development process, 81 issues were identified and largely resolved. After these revisions of the prototype, the number of problems found per tester decreased from 4.75 to 0.96, while the overall rating increased to 6.14 out of 7 points (SD = 1.2). These improvements demonstrate the value and efficiency of such a user-centered design process and illustrate that a user-friendly patient-facing digital data entry can replace preoperative paper questionnaires for anesthesiological management. [ABSTRACT FROM AUTHOR]

Using gene markers and other patient features to predict clinical outcomes plays a vital role in enhancing clinical decision making and improving prognostic accuracy. This work uses a large set of colorectal cancer patient data to train predictive models using machine learning methods such as random forest, general linear model, and neural network for clinically relevant outcomes including disease free survival, survival, radio-chemotherapy response (RCT-R) and relapse. The most successful predictive models were created for dichotomous outcomes like relapse and RCT-R with accuracies of 0.71 and 0.70 on blinded test data respectively. The best prediction models regarding overall survival and disease-free survival had C-Index scores of 0.86 and 0.76 respectively. These models could be used in the future to aid a decision for or against chemotherapy and improve survival prognosis. We propose that future work should focus on creating reusable frameworks and infrastructure for training and delivering predictive models to physicians, so that they could be readily applied to other diseases in practice and be continuously developed integrating new data. [ABSTRACT FROM AUTHOR]

Fine-meshed perioperative measurements are offering enormous potential for automatically investigating clinical complications during general anesthesia. In this study, we employed multiple machine learning methods to model perioperative hypoxia and compare their respective capabilities. After exporting and visualizing 620 series of perioperative vital signs, we had ten anesthesiologists annotate the subjective presence and severity of temporary postintubation oxygen desaturation. We then applied specific clustering and prediction methods on the acquired annotations, and evaluated their performance in comparison to the inter-rater agreement between experts. When reproducing the expert annotations, the sensitivity and specificity of multi-layer neural networks substantially outperformed clustering and simpler threshold-based methods. The achieved performance of our best automated hypoxia models thereby approximately equaled the observed agreement between different medical experts. Furthermore, we deployed our classification methods for processing unlabeled inputs to estimate the incidence of hypoxic episodes in another sizeable patient cohort, which attests to the feasibility of using the approach on a larger scale. We interpret that our machine learning models could be instrumental for computerized observational studies of the clinical determinants of post-intubation oxygen deficiency. Future research might also investigate potential benefits of more advanced preprocessing approaches such as automated feature learning. [ABSTRACT FROM AUTHOR]

The physiological age-related development of pediatric laboratory results interferes with pathological derangements, which can complicate the interpretation of test results. Recently proposed continuous reference intervals (RIs) promise to be beneficial, although their clinical use may depend on graphical presentations. To estimate the clinical utility of continuous RIs, we developed and evaluated an interactive visualization tool, and examined the differentiation of hemoglobinopathies that is attainable based on the underlying innovative RI model. The implemented web application allows users to easily enter laboratory test results, and displays various visualizations in conjunction with the corresponding RIs, such as charts and personalized Z-scores. To evaluate the usability of the visualization tool, we conducted concurrent think-aloud sessions with four physicians, who were prompted to solve a set of typical interpretation tasks, and acquired additional information through a questionnaire including the System Usability Scale (SUS). We used 85 de-identified clinical cases for an exemplified assessment of how well model-based interpretations of blood count parameters reproduced previously diagnosed hemoglobinopathies. Usability tests as well as questionnaire responses indicated that the developed tool was well received by the physicians. Results from the think-aloud evaluation revealed only minor problems and the tool reached an average SUS score of 86.9, suggesting good usability. Hemoglobinopathy discrimination depended on the considered subtype, although the overall performance of the novel method rivaled the one of the conventional approach. The interactive visualization of innovative continuous reference intervals demonstrated promising results, which justifies further testing on the path towards clinical routine. [ABSTRACT FROM AUTHOR]

Automated perioperative measurements such as cardiovascular monitoring data are commonly compared to established upper and lower thresholds, but could also allow for more complex interpretations. Analyzing such time series in extensive electronic medical records for research purposes may itself require customized automation, so we developed a set of algorithms for quantifying different aspects of temporal fluctuations. We implemented conventional measures of dispersion, summaries of absolute gradients between successive values, and Poincaré plots. We aggregated the severity and duration of hypotensive episodes by calculating the average area under different mean arterial pressure (MAP) thresholds. We applied these methods to 30,452 de-identified MAP series, and analyzed the similarity between alternative indices via hierarchical clustering. To explore the potential utility of these propositional metrics, we computed their statistical association with presumed complications due to cardiovascular instability. We observed that hierarchical clustering reliably segregated features that had been designed to quantify dissimilar aspects. Summaries of temporary hypotension turned out to be significantly increased among patient subgroups with subsequent signs of a complicated recovery. These associations were even stronger for measures that were specifically geared to capturing short-term MAP variability. These observations suggest the potential capability of our proposed algorithms for quantifying heterogeneous aspects of short-term MAP fluctuations. Future research might also target a wider selection of outcomes and other attributes that may be subject to intraoperative variability. [ABSTRACT FROM AUTHOR]

Discharge letters are an important means of communication between physicians and nurses from intensive care units and their colleagues from normal wards. The patient data management system (PDMS) used at our local intensive care units provides an export tool to create discharge letters by inserting data items from electronic medical records into predefined templates. Local intensivists criticized the limitations of this tool regarding the identification and the further processing of clinically relevant data items for a flexible creation of discharge letters. As our PDMS supports Arden Syntax, and the demanded functionalities are well within the scope of this standard, we set out to investigate the suitability of Arden Syntax for the generation of discharge letters. To provide an easy-tounderstand facility for integrating data items into document templates, we created an Arden Syntax interface function which replaces the names of previously defined variables with their content in a way that permits arbitrary custom formatting by clinical users. Our approach facilitates the creation of flexible text sections by conditional statements, as well as the integration of arbitrary HTML code and dynamically generated graphs. The resulting prototype enables clinical users to apply the full set of Arden Syntax language constructs to identify and process relevant data items in a way that far exceeds the capabilities of the PDMS export tool. The generation of discharge letters is an uncommon area of application for Arden Syntax, considerably differing from its original purpose. However, we found our prototype well suited for this task and plan to evaluate it in clinical production after the next major release change of our PDMS. [ABSTRACT FROM AUTHOR]

Digital individual participant data (IPD) from clinical trials are increasingly distributed for potential scientific reuse. The identification of available IPD, however, requires interpretations of textual data-sharing statements (DSS) in large databases. Recent advancements in computational linguistics include pre-trained language models that promise to simplify the implementation of effective classifiers based on textual inputs. In a subset of 5,000 textual DSS from ClinicalTrials.gov, we evaluate how well classifiers based on domain-specific pre-trained language models reproduce original availability categories as well as manually annotated labels. Typical metrics indicate that classifiers that predicted manual annotations outperformed those that learned to output the original availability categories. This suggests that the textual DSS descriptions contain applicable information that the availability categories do not, and that such classifiers could thus aid the automatic identification of available IPD in large trial databases.

Mapping clinical attributes from hospital information systems to standardized terminologies may allow their scientific reuse for multicenter studies. The Unified Medical Language System (UMLS) defines synonyms in different terminologies, which could be valuable for achieving semantic interoperability between different sites. Here we aim to explore the potential relevance of UMLS concepts and associated semantic relations for widely used clinical terminologies in a German university hospital. To semi-automatically examine a sample of the 200 most frequent codes from Erlangen University Hospital for three relevant terminologies, we implemented a script that queries their UMLS representation and associated mappings via a programming interface. We found that 94% of frequent diagnostic codes were available in UMLS, and that most of these codes could be mapped to other terminologies such as SNOMED CT. We observed that all examined laboratory codes were represented in UMLS, and that various translations to other languages were available for these concepts. The classification that is most widely used in German hospital for documenting clinical procedures was not originally represented in UMLS, but external mappings to SNOMED CT allowed identifying UMLS entries for 90.5% of frequent codes. Future research could extend this investigation to other code sets and terminologies, or study the potential utility of available mappings for specific applications.

Among medical applications of natural language processing (NLP), word sense disambiguation (WSD) estimates alternative meanings from text around homonyms. Recently developed NLP methods include word vectors that combine easy computability with nuanced semantic representations. Here we explore the utility of simple linear WSD classifiers based on aggregating word vectors from a modern biomedical NLP library in homonym contexts. We evaluated eight WSD tasks that consider literature abstracts as textual contexts. Discriminative performance was measured in held-out annotations as the median area under sensitivity-specificity curves (AUC) across tasks and 200 bootstrap repetitions. We find that classifiers trained on domain-specific vectors outperformed those from a general language model by 4.0 percentage points, and that a preprocessing step of filtering stopwords and punctuation marks enhanced discrimination by another 0.7 points. The best models achieved a median AUC of 0.992 (interquartile range 0.975 - 0.998). These improvements suggest that more advanced WSD methods might also benefit from leveraging domain-specific vectors derived from large biomedical corpora.

Background: Assessing the uncertainty of diagnostic findings is essential for advising patients. Previous research has demonstrated the difficulty of computing the expected correctness of positive or negative results, although clinical decision support (CDS) tools promise to facilitate adequate interpretations., Objectives: To teach the potential utility of CDS tools to medical students, we designed an interactive software module that computes and visualizes relevant probabilities from typical inputs., Methods: We reviewed the literature on recommended graphical approaches and decided to support contingency tables, plain table formats, tree diagrams, and icon arrays., Results: We implemented these functions in a single-page web application, which was configured to complement our local learning management system where students also access interpretation tasks., Conclusion: Our technical choices promoted a rapid implementation. We intend to explore the utility of the tool during some upcoming courses. Future developments could also model a more complex clinical reality where the likelihood of alternative diagnoses is estimated from sets of clinical investigations.

Preparations for anesthesiological management of patients build on preoperative patient self-reports concerning risk factors and comorbidities. In this setting, electronic documentation could facilitate innovative computerized functions, although patient-facing digital questionnaires require appropriate tools that patients can access effectively. To explore the feasibility of an electronic application for preoperative data acquisition directly from patients, a digital, tablet-based prototypical application has been developed within a user-centered design process in order to replace a previously used paper-based anamnesis sheet for perioperative risk evaluation. The implemented prototype has been extensively tested and iteratively improved to progressively provide an easy-to-use data entry function. To assess the suitability of this tool for everyday data acquisition by patients and physicians and to identify usability problems, the stepwise development process was accompanied by a heuristic evaluation as well as a think-aloud evaluation, while another 56 participating patients completed a feedback sheet according to ISO 9241/10. The latter method detected additional usability problems that occurred during the use of the application, which contributed to iterative improvements of the prototype. Throughout the development process, 81 issues were identified and largely resolved. After these revisions of the prototype, the number of problems found per tester decreased from 4.75 to 0.96, while the overall rating increased to 6.14 out of 7 points (SD = 1.2). These improvements demonstrate the value and efficiency of such a user-centered design process and illustrate that a user-friendly patient-facing digital data entry can replace preoperative paper questionnaires for anesthesiological management.