Dataset Information

Learning a Health Knowledge Graph from Electronic Medical Records.

ABSTRACT: Demand for clinical decision support systems in medicine and self-diagnostic symptom checkers has substantially increased in recent years. Existing platforms rely on knowledge bases manually compiled through a labor-intensive process or automatically derived using simple pairwise statistics. This study explored an automated process to learn high quality knowledge bases linking diseases and symptoms directly from electronic medical records. Medical concepts were extracted from 273,174 de-identified patient records and maximum likelihood estimation of three probabilistic models was used to automatically construct knowledge graphs: logistic regression, naive Bayes classifier and a Bayesian network using noisy OR gates. A graph of disease-symptom relationships was elicited from the learned parameters and the constructed knowledge graphs were evaluated and validated, with permission, against Google's manually-constructed knowledge graph and against expert physician opinions. Our study shows that direct and automated construction of high quality health knowledge graphs from medical records using rudimentary concept extraction is feasible. The noisy OR model produces a high quality knowledge graph reaching precision of 0.85 for a recall of 0.6 in the clinical evaluation. Noisy OR significantly outperforms all tested models across evaluation frameworks (p?

SUBMITTER: Rotmensch M

PROVIDER: S-EPMC5519723 | biostudies-literature | 2017 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Learning a Health Knowledge Graph from Electronic Medical Records.

Rotmensch Maya M Halpern Yoni Y Tlimat Abdulhakim A Horng Steven S Sontag David D

Scientific reports 20170720 1

Demand for clinical decision support systems in medicine and self-diagnostic symptom checkers has substantially increased in recent years. Existing platforms rely on knowledge bases manually compiled through a labor-intensive process or automatically derived using simple pairwise statistics. This study explored an automated process to learn high quality knowledge bases linking diseases and symptoms directly from electronic medical records. Medical concepts were extracted from 273,174 de-identifi ...[more]

PMID: 28729710

Similar Datasets

Project description:BackgroundDeep learning models have attracted significant interest from health care researchers during the last few decades. There have been many studies that apply deep learning to medical applications and achieve promising results. However, there are three limitations to the existing models: (1) most clinicians are unable to interpret the results from the existing models, (2) existing models cannot incorporate complicated medical domain knowledge (eg, a disease causes another disease), and (3) most existing models lack visual exploration and interaction. Both the electronic health record (EHR) data set and the deep model results are complex and abstract, which impedes clinicians from exploring and communicating with the model directly.ObjectiveThe objective of this study is to develop an interpretable and accurate risk prediction model as well as an interactive clinical prediction system to support EHR data exploration, knowledge graph demonstration, and model interpretation.MethodsA domain-knowledge-guided recurrent neural network (DG-RNN) model is proposed to predict clinical risks. The model takes medical event sequences as input and incorporates medical domain knowledge by attending to a subgraph of the whole medical knowledge graph. A global pooling operation and a fully connected layer are used to output the clinical outcomes. The middle results and the parameters of the fully connected layer are helpful in identifying which medical events cause clinical risks. DG-Viz is also designed to support EHR data exploration, knowledge graph demonstration, and model interpretation.ResultsWe conducted both risk prediction experiments and a case study on a real-world data set. A total of 554 patients with heart failure and 1662 control patients without heart failure were selected from the data set. The experimental results show that the proposed DG-RNN outperforms the state-of-the-art approaches by approximately 1.5%. The case study demonstrates how our medical physician collaborator can effectively explore the data and interpret the prediction results using DG-Viz.ConclusionsIn this study, we present DG-Viz, an interactive clinical prediction system, which brings together the power of deep learning (ie, a DG-RNN-based model) and visual analytics to predict clinical risks and visually interpret the EHR prediction results. Experimental results and a case study on heart failure risk prediction tasks demonstrate the effectiveness and usefulness of the DG-Viz system. This study will pave the way for interactive, interpretable, and accurate clinical risk predictions.

Project description:BackgroundWith the increasing incidences and mortality of digestive system tumor diseases in China, ways to use clinical experience data in Chinese electronic medical records (CEMRs) to determine potentially effective relationships between diagnosis and treatment have become a priority. As an important part of artificial intelligence, a knowledge graph is a powerful tool for information processing and knowledge organization that provides an ideal means to solve this problem.ObjectiveThis study aimed to construct a semantic-driven digestive system tumor knowledge graph (DSTKG) to represent the knowledge in CEMRs with fine granularity and semantics.MethodsThis paper focuses on the knowledge graph schema and semantic relationships that were the main challenges for constructing a Chinese tumor knowledge graph. The DSTKG was developed through a multistep procedure. As an initial step, a complete DSTKG construction framework based on CEMRs was proposed. Then, this research built a knowledge graph schema containing 7 classes and 16 kinds of semantic relationships and accomplished the DSTKG by knowledge extraction, named entity linking, and drawing the knowledge graph. Finally, the quality of the DSTKG was evaluated from 3 aspects: data layer, schema layer, and application layer.ResultsExperts agreed that the DSTKG was good overall (mean score 4.20). Especially for the aspects of "rationality of schema structure," "scalability," and "readability of results," the DSTKG performed well, with scores of 4.72, 4.67, and 4.69, respectively, which were much higher than the average. However, the small amount of data in the DSTKG negatively affected its "practicability" score. Compared with other Chinese tumor knowledge graphs, the DSTKG can represent more granular entities, properties, and semantic relationships. In addition, the DSTKG was flexible, allowing personalized customization to meet the designer's focus on specific interests in the digestive system tumor.ConclusionsWe constructed a granular semantic DSTKG. It could provide guidance for the construction of a tumor knowledge graph and provide a preliminary step for the intelligent application of knowledge graphs based on CEMRs. Additional data sources and stronger research on assertion classification are needed to gain insight into the DSTKG's potential.

Project description:Post-market medical device surveillance is a challenge facing manufacturers, regulatory agencies, and health care providers. Electronic health records are valuable sources of real-world evidence for assessing device safety and tracking device-related patient outcomes over time. However, distilling this evidence remains challenging, as information is fractured across clinical notes and structured records. Modern machine learning methods for machine reading promise to unlock increasingly complex information from text, but face barriers due to their reliance on large and expensive hand-labeled training sets. To address these challenges, we developed and validated state-of-the-art deep learning methods that identify patient outcomes from clinical notes without requiring hand-labeled training data. Using hip replacements-one of the most common implantable devices-as a test case, our methods accurately extracted implant details and reports of complications and pain from electronic health records with up to 96.3% precision, 98.5% recall, and 97.4% F1, improved classification performance by 12.8-53.9% over rule-based methods, and detected over six times as many complication events compared to using structured data alone. Using these additional events to assess complication-free survivorship of different implant systems, we found significant variation between implants, including for risk of revision surgery, which could not be detected using coded data alone. Patients with revision surgeries had more hip pain mentions in the post-hip replacement, pre-revision period compared to patients with no evidence of revision surgery (mean hip pain mentions 4.97 vs. 3.23; t?=?5.14; p?<?0.001). Some implant models were associated with higher or lower rates of hip pain mentions. Our methods complement existing surveillance mechanisms by requiring orders of magnitude less hand-labeled training data, offering a scalable solution for national medical device surveillance using electronic health records.

Project description:BACKGROUND:Administrative health records (AHRs) and electronic medical records (EMRs) are two key sources of population-based data for disease surveillance, but misclassification errors in the data can bias disease estimates. Methods that combine information from error-prone data sources can build on the strengths of AHRs and EMRs. We compared bias and error for four data-combining methods and applied them to estimate hypertension prevalence. METHODS:Our study included rule-based OR and AND methods that identify disease cases from either or both data sources, respectively, rule-based sensitivity-specificity adjusted (RSSA) method that corrects for inaccuracies using a deterministic rule, and probabilistic-based sensitivity-specificity adjusted (PSSA) method that corrects for error using a statistical model. Computer simulation was used to estimate relative bias (RB) and mean square error (MSE) under varying conditions of population disease prevalence, correlation amongst data sources, and amount of misclassification error. AHRs and EMRs for Manitoba, Canada were used to estimate hypertension prevalence using validated case definitions and multiple disease markers. RESULTS:The OR method had the lowest RB and MSE when population disease prevalence was 10%, and the RSSA method had the lowest RB and MSE when population prevalence increased to 20%. As the correlation between data sources increased, the OR method resulted in the lowest RB and MSE. Estimates of hypertension prevalence for AHRs and EMRs alone were 30.9% (95% CI: 30.6-31.2) and 24.9% (95% CI: 24.6-25.2), respectively. The estimates were 21.4% (95% CI: 21.1-21.7), for the AND method, 34.4% (95% CI: 34.1-34.8) for the OR method, 32.2% (95% CI: 31.8-32.6) for the RSSA method, and ranged from 34.3% (95% CI: 34.1-34.5) to 35.9% (95% CI, 35.7-36.1) for the PSSA method, depending on the statistical model. CONCLUSIONS:The OR and AND methods are influenced by correlation amongst the data sources, while the RSSA method is dependent on the accuracy of prior sensitivity and specificity estimates. The PSSA method performed well when population prevalence was high and average correlations amongst disease markers was low. This study will guide researchers to select a data-combining method that best suits their data characteristics.

Dataset Information

Learning a Health Knowledge Graph from Electronic Medical Records.

Publications

Learning a Health Knowledge Graph from Electronic Medical Records.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets