Abstract
BackgroundStructured electronic health records are a rich resource for identifying novel correlations, such as co-morbidities and adverse drug reactions. For drug development and better understanding of biomedical phenomena, such correlations need to be supported by viable hypotheses about the mechanisms involved, which can then form the basis of experimental investigations.MethodsIn this study, we demonstrate the use of discovery browsing, a literature-based discovery method, to generate plausible hypotheses elucidating correlations identified from structured clinical data. The method is supported by Semantic MEDLINE web application, which pinpoints interesting concepts and relevant MEDLINE citations, which are used to build a coherent hypothesis.ResultsDiscovery browsing revealed a plausible explanation for the correlation between epilepsy and inflammatory bowel disease that was found in an earlier population study. The generated hypothesis involves interleukin-1 beta (IL-1 beta) and glutamate, and suggests that IL-1 beta influence on glutamate levels is involved in the etiology of both epilepsy and inflammatory bowel disease.ConclusionsThe approach presented in this paper can supplement population-based correlation studies by enabling the scientist to identify literature that may justify the novel patterns identified in such studies and can underpin basic biomedical research that can lead to improved treatments and better healthcare outcomes.
Highlights
Structured electronic health records are a rich resource for identifying novel correlations, such as co-morbidities and adverse drug reactions
IL-1 beta and epilepsy The initial query to Semantic MEDLINE was (“interleukin 1” AND), which returned 2481 predications extracted from 240 citations
“Interleukin-1 beta-AFFECTS-Seizure” was extracted from Vezzani et al [44], which reports that intrahippocampal application of recombinant IL-1 receptor antagonist (IL-1ra) inhibits seizures experimentally induced by bicuculline methiodide in rodents
Summary
Structured electronic health records are a rich resource for identifying novel correlations, such as co-morbidities and adverse drug reactions. Structured data from electronic health records (EHRs) are increasingly mined to identify novel correlations, such as disease co-occurrences or adverse drug reactions [5] Such studies are sometimes highly localized, relying on Rindflesch et al Journal of Biomedical Semantics (2018) 9:25 data collected from a small set of institutions; they can violate some of the key assumptions made when using traditional statistical measures to determine significance, leading to false positive associations [6]. When performed with population-level data (e.g., Medicare claims data), these data mining studies can provide epidemiological evidence for co-morbidities and other biomedical phenomena; they alone are unable to elucidate the mechanisms involved in such phenomena or offer plausible explanations Such epidemiological evidence must be subjected to further analysis by scientists in order to generate viable hypotheses about the etiology of the observed correlations, a critical step for the development of safe and effective treatments
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
