Mapping of adverse events - identified by a high throughput data mining technology in a diabetes population - to an international consensus terminology
Translated title
Mapping af bivirkninger identificeret af data mining teknologi i en diabetes population til en international ordbog
Author
Palani, Fatima
Term
4. term
Publication year
2019
Submitted on
2019-09-30
Pages
38
Abstract
Formålet var at bruge automatiseret tekstmining til at opdage mulige bivirkninger i elektroniske patientjournaler og kortlægge dem til MedDRA, en international, flersproget ordbog til lægemiddelovervågning. Diabetespatienter får ofte mange lægemidler på samme tid (polyfarmaci), hvilket øger risikoen for interaktioner mellem lægemidler. Derfor er lægemiddelovervågning efter godkendelse vigtig. MedDRA organiserer termer i et fem-niveaus hierarki fra brede til specifikke begreber. Hver term har en 8-cifret kode, som er den samme på tværs af sprog, så viden kan deles internationalt. Vi analyserede 12.073 ustrukturerede elektroniske patientjournaler med naturlig sprogbehandling (NLP) for at strukturere, forstå og validere relevant information. De unikke udtryk, vi fandt, blev kortlagt til MedDRA ved hjælp af en intern ordbog, etableret af International Classification of disease (ICH), for at skabe et internationalt sammenligneligt datasæt. Analysen gav 3.830 unikke termer som mulige bivirkninger. Alle kunne kortlægges til MedDRA; simple ord var lettere at matche end mere komplekse formuleringer. Vi kunne også udtrække fænotypiske profiler samt de mest ordinerede lægemidler med tilknyttede bivirkninger. MedDRA kombineret med vores interne NER-tagger gjorde det muligt at lave en robust analyse, der kan anvendes i både skandinaviske og internationale sammenhænge. Tekstmining er et nyttigt værktøj til at finde mulige bivirkninger i store datasæt, som dem der bruges i sikkerhedsstudier.
This study used automated text mining to detect possible adverse events in electronic patient records and map them to MedDRA, an international, multilingual dictionary used in pharmacovigilance. People with diabetes often take many medicines at once (polypharmacy), increasing the risk of drug–drug interactions, so post-approval safety monitoring is important. MedDRA is organized as a five-level hierarchy from broad to specific concepts. Each term has an 8-digit code that is the same across languages, enabling international sharing of findings. We analyzed 12,073 unstructured electronic patient records with natural language processing (NLP) to structure, understand, and validate relevant information. The unique terms we detected were mapped to MedDRA using an in-house dictionary, established by the International Classification of disease (ICH), to produce an internationally comparable dataset. The analysis yielded 3,830 unique terms as candidate adverse events. All could be mapped to MedDRA; simple words were easier to match than more complex expressions. We also extracted phenotypic profiles and identified the most commonly prescribed drugs with their associated adverse events. Combining MedDRA with our in-house NER-tagger enabled robust analyses applicable in Scandinavian and international settings. Text mining is a useful tool for detecting possible adverse events in large datasets, such as those used in safety studies.
[This abstract was generated with the help of AI]
Documents