Mapping of adverse events - identified by a high throughput data mining technology in a diabetes population - to an international consensus terminology

Abstract

Formål: Formålet med dette studie er at foretage et kvalitets mapping ind til MedDRA, ved at, udfører tekst mining med formålet om opdage mulige bivirkninger, og dermed være i stand til at mappe termerne ind til en medicinsk terminologi. Introduktion: Polyfarmaci er den mest almindelige behandlingsstrategi hos diabetespatienter, som er forbundet med mange følgekomplikationer; Dog er den vigtigste interaktioner mellem to forskellige lægemidler. Pharmako-overvågelse vigtig efter et godkendt lægemiddel. Medical Dictionary for Regulatory Activity (MedDRA) blev oprettet for at dele og hente nye fund om et lægemiddel på kryds af forskellige lande. Den er opbygget af et 5 hierarkisk niveauer fra meget generelt til meget specifikt. Hvert term er karakteriseret med et 8-cifret nummer, og forbliver det samme tal på alle sprog. Denne form for kommunikation tillader videnskaber at forstå og dele nyt viden på tværs af forskellige sprog, hvilket leder til bedre sundhed globalt. Metode: En automatisk databehandlingsteknologi blev udført i 12.073 ustrukturerede elektroniske patientjournaler. Naturlig sprogbehandling (NLP) algoritmen blev udført for at strukturere, forstå og validere relevante information fra teksterne. Unikke tegn der blev opdaget blev oversat i MedDRA, der blev oprettet af International Classification of disease (ICD) i kombination med in-house ordbog for at være i stand til at sammenligne databasen på internationalt niveau. Resultater: Anvendelse af tekst mining i 12.073 EPR med NLP gav os 3.830 unikke tegn, hvilket kan være mulige bivirkninger. Alle termer var i stand til at blive oversat til MedDRA. Simple ord var nemmere at mappe sammenlignet med komplekse ord. Herudover var vi i stand til at trække fænotypes profiler samt de mest hyppige recept skrevne medikamenter samt relaterede bivirkninger. Konklusion: Baseret på vores resultater kan vi konkludere, at anvendelse af MedDRA som vores medicinske terminologi i kombination med vores in-house ordbog var muligt at udføre og kan give en sofistikeret analyse på et skandinavisk samt internationalt plan. Derudover var det muligt at udføre tekst mining til at opdage mulige unikke tegn, hvilket betyder at det er et godt værktøj til håndtering af store databaser, hvilket er tilfældet i de fleste kliniske sikkerhedsundersøgelser.

Aim: The purpose of this study is to create a quality mapping to MedDRA, by applying text-mining to detect possible adverse events (AE) and thereby be able to map the terms into an international medical dictionary. Introduction: Polypharmacy is the most common treatment strategy in diabetic patients, which is associated with various complications; however, the most important is drug-drug interactions (DDI). Pharmacovigilance (PV) is significant in drug development after an approved drug. The Medical Dictionary of Regulatory Activity (MedDRA) was developed to share findings and retrieve the latest information on drugs. A five level hierarchy, very general to very specific, structures it. Every term is identified by an 8-digit number, which describes the term of interest and remains the same number regarding of language. This kind of communications allows researcher to share and understand new findings across different languages, leading to better health globally. Methods: An automatic computational text mining technique was conducted in 12,073 unstructured electronic patient records (EPR). Natural language processing (NLP) algorithm was executed to structure, understand and validate relevant information. Detected unique terms were mapped into MedDRA in combination with the in-house dictionary, established by International Classification of disease (ICH) to achieve an internationally comparable dataset. Results: Application of text mining in 12,073 EPRs with NLP gave us 3,830 unique terms, which may be possible AEs. All unique terms were feasible to be mapped into MedDRA. Simple words were easier to be mapped compared to terms that were more complex. Moreover, extraction of the phenotypic profiles, as well as the most common prescribed drugs with associated AEs were able to be extracted. Conclusion: Based on our results, we conclude that the use of MedDRA as our dictionary in combination with the in-house NER-tagger were possible to use and can give a sophisticated analysis on a Scandinavian as well as international level. Moreover, application of text mining detect possible AE, and is a great tool when dealing with large databases, which is the case in most safety trails.

A master thesis from Aalborg University

Mapping of adverse events - identified by a high throughput data mining technology in a diabetes population - to an international consensus terminology

[Mapping af bivirkninger identificeret af data mining teknologi i en diabetes population til en international ordbog]