Nationwide adverse drug reaction screening using high process computing

Research output: Contribution to conferenceAbstract

Abstract

Background: Detecting adverse drug reactions has a very important role in minimizing drug‑related harms, particularly in older adults. Previous stud
focussed on ascertaining known adverse drug reactions (ADRs). We present here a sequence‑symmetry based approach aided by high process comp
systematically detect putative signals of abrupt ADRs and improve efficiency of screening of administrative health databas es.
Objectives: The aim of this study is to identify all signals of significant drug‑event associations that can be abrupt ADRs, in the New Zealand (NZ) 65y
parallelizing sequence‑symmetry calculations.
Methods: We sourced hospital events and prescription data (2005‑2016) from the NZ Ministry of Health. We processed the data using the HPC facility
and created a self‑matched cohort. The day an individual started a drug of interest is the index‑date, and this ascertains non‑exposure before the inde
a pre‑initiation window and a post‑initiation window before and after the initiation respectively. We identified a first‑time diagnosis of interest in both
lengths of 7, 15 and 30 days were considered. Short window lengths ascertain current or recent first‑time exposures at occurrences of new events, and
can be abrupt ADRs. We calculated adjusted sequence ratios (ASR) to measure the increased odds of first‑time event post‑initiation relative to the back
Hallas’ method, with modification that the calculations are within the pre‑post window‑pair. We stratified the cohort by drug and diagnosis code and us
ASRs in parallel. Starting with 7‑day windows, we detect strong signals with 100% odd increases. New ADR signals were identified with longer
Results: Our dataset has 2,170,393 new events, covering 9009 ICD‑10‑AM codes; and 37,359,480 drug initiations, covering 914 drugs. We found a to
drug‑event associations including well‑known ADRs, for example ibuprofen and GI‑bleeding (K92.2, log(ASR) = 0.79 [0.41‑1.16]), as well as warfar
haemorrhage (I62.0, log(ASR) = 1.94 [0.52‑3.36]). A number of less well‑known associations that can be ADRs were also detected, for example bend
hyponatremia (E87.1, log(ASR) = 1.89 [1.22‑2.56]), codeine and constipation (K59.0, log(ASR) = 0.90 [0.66‑1.13 ]).
Conclusions: With the aid of HPC, the symmetry‑based algorithm has the potential for the development of an efficient, big‑data ADR‑signal detection s
medication safety and advance post‑marketing surveillance.
Original languageEnglish
Pages401-401
Number of pages1
Publication statusPublished - 1 Aug 2019

Fingerprint Dive into the research topics of 'Nationwide adverse drug reaction screening using high process computing'. Together they form a unique fingerprint.

Cite this