Data-driven Identification of Factors That Determine the Quality of Spontaneous Reports for Malaysia: A 15-Year Analysis of QUEST and VigiBase

TMUIR > College of Medical Science and Technology > > Dissertations/Theses > Item 987654321/61503

Please use this identifier to cite or link to this item: http://libir.tmu.edu.tw/handle/987654321/61503

Title:	Data-driven Identification of Factors That Determine the Quality of Spontaneous Reports for Malaysia: A 15-Year Analysis of QUEST and VigiBase
Authors:	徐欣镁 MEI, CHOO SIM
Contributors:	醫學資訊研究所碩士班雪碧兒
Keywords:	data quality;global pharmacovigilance;spontaneous reporting;interpretable machine learning;hybrid feature selection;feature importance;time series analysis
Date:	2021-07-07
Issue Date:	2022-03-28 18:55:41 (UTC+8)
Abstract:	Background: Completeness of spontaneous reports is a necessary element to assess putative causal relationships. Little information is known regarding the root causes that have contributed to the spontaneous report quality. Malaysian spontaneous reports are above the global average ~0.44 completeness score, with an average score of 0.79 over the past five years – with 0.80 being a reference value for reports to be considered well-documented. Aims: The primary aim of this study is to employ a hypothesis-free, data-driven approach to explore the main drivers of the completeness of Malaysian spontaneous reports in VigiBase and to understand what actions were taken by the Malaysian authorities to achieve the relatively high completeness of reported data. A secondary aim is to identify target areas where the completeness of reported data to VigiBase could be improved. Methods: We employed the vigiGrade completeness score (C) method to evaluate the quality of spontaneous reports and factors associated with well-documented reports (C > 0.8). We studied 132,738 reports received in VigiBase as of February 2021, that occurred in Malaysia and received by the National Pharmaceutical Regulatory Agency (NPRA) between 2005 and 2019. Supplementary data were obtained from the Malaysian QUEST database. Given the distinctive difference in reporting elements, we further divided the dataset into INTDIS (63943 reports between 2005 and 2016) and E2B subsets (68795 reports between 2015 and 2019). We conducted time series analyses to study and summarise the chronological trends and impacts of key interventions on reporting quality. For machine learning analyses, we performed a two-stage feature selection followed by a random forest classifier algorithm to identify the top important features that determine the well-documented Malaysian reports. We then examined the magnitude, prevalence, and direction of feature effects using a Shapley additive explanations (SHAP) algorithm for a tree-based model (TreeExplainer). Results: More than two-thirds (67.1%) of E2B reports were classified as well-documented, while 16.7% of the historical INTDIS reports were well-documented. Higher staffing level of pharmacovigilance centre, reaction abated upon drug dechallenge, time-to-onset/drug duration within a day or a week, dosing interval within a day, reports from public specialist hospitals, reports by pharmacists, reporting via pharmacist information hospital system (PhIS)-integrated tool, and use of systemic antimicrobials were identified as the top important features that positively associated with well-documented Malaysian reports. Whereas the reports from product registration holders (PRHs) and other healthcare professional (HCPs) were found to negatively contribute to the quality of Malaysian reports. Several data quality issues upon report conversion and submission to VigiBase were also uncovered through vigiGrade. Conclusion: Multifaceted strategies and interventions comprising policy changes, continuity of education, and human resource development contributed to the foundation for spontaneous reporting in Malaysia, while technological infrastructure and enhancements on pharmacovigilance database and reporting tool bolstered both the quantity and quality of spontaneous reports. Findings from machine learning analysis indicated that the nature of drugs and reactions could further determine the quality of a report. Suboptimal contribution from private health sectors necessitates further research to better understand the behavioural and organisational barriers. Competency-based trainings, cultivation of data use culture, and quality assurance procedures are recommended to further enhance the quality of spontaneous data during generation, management, harmonisation and transmission.
Description:	碩士指導教授：雪碧兒委員：烏斯馬委員：楊軒佳
Data Type:	thesis
Appears in Collections:	[ ] Dissertations/Theses

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	123	View/Open

View Licence

Loading...