(341) Development and Evaluation of Python Language Processed Automated Disproportionality Analysis System for FAERS Database

Background: The existing limitations and time consuming process of disproportionality analysis of FDA-Adverse Event Reporting System (FAERS) data necessitates the requirement of automated tools for data mining and analysis.

Objectives: We aimed to develop an automated tool with the help of python language to perform the disproportionality analysis to assess an actual causal relationship between the exposed drug and the specific ADR.

Methods: This study follows a two-phase methodology which includes i) Development of automation using python language and signal detection; and ii) Validation through traditional disproportionality analysis. The publicly available international FDA Adverse Event Reporting System, now termed FAERS database (https://fis.fda.gov/extensions/FPD-QDE-FAERS/FPD-QDE-FAERS.html) was used for collecting the data. We have accessed the public version of quarterly extract files from 2004 (Q1; January-March)-2021 (Q4; October-December). The data files were retrieved from the year of the first marketing approval of the drug. All the reports were matched by various methods and the duplicates were removed and merged. After merging all the files the output file is generated with data elements. This data is further used to perform statistical analysis using a contingency table and to calculate ROR, PRR, Chi-square statistic and lower and upper limit 95% confidence intervals.

Results: We have taken “remdesivir” as an example to process the FAERS data trough automation and same was validated through traditional manual disproportionality analysis. The data of remdesivir from inception to December 2021 has been incorporated for data extraction. During the validation process, we were able to confirm that, the data retrieved from FAERS databases through the Python language automation and manual processing were similar. Additionally, the files removed through extreme deduplication process by automation are in accordance with the criteria and it has been confirmed manually. An ROR and PRR was calculated by the automated tool directly and the results were extracted to Excel files. We have got a total of 12777 adverse events (Pt’s) reported for Remdesivir and a total of 256 signals have been identified for Remdesivir.

Conclusions: The use of simple codes and languages, consumption of less storing space, cloud working with Google Collab, easy understanding of non-technical personals, less time consuming, tailor the drug reaction based on the investigator requirement, production of refined results and the automatic updating of data system is the advantage of this validated automated tool. This can be translated into practice through developing the suitable interfaces with desktop tools.