Abstract:
According to WHO, Adverse drug reactions (ADRs) can be considered as any harmful
and undesired consequences of a drug that occur after a patient takes the normal or
prescribed doses of drug used for prevention, diagnosis, and treatment. Once the drug
is completely developed, it goes through multiple phases of tests and trails to ensure
that drug is safe to use before it is release to the market. Although these tests play
a very vital role in drug safety assurance process, but these are not enough to ensure
that drug is completely safe to use for everyone. A large number of drug side effects
are identified after the drug is consumed by general public which were missed in the
pre-clinical tests. Many times the general public is not aware of drug’s negative impact
due to its under reporting. Thus, after the drug is delivered to the market, various
monitoring activities related to drug safety are performed. Major reliance of these activities
is on passive databases like FAERS. These database system are not up to date
and also take a lot of time to identify ADR. This results in huge number of unreported
ADRs. The damage caused by the under reporting of adverse drug reactions motivated
researchers to design system that uses automated tools and techniques for extracting
ADR mentions from biomedical text. Handling large amount of healthcare data, extracting
meaningful features, and choosing the most suitable and efficient model for
classification of ADRs are still the main concerns. Keeping these challenges in mind,
this research introduces topic modeling based approach for feature extraction and constructs
an end-to-end model for the identification of ADRs in medical text.
