Adverse drug reactions (ADR) are a major concern for both patient safety and the overall success of pharmacological treatments. These harmful—and potentially serious—effects can occur even when medications are taken at standard doses prescribed for the prevention, diagnosis, or treatment of illness.

Given their significant implications for public health, ADRs are subject to mandatory reporting in most countries. Healthcare professionals are legally obligated to report suspected cases to their respective national regulatory authorities. In Switzerland, this responsibility is overseen by Swissmedic, which operates through a network of accredited regional centers. In the Canton of Ticino, pharmacovigilance responsibilities are assigned to the Ente Ospedaliero Cantonale (EOC) and the Istituto di Scienze Farmacologiche Svizzera Italiana (ISFSI).

Within this context, the project NLP in Support of Pharmacovigilance: QUality Adverse Drug Reaction AcTIve Control (QUADRATIC) was established as a joint initiative between the Natural Language Processing Group at the Dalle Molle Institute for Artificial Intelligence (IDSIA USI-SUPSI) and EOC/ISFSI, with funding from the Swiss National Science Foundation (SNSF). The goal of the project is to explore how artificial intelligence methods can support and enhance pharmacovigilance activities—particularly the detection, assessment, understanding, and prevention of ADRs.

“Although reporting adverse drug reactions (ADRs) is a legal requirement, in practice it remains infrequent—especially for hospitalized patients,” explains Fabio Rinaldi, Senior Researcher at SUPSI and IDSIA. “The process is often time-consuming, demands specialized knowledge, and requires cross-checking multiple sources of information. All of this adds to the already heavy workload of healthcare professionals.”

To tackle these challenges, the project focused on the development of innovative tools, based on Natural Language Processing techniques and Artificial Intelligence technologies, to support and enhance the efficiency of the pharmacovigilance process.
 

Discharge Letters: a Key Resource for Smarter Pharmacovigilance

The experimental phase focused on hospital discharge letters, which provide a detailed overview of the key clinical events occurred during a patient’s hospital stay.

“Our goal is to provide effective, reliable support that eases the workload of pharmacovigilance experts,” says Fabio Rinaldi. “By automatically filtering out non-relevant cases and simplifying the extraction of essential information, we can significantly reduce the time and effort required for manual review.”

The project addressed two key areas:

1. Automatic Detection of Adverse Drug Reactions (ADRs): a system was developed to autonomously analyze discharge letters and identify potential ADRs. Suspected cases are flagged and forwarded to pharmacovigilance specialists for verification and validation.
    

2. Extraction of Essential Clinical Data: once an ADR is confirmed, the system extracts the core information needed for reporting to Swissmedic—including the names of the drugs involved, the described medical events, and related signs and symptoms.

By extending this AI-powered approach to other regional pharmacovigilance centers, it would be possible to significantly streamline reporting workflows, enhance post-marketing drug surveillance, and ultimately contribute to improved public health outcomes.