Investigating clinical adverse events from approved drugs: toward a harmonized approach in toxicological data integration

The increasing need for a unified and standardized approach to toxicological data has driven this thesis, which aims to improve the integration and understanding of human adverse events linked to approved drugs. Despite the availability of numerous sources of clinical trial data and toxicological data coming from in vitro and in vivo studies, it is difficult to understand according to which criteria these data are structured and managed. Actually, they are often scattered, heterogeneous, and difficult to access in a standardized form. This work aims help bridging the gap between in vivo and human studies by proposing a unified data model that connects and structures toxicological information coming from human studies. Ideally, this work aims to complement and integrate with the IUCLID drug dataset, which contains toxicological in vivo data. The final goal is creating a comprehensive resource that links human clinical trial information with preclinical animal study outcomes. The core of this research is to design a comprehensive data model, developed after an extensive review of accredited sources, including the FDA database and the European Union Clinical Trials Register, to ensure robust and reliable integration of clinical trial toxicological data. The selection targeted Phase 3 clinical trials, initially including at least one drug per ATC class. Next, a refined dataset of about twenty-five antidiabetic drugs was added for focused statistical analysis. Within the data model that was conceived, each adverse event is recorded as a single entry and is linked to its corresponding clinical trial and annotated using standardized ontological terms and codes. Crucial parameters such as dosage, administration route, and treatment duration are also included for each entry, along with the incidence rate of the event in each trial group (e.g., treatment vs. control). This structure allows for rapid access to specific, high-value toxicological information that would otherwise require considerable effort to be extracted, interpreted and organized manually. Although the proof-of-concept extracted data have been stored in a pilot Excel file, the model already demonstrates strong potential for scalability and future digital implementation. An overview of the results highlights that each drug presents a variable number of reported effects, without the identification of a minimum threshold to be included. Another piece of evidence is that most of the observed effects were classified as clinical signs, while laboratory parameters (urinalysis, clinical biochemistry, and haematology) were often overlooked. In addition, a more in-depth evaluation conducted on 3 drugs used in antidiabetic therapy revealed no significant correlation between the administered dose and the incidence of the observed effects. Evidence indicates that the database can be effectively applied to different therapeutic areas and to studies encompassing a broad range of trial designs. Most notably, the database emerges as a valuable toxicological tool that can support the development of New Approach Methodologies (NAMs), which aim to reduce animal experimentation by leveraging high-quality existing toxicological data.