Background: Drug interactions resulting in adverse drug reactions (ADRs) represent a major health problem both for individuals and society in general. Post-marketing pharmacovigilance reporting databases with compiled individual case safety reports (ICSRs) have been shown to be particularly useful in the detection of novel drug - ADR combinations, though these reports have not been fully used to detect adverse drug interactions. Aim: To explore the potential to identify drug interactions using ICSRs and to develop a method to facilitate the detection of adverse drug interaction signals in the WHO Global ICSR Database, VigiBase. Methods: All six studies included in this thesis are based on ICSRs available in VigiBase. Two studies aimed to characterise drug interactions reported in VigiBase. In the first study we examined if contraindicated drug combinations (given in a reference source of drug interactions) were reported on the individual reports in the database, and in the second study we examined the scientific literature for interaction mechanisms for drug combinations most frequently co-reported as interacting in VigiBase. Two studies were case series analyses where the individual reports were manually reviewed. The two remaining studies aimed to develop a method to facilitate detection of novel adverse drug interactions in VigiBase. One examined what information (referred to as indicators) was reported on ICSRs in VigiBase before the interactions became listed in the literature. In the second methodological study, logistic regression was used to set the relative weights of the indicators to form triage algorithms. Three algorithms (one completely data driven, one semi-automated and one based on clinical knowledge) based on pharmacological and reported clinical information and the relative reporting rate of an ADR with a drug combination were developed. The algorithms were then evaluated against a set of 100 randomly selected case series with potential adverse drug interactions. The algorithm’s performances were then evaluated among DDAs with high coefficients. Results: Drug interactions classified as contraindicated are reported on the individual reports in VigiBase, although they are not necessarily recognised as interactions when reported. The majority (113/123) of drug combinations suspected for being responsible for an ADR were established drug interactions in the literature. Of the 113 drug interactions 46 (41%) were identified as purely pharmacodynamic; 28 (25%) as pharmacokinetic; 18 (16%) were a mix of both types and for 21 (19%) the mechanism have not yet been identified. Suspicions of a drug interaction explicitly noted by the reporter are much more common for known adverse drug interactions than for drugs not known to interact. The clinical evaluation of the triage algorithms showed that 20 were already known in the literature, 30 were classified as signals and 50 as not signals. The performance of the semi-automated and the clinical algorithm were comparable. In the end the clinical algorithm was chosen. At a relevant level, 38% were of the adverse drug interactions were already known in the literature and of the remaining 80% were classified as signals for this algorithm. Conclusions: This thesis demonstrated that drug interactions can be identified in large post-marketing pharmacovigilance reporting databases. As both pharmacokinetic and pharmacodynamic interactions were reported on ICSRs the surveillance system should aim to detect both. The proposed triage algorithm had a high performance in comparison to the disproportionality measure alone.

Background: In paediatrics, the limited documentation to guide medication, the lack of suitable dosage forms, and the continuous development in childhood present a scenario where safety of medication is a particular challenge. Aim: To explore reported adverse drug reactions (ADRs) and the challenges in prescribing and administering medicines in paediatrics, in order to identify and suggest areas needing international surveillance within medication safety and improvement in the clinical setting. Methods: Four exploratory studies were conducted. Worldwide reporting of suspected ADRs (individual case safety reports, ICSR) with ages 0-17 years were examined overall. Twenty published case reports and ICSRs for adolescents, who developed a rare and incompletely documented ADR (rhabdomyolysis) during antipsychotic medicine use, were analysed in-depth. Prescribed doses of anti-inflammatory medicines were studied in a UK electronic health record database. Transcribed focus group interviews with 20 registered nurses from four paediatric wards in Sweden were analysed for factors that may promote or hinder safe medication practices. Descriptive statistics, multiple regression, and content analyses were used. Results: Although, skin reactions and anti-infective medicines were most frequently reported, and more reported in paediatric patients than in adults, medication errors and adverse reactions related to psychostimulant medicines were reported with increased frequency during 2005 to February 2010. The in-depth case analysis emphasised the need for increased vigilance following changes in patients’ medicine regimens, and indicated that ICSRs could contribute with clinically valuable information. Prescribed dose variations were associated with type of dosage form. Tablets and capsules were prescribed with a higher dose than liquid dosage forms. Six themes emerged from the interviews: preparation and administration was complex; medication errors caused considerable psychological burden; support from nurse colleagues was highly valued; unfamiliar medication was challenging; clear dose instructions were important; nurses handling medications needed to be accorded higher priority. Conclusions: Age-specific screening of ICSRs and the use of ICSRs to enhance knowledge of ADRs and medication errors need to be developed. Access to age-appropriate dosage forms is important when prescribing medicines to children. To improve medication safety practices in paediatric care, interdisciplinary collaborations across hospitals on national or even global levels are needed.

To increase medicine safety, researchers use adverse event reports to assess causal relationships between drugs and suspected adverse reactions. VigiBase, the world's largest database of such reports, collects data from numerous sources, introducing the risk of several records referring to the same case. These duplicates negatively affect the quality of data and its analysis. Thus, efforts should be made to detect and clean them automatically.  Today, VigiBase holds more than 3.8 million COVID-19 vaccine adverse event reports, making deduplication a challenging problem for existing solutions employed in VigiBase. This thesis project explores methods for this task, explicitly focusing on records with a COVID-19 vaccine. We implement Jaccard similarity, TF-IDF, and BERT to leverage the abundance of information contained in the free-text narratives of the reports. Mean-pooling is applied to create sentence embeddings from word embeddings produced by a pre-trained SapBERT model fine-tuned to maximise the cosine similarity between narratives of duplicate reports. Narrative similarity is quantified by the cosine similarity between sentence embeddings.  We apply a Gradient Boosted Decision Tree (GBDT) model for classifying report pairs as duplicates or non-duplicates. For a more calibrated model, logistic regression fine-tunes the leaf values of the GBDT. In addition, the model successfully implements a ruleset to find reports whose narratives mention a unique identifier of its duplicate. The best performing model achieves 73.3% recall and zero false positives on a controlled testing dataset for an F1-score of 84.6%, vastly outperforming VigiBase’s previously implemented model's F1-score of 60.1%. Further, when manually annotated by three reviewers, it reached an average 87% precision when fully deduplicating 11756 reports amongst records relating to hearing disorders.

This project aims to support pharmacovigilance, the science and activities relating to drug-safety and prevention of adverse drug reactions (ADRs). We focus on a specific ADR called QT prolongation, a serious reaction affecting the heartbeat. Our main goal is to group medicinal ingredients that might cause QT prolongation. This grouping can be used in safety analysis and for exclusion lists in clinical studies. It should preferably be ranked according to level of suspected correlation. We wished to create an automated and standardised process. Drug safety-related reports describing patients' experienced ADRs and what medicinal products they have taken are collected in a database called VigiBase, that we have used as source for ingredient extraction. The ADRs are described in free-texts and coded using an international standardised terminology. This helps us to process the data and filter ingredients included in a report that describes QT prolongation. To broaden our project scope to include uncoded data, we extended the process to use free-text verbatims describing the ADR as input. By processing and filtering the free-text data and training a classification model for natural language processing released by Google on VigiBase data, we were able to predict if a free-text verbatim is describing QT prolongation. The classification resulted in an F1-score of 98%. For the ingredients extracted from VigiBase, we wanted to validate if there is a known connection to QT prolongation. The VigiBase occurrences is a parameter to consider, but it might be misleading since a report can include several drugs, and a drug can include several ingredients, making it hard to validate the cause. For validation, we used product labels connected to each ingredient of interest. We used a tool to download, scan and code product labels in order to see which ones mention QT prolongation. To rank our final list of ingredients according to level of suspected QT prolongation correlation, we used a multinomial logistic regression model. As training data, we used a data subset manually labeled by pharmacists. Used on unlabeled validation data, the model accuracy was 68%. Analyzing the training data showed that it was not easily separated linearly explaining the limited classification performance. The final ranked list of ingredients suspected to cause QT prolongation consists of 1086 ingredients.