Eudravigilance database was established as a central repository for adverse event reports, which should be used for evaluating the benefits and risks of medicines during their development and monitoring their safety following their authorization in the European Economic Area (EEA). Although every Marketing Authorization Holder in the EEA is required to maintain an internal database of adverse events, data from Eudravigilance should help in the detection, and more importantly, validation of safety signals that arise during the postmarketing phase of a medicinal product.

To adequately evaluate such signals, it is of crucial importance that the data, extracted from Eudravigilance meets a certain degree of quality; if this is not the case, the results of analyses can be misleading, and can negatively affect the safety profile of the medicinal product in question. Additionally, the generation of irrelevant safety signals that are based on low-quality data, increases the workload in pharmacovigilance departments of Marketing Authorization Holders and regulators alike.

Our analysis randomly retrieved 500 reports of drug toxicity for five drugs that are most commonly implicated in accidental and intentional intoxications. Overall, the quality of the cases was relatively low, with mean quality category scores ranging from 10 to 13 out of 25. Especially two quality issues that were frequently encountered and have the potential to significantly impact the causality assessment: underlying conditions reported as adverse events, and concomitant drugs or drugs used to treat adverse events reported as suspect.

Additionally, since seriousness of a case is determined by assessing the seriousness of each reported adverse event, a wrong reporting can “upgrade” a case to a serious case, even if it is not. This in turn affects not only the quality of data, but also the timelines for reporting to other stakeholders, for example when such a case is obtained from Eudravigilance by means of ICSR download.

In the analysis of total quality scores, reports of paracetamol toxicity achieved higher mean scores compared to the other four evaluated substances, but statistical significance was reached only in comparison to reports of diazepam toxicity, which displayed the lowest total quality score. Distributional analysis showed that scores were not uniformly spread across quality categories, which likely reflects systematic reporting tendencies more than random variation. Category-level analyses revealed substantial variation in quality across reporting categories. Reports may excel in certain categories while underperforming in others, which indicates that reporting quality is multi-dimensional. Improving ICSR quality therefore requires both targeted (category-specific) improvements and a broader strategy to enhance overall completeness of reported information.

Many cases were identified where supporting/additional information clearly indicated which was the main drug that caused the event, but still irrelevant concomitant drugs are reported as suspect. If the reviewer does not perform a very thorough analysis (if only quantitative data are considered), the results of the analysis can be very wrong.

Nevertheless, there were several cases identified (over 20% of all evaluated cases) where the cited publication/background information reports much more information than what is actually included in the case. This could indicate that the person, who performed data entry into Eudravigilance, either did not have the required knowledge to adequately insert the case in the database or did not have the necessary time. In some cases, it is clearly evident that the reporter only inserted the basic necessary data for a case to be valid, not bothering to provide additional information that, when reviewed in further detail, painted a totally different picture of the event.

Another issue that arises from the evaluated reports is the applicability to registered medicinal products. For example, several cases were found to co-report fentanyl with heroin, cocaine and other illegal drugs of abuse, so it is plausible to assume that such cases are attributable to illegal fentanyl (produced by criminal organizations). It would therefore not be scientifically correct to “generalize” such findings to legal fentanyl-containing products (e.g. fentanyl patches) that are produced in a GMP-grade facilities.

Within Eudravigilance, there is the substance entry named FENTANYL (NON-PRESCRIPTION), which would, if reported correctly, enable a more adequate analysis of the safety of pharmaceutical grade, prescription fentanyl. Of the 100 evaluated cases, only one reported this substance as suspect, although most of them are clearly related to illicit fentanyl abuse, based on the context or based on additional information in the narrative or publication.

Overall, the quality of cases reported in Eudravigilance raises concerns for safety physicians that use these data to identify and assess safety signals. There are many pharmacovigilance processes in place that use this kind of data, including signal detection, benefit-risk analyses and periodic safety update reports. Despite being thoroughly described in the applicable guidelines, such as GVP, there is no comprehensive guidance provided that would warn the evaluators of this problem.

Our review clearly indicates that, in order to keep Eudravigilance as a valid data source, actions are needed to improve data quality, and these actions should be implemented by regulators and Marketing Authorization Holders alike.

In our opinion, the most important call to action is an overall mindset change of all stakeholders, involved in case reporting. Reporters should consider how their submissions will be interpreted by someone without prior knowledge of the case or access to supporting documentation.

This could be achieved by strengthening the cooperation between pharmacovigilance and medical departments, for example with regular trainings for data entry responsible persons, focusing on quality of reporting. Other possible approaches are to strengthen the quality control process, which should involve a medically trained person (this is especially important for cases, reported by consumers), promote the request of follow-up information whenever possible and to optimize the signal detection processes, which should consider data quality.

Our study has several limitations. The most important is the fact that the scoring is partially dependent on the subjective assessment by one physician only. The scoring system was developed to be as objective as possible, but since the assessed data are not fully standardizable (e.g., case narrative or literature publication from where the ICSR is reported), an element of subjectivity is always present. Nevertheless, this is the first application of the scoring system, and further work will be pursued to reduce as much as possible scoring based on subjective judgement and to validate the system on larger samples and multiple scorers.

In conclusion, a review of data quality of cases, reported in Eudravigilance, showed that there is plenty of room for improvement. The cases are very often reported with very basic errors (such as reporting underlying conditions or indications for use as adverse events), which creates confusion during medical assessment and can result in incorrect signal detection. In future, efforts to improve reporting of adverse events should be focused not only on quantitative compliance, but also on data quality and quality control.

It is the responsibility of us all that work in pharmacovigilance to share the highest possible level of data quality within public databases and thus enabling an adequate safety overview of all medicinal products that are available to our patients.