Using Big Data in Pharmacovigilance: Insights on Data Mining for Drug Safety

In the 1900s, pharmacovigilance looked very different. Healthcare professionals (HCPs) and clinical trial sponsors often looked at a patient’s symptoms, dosage, and prescription patterns to identify whether any adverse reactions were stemming from the medicine. However, today, pharmacovigilance relies on one major thing: data.

Today, adverse events and drug-drug interactions are studied from patient data. Data from sources such as clinical trial data, electronic health records, adverse event reporting systems, and social media forums is collected. Then, it is analyzed for the early detection of adverse drug reactions or for predicting drug-drug interactions. These predictions can help HCPs and clinical trial sponsors make better decisions for patient safety.

To obtain this data, often a simple process is employed: data mining. Data mining involves the analysis of large datasets to identify patterns, correlations, and insights. It uses various analytical techniques, like association rule mining, and technologies, like artificial intelligence (AI) and machine learning (ML), to identify and predict potential risks. Today, various regulatory-approved tools are available for data mining.

Data Mining Databases & Tools You Need for Pharmacovigilance

FAERS

The United States Food and Drug Administration’s adverse event reporting system (FAERS) is a public dashboard that supports the FDA’s post-market safety survey of all approved drugs and biologics. Data mining of the FAERS database can provide previously unknown clinically important associations, which can help in predicting drug-drug interactions.

Vigibase

Managed by the World Health Organization, Vigibase is a global database of adverse event reports related to medicines and vaccines. The general public can access some part of this data via the VigiAccess tool. The database is regularly updated and is linked to data management and quality assurance tools, allowing structured data entry, retrieval, and analysis.

EudraVigilance

EudraVigilance is a publicly accessible database of adverse reaction reports that is developed and managed by the European Medicines Agency (EMA). Its AI integration allows enhanced signal detection. The database also facilitates the electronic exchange of individual case safety reports between EMA, NCAs, MAHs, and clinical trial sponsors. The data in the EudraVigilance database can be mined for early detection and evaluation of possible safety signals.

BioBERT

Bidirectional Encoder Representations from Transformers for Biomedical Text Mining (BioBERT) is a biomedical text mining tool pre-trained on biomedical corpora. It outperforms various available text-mining tools in biomedical relation extraction, biomedical named entity identification, and biomedical question answering. This tool can be used for mining of pharma literature to study reported adverse events. It can also be used to find incident case study reports with specific symptoms or terminologies.

Oracle Argus Safety

Oracle’s Argus can be used for processing, analyzing, and reporting adverse events related to drugs, biologics, vaccines, devices, and combination products. It can be used to collect adverse event data from multiple sources, including electronic health records, clinical trials, and literature, which can be stored in a centralized data based and mined using machine learning and rule-based algorithms. Argus Safety is compliant with drug, vaccine, and medical device standards and regulations in all regions of the world.

MedDRA

While the MedDRA cannot be directly used for data mining, it certainly has some excellent applications. MedDRA stands for Medical Dictionary for Regulatory Activity, which was developed in the 1990s by the International Council for Harmonization of Technical Requirements for Pharmaceuticals for Human Use. It has a standardized adverse event classification and coding structure that is used for report generation across clinical trials, regulatory submissions, and post-market surveillance reports. It can be integrated with Argus Safety, Empirica Signal, and FAERS for real-time monitoring.

Empirica Signal

Oracle Empirica Signal is an advanced data mining and signal detection platform. It utilizes algorithms such as the proportional reporting ratio, reporting odds ratio, and multi-item gamma Poisson shrinker to detect safety signals. It also uses unsupervised learning to identify unknown relationships. It can detect latent patterns in drug safety data. It also includes AI models that can predict which drug combination poses high risks of adverse events.

These data mining tools are used by various pharmaceutical companies and researchers for various pharmaceutical goals.

Who’s Using Data Mining Tools for Pharmacovigilance?

Here are some examples of how these tools are currently used in the real world.

• FAERS: Recently, researchers mined the FAERS database to identify and analyze adverse drug events related to Vedolizumab. FAERS has also been used to mine data related to other drugs such as Nirmatrelvir, Ritonavir, axitinib, and dual orexin receptor antagonists.
• Bio-BERT: Researchers have used BERT-based language models for accurate adverse drug event extraction from social media data.
• Oracle Argus Safety: Yuhan Corporation, South Korea’s largest pharmaceutical company, uses Argus Safety for its pharmacovigilance operations.

These are just some examples of the companies and researchers who have used data mining tools for their pharmacovigilance operations; many more exist in the real world. However, the integration of such data mining tools has not been swift, primarily because there are many risks associated with these tools.



Risks Associated with the Use of Data Mining Tools

Data quality and accuracy issues

Many databases and data mining tools contain inconsistent or missing information, which can lead to incorrect conclusions. They may also contain duplicate entries, which leads to data skewing. Finally, many adverse drug events go unreported due to a lack of awareness, reporting complexity, and patients’ fear of legal consequences, which means most databases and mining tools do not have comprehensive information.

False positives and false negatives

Data mining tools may flag harmless drug associations as safety signals, resulting in false positives. By contrast, algorithms may also fail to detect rare or long-term adverse events due to weak statistical associations. These false positives and negatives can skew the data and provide an incomplete picture.

Data privacy concerns

Data privacy concerns persist in data mining as it can expose sensitive patient information. Pharmaceutical companies must adhere to strict data protection and privacy laws such as the GDPR and HIPAA. Unfortunately, this adherence limits access to detailed patient information, making it more difficult to find correlations.

The Future of Data Mining Tools

In the context of pharmacovigilance, data mining has excellent benefits, which have been realized by many pharma companies and governance agencies worldwide. Consequently, various tools are available for data mining, and more tools are still being researched. However, the implementation of such tools poses risks, primarily data quality and privacy concerns that need to be urgently addressed. These risks are the primary barriers to the widespread use of data mining tools in pharmacovigilance. They can be addressed through advanced analytics, expert validation, and ethical practices.

FAQs

1. Why is data mining used in pharmacovigilance?

Data mining can identify hidden patterns and previously unknown safety concerns regarding drug usage, which helps regulatory agencies and companies take proactive safety measures and manage risk.

2. How does data mining help detect adverse drug reactions?

Data mining techniques help identify unexpected or serious reactions by analyzing patterns in adverse event databases. These patterns help confirm whether the reaction is triggered by a specific stimulus.

3. Do regulatory agencies use data mining in their pharmacovigilance activities?

Yes, regulatory agencies use data mining to monitor post-market drug safety and issue safety alerts based on findings.