Innovations in Pharmacovigilance: Leveraging Artificial Intelligence for Enhanced Drug Safety Monitoring
 
Abdulaziz Ali Al Amri1*, Ahmed Ebrahem Al Thobaiti2, Fahad Sultan Al Zahrani3, Fahad Abdulali Alkharmani4, Abdulmuhsen Ghazy Alqethami5
1,2,3,4,5 Pharmacy Technician, Prince Mansour military Hospital, Taif, KSA
Email: Undertaker.810@hotmail.com
Abstract - The technique of pharmacovigilance, which involves keeping an eye on how medical pharmaceuticals are working after they've been approved for use, is vital in making sure that drugs are safe to use. Novel approaches brought about by the advent of Artificial Intelligence (AI) have the potential to greatly improve pharmacovigilance by making ADR identification, evaluation, and prevention much more effective. This study delves into the most recent advancements in AI-driven pharmacovigilance, shedding light on how drug safety monitoring is being revolutionised by machine learning algorithms, natural language processing, and big data analytics. In this article, we will go over the pros, cons, and possible next steps for pharmacovigilance frameworks that use AI.
Keyword: Pharmacovigilance, Innovations, Leveraging Artificial Intelligence, Enhanced Drug Safety Monitoring
INTRODUCTION
In order to protect the public's health, pharmacovigilance (PV) is crucial for detecting, assessing, and avoiding side effects linked to pharmaceuticals. Time is of the essence when it comes to traditional pharmacovigilance procedures, which are constrained by data silos and underreporting, and depend mostly on spontaneous reporting systems and manual review processes. More efficient and precise medication safety monitoring has been made possible by new options given by AI, which have emerged in recent years to overcome these restrictions (Kepuska, & Bohouta, 2018).
Artificial Intelligence Role in Pharmacovigilance
Machine learning (ML), big data analytics, and natural language processing (NLP) are all parts of artificial intelligence (AI) that have potential uses in pharmacovigilance. These technologies are very helpful for identifying ADRs and other safety signals because of their capacity to analyse massive datasets, discover trends, and make predictions (Salas, et al., 2022).
1. Machine Learning for Signal Detection
To improve ADR detection, machine learning algorithms have been used to sift through mountains of data from places like social media, spontaneous reporting systems, and electronic health records (EHRs). Both supervised and unsupervised ML models have the ability to detect ADRs that were not previously known, rank signals according to their importance for future research, and identify patient groups who are more likely to have adverse events (Liang, et al., 2022).
Example: One prominent use of ML in pharmacovigilance is the deployment of deep learning models to examine electronic health records (EHRs) for possible signals related to medication safety. To identify adverse drug reactions (ADRs) that could otherwise go unnoticed by conventional reporting methods, these algorithms can sift through unstructured data like patient medical records.
2. Natural Language Processing for Data Extraction
A branch of artificial intelligence, natural language processing (NLP) studies how machines and humans communicate via language. Medical literature, social media postings, and patient records are examples of unstructured text that pharmacovigilance uses natural language processing (NLP) methods to extract pertinent information from. Natural language processing (NLP) helps in ADR identification and automated signal detection by transforming text data into structured representations (Ball, & Dal Pan, 2022).
Example: Compared to more conventional approaches, natural language processing algorithms can quickly scour social media for postings on medication side effects, illuminating patients' experiences in real time and revealing any safety issues.
3. Big Data Analytics for Comprehensive Monitoring
Combining AI with big data analytics opens the door to analysing genetic data, information gathered from post-marketing monitoring, and results from clinical trials, among other various data sources. By looking at all of the data at once, we may see patterns and connections that might otherwise go unnoticed, providing a more complete picture of medication safety (Aronson, 2022).
Example: More targeted and less risky medication treatments are possible thanks to AI-driven big data analytics that combine genomic data with pharmacovigilance databases to find genetic markers linked to greater vulnerability to ADRs.
Benefits of AI in Pharmacovigilance
Several important advantages may be gained by integrating AI into pharmacovigilance (Bate, & Stegmann, 2023):
Challenges and Limitations
The use of artificial intelligence in pharmacovigilance has great promise, however there are many obstacles to overcome (Murali, et al., 2019):
Obstacles Facing AI-Reliant Pharmacovigilance in Constrained Environments
Pharmacovigilance (PV) Database Creation with the Help of AI
Artificial intelligence relies on data. Building a thorough PV database is, hence, crucial. The process by which each nation builds its PV database is distinct. Healthcare providers (HCPs) are often the first to report ICSRs on their own, marking the beginning of PV. The FDA Adverse Event Reporting System (FAERS) and the Vaccine Adverse Event Reporting System (VAERS) in the US, the pharmacovigilance database in France, China's pharmacovigilance system, VigiBase, and others in both developed and developing nations have been accumulating large-scale PV or PV system databases. The Uppsala Monitoring Centre is responsible for maintaining VigiBase, which includes information from over 150 different nations. The biggest problem with building a PV database for LMICs, however, is underreporting. Many things have contributed to this, such as inadequate funding, a lack of personnel and relevant rules, and a faulty reporting system architecture. This is shown by a set of numbers: In 2019, Africa produced only 0.9% of the ICSRs included in VigiBase, in contrast to the over 70% of data given by Western nations (e.g., the US and EU). In Kenya, there were 35.0 ICSRs per 100,000 people, in Ethiopia, there were 6.7 ICSRs per 100,000 people, and in Tanzania, there were 4.1 ICSRs per 100,000 people (Barry, et al., 2020). A significant amount of pharmaceutical usage in low- and middle-income countries (LMICs), especially in rural and isolated regions, goes unrecorded in medical records due to self-medication and underreporting by healthcare providers. Therefore, in settings with limited resources, the primary goal should be to find ways to apply AI to help increase the reporting rate of adverse drug events (ADEs). One such approach may be to extract unreported ADEs from the electronic health record (EHR) (Boland, & Tatonetti, 2015).
Problems in Human Resources: Education and Training
There is a dearth of PV HCPs in LMICs due to a lack of educational options and training. Experts in operations, surveillance, systems, and pharmacovigilance make up a conventional PV system's core team (QPPV). To complement the conventional PV workforce of drug safety supervisors, drug safety doctors, data/system security administrators, and quality assurance project managers, AI-based PV also necessitates engineers with expertise in natural language processing and machine learning techniques, among other areas of artificial intelligence. For LMICs, this presents yet another formidable obstacle. Training for AI-based PV is available in a limited number of countries, with few LMICs on the list. It is cross-specialty and even cross-lingual. Despite the usage of AI in PV, there aren't many courses focused on AI-based PV, even internationally. Inadequate funding for healthcare professionals in low- and middle-income countries may make it even more difficult for them to attend such courses. Regardless, healthcare providers in industrialised nations might get training in artificial intelligence technologies like data mining, deep learning, and natural language processing. On the other hand, there is a dearth of chances for AI training in LMICs. Artificial intelligence (AI)-based PV benefits from this training since it makes use of modern methods that aid in data processing and analysis.
It is typical in LMICs to have a lack of skilled workers when it comes to AI-powered photovoltaic systems. Raising the number of people versed in PV and AI might be achieved via educational initiatives. Undergraduates in relevant majors at educational institutions would benefit most from AI-based PV courses. More transdisciplinary experts might be produced and PV data could be better understood via education on AI-based PV. Although China has set up an online spontaneous self-reporting ADR monitoring system in recent years (Zhao, et al., 2018), a study of PV in Western China revealed that the ADR reporting rate was incredibly low due to unclear feedback pathways and a lack of understanding about the seriousness of ADRs (Chen, et al., 2021). This highlights the problem of inadequate education in the country. Rising college enrolment rates pose the greatest threat to artificial intelligence (AI) programs in low- and middle-income countries. Furthermore, AI-based PV's multidisciplinary character poses a challenge to AI education as it necessitates AI engineers to possess understanding of both AI and medicine.

Technological Challenges

When working with limited resources, artificial intelligence PV faces two technical hurdles: data integration and data annotation. There is no universally accepted model for medical data; instead, different medical institutions use different data models, and different types of data models use different terminology and value representations (Chen, et al., 2021). It is necessary for many institutions to combine and transform data from various sources into a common data model (CDM) compliant format. The data transformation process is labour-intensive and requires the involvement of both engineers and HCPs. The necessity to handle data in many local languages or expression habits makes language a potential obstacle to data transformation in LMICs.
Annotated data is crucial for AI-based PV because training models and algorithms employ annotated data to anticipate unannotated data. Research on AI-based PV is limited because to the expensive cost of annotation and the paucity of high-quality data for model training. Although structured databases include a huge number of ICSRs, the possible PV signals are not identified in these databases. Furthermore, PV information is enhanced more by unstructured data than structured data. Unstructured data includes things like medical records, social media postings, biological literature, and clinical notes. Research indicated that out of all the adverse responses to statins, only 28.6% were documented in a structured fashion in the HIS. The other adverse reactions were documented in unstructured clinical narratives (Skentzos, et al., 2011).
However, term variation remains the most significant technical hurdle for AI-based PV. Less formal and very varied terminology is utilised (e.g., medicine and illness names), side effect descriptions are not always clear, and spelling and grammatical errors are prevalent (Sloane, et al., 2015). Processing, integrating, and standardising data is complicated due to the variety of local languages. In addition to integrating different data sources, improving data quality, decreasing dimensionality, and increasing retrieval recall, concept normalisation helps with this problem. Having said that, standardising concepts is a laborious and time-consuming process. The use of concept normalisation is necessary for entity alignment in sources such as social media, medical records, and biological literature (Doğan, et al., 2014), due to the abundance of aliases, acronyms, and informal terminology present in these sources. However, there are significant gaps due to the linguistic variety in LMICs and the absence of a mapping between local languages and standard vocabulary. This hinders the transfer of advanced technological platforms in industrialised nations.

Regulations and Funding for Pharmacovigilance

Decisions in AI-based PV are heavily influenced by the government. Stronger rules are needed to involve local industry and HCPs in most LMICs' PV systems since these systems are immature and not supported by strong legal or regulatory measures (Olsson, et al., 2015). The regulatory foundation for the PV system isn't always strong enough for the power to execute regulatory measures, even in resource-limited nations with rather well-functioning PV systems, despite the fact that several of these nations have established PV standards (Olsson, et al., 2015). Data privacy and security rules are additional ones that pertain to AI-based PV. This is critical because artificial intelligence systems need ccess to vast amounts of personal data, the disclosure of which might have disastrous effects.
Another obstacle is the absence of funding. Existing system support, information infrastructure upgrade, education, compensation for HCPs and AI engineers, and so on all fall under this category. Even if a lot of nations have installed PV systems, they won't be able to keep them running smoothly over the long haul unless their budgets are stable. Here, we may look at how China and India have benefited from financial assistance to set up sustainable PV systems (Olsson, et al., 2015). Furthermore, AI research needs funding for software and hardware. The use of artificial intelligence software brings the system to life, while hardware serves as the substrate upon which software runs. Given the often-inadequate equipment un LMICs, this may end up costing a pretty penny. Therefore, governments should establish appropriate laws to provide funding and support the training of healthcare providers and artificial intelligence developers.

Emerging Technology's Potential in PV and Safety Surveillance

Ensuring the safety of medications sold in the United States is a long-time priority of the FDA. This commitment is maintained by continuing monitoring and research after approval. Submit reports to the FDA after reviewing any adverse drug experience information received or otherwise collected in the postmarket context. This is required of regulated industries under 21 CFR 314.80, 314.98, and 600.80. The ever-increasing case numbers, together with the difficulties of efficiently collecting, processing, and evaluating both individual and aggregate patient safety data, pose a threat to both business and regulatory bodies. Some of these problems may be amenable to future technological advancements that improve PV system efficiency. For instance, in an effort to reduce administrative load and expenses, early adopters of AI are using these new technologies to automate basic operations like adverse event intake, data input, and processing. By collecting, combining, and analysing more extensive and varied datasets, these technologies may also improve the efficacy and efficiency of safety monitoring. 
AI and IA Benefits in PV
With the help of AI and ML, the IA revolution is about to revolutionise the PV game. The following are just a few of the ways in which these technologies have the potential to revolutionise PV (Olsson, et al., 2015):
1. Improving Data Collection and Analysis: The sheer amount and complexity of data is one of PV's greatest problems. It might be difficult to manage information overload due to the variety of sources used to gather safety data. Faster and more accurate assessments of adverse events are possible with the use of AI and IA systems that can discover and extract relevant data from multiple sources, including social media, electronic health records (EHRs), and medical literature. In addition, these instruments may examine the data with the use of machine learning algorithms to spot trends and patterns that might point to medication interactions or side effects.
2. Prediction of Adverse Events: Artificial intelligence methods may be used to create models that can foretell which patients are most likely to have adverse medication responses. It may also aid in the identification of patient populations that may need diligent post-treatment monitoring. As a result, patients using drugs would be far less likely to have adverse events, leading to better safety results overall.
3. Oversee Case Triage and Prioritisation: Artificial intelligence systems may help sort adverse event instances according to their severity, possible cause, and effect. This allows us to prioritise instances that need quick attention and allocate resources more effectively.
4. Improving Drug Adverse Effects Reporting and Processing: Drug adverse effect reports are collected and evaluated individually as part of case processing. It may take a lot of time and effort to do this manually. Data input, case triage, and causality assessment are just a few examples of the kinds of jobs that may be automated with the use of AI and IA technologies. Enabling PV teams to simplify their reporting procedures and meet with regulatory standards, these technologies may also provide standardised reports. By using natural language processing methods to automatically extract essential information from free-text submissions, IA may simplify the case processing process. This has the potential to enhance medication safety profiles generally while cutting costs, saving time, and increasing the precision and completeness of safety data.
5. Enhancing Signal Detection and Management: The discovery of novel or potentially harmful pharmacological effects is known as signal detection. AI and IA technologies can analyse massive volumes of data from social media, electronic health records, and other pertinent sources, allowing for accurate and rapid signal detection.
These technologies may assist in sorting signals according to their probability and severity, allowing PV teams to concentrate on the most critical ones. By sifting through mountains of data in search of promising signals using statistical algorithms and ML models, it may automate this process.
6. Detect Possible Drug Interactions: AI may aid in avoiding patients taking drugs that potentially have adverse interactions. To further enhance patient safety and facilitate individualised therapy, it may also estimate the likelihood of adverse events (AEs) for specific patients.
Automatic screening and evaluation of scientific literature for pertinent safety information is a capability of AI. This brings us to our seventh point, literature screening and review. Data from clinical trials, case reports, scholarly papers, and other sources may be extracted using natural language processing algorithms, which can help identify any safety risks.
7. Assurance of PV Data Quality: Data cleansing, deduplication, and standardisation are just a few examples of how AI and IA approaches may improve and verify PV data. This guarantees that the data used for analysis and decision-making is correct and dependable.
7. Data Mining and Pattern Recognition: Artificial intelligence and image recognition may be used to sift through PV datasets, revealing connections and patterns that could otherwise go unnoticed using more conventional approaches. It may aid in the detection of safety signals, adverse event connections, and medication interactions that were previously unknown.
Managing risks and ensuring regulatory compliance are of the utmost importance in the highly regulated profession of pharmacovigilance. By automating processes like adverse event reporting, regular safety updates, and risk management strategies, AI and IA solutions may aid in assuring compliance. Additionally, pharmacovigilance teams may take prompt action and reduce risks with the help of these systems' real-time monitoring and alarms.
The Use of AI and IA in Pharmacovigilance: A Comprehensive Review
The healthcare and life science sectors stand to gain a great deal from PV that incorporates AI and IA. Optimisation of labour-intensive processes may be achieved via the use of AI, resulting in improved consistency and less bias. Improved data collecting, analysis, and adverse event prediction made possible by AI and IA techniques allows for more rapid and precise assessments of medication safety. Improved patient care and overall safety are results of these technologies' simplification of case processing, reporting, signal detection, and management. Beyond its obvious use in PV, AI and IA also aid in managing risks, ensuring data quality, and complying with regulations.
Once your plan and goals for a sustainable future are determined, the Apexon sustainability framework may help you turn your net-zero path faster. Internal IT, corporate IT, and customer operations are all part of our holistic strategy for transforming the IT value chain. By using trustworthy data, Apexon gives you complete visibility into sustainability objectives. Sustainable acts are powered by AI. Incorporating sustainability data and performance into every aspect of a company's operations is our 5-step process that helps organisations quantify their value to stakeholders and their impact on the environment (Salvo, et al., 2023).

Regulations and Funding for Pharmacovigilance

The government has a significant impact on decisions in PV that are based on AI. Because most PV systems in LMICs are still in their early stages and lack robust legal and regulatory backing, more regulations are required to engage local businesses and HCPs in these systems (Olsson, 2015). Although some resource-limited nations have set PV requirements, the regulatory framework for the PV system isn't always robust enough to implement regulatory measures, even in countries with very well-functioning PV systems (Olsson, 2015). Additional regulations pertaining to AI-based PV include those pertaining to data security and privacy. This is of the utmost importance since AI systems need access to large volumes of sensitive personal information, the exposure of which might lead to catastrophic consequences.
The lack of financial resources is another problem. Included in this area are current system support, education, upgrades to the information infrastructure, remuneration for healthcare providers and artificial intelligence developers, and so on. If governments do not have steady funding, it will be impossible for them to maintain PV systems even if many have installed them. As an example, we may examine the positive effects of financial aid on the installation of sustainable PV systems in India and China (Olsson, 2015). Additionally, software and hardware funding is necessary for AI development. While hardware provides the necessary structure for software to operate, artificial intelligence software is what really makes the system tick. This may wind up costing a considerable sum, considering the often poor equipment in LMICs. As a result, governments should pass suitable legislation to finance and facilitate the education of healthcare practitioners and developers of artificial intelligence.

A Safer Future with AI-Powered Pharmacovigilance

A new age of proactive, data-driven drug safety monitoring is being ushered in by the integration of AI and technical advances, which is reshaping pharmacovigilance. Advancements in natural language processing, big data analytics, RWD, and blockchain are opening up new possibilities for thorough safety monitoring, while AI-powered solutions are improving signal detection, risk assessment, and predictive modelling. We should anticipate a sea change in medication safety as AI becomes more integrated into the pharmacovigilance ecosystem, guaranteeing a safer future for patients all around the globe. Pharma organisations may benefit from working with Freyr, an experienced partner, since they can help them use AI and technology to improve pharmacovigilance processes. This will boost product safety, safeguard patient health, and build confidence in the pharmaceutical business.
Future Directions
Several new developments in artificial intelligence (AI) for pharmacovigilance hold great promise for improving drug safety monitoring in the future:
CONCLUSION
By improving ADR identification, evaluation, and prevention, artificial intelligence (AI) may revolutionise pharmacovigilance. The advantages of AI-driven pharmacovigilance, including real-time monitoring, improved efficiency, and accuracy, make it a valuable tool for enhancing medication safety, despite the hurdles it faces. The full promise of artificial intelligence in pharmacovigilance can only be realised with ongoing innovation, strong regulatory frameworks, and careful consideration of ethical issues.
REFERENCES
  1. Olsson S, Pal SN, Dodoo A. Pharmacovigilance in resource-limited countries. Expert Rev Clin Pharmacol. 2015;8(4):449–460.
  2. Salvo, F., Micallef, J., Lahouegue, A., Chouchana, L., Létinier, L., Faillie, J. L., & Pariente, A. (2023). Will the future of pharmacovigilance be more automated?. Expert Opinion on Drug Safety, 22(7), 541-548.
  3. Olsson, S., Pal, S. N., & Dodoo, A. (2015). Pharmacovigilance in resource-limited countries. Expert review of clinical pharmacology, 8(4), 449-460.
  4. Doğan, R. I., Leaman, R., & Lu, Z. (2014). NCBI disease corpus: a resource for disease name recognition and concept normalization. Journal of biomedical informatics, 47, 1-10.
  5. Sloane, R., Osanlou, O., Lewis, D., Bollegala, D., Maskell, S., & Pirmohamed, M. (2015). Social media and pharmacovigilance: a review of the opportunities and challenges. British journal of clinical pharmacology, 80(4), 910-920.
  6. Skentzos, S., Shubina, M., Plutzky, J., & Turchin, A. (2011). Structured vs. unstructured: factors affecting adverse drug reaction documentation in an EMR repository. In AMIA annual symposium proceedings (Vol. 2011, p. 1270). American Medical Informatics Association.
  7. Chen, R., Zhang, Y., Dou, Z., Chen, F., Xie, K., & Wang, S. (2021). Data sharing and privacy in pharmaceutical studies. Current Pharmaceutical Design, 27(7), 911-918.
  8. Chen, Y., Wang, Y., Wang, N., Xiang, Y., Zhang, R., Xiao, J., ... & Feng, B. (2021). Knowledge, attitude, and practice regarding pharmacovigilance among the general public in Western China: a cross-sectional study. Current Medical Research and Opinion, 37(1), 101-108.
  9. Zhao, Y., Wang, T., Li, G., & Sun, S. (2018). Pharmacovigilance in China: development and challenges. International journal of clinical pharmacy, 40, 823-831.
  10. Boland, M. R., & Tatonetti, N. P. (2015). Are all vaccines created equal? using electronic health records to discover vaccines associated with clinician-coded adverse events. AMIA Summits on Translational Science Proceedings, 2015, 196.
  11. Barry, A., Olsson, S., Minzi, O., Bienvenu, E., Makonnen, E., Kamuhabwa, A., ... & Aklillu, E. (2020). Comparative assessment of the national pharmacovigilance systems in East Africa: Ethiopia, Kenya, Rwanda and Tanzania. Drug safety, 43(4), 339-350.
  12. Murali, K., Kaur, S., Prakash, A., & Medhi, B. (2019). Artificial intelligence in pharmacovigilance: Practical utility. Indian Journal of Pharmacology, 51(6), 373-376.
  13. Bate, A., & Stegmann, J. U. (2023). Artificial intelligence and pharmacovigilance: What is happening, what could happen and what should happen?. Health Policy and Technology, 12(2), 100743.
  14. Liang, L., Hu, J., Sun, G., Hong, N., Wu, G., He, Y., ... & Gong, M. (2022). Artificial intelligence-based pharmacovigilance in the setting of limited resources. Drug Safety, 45(5), 511-519.
  15. Salas, M., Petracek, J., Yalamanchili, P., Aimer, O., Kasthuril, D., Dhingra, S., ... & Bostic, T. (2022). The use of artificial intelligence in pharmacovigilance: a systematic review of the literature. Pharmaceutical medicine, 36(5), 295-306.
  16. Kepuska, V., & Bohouta, G. (2018, January). Next-generation of virtual personal assistants (microsoft cortana, apple siri, amazon alexa and google home). In 2018 IEEE 8th annual computing and communication workshop and conference (CCWC) (pp. 99-103). IEEE.
  17. Ball, R., & Dal Pan, G. (2022). “Artificial intelligence” for pharmacovigilance: ready for prime time?. Drug safety, 45(5), 429-438.
  18. Aronson, J. K. (2022). Artificial intelligence in pharmacovigilance: an introduction to terms, concepts, applications, and limitations. Drug Safety, 45(5), 407-418.