Impact of AI on Pharmacovigilance: A Systematic Review

Harshal G. Patil*, Vinit S. Khairnar

Ahinsa Institute of Pharmacy, Dhule road, Dondaicha, Tal. Shindkheda, Dist. Dhule, Maharashtra

 *Correspondence: harshpatil6006@gmail.com; Tel.: 9561957379

DOI: https://doi.org/10.71431/IJRPAS.2025.4503  

INTRODUCTION

Regulatory agencies around the world, including those of the FDA, EMA, and WHO, are proactively exploring the use of AI technology to enhance work efficiency, accelerate drug review and approval processes, and expedite the market launch of drugs. For instance, FDAs AI/ML Action Plan for 2024: The focus of the plan is to monitor and manage the lifecycle of AI/ML-based medical software (SaMD). Tasks include guidance for adaptive algorithms, providing transparency to users, and piloting using real-world performance monitoring. Various regulatory agencies have realized the great potential of AI in the field of drug regulation; however, its application still lags behind its established use in drug research and development and remains in the early stages of exploration.

In the history of (AI), ophthalmology emerged as an early adopter of AI technology. A notable example is the development of the CASNET/glaucoma model in 1976, which utilized stored patient data to make informed decisions based on physiological parameter changes, clinical manifestations, and treatment outcomes. This model showed the potential of AI in ophthalmology and paved the way for further advancements in the field. A similar system known as the INTERNIST-1 was developed to assist physicians in accurately diagnosing complex internal medicine diseases. [2]

Understanding Pharmacovigilance

The WHO defines Pharmacovigilance as a science which entails monitoring for evaluating, understanding, and preventing adverse drug effects or other drug-related issues. It aids in ensuring the safety in public health and safety medication utilization. Hand reporting along manual organism and regulation databases such as WHO VigiBase and FDA’s Adverse Event Reporting System (FAERS) form the foundation of traditional pharmacovigilance techniques. Nonetheless, these methodologies incur reporting lags as well as delayed signal detection. The developments in pharmacovigilance AI technology has improved and accelerated processes such as signal detection, ADR identification, and case processing beyond any level of precision and speed previously attained. [3]           

Through the use of machine learning, natural language processing, and predictive analytics, AI is transforming pharmacovigilance by enhancing the speed and precision with which adverse events are detected and signals are identified. Nonetheless, such implementation comes with significant risks including algorithmic discrimination, siloes in regulatory oversight, and inequitable data distribution—these challenges may diminish the prospect of enhancing factors of safety across medications. This systematic review aims to evaluate the multifaceted impacts of AI on pharmacovigilance while also analysing the systemic barriers to its adoption and, propose initiatives that encourage collaboration and focus on equity to ensure trustful and responsible AI use in pharmacovigilance. [4,5]

Application of AI in Pharmacovigilance

 Automated Adverse Event Detection

Automated adverse drug reaction (ADR) detection is an important step in pharmacovigilance. It may solve a series of critical shortcomings present in traditional spontaneous reporting systems (SRS). SRS usually involves voluntary information provided by healthcare professionals or patients. However, many SRS currently suffer from chronic underreporting (with an estimated 90–95% underreporting rate in many drugs) and slow detection of safety signals which can take months or years to be detected at the regulatory level.[8]

AI-powered natural language processing (NLP) overcomes those limitations by systematically extracting from unstructured datasets from an array of sources, including EHRs, clinical narratives, and social media.

• Improved reporting performance: Liang et al. reported 47% better adverse event (AE) detection rate when using NLP as compared with manual SRS, which could be explained by automated extraction from clinical notes.

• Quality of EHR Analysis: Liu et al. (2021) study used a domain-specific language model, BioBERT, to identify ADRs such as drug-induced renal toxicity in EHRs with 30–50% more sensitivity than rule-based methods.

• Identification of Rare AEs in Unstructured Data Guellil et al.  study developed NLP pipelines to isolate rare cases of vaccine-associated myocarditis in unstructured social media posts with respect to slang and typos using advanced pre-processing and contextual analysis. [6,7,21]

Figure no. 1: AI-driven adverse event detention model.

SIGNAL DETECTION and CASE PROCESSING

Signal detection refers to the identification of potential causal relationships between a drug and an adverse event that requires further investigation. Traditionally, regulatory authorities manually analysed adverse event reports submitted to pharmacovigilance databases. This process was slow, and people often made mistakes during it because it relied on humans to do the work.

AI has revolutionized signal detection. Artificial intelligence has greatly improved the way we detect signals. It uses machine learning algorithms to discover hidden patterns in vast amounts of data. This means that tasks involving signal detection are now more efficient and accurate because AI can analyse data much faster and find patterns that humans might miss. For example, the Bayesian Confidence Propagation Neural Network (BCPNN) used by the WHO VigiBase database has demonstrated a 60% reduction in case processing time by automatically detecting safety signals from adverse event reports. Similarly, deep learning models can analyse millions of EHR records to detect unknown ADRs with high accuracy. [9,10,11]

Machine learning (ML) and natural language processing (NLP) are used to classify, validate, and submit for review adverse event reports sent to regulatory authorities. With platforms like IQVIA's Vigilance Platform reducing case processing times from 5 days to under 1 day (80% cut) by automating processes including adverse event detection and data validation, studies show this automation lowers human workload by 75% in narrative generation and speeds regulatory response times (IQVIA, 2023). Supported by AI's capacity to extract safety signals from unstructured data with F1 scores of 0.72–0.74 (a measure of accuracy), industry reports also draw attention to 50% cost savings in manual reviews and 35–65% efficiency improvements in data entry (IQVIA, 2023; Indegene, 2022). [12,13,14]

Figure no. 2: Comparative analysis of manual vs AI-driven case processin.

Predictive Pharmacovigilance and Risk Management

By modelling past data, predictive modelling allows artificial intelligence systems to predict the effect of a given drug on a human's body. Predictive modelling then allows medical practitioners to intervene in advance to prevent harm from occurring, thereby attenuating the risks associated with prescription drug use. Advanced medicate security methodologies progressively utilize AI and machine learning to foresee hurtful side impacts, in spite of the fact that precision changes. For case, instruments like XGBoost identify extreme responses in cancer drugs 55.6% of the time but affirm them accurately as it were 38.5% of the time, whereas broader thinks about appear moved forward exactness (78–81.5% affectability, 70–79.5% specificity) with thorough testing. Real-world applications, like AbbVie's Medicate X trial, hailed dangers six months speedier than manual audits, and instruments such as VigiRank prioritize high-risk pediatric cases utilizing measurable investigation. AI empowers personalized dosing (e.g., Tacrolimus alterations) and mechanizes post-market checking by means of databases like FAERS, whereas cutting human blunders by 30% with language-processing apparatuses. In spite of challenges like divided information and constrained outside approval (as it were 23% of considers), combining hereditary experiences with crossover AI-human surveys and versatile administrative plans for biosimilar offers a way toward more secure, proactive medicate administration. [15,16]

Multi-Source Data Integration

Another significant application of AI in pharmacovigilance is multi-source data integration. Traditionally, pharmacovigilance databases relied heavily on structured data collected from healthcare institutions and regulatory bodies. However, a large portion of adverse event data exists in unstructured formats such as social media, patient forums, biomedical literature, and public health records. By analysing these assorted inputs, their models diminished false-positive liver damage signals by 40%, outflanking conventional animal-based strategies that frequently come up short to anticipate human-specific harmfulness. For case, DILIPredictor distinguished hepatotoxicity dangers in a novel antiviral sedate that were undetected in preclinical trials, empowering early relief. This approach not as it were upgrades preclinical security appraisals but moreover diminishes expensive late-stage clinical trial disappointments, clearing the way for worldwide administrative selection of AI-driven systems to streamline medicate improvement and make strides quiet security. Their discoveries emphasize the basic part of multi-source information integration in progressing proactive pharmacovigilance and setting up vigorous, prescient security analytics. [17,18]

Figure no. 3: AI-enabled multi-source data integration for pharmacovigilance

Challenges in Implementing AI in pharmacovigilance:

Data Quality and Availability:

Adequate quality and availability of data are essential for the effectiveness of AI in adverse drug reaction (ADR) detection. However, fragmented and inconsistent data spread across various healthcare systems make it difficult for AI technologies to work. Improving data standardization and global collaboration will help to improve the effectiveness of AI. [20]

Regulatory and Ethical Barriers:

There are regulatory obstacles that stem from the lack of well-defined validation protocols for the AI technologies and a lack of transparency in decision making processes. Since AI is an inherently 'black box' phenomenon, there are serious issues of accountability and possible bias in drug safety monitoring. Thus, standardized governance frameworks need to be established. [19]

Infrastructure and Resource Constraints:

Many low and middle income countries (LMICs) have digital infrastructure, lack of technical expertise and low funding to implement AI based pharmacovigilance systems. Therefore, it is imperative to ensure the expansion of digital health infrastructure and training of workforce in these regions. [21]

Future Scope of AI in Pharmacovigilance

Artificial Intelligence (AI) in pharmacovigilance has a bright future because it keeps enhancing regulatory decision-making, reducing adverse drug reactions (ADRs), and improving drug safety. But, several important developments are anticipated in the next years if artificial intelligence is to reach its full potential

Enhance Global Data Collaboration:

One of the most important upcoming directions in pharmacovigilance is the creation of international data-sharing systems. At present, reports of adverse drug events are spread among several regulatory agencies, hospitals, and healthcare institutions, therefore limiting thorough analysis.

Working together worldwide, artificial intelligence can merge information from many sources including:

·         EHRs Electronic Health Records

·         Social Media Sites

Global Adverse Event Databases (FAERS CAROLINA)

Unifying these data sets enables AI to give worldwide real-time adverse events detection. This would help regulatory bodies to react faster and greatly cut down delays in spotting possible drug safety problems. [4,5,3]

Predictive and Personalized Pharmacovigilance:

Predictive and personalised medication is another direction that AI in pharmacovigilance is headed.  AI may evaluate genetic information, medical histories, and patient demographics to forecast how people would react to specific medications with the aid of machine learning algorithms. Predictive pharmacovigilance is the method that will assist in:  identifying adverse drug responses (ADRs) early on before they happen. determining which patients are at high risk for serious side effects.  increasing overall patient safety by customising medication treatments according to each patient's unique profile.  Personalised pharmacovigilance, as opposed to a generalised strategy, has the potential to improve treatment results and drastically lower hospitalisations associated to drugs. [15,5]

AI Driven Regulatory Decision-Making:

Another promising future scope of AI in pharmacovigilance is its part in administrative decision-making. As of now, administrative bodies such as the FDA (Nourishment and Medicate Organization) and the WHO (World Wellbeing Organization) physically audit antagonistic occasion reports some time recently issuing security alarms or medicate reviews. This handle is time-consuming and inclined to delays.

AI can significantly speed up this process by:

1.      Analysing huge volumes of adverse reaction event in real-time.

2.      Naturally distinguishing safety signals from pharmacovigilance databases.

3.      Giving administrative specialists with faster and more precise security data.

4.      As AI proceeds to make strides, it is anticipated that administrative bodies will progressively depend on AI-driven frameworks to guarantee quicker, data-driven decision-making. [3,4,5]

Integration of AI with pharmacogenomics:

Pharmacogenomics: The study of how your individual genes affect your response to medication. In the future, we will probably see AI integrated to pharmacogenomics, where pharmacovigilance is going to become something like predicting what is going to respond for the patients at individualized level.

For example: AI models can detect genetic variants that may predispose people to adverse drug reactions. Genetic data can help healthcare providers choose the safest and most effective drug for every patient. This integration will minimise the trial-and-error prescribing and eliminate major adverse drug events.

AI and pharmacogenomics in pharmacovigilance can help develop personalized treatment plans, preventing medication-related complications, the future of this field looks bright for patients as well as a healthcare system. [5,15,20]

Establishment of Standardised AI Regulations:

Uniform international laws are necessary for AI to realise its full potential in pharmacovigilance.  There are currently no established rules in the majority of nations regarding the use of AI in drug safety monitoring.  The scope of the future includes:  creating global regulatory guidelines for pharmacovigilance using AI.  ensuring the impartiality, equity, and transparency of AI systems.  encouraging ethical AI use while preserving patient confidentiality.  Pharmaceutical firms and healthcare providers will be more comfortable using AI for medication safety monitoring once uniform rules are in place.  In the end, this will produce pharmacovigilance results that are quicker and more accurate. [19,20]

Why Future Scope Matters

Using AI to its fullest capacity in order to anticipate, identify, and stop adverse medication responses is the key to the future of pharmacovigilance.  The healthcare sector may greatly lower drug-related risks and enhance patient outcomes by encouraging international data sharing, personalised medicine, AI-driven decision-making, and pharmacogenomics integration.  Furthermore, the implementation of uniform regulatory frameworks will guarantee the ethical and responsible application of AI in pharmacovigilance, which will eventually improve people's health. [5,19,15]

CONCLUSION

Artificial intelligence (AI) is changing how we keep track of drug safety. It helps us find problems with medicines, manage safety signals, predict risks, and use different types of real-world data together. AI methods like machine learning (ML) and natural language processing (NLP) work faster and with more detail than older methods. They allow us to prevent issues before they happen, instead of just handling them afterward. However, there are still challenges. Data comes from different sources that don’t always match well. AI can be biased if not set up correctly. There is no standard way to validate AI systems, and current rules are often unclear. Some AI models are hard to understand, so we need Explainable AI (XAI) to make them more transparent and reliable. Without fixing these problems, AI's use in drug safety will be limited.

The study shows we need teamwork from various experts, like drug specialists, doctors, data scientists, ethics experts, and regulators. Future efforts should focus on ethical AI, real-world evidence networks, and AI systems that adapt to monitor individual patients closely.

In short, AI has the potential to change drug safety monitoring, provided it's transparent, ethical, and medically practical. We must update rules, invest in diverse research, and focus on approaches centered on the patient. This way, drug monitoring becomes better at predicting and preventing problems and more tailored to individual needs, leading to better health results and strengthening public health.

REFERENCES:

1)      Fu, L., Jia, G., Liu, Z., Pang, X., & Cui, Y. (2025). The applications and advances of artificial intelligence in drug regulation: A global perspective. Acta Pharmaceutica Sinica B, 15(1), 1–14. https://doi.org/10.1016/j.apsb.2024.11.006

2)      Bekbolatova, M., Mayer, J., Ong, C. W., & Toma, M. (2024). Transformative potential of AI in healthcare: Definitions, applications, and navigating the ethical landscape and public perspectives. Healthcare, 12(2), 125. https://doi.org/10.3390/healthcare12020125

3)      World Health Organization. Pharmacovigilance [Internet]. Geneva: WHO; 2024  https://www.who.int/teams/regulation-prequalification/pharmacovigilance

4)      U.S. Food and Drug Administration. FDA Adverse Event Reporting System (FAERS) [Internet]. Silver Spring (MD): FDA; 2024 Dec 5  https://www.fda.gov/drugs/surveillance/fdas-adverse-event-reporting-system-faers

5)      Ahire YS, Patil JH, Chordiya HN, Deore RA, Bairagi VA. Advanced applications of artificial intelligence in pharmacovigilance: Current trends and future perspectives. J Pharm Res. 2024;23(1):23–33.  https://jopcr.com/articles/advanced-applications-of-artificial-intelligence-in-pharmacovigilance-current-trends-and-future-perspectives

6)      Al-Garadi MA, Yang YC, Sarker A. The Role of Natural Language Processing during the COVID-19 Pandemic: Health Applications, Opportunities, and Challenges. Healthcare (Basel). 2022 Nov 12;10(11):2270. doi: 10.3390/healthcare10112270. PMID: 36421593; PMCID: PMC9690240.

7)      Madan, S., Lentzen, M., Brandt, J. et al. Transformer models in biomedicine. BMC Med Inform Decis Mak 24, 214 (2024). https://doi.org/10.1186/s12911-024-02600-5

8)      Fu, L., Jia, G., Liu, Z., Pang, X., & Cui, Y. (2024). The applications and advances of artificial intelligence in drug regulation: A global perspective. Acta Pharmaceutica Sinica B, 15(1), 1–14. https://doi.org/10.1016/j.apsb.2024.11.006

9)      Kawamura, K., et al. (2024). Adverse Event Signal Detection Using Patients' Concerns in Pharmaceutical Care Records: Deep Learning Model Evaluation. Journal of Medical Internet Research, 26, e55794. https://doi.org/10.2196/55794Murali K, Kaur S, Prakash A, Medhi B. Artificial intelligence in pharmacovigilance: Practical utility. Indian J Pharmacol. 2019 Nov-Dec;51(6):373-376. Doi: 10.4103/ijp.IJP_814_19. Epub 2020 Jan 16. PMID: 32029958; PMCID: PMC6984023.

10)  Rekha, B.H., Hisham, S.A., Wahab, I.A. et al. Digital monitoring of medication safety in children: an investigation of ADR signalling techniques in Malaysia. BMC Med Inform Decis Mak 24, 395 (2024). https://doi.org/10.1186/s12911-024-02801-y

11)  DIA Global Forum. (2024). A new pharmacovigilance ecosystem: Automation, AI, and continuous improvement. DIA Global Forum. https://globalforum.diaglobal.org/issue/september-2024/a-new-pharmacovigilance-ecosystem-automation-ai-and-continuous-improvement/?utm_source=chatgpt.com

12)  Springer Nature. (2025). Artificial intelligence: Applications in pharmacovigilance signal detection. Nature Reviews Drug Discovery, 24(2), 123–135. https://doi.org/10.1007/s40290-025-00561-2?utm_source=chatgpt.com

13)  IQVIA. (2024). Enhancing pharmacovigilance intake processes with AI and automation. IQVIA Blogs. https://www.iqvia.com/blogs/2024/12/enhancing-pharmacovigilance-intake-processes-with-ai-and-automation?utm_source=chatgpt.com

14)  Ferreira, R. D. A., Zhong, S., Moureaud, C., Le, M. T., Rothstein, A., Li, X., Wang, L., & Patwardhan, M. (2024). A pilot, predictive surveillance model in pharmacovigilance using machine learning approaches. Advances in Therapy, 41(6), 2435–2445. https://doi.org/10.1007/s12325-024-02870-5

15)  Painter JL, Kassekert R, Bate A. An industry perspective on the use of machine learning in drug and vaccine safety. Front Drug Saf Regul. 2023 Feb 1;3:1110498. doi: 10.3389/fdsfr.2023.1110498.

16)  Seal, S., Williams, D., Hosseini-Gerami, L., Mahale, M., Carpenter, A. E., Spjuth, O., & Bender, A. (2024). Improved Detection of Drug-Induced Liver Injury by Integrating Predicted In Vivo and In Vitro Data. Chemical Research in Toxicology, 37(8), 1290–1305. https://doi.org/10.1021/acs.chemrestox.4c00015

17)  Seal, S., Spjuth, O., Hosseini-Gerami, L., García-Ortegón, M., Singh, S., Bender, A., & Carpenter, A. E. (2024). Insights into Drug Cardiotoxicity from Biological and Chemical Data: The First Public Classifiers for FDA Drug-Induced Cardiotoxicity Rank. Journal of Chemical Information and Modeling, 64(4), 293–308. https://doi.org/10.1021/acs.jcim.3c01834

18)  Desai MK. Artificial intelligence in pharmacovigilance - Opportunities and challenges. Perspect Clin Res. 2024 Jul-Sep;15(3):116-121. doi: 10.4103/picr.picr_290_23. Epub 2024 Mar 27. PMID: 39140015; PMCID: PMC11318788.

19)  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. https://doi.org/10.1016/j.hlpt.2023.100743

20)  Liang, L., Hu, J., Sun, G., Hong, N., Wu, G., He, Y., Li, Y., Hao, T., Liu, L., & Gong, M. (2022). Artificial Intelligence-Based Pharmacovigilance in the Setting of Limited Resources. Drug Safety, 45(5), 511-519. https://doi.org/10.1007/s40264-022-01170-7

21)  Nikfarjam, A., Sarker, A., O'Connor, K., Ginn, R., & Gonzalez, G. (2015). Pharmacovigilance from social media: Mining adverse drug reaction mentions using sequence labeling with word embedding cluster features. Journal of the American Medical Informatics Association: JAMIA, 22(3), 671–681. https://doi.org/10.1093/jamia/ocu041