Media Farmasi Indonesia Vol 20 No 2 | DOI 10.
53359/mfi.
Transformation in Pharmacovigilance Signal Detection: From Spontaneous Reporting System to Artificial Intelligence Integration Novita Diana Ayu Candra*.
Nur Muhammad Herunda Putra.
Endang Darmawan Faculty of Pharmacy.
Universitas Ahmad Dahlan email: ndacnovita@gmail.
Abstract Adverse drug reactions (ADR.
are a global health problem that has a significant impact on patient safety and healthcare system efficiency.
Spontaneous reporting systems (SRS) have been the main approach in pharmacovigilance signal detection, but still face obstacles such as underreporting, reporting delays, and reporter bias.
Meanwhile, the development of artificial intelligence (AI) offers transformational potential through big data-based automated ADR detection.
This review aims to comprehensively evaluate the differences and potential integration between SRS and AI in pharmacovigilance signal detection.
The method used was a narrative review of original literature published between 2015-2025 from PubMed.
BMC, and Google Scholar databases.
A total of 63 articles were screened, and 11 key studies were analyzed in depth.
Results showed that although various interventions were able to improve ADR reporting through SRS, their effectiveness was limited.
In contrast.
AI demonstrated high capability in detecting ADRs from unstructured data with superior accuracy and speed.
In-depth discussions highlighted that AI approaches can strengthen pharmacovigilance systems that have relied on manual reporting.
In conclusion, the integration of SRS and AI is a promising strategy to address modern pharmacovigilance challenges, and should be adopted through a national system supported by regulation, data digitization, and human resource capacity strengthening.
Keywords: pharmacovigilance, adverse drug reactions, spontaneous reporting system, artificial intelligence INTRODUCTION Pharmacovigilance (PV) is a scientific and regulatory activity related to the detection, assessment, understanding, and prevention of adverse drug events or other problems associated with drug use.
Pharmacovigilance can help in detecting new or unknown signals between drugs and adverse reactions.
Adverse drug reactions (ADR.
are a public health problem that requires the attention of all stakeholders regardless of the practice environment (Ampadu et al.
, 2.
The incidence of ADRs in the world represents a significant burden to the global health In high-resource countries.
ADR reporting reaches between 3 - 613 reports per one million population per year (Aagaard et al.
, 2.
Developing countries show wide variations in reporting, ranging from zero to 50,000 ADR reports per year.
Thailand.
Malaysia and Singapore have the highest reporting rates in the Southeast Asian region.
Patient self-reporting of ADRs is still very low, with the highest rate recorded in Denmark Media Farmasi Indonesia Vol 20 No 2 | DOI 10.
53359/mfi.
reports per million populatio.
, while in Asian countries such as Malaysia and the Philippines it only reached 0.
86 and 0.
01 reports per million population (Worakunphanich et , 2.
Globally, more than 43,000 reports of ADR deaths were recorded in the WHOVigiBase database during the period 2010 to 2019 (Montastruc et al.
, 2.
These data confirm that ADRs pose a serious threat to patient safety and healthcare efficiency Healthcare workers are the focal point in operationalizing a pharmacovigilance system capable of managing ADRs (Gyner & Ekmekci, 2.
With insufficient knowledge, health workers may not be able to report ADRs appropriately (De Angelis et al.
, 2.
However, they are required to have a good understanding and skills in the area of drug safety related to detecting, recognizing, managing and reporting suspected adverse drug reactions .
an Eekeren et al.
, 2.
SRS have long been used as a traditional method of detecting pharmacovigilance signals (WHO, 2.
However, as information technology develops.
AI methods are being used to improve the efficiency and accuracy of signal detection (Warner et al.
, 2.
The development of AI technology opens up opportunities to strengthen PV systems through automatic detection and analysis of unstructured data.
Unfortunately, to date, there are not many studies that systematically compare the strengths and limitations of each SRS and AI approach, especially in the context of integration for national systems.
Therefore, this study aims to critically analyze the differences and potential synergies between SRS and AI in pharmacovigilance signal detection, and evaluate their relevance to the challenges of PV implementation in the digital era.
METHOD
This review is a narrative review of the literature organized following the PRISMA guidelines for narrative reviews.
Literature was searched through online databases such as PubMed.
BMC, and Google Scholar using the keywords: "adverse drug reactions", "pharmacovigilance", "signal detection", "spontaneous reporting system", "artificial intelligence", and "machine learning".
Inclusion criteria included original articles, in English, published between 2015 and 2025 and relevant to ADR signal detection via SRS or AI.
From a total of 147 articles found, 63 articles met the inclusion criteria after screening based on title, abstract, and full text.
A total of 11 key studies were reviewed in depth as they specifically compared or addressed the effectiveness and challenges of ADR reporting through SRS and AI.
Data were analyzed and presented in a descriptive-comparative manner.
Media Farmasi Indonesia Vol 20 No 2 | DOI 10.
53359/mfi.
RESULTS AND DISCUSSION
According to the results of a study conducted by that SRS faces challenges in timeliness and data completeness, so the application of AI models successfully identified more potential ADR cases although it still requires expert judgment at the clinical validation stage to interpret the results (Crisafulli et al.
, 2.
As technology develops.
AI integration is being applied to overcome these shortcomings and improve signal detection more quickly and accurately.
The results of studies related to the development of ADR detection systems from conventional methods to the utilization of AI are presented in Table 1 below.
Table 1.
ADR Pharmacovigilance Studies on SRS and AI No.
Author Title Result Asiamah et al.
Spontaneous reporting of 1.
Factors that significantly influenced adverse drug reactions among spontaneous ADR reporting were age health professionals in Ghana (AOR = 2.
, pharmacovigilance training (AOR = 18.
, and barriers such as legal fear (AOR = 0.
, time constraints (AOR = 0.
, and unavailability of reporting forms (AOR = 0.
The proportion of spontaneous reporting of ADRs by health workers in Kpone-Katamanso District is low, thus requiring a mandatory reporting policy for health workers.
Fang et Multifaceted interventions for 1.
Physician training.
KAP education, and economic incentives improved SRS reporting of adverse drug compliance, while changes in drug reactions in a general hospital guidelines affected ADR patterns and in China SRS adherence improved through multifaceted interventions.
Drug use patterns affect ADRs, so rational use programs are important to strengthen SRS and pharmacovigilance development in China.
Yu & Lee, .
Enhanced knowledge of 1.
Pharmacists with private access were spontaneous reporting with less likely to correctly identify SRS content than those who attended Korean structured programs .
<0.
community pharmacists: a 2.
Pharmacists' knowledge was lower cross-sectional study regarding reporting of non-prescription products, supplements, and hygiene products than prescription drugs Media Farmasi Indonesia Vol 20 No 2 | DOI 10.
53359/mfi.
<0.
Structured education programs, alone or in combination, improved SRS Ma et Immune checkpoint inhibitors 1.
Signal detection indicating Avelumab-related related myocarditis was highest with patients with cancer: an the ROR and PRR methods, while the analysis of international strongest signal for Ipilimumab was detected with the BCPNN method.
This study highlights the potential risk of myocarditis due to ICI use, in line with previous clinical trials, and may serve as a reference for clinical personnel in its use.
Bukic et al.
Analysis of spontaneous 1.
ADRs were most common in patients reporting of suspected adverse Ou70 years .
%), and 5% of reports drug reactions for nonwere from accidental exposure in over-the-counter drugs from 2008 to 2017 Pharmacists most commonly report ADRs of over-the-counter drugs, and consumer awareness is increasing.
Health workers need education on ADR reporting, especially regarding the safety of over-the-counter drugs in the elderly and children.
Kim et A cross-sectional survey of 1.
Positive attitude.
SRS awareness, selfknowledge, attitude, and efficacy, and ADR counseling willingness to engage in experience significantly influenced spontaneous reporting of reporting intention .
OR >1, p < 0.
adverse drug reactions by 2.
The Korean consumers.
consumer vigilance and empowering self-reporting to strengthen drug safety.
Martin et al.
Validation Artificial 1.
The AI model showed promising Intelligence to Support the performance in automatically coding Automatic Coding of Patient ADR reports, with consistent results Adverse Drug Reaction across different approaches.
Reports.
Using Nationwide 2.
The system has been used by French Pharmacovigilance Data health authorities since January 2021 to support pharmacovigilance of COVID19 vaccines.
Further validation is needed in other settings.
Media Farmasi Indonesia Vol 20 No 2 | DOI 10.
53359/mfi.
Lytinier et al.
Artificial Intelligence for 1.
The best model for ADR identification Unstructured Healthcare was LGBM with AUC 0.
93 and FData: Application to Coding external validation of Patient Reporting of showed AUC 0.
91 and F-measure 0.
Adverse Drug Reactions AI is capable of learning from unstructured data, supporting further studies and development of practical tools for drug safety management.
Destere et al.
An artificial intelligence 1.
dLBM identifies drug-ADR specific algorithm for co-clustering to associations, such as antiplatelet and help in pharmacovigilance anticoagulants with bleeding.
before and during the COVID- 2.
Together with co-clustering, dLBM is a 19 pandemic promising tool for unsupervised detection and exploration of safety signals in large-scale pharmacovigilance Roosan et al.
Artificial Intelligent Context- 1.
The AI algorithm aTarantula was Aware Machine-Learning successfully developed to detect Tool to Detect Adverse Drug warfarin-related ADEs from online Events from Social Media forums, with a sensitivity of 84.
2% and Platforms specificity of 98%.
This study demonstrates the potential of aTarantula, which can be further validated on more diverse data.
Gordo et al.
Root causes of adverse drug events in hospitals and capabilities for prevention The analysis shows that AI's capabilities in identification and readout can help prevent ADEs, by addressing misidentification as a major root cause of adverse drug events.
ADR reporting with SRS Most studies show that ADR reporting through the SRS remains low, influenced by personal, institutional and systemic factors.
Asiamah et al.
, .
noted that PV training increased the likelihood of reporting (AOR = 18.
, while legal fear (AOR = 0.
and time constraints (AOR = 0.
decreased it.
Fang et al.
, .
showed that training, education, and incentives increased compliance with the SRS.
Meanwhile, they found that structured education programs contributed to the improvement of reporting knowledge by community Ma et al.
, .
analyzed data from international reporting systems and identified myocarditis signals associated with the use of immune checkpoint inhibitors (ICI.
such as Avelumab and Ipilimumab.
Bukic et al.
, .
noted that the age group Ou70 years and children are vulnerable populations to ADRs from over-the-counter drugs.
Kim et al.
asserted that positive attitude and counseling experience increase the willingness of ADR reporting by consumers.
Media Farmasi Indonesia Vol 20 No 2 | DOI 10.
53359/mfi.
ADR Reporting with AI AI approaches in PV provide promising results in signal detection.
Martin et al.
, .
validated an AI model for automatic coding of patient ADR reports nationwide in France, showing high performance in data classification.
Lytinier et al.
, .
noted that the LGBM model achieved an AUC of 0.
93 and F-measure of 0.
72 in detecting ADRs from unstructured Destere et al.
, .
developed dLBM to identify drug-ADR relationships in large databases, especially during the COVID-19 pandemic.
Roosan et al.
, .
developed the aTarantula algorithm that detects ADEs from social media with 84.
2% sensitivity and 98% Discussion The findings in this study emphasize the fundamental differences between SRS systems and AI approaches, in terms of methodology, effectiveness, and data coverage.
SRS
systems, as a traditional approach, have the strength of directly documenting reports from clinical practice, but are limited by reliance on human reporters, risk of bias, delay, and underreporting (Ampadu et al.
, 2.
Gyner & Ekmekci, 2.
These challenges are consistent globally, as revealed by Aagaard et al.
, .
who reported that low-resource countries show very low reporting rates.
Although interventions such as training and incentives improve reporting Fang et al.
and Yu & Lee .
the sustainability of interventions and their long-term effects have not been fully measured.
Even in countries with developed health systems, patient selfreporting remains low (Worakunphanich et al.
, 2.
This suggests that the SRS approach is not sufficient to detect the entire spectrum of ADRs, especially in vulnerable populations or in the context of new drug use.
In contrast.
AI approaches have significant advantages in terms of speed, efficiency, and data coverage.
AI models such as LGBM, dLBM, and aTarantula demonstrate better ADR detection capabilities than traditional methods, especially from unstructured data such as online forums and electronic medical records (Martin et al.
, 2022.
Roosan et al.
, 2.
This capability is important as a lot of ADR data is hidden in narrative medical records or informal platforms that are not reached by SRS systems.
Furthermore.
AI enables predictive analysis and early detection, which is particularly relevant in emergency situations such as the COVID-19 pandemic, where the use of new or off- label drugs is drastically increasing (Destere et al.
, 2.
This strengthens the argument that the integration of SRS and AI can create a PV system that is more responsive and Media Farmasi Indonesia Vol 20 No 2 | DOI 10.
53359/mfi.
adaptive to clinical and global dynamics.
However, it should be noted that the use of AI is not without challenges.
Issues such as limited external validation, potential algorithm bias, as well as the need for ethical and regulatory oversight, remain a concern (Gordo et al.
The development of a hybrid system between SRS and AI needs to be supported by national policies, digital infrastructure, and increased human resource capacity in the health
CONCLUSIONS
Detection of pharmacovigilance signals through SRS still faces challenges of underreporting, reporting delays, and reporter bias.
The integration of AI offers an innovative solution with big data analysis capabilities, high accuracy, and detection from unstructured data automatically and in real-time.
Evidence from various studies shows that the combination of SRS and AI approaches can complement each other to improve the effectiveness and accuracy of ADR detection.
In the future, the development of a hybrid pharmacovigilance system needs to be supported by national policies, health worker training, and a strong digital infrastructure.
This strategy is important to realize an adaptive, precise, and proactive drug safety system in the era of digital transformation of healthcare.
REFERENCES