A Systematic Review of Artificial Intelligence (AI) and Machine Learning (ML) in Pharmaceutical Supply Chain (PSC) Resilience: Current Trends and Future Directions

Abstract

The resilience of the pharmaceutical supply chain (PSC) is crucial to ensuring the availability of medical products. However, increasing complexity and logistical bottlenecks have exposed weaknesses within PSC frameworks. These challenges underscore the urgent need for more resilient and intelligent supply chain solutions. Recently, Artificial Intelligence and machine learning (AI/ML) have emerged as transformative technologies to enhance PSC resilience. This study presents a systematic review evaluating the role of AI/ML in advancing PSC resilience and their applications across PSC functions. A comprehensive search of five academic databases (Scopus, the Web of Science, IEEE Xplore, PubMed, and EMBASE) identified 89 peer-reviewed studies published between 2019 and 2025. PRISMA 2020 guidelines were implemented, resulting in a final dataset of 32 studies. In addition to analyzing applications, this study identifies the AI/ML grouped into five main categories, providing a clearer understanding of their impact on PSC resilience. The findings reveal that despite AI/ML’s promise, significant research gaps persist. Particularly, AI/ML-driven regulatory compliance and real-time supplier collaboration remain underexplored. Over 59.3% of studies fail to address regulatory frameworks and ethical considerations. In addition, major challenges emerge such as the limited real-world deployment of AI/ML-driven solutions and the lack of managerial impacts on PSC resilience. This study emphasizes the need for stronger regulatory frameworks, broader empirical validation, and AI/ML-driven predictive modeling. This study proposes recommendations for future research to foster more efficient, transparent and ethical PSCs capable of navigating the complexities of global healthcare.

1. Introduction

The pharmaceutical supply chain (PSC) is a critical pillar of global healthcare, ensuring the availability of essential medicines, vaccines, and medical products. However, increasing complexity and exposure to disruptions, such as pandemics, geopolitical instability, logistical constraints, and natural disasters, have revealed significant vulnerabilities in existing PSC frameworks. The COVID-19 pandemic, in particular, exposed global supply chain fragility, leading to severe shortages of life-saving products and underscoring the need for more resilient systems [1].

PSC resilience is broadly defined as the capacity to anticipate, respond to, and recover from disruptions while maintaining continuous supply and service [2]. Nonetheless, traditional approaches to enhancing PSC resilience often lack the responsiveness needed to address large-scale, dynamic disruptions, as demonstrated during the COVID-19 pandemic [3,4]. This has prompted growing interest in leveraging advanced technologies, notably artificial intelligence (AI) and machine learning (ML), to strengthen resilience.

While academic discussions on AI/ML in PSCs remain limited, real-world implementations have begun to emerge in areas such as vaccine distribution and cold chain management mainly in light of the COVID-19 pandemic [5]. AI-driven predictive analytics support accurate vaccine demand forecasting, while machine learning algorithms help optimize delivery routes and logistics. Major pharmaceutical companies, such as Pfizer and Merck, have implemented AI solutions to streamline vaccine distribution, enhancing both efficiency and regulatory compliance [6]. Additionally, AI/ML-enabled cold chain monitoring systems have helped maintain temperature integrity and improve supply chain transparency [5]. These applications illustrate the practical potential of AI/ML technologies in enhancing PSC operations, particularly in time-sensitive and high-risk contexts.

Recent systematic reviews have attempted to consolidate knowledge in this emerging field. For example, Katakam et al. [7] examined ethical and practical challenges in AI/ML integration, while Smyth et al. [8] identified significant knowledge gaps in AI’s contributions to PSC resilience through a comprehensive review of 76 primary studies. Similarly, Sharma et al. [3] analyzed 67 risks in Indian PSCs using fuzzy evaluation methods, and Emmanuel et al. [9] emphasized the importance of real-time data and blockchain in PSC management using artificial neural networks. However, these studies often focus on risk assessment and theoretical applications, with limited attention on practical implementation, real-world testing, or regulatory compliance. Recently, Weraikat and Al-Hourani [10] reviewed two major databases, the Web of Science (WoS) and IEEE Xplore, and identified only seven articles explicitly discussing AI/ML applications for enhancing PSC resilience.

A consistent theme highlighted across these reviews is the lack of empirical validation, narrow methodological scope, and under-representation of advanced techniques such as reinforcement learning (RL) and Natural Language Processing (NLP). Moreover, few studies consider ethical implications or alignment with industry regulations like Good Distribution Practices (GDPs), Good Manufacturing Practices (GMPs), or the General Data Protection Regulation (GDPR) [11].

The objective of this review is to systematically analyze existing research on the application of AI and ML in enhancing pharmaceutical supply chain resilience. Based on 32 primary studies, it categorizes AI/ML techniques into five core methodological classes and examines five key dimensions: resilience strategies, AI/ML applications, functional areas, study types, and regulatory challenges. By identifying current trends, underexplored areas, and critical research gaps, this review bridges theoretical advancements and practical implementations. Ultimately, it provides a comprehensive roadmap to guide future interdisciplinary research towards developing more adaptive, ethical, and resilient pharmaceutical supply chains.

The remainder of this article is structured as follows. Section 2 describes the methodology employed in this study. Section 3 presents the results and their analysis. A detailed discussion of the findings is provided in Section 4. Finally, Section 5 concludes the paper with future insights.

2. Methodology

With the increasing reliance on AI/ML technologies in SCM, it is essential to consolidate current knowledge, identify key trends, highlight gaps, and suggest future research directions. This section describes the methodologies used to identify the relevant articles.

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology [12] was employed to identify relevant articles. PRISMA provides a structured and transparent process for identifying, screening, and analyzing the relevant literature, ensuring scientific credibility and reproducibility. This research was conducted in three main stages: first, an extensive literature search across multiple academic databases; second, a structured selection process to refine the dataset; and finally, a detailed analysis of research themes, challenges, and gaps.

A full-text review followed, leading to the exclusion of 9 additional studies due to insufficient empirical validation, the lack of AI/ML integration, or irrelevance to PSC resilience. By applying these criteria, after independent screening by each researcher, followed by cross-validation for consistency, 32 peer-reviewed studies were included in the final dataset, forming the foundation of this systematic review. A PRISMA flow diagram, as shown in Figure 1, visually represents this process, showing the progression from initial search results to the final selection of studies.

2.1. Data Sources and Search Strategy

A structured search was conducted using five major academic databases: Scopus, the Web of Science (WOS), IEEE Xplore, PubMed, and EMBASE. The databases were selected for their comprehensive indexing of high-quality peer-reviewed research in AI, SCM, and pharmaceutical logistics. The search strategy was formulated to capture a wide range of relevant AI/ML applications in PSC resilience. A Boolean search string was applied uniformly across all databases to ensure consistency:

(“Resilien*” OR “Adapt*” OR “flexibil*”) AND (“Pharmaceutical” OR “Pharmac*” OR “Drug” OR “Vaccin*” OR “Medic*”) AND (“Supply Chain” OR “Demand Forecasting” OR “Inventory Management” OR “Supplier Selection”) AND (“Logistics” OR “Transportation” OR “Distribution”) AND (“AI” OR “Artificial Intelligence” OR “Machine Learning” OR “ML” OR “Neural Networks”).

This search strategy was carefully developed to ensure relevant studies were retrieved while minimizing the inclusion of unrelated research.

2.2. Inclusion and Exclusion Criteria

To ensure a well-structured and high-quality selection of studies, inclusion and exclusion criteria were carefully defined. These criteria helped to filter out irrelevant studies while retaining only those that were scientifically directly relevant to AI/ML applications in PSC resilience.

Studies were included in the review if they explicitly integrated AI/ML applications within the context of enhancing supply chain resilience, as identified through the Boolean search strategy. Only peer-reviewed journal articles and conference proceedings were considered to maintain scientific credibility. Additionally, only studies published in English were included to ensure consistency in data analysis and interpretation.

Studies that did not directly align with the research focus were excluded. This included research that broadly addressed SCM without a specific focus on AI/ML or the pharmaceutical sector. Studies centered on industries outside of healthcare and pharmaceuticals, such as retail or general logistics, were also removed. Furthermore, duplicate studies were excluded, with only the most recent or comprehensive version retained. Non-peer-reviewed sources, including book chapters, press articles, and opinion pieces, were omitted as they did not meet the methodological standards for systematic reviews.

To capture recent advancements in AI/ML, studies published before 2019 were excluded. The articles were collected during the fourth week of December 2024, covering research published over approximately the past five years. Following this structured screening process, the initial dataset of 89 studies was refined to 32 peer-reviewed articles, providing a strong foundation for analyzing AI/ML applications, research methodologies, and challenges in PSC resilience.

2.3. Research Dimensions and Classification Approach

Following the dataset finalization, the selected studies were examined using five research dimensions developed through an inductive and iterative synthesis process. These dimensions were not adopted from pre-existing classification frameworks but emerged from a detailed thematic analysis of the 32 articles included in this review. While the construction of these dimensions was informed by the foundational literature on supply chain resilience (e.g., [1,2]), their final structure reflects recurring patterns, emerging priorities, and key thematic elements relevant to AI/ML applications in the PSC context.

Below, each dimension is outlined with its conceptual foundation (see Figure 2):

1.

Resilience Strategies

This dimension examines how AI/ML contributes to PSC resilience by consolidating related constructs into three integrated categories:

• Agile Flexibility, combining agility and flexibility, focuses on systems designed to adapt quickly to disruptions and changing conditions through techniques such as dynamic demand forecasting, alternative sourcing, and real-time logistics adjustments. For instance, some studies applied AI/ML to reroute shipments dynamically in response to disruption alerts.
• Collaborative Visibility, merging supply chain visibility and inter-organizational collaboration, captures the role of AI/ML in enabling real-time tracking and coordinated response. Several studies demonstrated the integration of AI-powered tracking tools with supplier platforms to synchronize inventory and delivery operations.
• Resilient Risk Management, combining redundancy and risk mitigation, includes predictive risk identification and the proactive deployment of contingency mechanisms such as backup suppliers or safety stock. Predictive analytics models were commonly used in reviewed studies to anticipate potential supply failures and trigger alternative sourcing protocols.

This grouping aligns with concepts in the resilience literature [13,14] and reflects the practical applications observed in AI/ML-enhanced PSCs.

2.

AI/ML Applications

This dimension identifies the primary functional roles AI/ML systems fulfill in PSC contexts:

• Demand forecasting involves predicting future demand using time-series analysis, neural networks, or hybrid methods. For example, Long Short-Term Memory (LSTM) models have been used to anticipate vaccine demand in fluctuating market environments.
• Inventory optimization refers to models designed to balance stock levels, improve reorder efficiency, and reduce holding costs. In one study, reinforcement learning was applied to optimize inventory replenishment across multiple distribution centers.
• Resilience risk management includes the identification and mitigation of operational risks using AI/ML. Several studies employed decision trees and ensemble algorithms to detect potential disruptions, such as unreliable suppliers or route failures, before they could impact operations.

These categories align with application domains in AI/ML research across logistics and healthcare [15].

3.

Functional Areas

This dimension categorizes the functional domains within the PSC where AI/ML tools are implemented:

• Supply Planning and Procurement includes sourcing, supplier selection, and purchasing strategies. One example involved the use of AI models to evaluate supplier reliability and optimize contract allocations.
• Logistics and Inventory Management encompasses warehousing, transportation, and inventory control. Multiple studies applied AI/ML to optimize delivery routes and warehouse layouts based on demand variability.
• Customer Service focuses on end-user engagement, delivery responsiveness, and access assurance. A number of studies illustrated AI-driven chatbot systems for real-time order tracking and patient communication in last-mile logistics.

4.

Study Types

Studies were classified based on the methodological approach:

Empirical/Case Studies include real-world data analysis, observational studies, or practical implementations. For instance, one study detailed AI deployment in a hospital pharmacy’s inventory system during a regional health crisis.

Theoretical Studies present conceptual frameworks or analytical models without field application. An example includes a proposed model linking AI to enhanced transparency in PSC monitoring systems.

Simulation Studies use synthetic data or computational modeling to test system performance under various scenarios. One study employed agent-based simulation to assess the robustness of AI-assisted logistics under pandemic conditions.

This typology aligns with common practices in systematic literature reviews [16].

5.

Regulatory Challenges

This dimension addresses governance, compliance, and ethical considerations in the application of AI/ML to PSC operations:

• Data Privacy and Security concerns were noted in studies that discussed the need for compliance with regulations such as GDPR and the HIPAA when handling health-related data.
• Ethical Considerations focused on the fairness, transparency, and explainability of AI models, especially in contexts where algorithmic decisions impact medicine accessibility.
• Regulatory Compliance referred to adherence to established industry standards, including Good Distribution Practice (GDP) and Good Manufacturing Practice (GMP), particularly in automated systems for handling temperature-sensitive products.

Although relatively few studies directly addressed regulatory issues, the growing prominence of legal and ethical concerns in digital healthcare logistics was noted across multiple papers [17].

By structuring the review around these five dimensions, this study offers a holistic perspective on AI/ML applications in PSC resilience while identifying major research gaps and areas for future exploration.

A cross-sectional analysis was conducted to classify the specific AI/ML models used across the studies, such as supervised, unsupervised, deep learning, and hybrid techniques, providing deeper insight into the methodological landscape of AI/ML implementation in PSCs to enhance resilience.

3. Results and Analysis

This systematic review provides a comprehensive overview of how AI/ML technologies contribute to enhancing PSC resilience. An analysis of the 32 peer-reviewed studies shows significant contributions in demand forecasting, inventory optimization, and supply chain risk management. However, gaps remain in regulatory compliance, empirical validation, and large-scale real-world implementation. While some studies present promising AI/ML models, few have been tested or deployed at scale. Additionally, data privacy, security, and ethical considerations remain underexplored.

The review further analyzes publication trends over time, keyword usage patterns, and the classification of AI/ML models used, including supervised, unsupervised, deep learning, and hybrid approaches. It also structures the analysis around five core dimensions, applying a sub-criterion to classify topics as either the primary or secondary focus of each article.

Table 1. Classification of 32 peer-reviewed articles based on research dimensions.

Reference	Resilience Dimensions			AI/ML Application			Functional Areas			Study Types			Regulatory Challenges
	1 a	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16
[7]	*					*	*				*					*
[11]	*					*	*				*					*
[3]			*			*		*			*					*
[18]	*					*	*				*					*
[9]			*	*			*				*					*
[19]			*		*			*				*				*
[20]		*				*	*				*					*
[21]			*			*		*			*					*
[22]	*				*			*		*						*
[23]			*	*				*				*				*
[24]			*	*				*		*						*
[25]			*	*			*			*						*
[26]			*	*			*				*			*
[4]			*			*	*			*				*
[27]			*			*		*		*				*
[28]			*			*	*				*			*
[29]	*					*		*		*						*
[30]		*				*		*			*					*
[31]			*			*	*				*					*
[32]		*				*		*		*				*
[33]			*			*		*		*				*
[34]			*			*		*		*				*
[35]			*			*		*		*			*
[36]			*			*		*				*	*
[15]			*			*		*		*			*
[37]			*			*	*					*	*
[38]			*			*		*		*			*
[39]		*				*		*				*		*
[40]			*			*		*				*				*
[41]			*	*			*			*						*
[42]		*				*		*		*						*
[43]			*			*	*					*				*

^a The titles of the sub-dimensions numbered 1 to 16 are provided in Figure 2; * provides the primary classification of the 32 studies across these dimensions.

provides the primary classification of the 32 studies across these dimensions.

3.1. Publication Growth over Time

The growing interest in AI/ML applications for PSC resilience is evident from the publication trends over the past five years. As shown in Figure 3, there was no research on this topic in 2019, and after that, the number of publications gradually increased, reflecting a shift in focus toward technology-driven solutions for supply chain challenges. The COVID-19 pandemic acted as a wake-up call for the pharmaceutical industry, exposing major weaknesses in global supply chains. The sudden demand fluctuations, logistical bottlenecks, and disruptions in drug distribution created an urgent need for data-driven forecasting and AI-enhanced decision-making tools.

The research period was limited to studies published from 2019 onward. This timeframe was selected based on initial findings indicating that significant research on AI/ML applications for PSC resilience began to emerge primarily during and after the COVID-19 pandemic. This explains the emergence of AI/ML-based PSC resilience research in 2020 and its steady growth in the following years. In 2021, six studies explored AI’s role in managing supply chain disruptions, while 2022 saw a slight dip with four publications, likely due to shifting research priorities as industries adapted to pandemic-related challenges. The notable increase in 2023 (9 publications) and 2024 (12 publications) suggests that AI/ML in PSC resilience is no longer just an emerging topic but areas of serious academic and industry exploration.

This growth reflects a shift from theoretical discussions to more practical applications, with researchers focusing on AI-powered demand forecasting, automated inventory management, and real-time logistics optimization. The peak in 2024 reflects a growing research interest in AI/ML as promising tools for improving adaptability and resilience in PSCs

Despite the encouraging rise in research, one critical question remains: how much of this knowledge is being implemented in real-world supply chains? While AI/ML has gained traction in academic discussions, its widespread adoption across pharmaceutical industries is still limited, often due to regulatory challenges, data privacy concerns, and integration issues with legacy systems.

As illustrated in Figure 3, the upward trend in AI/ML-related PSC resilience research underscores the growing recognition of AI’s role in overcoming supply chain challenges. However, to fully realize its potential, future research should focus on developing new AI/ML models and addressing practical implementation challenges, regulatory compliance, and long-term scalability.

3.2. Keyword Analysis in AI/ML and PSC Research

To better understand the main research themes in AI/ML applications for PSC resilience, we performed a keyword analysis based on the titles and abstracts of the 32 selected studies. This analysis helped identify the most frequently used terms, revealing key focus areas and emerging trends in the field.

We adopted a frequency-based approach, utilizing Microsoft Excel and Python version 3.10.12 to extract and count the appearance of key terms. A predefined list of AI/ML and supply chain-related keywords was initially developed through a preliminary literature scan and refined iteratively during the article review. This method enabled a structured yet flexible analysis of thematic emphasis across studies.

As illustrated in Figure 4, the term “supply chain” appeared most frequently, highlighting the central role of AI/ML in improving supply chain efficiency, resilience, and optimization. Keywords such as “vaccine” and “health” were also common, reflecting the impact of the COVID-19 pandemic and the heightened focus on healthcare logistics and emergency preparedness.

Terms like “pharma,” “drug,” and “pharmaceutical” further suggest a strong research emphasis on AI/ML applications in medication production, distribution, and regulatory compliance. The prominence of “machine learning” and “artificial intelligence” underscores the growing relevance of these technologies in forecasting, risk detection, and operational automation.

Additionally, keywords such as “distribution,” “resilience,” and “logistics” indicate continued interest in disruption management and supply chain visibility. However, terms like “risk management” and “adaptability” were mentioned less often, pointing to underexplored opportunities for AI-driven strategies in proactive risk mitigation and dynamic response mechanisms.

3.3. Research Dimensions

3.3.1. Resilience Dimensions in AI/ML-Driven PSC Research

Resilience in PSCs is crucial for ensuring stability, responsiveness, and risk mitigation, particularly in the face of disruptions such as pandemics, geopolitical conflicts, and supply shortages. AI/ML applications contribute significantly to enhancing resilience in PSCs by improving real-time decision-making, visibility, and risk management. To systematically analyze how AI/ML contributes to PSC resilience, this study categorized resilience strategies into three primary dimensions:

• Agile Flexibility: This category focuses on adaptability and responsiveness to disruptions, including dynamic demand forecasting, adaptive decision-making, and flexible logistics networks. AI/ML techniques in this area support rapid adjustments to supply chain fluctuations and enhance overall operational agility.
• Collaborative Visibility: This dimension emphasizes real-time tracking, monitoring, and information sharing within the supply chain. Technologies such as blockchain, the Internet of Things (IoT), and AI-powered supplier collaboration tools enhance supply chain transparency, stakeholder coordination, and predictive insights into potential disruptions.
• Resilient Risk Management: This category focuses on risk identification and mitigation, incorporating strategies such as predictive risk modeling, safety stock management, and backup supplier strategies. AI/ML models enable the early detection of vulnerabilities and the development of proactive contingency plans to maintain supply chain stability.

To evaluate how AI/ML research aligns with resilience strategies in PSCs, we categorized the 32 selected studies based on their primary and secondary thematic alignment. Studies that primarily focused on one resilience category were assigned a primary classification, while those that incorporated multiple resilience dimensions but had a stronger emphasis on one were assigned a secondary classification. The classification results, as illustrated in Figure 5, indicate a clear preference for risk management strategies, with 22 studies primarily focusing on AI-driven risk identification and mitigation, while only 5 studies each prioritized agile flexibility and collaborative visibility as their main themes. However, 14 studies considered agile flexibility as a secondary theme, and 12 studies integrated collaborative visibility as a supporting factor.

These findings suggest that AI/ML applications in PSCs are predominantly geared toward predictive risk management, helping supply chains anticipate and mitigate disruptions before they escalate. The significant emphasis on risk identification, safety stock management, and predictive modeling highlights a strong industry-wide focus on crisis prevention and resilience planning. While risk management is a crucial aspect of PSC resilience, the relatively lower focus on agile adaptability and collaborative transparency suggests a potential imbalance in research priorities.

Although agile flexibility appeared in 14 studies as a secondary research theme, only 5 studies positioned it as their primary focus. This suggests that most AI/ML research acknowledges the importance of adaptability but does not treat it as a central component of supply chain resilience. Agile flexibility in PSCs is critical for real-time adjustments, dynamic demand forecasting, and AI-assisted decision-making, yet existing studies appear to integrate it only as a supporting element rather than a primary objective. This highlights a potential gap in research that focuses on developing AI-driven supply chain models capable of responding dynamically to shifting demands, disruptions, and logistical challenges.

Similarly, collaborative visibility was primarily explored in only 5 studies, though 12 additional studies incorporated it as a secondary focus. This suggests that real-time tracking, supplier collaboration, and AI-enhanced information-sharing mechanisms are still emerging rather than well-established applications in PSC resilience. While AI-powered supplier networks, decentralized tracking systems, and blockchain integration have been widely adopted in other industries, their role in PSCs remains underdeveloped. This represents a valuable opportunity for future research to explore how AI/ML can enhance supply chain transparency, improve supplier collaboration, and strengthen real-time decision-making frameworks.

A key takeaway from these findings is that most studies address resilience as a fragmented concept rather than an interconnected system. While risk management is widely recognized as critical, it cannot function in isolation. Effective PSC resilience requires a synchronized approach where AI/ML-driven risk modeling is seamlessly integrated with agile adaptability and real-time supply chain coordination. AI/ML applications that focus solely on predictive risk assessment without incorporating real-time adaptability and collaborative decision-making mechanisms may fail to provide comprehensive supply chain resilience solutions. Future research should focus on integrating all three resilience dimensions within AI/ML-based frameworks rather than treating them as separate research areas.

Another emerging insight from this classification is that AI/ML applications in PSCs tend to emphasize predictive modeling rather than real-time adaptability. Many studies focus on AI’s role in forecasting supply chain disruptions and optimizing risk mitigation strategies, yet fewer explore its ability to support real-time decision-making, automated logistics adjustments, and decentralized supply chain coordination. While predictive analytics remains a cornerstone of AI/ML resilience strategies, there is an opportunity to move beyond forecasting toward AI-assisted, self-correcting supply chain models. This shift would enable greater responsiveness, automation, and supply chain agility, ensuring that PCSs can adapt dynamically to evolving challenges rather than simply predicting risks in advance.

As illustrated in Figure 5, the uneven distribution of research focus highlights the need for a more balanced and integrated approach to AI-driven PSC resilience. The overwhelming focus on risk management indicates a missed opportunity to explore AI/ML applications that enhance real-time adaptability and supply chain transparency through agile flexibility and collaborative visibility. Future research should prioritize AI-driven adaptability frameworks, develop automated supply chain coordination models, and explore decentralized AI-powered supplier collaboration mechanisms to ensure a more comprehensive and sustainable approach to PSC resilience.

In conclusion, this analysis reveals that while AI/ML research in PSC resilience has made significant progress, it remains heavily skewed toward risk management applications, with less emphasis on adaptability and real-time visibility. To enhance resilience in a meaningful way, future research should focus on developing more adaptive supply chains using these technologies that integrate real-time responsiveness with risk management and visibility strategies. By moving beyond predictive models and embracing this decision-making, PSCs can transition from risk-prevention strategies to truly resilient, intelligent, and self-optimizing systems capable of responding dynamically to disruptions.

3.3.2. AI/ML Applications in PSC Resilience

The use of AI/ML in PSC resilience has gained increasing attention, offering solutions to enhance forecasting accuracy, inventory management, and risk mitigation. AI/ML-driven models are transforming the way supply chains operate by enabling better decision-making, reducing inefficiencies, and improving resilience against disruptions. To better understand how AI/ML technology is being applied in PSC resilience, this study classified its use into three key functional areas:

• Demand Forecasting: AI/ML-powered models are helping supply chains predict demand fluctuations more accurately by analyzing historical trends, market shifts, and external factors. This allows for better production planning, efficient distribution, and proactive decision-making, ensuring that medicines and vaccines are available where and when they are needed.
• Inventory Optimization: AI/ML applications are also being leveraged to optimize stock levels and automate replenishment strategies, reducing the risks of overstocking or shortages. AI-driven inventory management helps minimize waste, improve efficiency, and ensure a stable supply of pharmaceutical products.
• Resilient Risk Management: One of AI/ML’s strongest contributions to PSC resilience is its ability to predict and mitigate risks before they escalate. Supply chain vulnerabilities and anticipated disruptions can be predicted through AI/ML technology. In addition, proactive responses can be enabled to ensure supply chain stability.

As illustrated in Figure 6, the results show a clear dominance of AI-driven risk management strategies, with 24 studies primarily focused on this area, compared to only 6 studies on demand forecasting and just 2 on inventory optimization. However, 16 studies included demand forecasting as a secondary theme, and 12 studies incorporated inventory optimization as a secondary consideration, suggesting that these areas are often studied in combination with broader risk management frameworks.

The overwhelming focus on Resilient Risk Management reflects a growing industry-wide concern with identifying vulnerabilities, preventing disruptions, and strengthening supply chain security. This finding is consistent with our earlier analysis of resilience dimensions, where risk management was also the dominant theme. Many studies have explored how AI/ML can improve early warning systems, predictive risk assessments, and supply chain contingency planning, demonstrating AI’s potential in creating more proactive, disruption-resistant supply chains. The low number of secondary classifications (only four studies) further suggests that risk management is often studied as a standalone function rather than as part of a holistic AI/ML framework for PSC resilience.

In contrast, demand forecasting was the primary focus in just 6 studies but was included as a secondary theme in 16 studies, indicating that it is frequently integrated into broader AI/ML supply chain models rather than being the main focus of research. Demand forecasting is essential for ensuring stable production, reducing stockouts, and preventing overstocking, yet the low number of primary studies suggests that AI’s full potential in this area has not yet been fully explored. Given AI’s proven success in forecasting for other industries, PSCs could benefit from deeper research into AI-driven demand prediction models tailored to their unique challenges.

Inventory optimization was the least explored AI/ML application, with only 2 studies focusing on it primarily, and 12 studies incorporated this dimension as a secondary theme. This suggests that while AI/ML’s role in stock management is acknowledged, it is rarely the central topic of research in PSC resilience. Effective inventory management is crucial in the pharmaceutical industry, given the high costs, regulatory constraints, and perishability of certain drugs. However, most existing research appears to address inventory optimization within broader frameworks, rather than as a standalone area of innovation. Future research could explore how AI/ML-driven automation, smart warehouses, and predictive restocking models can enhance supply chain efficiency and reduce pharmaceutical waste.

One of the key insights from this classification is that AI/ML applications in PSC resilience are often viewed primarily through a risk management lens, rather than as integral component of a broader strategy for optimizing supply chain operations. While risk mitigation is undeniably critical, demand forecasting and inventory optimization also play fundamental roles in ensuring a stable, responsive, and resilient supply chain. AI/ML’s full potential in PSCs should extend beyond preventing disruptions and be about creating supply chains that can self-adjust, anticipate future trends, and operate with greater efficiency in real time. However, the current research landscape appears to prioritize crisis prevention over long-term supply chain adaptability and operational efficiency, highlighting an opportunity for future studies to bridge this gap.

As Figure 6 illustrates, AI/ML research in PSC resilience is heavily concentrated on risk management, while demand forecasting and inventory optimization remain relatively underexplored as primary research areas. This suggests that there is a significant opportunity to expand research on AI/ML’s role in real-time adaptability, proactive inventory planning, and automated supply chain coordination. Future studies should focus on developing AI-driven forecasting systems that can dynamically adjust to demand shifts, AI-powered inventory optimization models that reduce waste, and integrated AI/ML frameworks that unify risk mitigation, forecasting, and inventory control into a seamless, automated decision-making system.

In conclusion, while AI/ML research has made significant contributions to enhancing PSC resilience, it has largely focused on risk management, with less attention given to predictive demand forecasting and inventory optimization. To fully harness AI’s potential in PSCs, future research should move beyond risk prevention and develop AI-driven supply chain solutions that enable real-time adjustments, predictive forecasting, and automated inventory management. By expanding AI/ML applications beyond risk detection and incorporating demand-driven and inventory-based optimization strategies, PSCs can transition into more resilient, intelligent, and data-driven systems, capable of dynamically adapting to disruptions and evolving global market demands.

3.3.3. Functional Areas of AI/ML in PSC Resilience

The application of AI/ML in PSCs extends across various functional areas, enhancing efficiency, coordination, and adaptability. AI/ML technologies are being leveraged to improve procurement strategies, logistics operations, and overall supply chain planning, ensuring a more resilient and responsive system. For this study, AI/ML applications were categorized into three key functional areas:

• Supply Planning and Procurement: AI/ML is being used to optimize supplier selection, enhance demand forecasting, and streamline procurement strategies. These tools help pharmaceutical companies anticipate resource needs, reduce costs, and maintain a steady supply of essential medicines and materials.
• Logistics and Inventory Management: AI/ML-driven logistics solutions improve transportation efficiency, warehouse operations, and stock management, helping to reduce delays, prevent shortages, and minimize wastage.
• Customer Service: AI/ML has the potential to enhance last-mile logistics, improve patient access to medicines, and ensure timely delivery.

As shown in Figure 7, research is largely concentrated on Supply Planning and Procurement and Logistics and Inventory Management, with 13 and 19 studies, respectively, identifying them as a primary focus. The overlap is evident in secondary classifications as well, where Supply Planning and Procurement appeared as a secondary theme in 19 studies and Logistics and Inventory Management in 13 studies. Interestingly, Customer Service was not identified as a primary or secondary focus in any study, suggesting that AI’s role in enhancing pharmaceutical customer service remains largely unexplored.

The strong emphasis on Logistics and Inventory Management (19 primary and 13 secondary) underscores AI’s role in optimizing transportation networks, automating warehouse processes, and managing stock levels with predictive analytics. Given the pharmaceutical industry’s reliance on efficient and time-sensitive logistics, AI-driven solutions help forecast supply chain disruptions, optimize routing, and improve distribution reliability.

Similarly, Supply Planning and Procurement (13 primary, 19 secondary) reflects a significant focus on AI-driven demand forecasting, sourcing decisions, and procurement automation. In an industry where supply shortages can have life-threatening consequences, AI/ML models play a crucial role in anticipating supply chain risks, improving sourcing strategies, and strengthening supplier relationships. The fact that this category frequently appears as a secondary research theme suggests that supply planning is often integrated with logistics AI/ML applications, reinforcing the interdependence between procurement strategies and distribution efficiency.

The absence of research on Customer Service as either a primary or secondary focus is a notable gap. While AI/ML has transformed customer engagement in industries such as retail and finance, its role in PSCs remains largely untapped. AI-driven tools could be used to enhance patient access to medicines, optimize last-mile delivery, and provide real-time support for healthcare providers and distributors. Future research should explore how AI/ML can improve pharmaceutical distribution networks, enhance patient engagement, and integrate predictive analytics into customer service models.

In conclusion, the current research landscape in AI/ML for PSC resilience is primarily concentrated on logistics, inventory management, and procurement planning, while customer service remains an overlooked area. While optimizing procurement and logistics operations is essential, future research should explore AI’s potential in patient-centered service delivery, last-mile tracking, and healthcare provider engagement. By bridging this gap, PSCs can evolve into more adaptive, efficient, and patient-focused systems, ensuring timely access to life-saving medicines and improved healthcare outcomes.

3.3.4. Study Types in AI/ML Research for PSC Resilience

The way AI/ML research is conducted in PSCs plays a crucial role in shaping how these technologies are applied in real-world scenarios. Different studies take different methodological approaches, ranging from real-world case studies to theoretical frameworks AI/ML and simulations. Understanding research methods clarifies whether AI/ML in PSC resilience is practically implemented or remains largely conceptual. For this study, we classified AI/ML research into three key study types:

• Empirical/Case Study: Research in this category is based on real-world data, experiments, and case studies, offering practical insights into AI/ML applications in PSCs. These studies provide evidence-based validation of AI/ML models, assessing their impact on real supply chain operations.
• Theoretical: Studies in this category develop conceptual models, frameworks, and hypotheses without directly applying them to real-world data. These works contribute to theoretical advancements in AI-driven PSC resilience but may lack empirical validation.
• Simulation: Research in this category employs computational models and scenario testing to evaluate how AI-powered solutions can optimize supply chain processes, mitigate risks, and improve efficiency. Simulations help researchers explore “what-if” scenarios to predict outcomes under various supply chain conditions.

Some studies predominantly focused on one methodology, while others integrated multiple approaches, prioritizing one over the others. As shown in Figure 8, Empirical/Case Study research emerged as the most dominant category, with 14 studies primarily based on real-world data, while 13 others incorporated empirical methods as a secondary approach. Theoretical studies were less common, with 11 studies primarily developing AI/ML frameworks and 4 others using theoretical models as a secondary theme. Simulation-based research was the least frequent as a primary focus, with only 7 studies, but was widely used as a secondary approach in 15 studies, highlighting its role as a supporting method in AI/ML research.

These findings reveal some interesting patterns in how AI/ML is being studied in PSC resilience. The strong presence of empirical research suggests that many AI/ML applications are being tested in real-world PSCs, helping validate their effectiveness and impact. However, the fact that many studies integrate empirical methods as a secondary focus suggests that some AI/ML models are first developed conceptually or tested in simulations before being applied in practical settings. This means that while real-world validation is happening, a significant portion of AI/ML research still remains in the testing phase before it reaches practical implementation.

The relatively high number of theoretical studies shows that AI/ML in PSCs is still in a stage of active framework development, where researchers are working on defining the best models and methods to apply AI/ML in supply chain resilience. Many AI-driven decision-making models, optimization strategies, and predictive risk frameworks are being developed, but not all of them have been tested with real-world data yet. This suggests a need to bridge the gap between theoretical research and practical application, ensuring that AI/ML concepts are validated and refined through empirical testing.

The strong presence of simulation-based research as a secondary focus highlights an important trend: AI/ML models are frequently tested in controlled environments before being applied to real-world supply chains. Simulations allow researchers to model different scenarios, assess AI’s performance in stress conditions, and refine predictive algorithms before deploying them in real supply chain operations. This approach is particularly useful for PSCs, where disruptions can have severe consequences, and AI-driven strategies need to be thoroughly tested before being implemented in high-risk environments.

Looking at the overall distribution, it is clear that AI/ML research in PSC resilience is making progress toward real-world applications, but there is still work to be carried out to ensure that more AI/ML models transition from theoretical development and simulations into real-world implementation. Future research should focus on strengthening the link between AI-driven theoretical models, simulation testing, and empirical validation to create more reliable, data-driven, and adaptable AI/ML applications in PSCs. By combining these research approaches effectively, AI/ML can play a transformative role in making PSCs more resilient, efficient, and adaptive to future disruptions.

3.3.5. Regulatory Challenges in AI-Driven PSC

As AI/ML applications become more integrated into PSCs, regulatory challenges have emerged as a key concern in their real-world adoption. Unlike traditional SCM systems, AI-driven approaches involve data-intensive automation and predictive modeling, raising critical concerns around compliance, privacy, and ethics. If AI/ML technology is to play a lasting role in PSC resilience, its implementation must align with industry regulations, data protection laws, and ethical standards to ensure transparency and accountability. To understand how AI/ML research approaches these regulatory issues, the 32 selected studies were examined across three key areas:

• Alignment and Compliance: AI/ML must operate within the strict regulatory frameworks that govern PSCs, such as GDP, GMP, and global pharmaceutical regulations. For example, AI-driven quality control systems can identify defects in production processes, ensuring compliance with GMP. Similarly, AI-powered real-time tracking systems can improve traceability and transparency in distribution, aligning with GDP requirements. Ensuring compliance with such standards is critical for AI/ML adoption in real-world pharmaceutical operations.
• Privacy Levels: AI/ML systems in PSCs rely heavily on real-time tracking, supplier data sharing, and predictive analytics, making data privacy and security a major issue. PSCs handle sensitive supplier contracts, distribution records, and, in some cases, patient data, raising concerns about data protection, unauthorized access, and regulatory compliance with global privacy laws.
• Ethical Considerations: AI-driven decision-making can raise questions of fairness, accountability, and bias in supply chain operations. As AI/ML systems begin to influence drug distribution, supplier selection, and risk assessment, future research should prioritize ethical AI/ML development, ensuring that AI-driven decision-making remains fair, accountable, and unbiased.

By analyzing how these concerns were addressed in AI/ML research for PSC resilience, a clear gap emerged. As illustrated in Figure 9, the majority of studies (19 out of 32) did not address regulatory concerns at all. This suggests that most AI/ML research in PSC resilience is focused on technical advancements rather than the legal and ethical challenges that could impact real-world adoption.

Among the studies that did consider regulatory issues, privacy emerged as the most frequently discussed concern, with eight studies primarily focused on data protection and five others mentioning it as a secondary theme. This trend makes sense given the increasing digitization of supply chains, where real-time tracking and AI-driven data analytics introduce significant privacy risks.

However, compliance concerns were explored far less often, with only five studies treating it as a primary focus and four others addressing it as a secondary issue. Given that AI/ML must align with strict pharmaceutical regulations before it can be widely implemented, this limited attention to compliance raises concerns about the real-world feasibility of AI-driven PSC systems. AI/ML models that fail to account for industry regulations may struggle with adoption, even if they demonstrate strong technical performance.

The most surprising finding was that none of the studies explicitly examined ethical concerns related to AI/ML in PSC resilience. This is particularly striking given the potential risks of AI-driven decision-making, such as algorithmic bias in supplier selection, unfair drug distribution models, or the lack of accountability in automated processes. As AI/ML continues to shape PSCs, failing to address ethical considerations could lead to unintended negative consequences, particularly in supply chain prioritization and resource allocation.

The fact that 19 studies did not mention regulatory concerns at all suggests that many AI/ML applications in PSCs are being developed without considering the legal, privacy, and ethical challenges they may face in practice. While AI/ML offers significant potential for optimizing supply chains, reducing disruptions, and improving efficiency, its widespread adoption will depend on how well it aligns with existing pharmaceutical regulations and ethical guidelines.

Publication trends in Figure 10 indicate that regulatory concerns are gaining greater attention in PSC research. This increased focus may be driven by the widespread use of AI/ML applications without clear guidelines or policies to protect company data. Therefore, regulatory considerations should be integrated into AI/ML adoption from the outset rather than being treated as an afterthought. Ensuring compliance with pharmaceutical regulations within AI/ML model design is critical for maintaining safety and effectiveness. Moreover, data privacy protections must evolve alongside AI-driven analytics to achieve supply chain transparency without compromising security.

Collaboration among regulators, industry, and academia is essential to develop ethical AI/ML frameworks. Regulators can provide clear adoption guidelines, industry can share practical insights and best practices, and academia can advance research on ethical AI/ML while designing relevant training programs for professionals. Ultimately, ethical AI/ML development must remain a priority to ensure that AI-driven decisions are fair, transparent, and unbiased.

3.4. Common AI/ML Models Applied in PSC Resilience

This section synthesizes AI/ML techniques prevalent in PSC resilience from the 32 reviewed studies, categorized into supervised learning, unsupervised learning, reinforcement learning, deep learning, Natural Language Processing (NLP), and hybrid models (Figure 11).

Supervised learning dominates, employing algorithms such as Random Forest (RF), Support Vector Machines (SVMs), Logistic Regression, Gradient Boosting (XGBoost, LightGBM), and neural networks (Multilayer Perceptron). These models address demand forecasting, disruption prediction, risk classification, and supplier evaluation. RF’s robustness and handling of high-dimensional data are evidenced in studies like Mariappan et al. [27] and Mahdavimanshadi et al. [40]. Gradient Boosting models excel in inventory forecasting and resource optimization [41], while Logistic Regression supports interpretable classification in regulated contexts [4].

Unsupervised learning, notably K-Means Clustering, is applied for scenario reduction and risk factor prioritization [23,34].

Deep learning techniques, including artificial neural networks (ANNs) and LSTM networks, enable the modeling of complex nonlinear patterns for demand forecasting and anomaly detection [25].

Hybrid approaches integrate ML with optimization and fuzzy logic to address uncertainty and multi-objective problems. Examples include fuzzy logic combined with Naïve Bayes classifiers for inventory management under uncertainty [19] and stochastic optimization coupled with ML for cost-resilience trade-offs [21,22]. Metaheuristics (NSGA-II and GWO) facilitate routing and scheduling optimizations [36,37].

Reinforcement learning (RL) and NLP remain largely unexploited due to PSC data characteristics (structured numerical data) and the need for model interpretability in regulatory environments. Their future applicability is anticipated with the increasing digitalization and availability of unstructured data.

Emerging technologies such as blockchain enhance traceability and compliance [20], while Fuzzy Cognitive Maps (FCMs) support decision-making under uncertainty [31].

In sum, supervised learning methods prevail due to predictive accuracy and scalability, unsupervised techniques aid exploratory analysis, and hybrid/optimization models address real-world complexity. RL and NLP, though underutilized, represent promising avenues for future PSC resilience research as digital transformation progresses.

4. Discussions and Research Gaps

This study provides a detailed analysis of how AI/ML is being applied to enhance PSC resilience. While the findings reveal promising advancements in predictive analytics, risk management, and supply chain optimization, they also highlight critical gaps that must be addressed for AI/ML to be effectively integrated into real-world supply chain operations. This section reflects on the key takeaways from the analysis and outlines areas where further research and development are needed.

4.1. Closing the AI/ML Implementation Gap: From Concept to Practical Solutions

A recurring theme in this review was the disconnect between AI/ML technology development and practical implementation in PSCs to enhance resilience. While numerous studies propose advanced AI/ML-driven frameworks, many remain in the theoretical or simulation phase. The predominance of conceptual and computational modeling over empirical case studies suggests that AI/ML’s potential in PSCs is still being explored in controlled environments rather than in active supply chains.

Furthermore, the analysis indicates that the current research has primarily focused on logistics, inventory management, and procurement planning, while customer service and last-mile distribution remain largely unexplored. Given the critical role of timely and efficient medicine distribution, this gap represents a missed opportunity for AI/ML-driven improvements in patient accessibility, healthcare supply chains, and last-mile pharmaceutical logistics.

For AI/ML to deliver tangible improvements in PSC resilience, more research should focus on real-world deployment, pilot programs, and industry collaborations. Pharmaceutical companies, logistics providers, and policymakers need to work together to test, refine, and scale AI/ML applications in operational settings. Without this practical validation, AI/ML remains an untapped resource with limited impact beyond academic research.

In addition, future research should explore how AI/ML can be leveraged to enhance pharmaceutical distribution networks, improve delivery efficiency, and streamline patient access to essential medicines. Solutions such as predictive delivery tracking, automated supply chain adjustments, and patient-centered logistics platforms could significantly enhance the responsiveness and effectiveness of PSCs.

Another prominent trend in the reviewed studies is the strong focus on AI/ML-driven risk management strategies, with comparatively fewer studies addressing supply chain agility and real-time adaptability. While risk mitigation is crucial, true resilience in PSCs requires more than just predicting and preventing disruptions; it also demands the flexibility to rapidly adapt to unexpected challenges as they arise.

AI/ML models should focus on predicting risks and provide adaptive solutions that allow PSCs to adjust dynamically. Future research should prioritize AI-driven agile decision-making frameworks, self-adjusting supply networks, and AI-powered demand elasticity models to ensure that PSCs can react swiftly and effectively to disruptions, rather than merely predicting them.

4.2. Regulatory and Ethical Considerations: A Critical Oversight

One of the most striking findings of this review is that compliance, privacy, and ethical concerns are significantly under-represented in AI/ML research on PSC resilience. Given the strict regulatory environment of the pharmaceutical industry, the lack of discussion on AI/ML’s alignment with industry regulations is concerning.

While some studies address data privacy concerns, very few focus on regulatory compliance frameworks or ethical risks associated with AI/ML-driven decision-making. Ethical concerns, such as bias in supplier selection using AI/ML technologies, fairness in drug distribution algorithms, and transparency in automated decision-making remain critically underexplored.

To ensure the widespread adoption and acceptance of AI/ML in PSCs, future research must address regulatory alignment, data governance, and ethical development. Collaboration with regulatory agencies, compliance bodies, and technology ethics experts is essential to ensure that AI/ML-powered PSC solutions meet legal standards while maintaining transparency and accountability.

AI/ML research in PSC resilience must move beyond purely technical studies and incorporate insights from SCM, regulatory sciences, and healthcare policy. The effective adoption of advanced technologies requires a multidisciplinary approach, integrating expertise from data scientists, pharmaceutical professionals, policymakers, and logistics experts.

Future studies should actively involve industry stakeholders in AI/ML development, ensuring that advanced solutions align with real-world PSC challenges. Without this interdisciplinary collaboration, AI/ML risks remaining a conceptual innovation rather than a transformative force in PSC resilience.

4.3. Managerial Implications for AI/ML Adoption

The findings of this review have several practical implications for PSC managers, policymakers, pharmaceutical companies, and academic researchers:

• PSC Managers: Our review highlights the prevalent use of supervised learning models (e.g., RF, SVMs) for demand forecasting and inventory optimization. Managers should prioritize implementing these proven AI-driven tools to reduce costs, minimize stockouts, and improve service levels. For example, integrating RF-based demand prediction models with existing ERP systems can enable dynamic inventory adjustments aligned with demand variability.
• Policymakers: The review identified a critical gap in addressing regulatory, ethical, and compliance considerations in AI/ML applications. Policymakers should develop sector-specific guidelines that facilitate AI/ML adoption while ensuring adherence to GDP, GMP, and GDPR standards. This includes establishing data-sharing protocols that balance innovation with patient privacy and security concerns.
• Pharmaceutical Companies: Given the limited real-world validation of AI/ML models found in our review, companies should initiate pilot projects that incorporate field testing and stakeholder feedback loops. For instance, piloting AI-driven last-mile delivery optimization in urban areas can demonstrate tangible benefits in delivery time and cost reduction before wider implementation.
• Academic Researchers: The dominance of simulation-based studies indicates a need for more empirical, practice-oriented research. Researchers should design studies that involve direct collaboration with industry partners to test AI/ML models in operational settings. For example, co-developing reinforcement learning models for dynamic transportation routing with logistics providers can generate realistic, implementable solutions while safeguarding data confidentiality through anonymization techniques.

By addressing these implications and fostering targeted collaborations between academia, industry, and policymakers, AI/ML technologies can transition from conceptual models to practical, scalable solutions that enhance the resilience and performance of PSCs.

4.4. Diversity and Gaps in AI/ML Models: A Missing Layer of Maturity

While AI/ML applications in PSCs are steadily expanding, a deeper examination of the models employed reveals a fragmented and uneven landscape. Across the 32 reviewed studies, most applications relied on traditional supervised learning approaches, particularly for demand forecasting and classification tasks. However, more advanced techniques, such as deep learning, unsupervised learning and hybrid frameworks were applied inconsistently, while RL was entirely absent.

NLP, despite its relevance in handling unstructured supply chain data, was notably underutilized, appearing in only one study. This limited adoption of RL and NLP can be attributed to both regulatory restrictions and data structure challenges prevalent in the pharmaceutical sector. RL techniques typically require dynamic, real-time feedback loops to function effectively conditions that are often absent in the highly static and tightly controlled environments of pharmaceutical supply chains. Similarly, NLP thrives on large volumes of unstructured textual data, such as regulatory documents or clinical notes, whereas most PSC datasets remain structured and numerical in nature. Additionally, strict data protection and compliance regulations, such as GDPR and FDA requirements, constrain the deployment of AI techniques that operate as “black boxes” or require access to sensitive data. These factors collectively limit the feasibility of deploying RL and NLP, despite their promising capabilities [44,45,46].

Conversely, hybrid models and optimization-based techniques showed a stronger presence, reflecting a trend toward integrating AI/ML with operations research and simulation. Yet, the lack of standardization in these methods raises concerns about replicability and scalability.

These patterns suggest that while AI/ML is gaining attraction in PSC resilience research, the overall maturity of model deployment remains limited. Future work should emphasize not only diversity in methods but also their practical relevance, validation, and real-world adaptability, particularly in advancing underexplored areas like RL, NLP, and integrated decision-support systems.

To conclude and ensure clear alignment between identified gaps and proposed future directions,

Table 2. Identified gaps and recommendations.

Identified Gap	Corresponding Recommendation
Predominant use of supervised learning models; minimal use of deep learning, RL, and NLP (Section 4.4)	Expand exploration of advanced AI/ML techniques such as RL, NLP, and hybrid frameworks.
Limited real-world deployment; predominance of theoretical and simulation studies (Section 4.1)	Strengthen practical implementation through pilot programs and industry collaborations.
Narrow focus on risk management; underexploration of agile flexibility and collaborative visibility (Section 4.1 and Section 3.3.1)	Develop AI models targeting supply chain agility and collaborative visibility.
Insufficient regulatory and ethical considerations (Section 4.2)	Incorporate regulatory compliance and ethical frameworks in AI/ML research design.

summarizes direct corresponding recommendations. For instance, the limited use of reinforcement learning informs the recommendation to explore advanced AI techniques, while the lack of empirical deployment underpins the call for real-world pilot studies and industry collaborations.

5. Conclusions

PSC resilience is essential to ensuring medical product availability, yet increasing complexity and logistical challenges expose critical vulnerabilities. AI/ML models offer transformative potential for demand forecasting, risk mitigation, and operational efficiency, with a limited focus on areas like last-mile delivery, customer service, and end-user engagement. This review finds their application in PSC resilience remains uneven and often misaligned with real-world practice.

This review aimed to systematically examine existing research on the application of AI/ML in enhancing PSC resilience, with a focus on identifying prevailing trends, underexplored areas, and critical research gaps. The findings demonstrate that although AI/ML models hold significant promise for improving demand forecasting, risk mitigation, and operational efficiency, their current application within PSCs remains uneven and misaligned with practical needs.

Across the 32 reviewed studies, AI/ML techniques improved predictive accuracy and resource planning, with supervised methods like logistic regression, RF, and SVMs widely used for structured tasks. Most studies favored hybrid models, combining AI/ML with optimization, fuzzy logic, or simulation, to address complex, multi-objective challenges in PSC resilience. However, few studies involved real-world implementation; most relied on simulations or case studies, highlighting a gap between theory and practice. Barriers such as data confidentiality, the lack of benchmarks, and limited real-time data access hinder the broader adoption of AI/ML.

The methodological scope was also narrow. Promising approaches like RL and NLP were absent, despite their potential for dynamic planning and unstructured data analysis. Regulatory, ethical, and compliance considerations were rarely addressed, even though PSCs operate under strict standards (e.g., GDP, GMP, GDPR). Issues such as transparency, bias, and accountability in AI/ML decisions remain underexplored. These gaps suggest a need to shift from model development to real-world validation and system integration to fully realize AI/ML’s potential in PSC resilience.

To fully realize the potential of AI/ML in PSC resilience, we propose four strategic directions: (1) bridging the implementation gap through field-tested applications, cross-sector collaboration, and real-time deployment scenarios; (2) expanding the methodological toolkit to include RL, NLP, and ensemble methods that support adaptability and unstructured data analysis; (3) incorporating regulatory and ethical dimensions into AI/ML model design to support regulatory compliance and build public trust; and (4) extending AI/ML applications beyond logistics and planning to underexplored areas like last-mile delivery, customer service, and healthcare access optimization.

By addressing these challenges, AI/ML can evolve from conceptual innovations into practical, scalable, and ethically grounded tools that truly strengthen PSC resilience. This systematic review lays the foundation for future interdisciplinary efforts that combine technical excellence with policy, ethics, and operational relevance. AI/ML not only supports PSC resilience performance but also upholds the integrity, safety, and accessibility of essential healthcare systems worldwide.