Artificial Intelligence in Medical Affairs: A New Paradigm with Novel Opportunities

Abstract

The advent of artificial intelligence (AI) revolutionizes the ways of working in many areas of business and life science. In Medical Affairs (MA) departments of the pharmaceutical industry AI holds great potential for positively influencing the medical mission of identifying and addressing unmet medical needs and care gaps, and fostering solutions that improve the egalitarian and unbiased access of patients to treatments worldwide. Given the essential position of MA in corporate interactions with various healthcare stakeholders, AI offers broad possibilities to support strategic decision-making and to pioneer novel approaches in medical stakeholder interactions. By analyzing data derived from the healthcare environment and by streamlining operations in medical content generation, AI advances data-based prioritization and strategy execution. In this review, we discuss promising AI-based solutions in MA that support the effective use of heterogenous information from observations of the healthcare environment, the enhancement of medical education, and the analysis of real-world data. For a successful implementation of such solutions, specific considerations partly unique to healthcare must be taken care of, for example, transparency, data privacy, healthcare regulations, and in predictive applications, explainability.

Key Points

Implementing artificial intelligence (AI) in Medical Affairs (MA) departments of pharmaceutical companies offers a unique opportunity to beneficially impact their mission of identifying and addressing care gaps and fostering solutions that enhance the egalitarian and unbiased access of patients to treatments worldwide.

Promising AI-supported solutions in MA are focusing on the effective utilization of heterogenous information from observations of the healthcare environment, the enhancement of medical education, and the analysis of real-world data.

To leverage AI's capabilities in data analysis and content generation, MA must address specific ethical, legal, and regulatory considerations, partly unique to healthcare, to ensure a responsible and reliable use of AI-based methods.

Introduction

The critical importance of Medical Affairs (MA) departments within pharmaceutical companies is emphasized by its classification as a third key realm of operations, alongside Research and Development (R&D) and Commercial departments [1]. Applications of artificial intelligence (AI) have rapidly extended to a plethora of professional domains including healthcare and the pharmaceutical industry [2, 3].

Future trajectories of MA must (i) adapt to the rapidly growing integration of AI-based systems and digitalization in clinical and corporate practice, and (ii) implement AI-based innovations of processes and approaches into workflows of MA departments themselves.

There is an inherent affinity of MA to AI, considering the escalating volume of data and the fast-paced progress and transformations in the overall medical domain. A wide range of AI use cases in MA are anticipated, with some already being implemented [4–7]. However, few publications address this fast-paced transformation and its upcoming implications for MA [1, 2, 4, 5, 8–11]. In this article, we describe a practice-oriented perspective on how AI will impact the MA function in several beneficial ways.

A 101 of AI

AI can be defined as the emulation of human intelligence in machines programmed to perform tasks that otherwise require higher-level cognitive abilities. The tasks to be covered include problem solving, learning, planning, natural language processing, as well as image analysis and interpretation [12].

AI encompasses highly diverse but potentially complementing sub-areas (Fig. 1). AI-based systems have the ability to utilize algorithms and computational models to process and interpret data, learn from content and patterns, draw conclusions, derive hypotheses, and assist in decision-making [13].

Fig. 1.

Artificial intelligence (AI): A schematic representation of functional principles.

Adapted from Pettit et al. [20]

Machine learning, a subset of AI, enables systems to increase their performance over time without explicit human programming, leveraging input data patterns and processing feedback (Table 1) [13].

Table 1.

Artificial intelligence (AI)-based technologies: description and examples

Technology	Description	Examples
AI	AI, a subset of computer science, uses a range of techniques such as computational algorithms, machine learning, natural language processing, and neural networks to learn from large datasets and make decisions [81, 82]	AI is used to operate autonomous vehicles [83]
Machine learning (ML)	ML refines algorithms with data, covering supervised, unsupervised, and reinforcement learning in structured data, often involving feature engineering [84]	A model estimating the probability of diabetes mellitus from patient profiles [85]
Deep learning (DL)	Deep learning, a variant of machine learning, employs multiple-layer neural networks for complex data analysis, avoiding extensive manual feature engineering [86]	A deep learning model was used for virtual screening of novel 3C-like protease enzyme inhibitors of SARS-CoV-2 [87]
Artificial neural networks (ANNs)	ANNs, inspired by the network architecture of the human brain, are a subset of ML. They approximate linear and nonlinear functions through a network of connected calculation steps, adjusting model parameters based on data. DL employs such neural networks with multiple layers for complex data analysis, mimicking the intricate processes in the human brain [88, 89]	Within an ANN framework, five models with multiple layers and variable predictive parameters were tested to identify the optimum setup for the prediction of prenatal metabolic bone disease [90]
Natural language processing (NLP)	NLP, a subset of AI, analyzes unstructured data, automatizes question answering, and conducts sentiment analysis. It uses machine learning, neural networks, and text mining [91]	Language processing of patient-initiated electronic health record messages to identify patients with COVID-19 infection for rapid therapeutic response [36]
Large language models (LLMs)	LLMs working with billions of parameters use NLP, machine translation, and question-answering approaches to generate human-like text and content [36]	Medical reports explained by LLMs help to make health information more comprehensible for patients without medical background knowledge [92]
Explainable artificial intelligence (XAI)	Explainable AI (XAI) fosters AI system transparency by providing clear reasons for their behavior, demystifying the decision-making process [93]	XAI that provides physicians with accurate predictions and understandable explanations [94]
Generative AI (GenAI)	The term GenAI refers to a subset of AI systems trained with existing datasets to generate content – such as text, image, music, or video – emulating the style and patterns of the input data. Utilizing advanced AI techniques (such as NLP or DL), generative AI can produce output that closely resembles work and content produced by humans [95]	Utilizing the chatbot ChatGPT in breast cancer management [96]

AI has applications across multiple industries including healthcare. It is already used widely in virtual assistants, image analysis, speech recognition systems, chatbot engines, autonomous devices, robots, and vehicles, as well as in various other domains (Table 1) [14–19].

The use and capacities of AI in multiple healthcare domains is expanding rapidly, leading to increasingly sophisticated, capable, and useful systems [2].

Functions and Requirements of MA in a Rapidly Evolving Environment

Unlike the clearly defined responsibilities of physicians or pharmacists, the term "Medical Affairs" often is poorly understood beyond the boundaries of the pharmaceutical industry [20]. However, MA plays a pivotal role in the overall healthcare sector [21].

The primary missions of MA teams in pharmaceutical companies are (i) to effectively convey and contextualize the clinical value of innovative medicines to stakeholders, and (ii) overcome barriers impeding the use of those medicines to their full potential to improve patients’ lives [22]. The target groups of MA encompass healthcare professionals (HCPs), medical opinion leaders, medical associations, academic institutions, scientific experts, regulatory authorities, health technology assessment agencies, and other players involved in decisions in the overall healthcare field.

Key competencies of MA staff include robust medical knowledge and clinical understanding, particularly in the area of companies’ products, the competitive landscape, and the respective therapeutic areas, as well as comprehensive familiarity with regional healthcare structures and players, and proficient management skills. In many companies, MA staff oversee external scientific and policy-related engagements with heterogeneous stakeholders and ensure various internal collaborations amongst other tasks.

The core roles in MA – Medical Advisors with their primarily strategic and content-generating responsibilities, and Medical Science Liaisons (MSL) with their predominantly external scope and customer-facing activities – involve professionals with a diverse spectrum of academic backgrounds, including medicine, pharmacology, and life sciences [22–25].

Their collective expertise and analytic mindsets enable a range of collaborative engagements serving the common mission of (i) identifying and addressing unmet medical needs and care gaps to drive innovation, and (ii) fostering solutions that improve the egalitarian and unbiased access of patients to state-of-the-art treatments worldwide.

Concepts that Define the Tasks of MA

Farrington et al. describe two basic functional modalities in MA: inside-out and outside-in [22]. We propose a third modality, i.e., internal, that forms the integrating capacity, support structure, and knowledge host for both directional modalities (Table 2).

Table 2.

Modalities of Medical Affairs (MA) tasks with potential benefit from artificial intelligence (AI)-based approaches

Outside-in modalities	AI facilitates and accelerates an enhanced understanding of disease and treatment patterns, supports research updates, and assists in detecting local, regional, or global trends It also helps to identify shortcomings and care gaps in specific therapeutic areas AI-based support of data, facts, and observations (DFO) documentation of medical fields force and real-time mining of publications enhances and accelerates the generation of actionable insights Re-evaluation and adaptation of medical strategies and tactics, planning of studies, and targeted generation of real-world insights are greatly enhanced by AI AI-assisted analyses of patient monitoring data contribute to improved study site recruitment Preparation of summaries and extracts from meeting content matters and congress proceedings are greatly facilitated Drawing insights from AI-based analyses of post-marketing studies and observational data may show opportunities for label extensions and for a broadened use of a drug in the existing indications
Internal modalities	Management and sharing of knowledge are facilitated through AI-based searches, categorizations, and summaries. AI supports real-time access to internal knowledge sources on specific issues and subjects, supporting meaningful healthcare professional (HCP) interactions and accelerated responses to medical inquiries Finding the right person inside the organization at the right time opens up previously unknown or underused opportunities for collaborative efforts AI-assisted learning and staff development initiatives include personalized platforms for onboarding, continuous education, and personal growth
Inside-out modalities	AI can enhance and support target group and patient-centric communication strategies in tailoring summaries of data and studies specifically to HCPs with diverse specializations. Materials directed at patients can be prepared by AI in multiple languages and complexity levels to enhance accessibility and understanding Strategies are implemented to identify optimal modalities for engaging with key opinion leaders (KOLs), considering factors such as fields of interest, locations, and timing to enhance stakeholder engagement. Acceleration and optimization of KOL profiling can be substantially supported by using AI-based models AI approaches facilitate the enhanced use of real-time data to support the performance of MA teams. MSLs are enabled to better engage healthcare providers with data-driven, current information, adding significant value to their interactions e.g., with KOLs

Internal Modalities

The focus of the internal modality is information management and knowledge sharing inside the company – with the intention of harnessing synergies and foster innovative ideas. One of its central tasks is the continuous provision of contextualized up-to-date knowledge on developments and changes in the medical environment. This may stem from the latest insights on changes or ongoing developments in medical practice, from data presented at medical congresses, or from direct interaction with external stakeholders.

The interpretation of novel information from the field and its implications for corporate strategies and activities is a pivotal success factor provided by MA staff. Real-time internal knowledge sharing and its multimodal management is vital in the light of an exponential accumulation of medical data – with doubling times in the range of a few years [26, 27].

Inside-Out Modalities

Regarding inside-out modalities, the focus is on ensuring that appropriate internal data and information reaches the right external target group at the right time. This involves fostering the participation of HCPs in clinical trials or real-world studies, preparing healthcare technology assessment dossiers, and supporting the interactions of commercial functions with national payer entities to build the foundation for patient access to innovative medicines. Furthermore, MA engages in specialty-targeted and disease-centric communication through various channels, disseminating scientific data, raising awareness, and providing research-based insights and medical education of whole therapeutic fields – besides comprehensive product-related information [22].

Outside-In Modalities

On the outside-in interface, the tasks of MA revolve around understanding specific healthcare environments through active listening and targeted engagement with relevant stakeholders. This includes (i) gathering medical and clinical insights to comprehend external needs, and (ii) developing, evaluating, and communicating solutions for these demands. It is important to focus on actionable insights to generate impactful and valuable output. The enhanced understanding informs the development of corporate measures to address unmet medical needs and gaps in patient care.

The pursuit of partnerships with stakeholder groups is another important approach to collaboratively create and communicate sustainable healthcare solutions, fostering patient centricity and integrating input from the community [22].

Benefits and Visions for AI in MA

AI is transforming MA with groundbreaking innovations in evidence generation, the identification of unmet needs in defined patient cohorts and addressing them prioritized based on data-driven impact analyses. It will also accelerate and enhance the access to top-tier medical insights, by supporting documentation and analysis processes. Additionally, AI offers the opportunity to revolutionize external engagement by generating and coordinating tailored medical content across all channels, thereby enhancing decision-making for healthcare providers and patients [4].

In a rapidly changing environment, MA should play a part in evaluating how large volumes of complex data can be used to support clinical decision-making and enhance improvements in patient outcomes. MA should focus on developing comprehensive capabilities of data analysis and interpretation – embracing AI-based automation for scope, efficiency, and analytical potential. AI may foster novel ways of cross-functional collaboration to create and adapt data-driven medical strategies. Overall, these developments indicate a growing role of MA as a pivotal function in translating scientific progress into advancements of medical care and patient benefit.

The currently overwhelming surge of information in medicine is both a challenge and an opportunity. With an unprecedented volume of data and AI-based analytics at their disposal, companies have the capacity to transform the impact of MA into a more measurable asset [11]. All three modalities of MA include areas with the potential of major benefits from the implementation of AI based approaches (Table 2).

The incorporation of AI into MA has introduced a transformative era, revolutionizing key corporate functions. AI tools excel at extracting and deriving insights from diverse data sources, thereby enhancing strategic development by suggesting best practices [6].

In the domain of evidence generation, MA leverages AI to improve hypotheses and study designs supporting the drug lifecycle management or to identify and to fill data gaps in the clinical practice landscape [28–30].

AI optimizes stakeholder engagement via support in planning and selecting tailored measures. It streamlines communication, enabling real-time customization of messages to key stakeholders [31, 32]. In the following, we discuss use cases of AI according to their potential benefits in MA activities and tasks.

Generation and Analysis of Insights

The ability to capture relevant data, information, or facts using different external engagement strategies is commonly used by MA to identify insights with a potential outcome that enhances patients’ disease management. Insights are generated to understand the reasons behind HCP inquiries and anticipate subsequent actions that could close the identified care gaps [33].

Utilizing AI for the processing of medical information data delivers actionable intelligence, clusters inquiries for insights, and facilitates tailored content delivery.

AI can assist in understanding the drivers of trends and patterns. It supports insight-generating processes based on data provided by the current healthcare landscape across a range of internal and external sources [4, 6]. AI is highly effective in the evaluation of large volumes of publications and medical research data [34]. By evaluating bulk data, AI supports MA in capturing key topics of interest for HCPs or identifying gaps in healthcare provision. AI methods can also support in detecting safety signs by identifying patterns and anomalies that may indicate potential risks or adverse events, or can strengthen the evidence of safety profiles [35]. Further important applications include the detection of unmet needs, further strategic opportunities, and risks for the pharmaceutical company [4, 36].

With fully developed AI-based insight generation, the opportunity arises to better understand present requirements and proactively prepare for future needs. This can involve refining processes, devising medical strategies, or accumulating data to inform sound decisions with a focus on improving patient outcomes. Connecting medical insights to planning of activities and a strategic framework is key in this context [6].

Example: Enhancing Insight Generation with Machine Learning Techniques

MUFASA (MUltimodal Fusion Architecture SeArch (for Electronic Health Records); Medical Information Data Uses For AI Semantic Analysis) is a machine-learning tool developed to support MA activities. It leverages AI technologies such as the Sentence Transformer Library, along with clustering, semantic research, and visualization methods, aiming to boost the efficiency and effectiveness of MA intelligence.

It facilitates precisely targeted content delivery to HCPs, streamlining the management and distribution of medical information data. MUFASAs functionalities allow an effective understanding of inquiries, as demonstrated through 3D vector mapping and clustering tests.

MUFASA uses a clustering approach to uncover insights and identify actionable issues from large inquiry data sets. It provides supportive function via semantic search graphs, helping evidence-based decision-making by tracking effectiveness of initiatives and by monitoring of trends. For example, the system saves each MSL team member hours of work per week by efficiently clustering responses to similar inquiries.

Overall, these features enhance strategic decision-making based on a deep analysis of unsolicited data, cultivate actionable insights, and enhance engagement with HCPs [6].

Generation and Analysis of Real-World Evidence

Success in the rapidly evolving healthcare landscape hinges on the effective use of scientific capabilities, especially the ability to integrate, analyze, and interpret diverse datasets. This is a vital requirement to inform interactions with stakeholders and ultimately enhance patient outcomes [37].

The term real-world evidence (RWE) refers to healthcare information gathered from a variety of sources not only derived from clinical research environments. These sources include electronic health records, insurance claims, disease registries, and data collected from personal devices or health apps [38]. RWE offers valuable insights that complement clinical trials, particularly by addressing diverse patient populations and healthcare environments [39]. It contributes to several aspects of healthcare, including the development of novel therapeutics, patient care, and safety surveillance [38]. Recently, a trend is notable towards utilizing AI to process real-world data (RWD), as it facilitates rapid translation of study findings into medical practice and improved alignment of guidelines and their recommendations with the real-life patient diversity [40].

Observational data from cohort and case-control studies provide valuable insights, for example, regarding the clarification of potential safety issues that often require larger sample sizes from post-approval real-life patient cohorts [41]. Advanced techniques, such as AI-based propensity score matching, enhance the validity of non-experimental studies, and significantly broaden the spectrum of study data available to inform treatment decisions [42].

However, a large part of patient data is available as unstructured text in RWD sets, requiring curation for analysis [43, 44]. Vital information on patient characteristics, disease progression, and outcomes is embedded in clinical notes within electronic health records [45]. The conventional methods of hands-on data extraction and curation are resource-intensive and time-consuming, restricting the pool of patient information actually available for RWE generation [46]. Therefore, the use of Natural Language Processing (NLP) extraction techniques to support large-scale RWE generation from electronic healthcare records is growing [47].

NLP techniques are employed to systematically convert unstructured data sources, such as clinical notes, into structured formats suitable for analysis [48]. This workflow encompasses several stages, including data cleaning to remove inconsistencies, tokenization to break down text into manageable units, and the application of advanced methods for text classification to identify and categorize relevant data. Once the data have been standardized and structured, data are subjected to analysis using AI technologies, including Machine Learning (ML) and Deep Learning (DL), which are capable of detecting complex patterns and extracting significant insights [49]. This approach facilitates the generation of robust and actionable insights for clinical research, ultimately informing and improving healthcare decision-making.

By leveraging AI-driven data analysis, MA is able to make more data-driven informed decisions and optimize its workflow to increase the value of medicines in real-life settings. For example, ML can be used to detect factors involved in the divergences of real-life outcomes from clinical trial results [7].

Example: Machine Learning Bridges the Gap Between Non-interventional Studies and Randomized Controlled Trials: The Case of Neovascular Age-Related Macular Degeneration

In real-world settings, the overall effectiveness of intravitreal anti-vascular endothelial growth factor antibodies is usually lower as compared with randomized clinical trials.

Based on machine-learning principles, a clinical decision model was developed based on ranibizumab real-world patient data from the USA validated with data from Australia and the UK. The model learned to identify the most influential factors (out of 59 initial variables) in a manner that they effectively predict the change in visual acuity (VA) over 12 months. These factors were baseline VA, presence of subretinal fluid, and administration of three loading doses by day 90 from treatment initiation.

When applying these criteria, real-world outcomes became similar to those obtained in published randomized controlled trials (RCTs). The example shows that machine learning can be used to classify real-world cohorts and identify subsets of patients whose benefit is equivalent to the results from RCT populations. This methodology may support the translation of findings from clinical trials into clinical practice settings to enhance individual treatment benefits [7].

Medical Education and Content Generation

Introducing innovative medicines demands up-to-date knowledge. The dynamic changes in the knowledge landscape, accelerated by digitalization and AI, intensify this need. Practical examples in the following section showcase areas where AI tools can be employed in MA for effective medical education.

Internal and external medical education is a central component of work within MA. For the transfer of knowledge, which is often based on complex information and data, it is essential to convey the extracted key aspects in an engaging and straightforward manner.

As discussed, NLP gathers information out of unstructured content of various sources such as publications, patent specifications, and healthcare documents [48]. Analyzing these data can unveil associations and ease researcher workload [46].

For example, in the vast realm of COVID-19 research literature, text mining has become indispensable, given the unrelenting surge of publications. Leveraging the COVID-19 Open Research Dataset (CORD-19), text-mining models using NLP facilitate a range of tasks. These include summarization, visualization, extraction, and streamlining of relevant information. These tools enable researchers to effectively derive meaningful insights from the rapidly evolving landscape of COVID-19 literature, overcoming the challenges of information overload [50]. Implementing NLP-based tools in systematic reviews significantly reduces screening time. They improve efficiency without compromising accuracy [51].

Automated text summarization supports the research and medical community by identifying and extracting essential information from large numbers of articles. These tools generate condensed versions of documents, helping users to locate crucial information in the original text more efficiently [31].

Recently, plain language summaries have been implemented in the scientific community to make the content accessible to a broader non-scientific audience [52]. AI-supported generation of lay summaries might enhance trust and transparency into research and clinical study results in a timely and resource-saving process, resulting in more well-informed patients [53].

Automated slide show generators leverage AI to read and interpret text, extracting key points to create slides with customizable designs, diagrams, images, and flow [54].

AI will soon play a crucial role in supporting MA in its mission to improve patient care by addressing care gaps with advanced medical education. By rapidly providing relevant information, structured data, and sophisticated analyses, AI enables faster and more accurate processes, contributing to overall operational effectiveness [55]. Furthermore, generative AI will empower MA to specifically tailor its educational content and strategies to the diverse needs of their audience. Customized content could be created and updated without producing a high workload for MA staff, unlocking valuable capacities for strategic responsibilities and personal customer engagements.

Possible Reasons for Unused Potential of AI in MA

As described in a recent report prepared by the management consulting firm McKinsey & Company, only 10% of life science companies are actively investing in AI solutions for MA [11]. Current approaches aim to enable real-time communication between corporate headquarters and field medical teams or provide real-time access to information for MSLs and KOLs [11].

The challenges in implementing AI-technologies in the field of MA revolve around a few core factors. Analyzing data on social interactions is complex due to the heterogeneous nature of these data, specialized terminology, but also contextual nuances and implicit meanings [56].

Further complexities arise from accurate integration of social data into models [57, 58]. Advanced AI techniques are required to effectively analyze and integrate the heterogenous data from medical records, patient surveys, and customer feedback, while ensuring data privacy and security [59].

Additionally, stringent privacy regulations in healthcare substantially limit the availability of training data for developing specific AI learning capabilities that may be useful for successful implementations of AI in MA. These restrictions could be overcome or at least eased by privacy-preserving techniques such as federated learning and hybrid techniques [60].

However, the increasing digitalization of healthcare data provides more structured and easier accessible information for AI applications [61]. Furthermore, the development of explainable AI helps to build trust in suggestions or even decisions made by AI systems [62]. Further improvements could stem from advances in natural language comprehension, for example, offering enhanced interpretation of proceedings from meetings and spoken content from external communications with stakeholders as input for augmented analytics [63].

Risks and Challenges of AI in MA

Potential risks associated with the application of AI in the pharmaceutical industry include data privacy concerns and security [64], new regulations (see chapter 6.1), and concerns regarding transparency [65].

Large Language Models (LLMs), for example, exhibit hallucinations, where the model generates plausible sounding, but incorrect or invented information presented as facts. Hallucinations arise from issues in data quality, training inconsistencies, and biases such as over-reliance on memorized data, turning them into a challenge [66, 67]. An LLM may also point to false references citing non-existent resources [68]. In medical contexts including MA, validating information derived from LLMs can be time-consuming, posing challenges to ensure the validity of information provided to healthcare professionals and patients. Retrieval-augmented generation (RAG) may significantly enhance the accuracy and relevance of content by combining real-time information retrieval with generative models [69].

The increasing implementation of machine-learning algorithms in healthcare raises concerns about fairness and biases in their development, jeopardizing their utilization for clinical decisions and evaluations [70]. AI methods, often developed outside the medical realm, may face more intense skepticism in medical science. Trust in the robustness of outputs is crucial for the adoption and safe use of AI in the clinical and medical science community, where the conclusions impact clinical decisions and ambiguities drain resources [71]. In medical science and practice, the stakes are high. Trust in the robustness and stability of AI-based analyses, reporting, and recommendations is vital for the medical science community to use AI efficiently and safely [69].

An impactful challenge which comes along with the increasing implementation of AI in healthcare could be an over-reliance on AI. So-called automation biases could adversely impact humans’ autonomy due to a loss of skills resulting in decreased quality in medical decision-making. The risk to preclude further evidence checks after machine-generated output can lead to fatal consequences in patient outcomes [72]. Next to AI regulations from health authorities, risk-mitigating strategies specifically in the pharmaceutical industry such as industry self-governance solutions can support responsible handling with AI [73].

As AI becomes an integral component of corporate decision processes, the necessity for AI explainability has gained importance. Machine-learning and deep-learning models usually operate as black or gray boxes that draw conclusions without explicitly communicating the underlying process and the handling of input parameters [74].

An absence of explainability in decision support systems would counteract fundamental principles of good healthcare practice, potentially leading to adverse outcomes regarding corporate goals and policies [75]. It is crucial to comprehend the factors influencing AI-based decisions, particularly in scenarios where these choices have major implications for corporate strategies [46, 76]. Therefore, the explainability of AI-based approaches and their rationales is a critical success factor for the implementation of systems influencing healthcare decisions.

Regulation of AI in Medical Realms

The rapid deployment of AI technologies in medicine raises ethical concerns such as data privacy, cybersecurity threats, and biases. Further areas of concern include transparency, risk management, external validation, data quality, or in specific cases of direct use in humans, compliance with medical device regulations [77, 78].

The World Health Organization (WHO) has issued comprehensive regulatory considerations for the implementation of AI in healthcare. It underscores the significance of ensuring safety, efficacy, and accessibility of AI systems while advocating productive dialogues among stakeholders [79]. The guidance emphasizes the need for regulating biases and errors in training data through enhanced review and requirements of consent. WHO will provide essential principles for governments and regulatory bodies to develop or adapt guidelines on AI use at national or regional levels. For the European Union (EU) the use of AI will be regulated by the Artificial Intelligence Act that was approved by the EU Council in May 2024 [80].

Conclusion

AI is currently transforming MA with a great potential to increase efficiency, enhance stakeholder engagement, facilitate the generation and analysis of insights and real-world evidence, and leverage its value as a function within the pharmaceutical industry and the healthcare sector. Thus, AI promises a vast potential to support and enhance the MA mission in creating innovations that close care gaps and foster patient access to breakthrough advances in healthcare worldwide. However, it is important to acknowledge that implementing AI in MA presents significant challenges, including the necessity of ensuring the accuracy and reliability of data, as well as complex ethical, legal, and regulatory considerations.

Declarations

Conflict of Interest

All authors are employees at Pfizer Pharma GmbH. The views and opinions expressed in this review article are those of the authors and do not represent or reflect in any way the official policy/position of the company.

Funding

Funding for this review including medical writing was provided by Pfizer Pharma GmbH.

Author Contributions

Emma Fröling, Dina Domrös-Zoungrana, and Christian Lenz were responsible for the idea and structure of this article. Emma Fröling, Neda Rajaeean, and Klara Sonnie Hinrichsmeyer mainly conducted the literature search. The technical part on artificial intelligence and the conception of the graphics were created by Neda Rajaeean. All authors contributed to writing the manuscript, reviewed it carefully, and approved the final submitted manuscript.

Ethics approval

Not applicable.

Consent to participate

Not applicable.

Consent for publication

Not applicable.

Availability of data and material

Not applicable.

Code availability

Not applicable.