Applications of artificial intelligence in clinical research: a narrative review of recent advances and challenges

Introduction

The integration of artificial intelligence (AI) into healthcare has ushered in a transformative era for clinical research and drug development. Clinical trials, while central to evidence-based medicine, have long faced persistent challenges—such as high operational costs, lengthy timelines, low patient enrollment, and high failure rates due to inadequate recruitment or poor protocol adherence (1,2). The deployment of AI technologies offers a promising opportunity to address these limitations by automating labor-intensive tasks such as data collection and initial analysis, streamlining decision-making processes, enhancing patient safety monitoring, and uncovering novel insights from complex, multidimensional datasets (3). While previous reviews have explored AI in specific niches of healthcare or drug development [e.g., on AI in drug discovery (4); with techniques like graph neural networks predicting drug-target interactions (5); on machine learning (ML) for dermatological diagnosis (6)], this review offers a contemporary and broad perspective across the entire clinical trial lifecycle. We specifically focus on advancements from 2020 to 2025, integrating recent developments in generative AI, evolving regulatory landscapes, and the increasing momentum of AI-driven decentralized trials and real-world evidence (RWE) analysis, areas that are rapidly evolving.

Recent advances in deep learning (DL) and predictive analytics have enabled AI systems to identify patterns in historical trial data and electronic health records (EHRs), facilitating improved patient stratification, optimized endpoint selection, and dynamic protocol adjustments (3). These enhancements not only accelerate the execution of trials but also improve their overall robustness and adaptability to real-world clinical environments. Additionally, the application of AI in participant recruitment has demonstrated a tangible ability to increase enrollment rates by better matching patients with trial eligibility criteria, thereby reducing dropouts and enhancing trial efficiency (2,7).

Another area where AI has gained traction is the generation of synthetic medical data. With increasing concerns around data privacy and limited access to large, diverse datasets, generative AI models—such as Generative Adversarial Networks (GANs) and transformer-based architectures—are being used to simulate realistic patient data across multiple modalities, including imaging, genomics, and clinical notes (8,9). These synthetic datasets support model training, feasibility testing, and bias evaluation in a secure and scalable manner. The emergence of large language models (LLMs) and multimodal AI platforms has further expanded the scope of AI in medicine by enabling natural language understanding, integration of multimodal inputs, and decision support across diverse clinical contexts (10).

Despite these promising developments, the integration of AI into clinical trials requires a careful examination of regulatory frameworks to ensure patient safety, model transparency, and compliance with ethical standards. Both the European Medicines Agency (EMA) and the U.S. Food and Drug Administration (FDA) have released guidance documents outlining the lifecycle management, validation, and monitoring requirements for AI-based medical technologies in clinical research (11-13). These evolving regulatory landscapes underscore the importance of continuous oversight and cross-disciplinary collaboration between technologists, clinicians, and policymakers.

This narrative review aims to provide a broad overview of how AI is transforming each phase of the clinical trial process. Drawing on peer-reviewed literature published primarily between 2020 and 2025, as well as regulatory documents, we examine key innovations, challenges, and opportunities associated with AI deployment in clinical research. We also explore current regulatory and ethical considerations, identify gaps in the literature, and offer future directions for responsible and effective AI adoption. We present this article in accordance with the Narrative Review reporting checklist (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-114/rc).

Methods

This article is a narrative review of the applications of AI in clinical research. We conducted a non-systematic search of literature from databases including PubMed, Google Scholar, and official publications and guidance documents from regulatory agencies such as the U.S. FDA and the EMA. Table 1 provides the detailed search strategy used in performing the review. Search terms included combinations of “artificial intelligence”, “machine learning”, “deep learning”, “natural language processing”, “generative AI”, “LLMs”, “clinical trial”, “drug development”, “patient recruitment”, “trial design”, “synthetic data”, “ethics”, “regulation”, “FDA”, and “EMA”. The primary focus was on literature published between 2020 and 2025 to capture recent advances, although seminal earlier works were included for foundational concepts. Articles were selected based on their relevance to AI applications across the clinical trial lifecycle, novel technological advancements, discussions of significant challenges (ethical, regulatory, practical), and future perspectives. Only English-language publications were considered. The aim was to synthesize current knowledge and trends rather than perform a meta-analysis of specific outcomes.

Overview of AI techniques in clinical research

AI techniques are varied and robust, allowing clinical researchers to leverage algorithms for problem-solving across multiple aspects of clinical trials. Key techniques include ML, natural language processing (NLP), DL, federated learning, and generative AI. Table 2 provides a summary of these techniques and their applications.

ML

ML encompasses a broad range of algorithms that enable systems to learn from data without being explicitly programmed, making them highly effective for predictive analytics using structured and unstructured data and fundamentally reshaping many areas of medicine (23). In clinical trials, ML is widely used to improve patient recruitment, identify eligible participants, predict clinical outcomes, and optimize trial protocols (24). These models typically function by learning patterns from historical patient data (including demographics, diagnoses, procedures, lab values, and even unstructured clinical notes processed via NLP) to identify features correlated with trial eligibility criteria or outcomes. For example, an ML model might be trained to recognize complex inclusion/exclusion criteria that are difficult for manual reviewers or simple keyword searches to capture efficiently, thereby accelerating the identification of potentially eligible candidates from large patient repositories (25).

Predictive modeling using ML has been employed to forecast the likelihood of adverse reactions to treatments based on patient data, enabling researchers to intervene preemptively and minimize risk (26). For instance, Support Vector Machines (SVMs) have been employed to predict patient adherence to trial protocols based on demographic and behavioral data, allowing for targeted interventions (27). Similarly, Random Forest models have shown utility in identifying complex biomarker signatures predictive of treatment response, such as in oncology, thereby refining patient stratification methodologies (28). Recent studies demonstrated that an ML algorithm for predicting patient dropout risk in a phase III oncology trial achieved an area under the curve (AUC) of 0.78, allowing for proactive retention strategies (15). Additionally, ML models have been critical in streamlining patient selection for clinical trials, which has historically been a bottleneck for the industry.

NLP

NLP, a specialized field within AI and ML, enables computers to interpret, analyze, and extract meaningful information from human language, such as that found in EHRs, clinical trial reports, scientific literature, and even patient-generated text from social media.

NLP algorithms can scan unstructured text in EHRs to identify patients who are eligible for clinical trials based on detailed medical histories, phenotypes, and disease characteristics. This significantly accelerates the recruitment process and helps ensure that the most relevant candidates are selected for trials (7). Recent studies utilizing Bidirectional Encoder Representations from Transformers (BERT)-based models on EHR data have reduced manual chart review time for identifying eligible patients while maintaining high accuracy in identifying truly eligible patients (16). In oncology, NLP has been used to extract detailed data such as cancer stage, histology, and biomarker status from pathology reports with accuracies exceeding 95%, ensuring accurate categorization for trial inclusion (29). Beyond oncology, NLP has proven valuable in rheumatology for extracting disease activity scores from patient narratives in EHRs for longitudinal studies (30) and in pharmacovigilance for identifying potential adverse drug events from social media posts and medical forums, for instance, using advanced models like BERT (17,31).

DL

A subset of ML, DL utilizes complex artificial neural networks with multiple layers (deep neural networks) to analyze vast and high-dimensional datasets, such as medical images, genomic sequences, and longitudinal patient records, to uncover intricate patterns that may be too complex for traditional ML algorithms or human analysis.

In clinical trials, DL is particularly powerful for image-based diagnostics and the analysis of complex longitudinal data. For instance, Convolutional Neural Networks (CNNs) have been utilized for the automated detection of lung cancer in computed tomography (CT) scans, often achieving performance comparable to experienced radiologists (32,33). DL has also proven effective in analyzing histopathology images for cancer classification and mutation prediction (19). Similar successes have been reported in breast cancer, where CNNs analyze mammograms to detect early-stage tumors with high accuracy, potentially improving screening outcomes (34), and in gastroenterology, where DL models assist in identifying pre-cancerous polyps during colonoscopies with improved detection rates compared to standard colonoscopy alone (35). The groundbreaking DeepMind collaboration with Moorfields Eye Hospital demonstrated a DL system diagnosing diabetic retinopathy from retinal scans with accuracy comparable to human ophthalmologists (18), showcasing its potential in triaging patients for specialist review. For longitudinal data, DL-based sequence models like Recurrent Neural Networks (RNNs) or Transformers have been used to predict disease progression in neurodegenerative conditions like Alzheimer’s disease using time-series magnetic resonance imaging (MRI) data and clinical assessments (36), which is essential for evaluating long-term treatment efficacy in trials.

Federated learning

Federated learning is a decentralized ML approach that enables models to be trained across multiple data sources (e.g., hospitals, research institutions) without the need to transfer raw patient data to a central server, thereby preserving patient privacy and data security. Instead, model updates are shared and aggregated to create a global model. It has been applied to multi-center clinical trials, allowing institutions across different regions or countries to collaborate on developing more robust and generalizable models without directly sharing sensitive patient data (21,37).

Beyond preserving patient privacy, federated learning facilitates the training of more robust and generalizable models by leveraging diverse datasets from multiple institutions without the substantial costs and logistical hurdles of centralizing data. This approach can also accelerate research by enabling collaboration among institutions that may otherwise be restricted by data governance policies [e.g., General Data Protection Regulation (GDPR)] or competitive concerns (38). For example, the MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) consortium successfully used federated learning to train drug discovery models across data from ten pharmaceutical companies (21). Similarly, federated learning has been used to develop predictive models for coronavirus disease 2019 (COVID-19) outcomes by training on data from diverse patient populations across multiple international hospitals (20).

Generative AI and synthetic data

Generative AI, encompassing techniques like GANs, variational autoencoders (VAEs), and more recently, diffusion models, has emerged as a powerful tool for creating synthetic patient data that statistically mimics real-world distributions without exposing individual private information.

Synthetic data is increasingly explored in clinical research to augment limited patient datasets, especially in rare disease trials where patient numbers are insufficient for robust model training or for creating control arms (9,39). For example, GANs can generate synthetic imaging data (e.g., CT scans, MRIs) that can be used to train diagnostic models without requiring real patient images, which helps reduce ethical concerns related to patient privacy and can help balance datasets (22). While promising, the clinical utility and regulatory acceptance of synthetic data hinge on rigorous validation. Recent studies focus on quantitative metrics to assess fidelity (how well synthetic data mimics real data distributions using statistical tests and distributional overlaps) and utility (evaluating the performance of models trained on synthetic vs. real data) (40,41). Regulatory bodies like the FDA are actively developing frameworks for evaluating synthetic data, emphasizing the need for transparency in generation methods, thorough validation protocols, and clear use cases, particularly for pivotal trials (42).

Applications across the clinical trial lifecycle

AI has found applications at every stage of clinical trials, from design and recruitment to monitoring, analysis, and post-market surveillance (Figure 1). These applications not only improve the efficiency of trials but also contribute to their success and the robustness of their findings.

Figure 1 Overview of AI integration across the clinical trial lifecycle. AI supports clinical trials from start to finish. In the design phase, it aids planning with predictive tools. During recruitment, NLP helps match patients efficiently. Monitoring uses wearables and AI to track health and detect issues in real time. Analysis involves predictive models and synthetic data generation. In the post-market phase, AI ensures data security and supports ongoing treatment improvements. AI, artificial intelligence; GAN, generative adversarial network; NLP, natural language processing.

Trial design and feasibility

AI systems can simulate trial protocols using historical datasets and real-world data (RWD) to test their feasibility and optimize the design process. Predictive models assess the likelihood of recruitment success based on protocol complexity and site characteristics, identify potential risks such as high dropout rates, and suggest modifications to trial protocols, potentially reducing costly amendments. For example, AI platforms can analyze past trial data to identify optimal inclusion/exclusion criteria that balance specificity with recruitment feasibility (43). Moreover, digital twins—virtual models that simulate a patient’s health trajectory based on their individual data and known biological pathways—are being explored to personalize trial designs and predict individual responses to interventions (44,45). For instance, companies like Unlearn.AI utilize digital twins to create prognostic scores for individual participants, which can be used to reduce sample size requirements or increase statistical power in randomized controlled trials by augmenting control arms (46). AI-driven models also help predict how treatments will perform in specific patient populations by analyzing omics data, imaging features, and clinical variables, assisting in determining the appropriate sample size and ensuring a more precise analysis.

Patient recruitment and retention

Recruitment remains one of the most significant challenges in clinical trials, with up to 80% of trials experiencing delays due to difficulties in finding suitable participants (1). AI-driven tools can efficiently sift through vast databases of EHRs, patient registries, and even social media to match participants to trials based on complex eligibility criteria, often leveraging NLP to interpret unstructured clinical notes. An AI-based system identifying eligible patients for oncology trials from unstructured electronic health records reported identifying 25.2% more eligible patients than the manual review process (47).

AI chatbots also play an increasing role in improving patient engagement and retention. These tools can engage participants throughout the trial process by providing information, reminding them of appointments or medication schedules, collecting patient-reported outcomes, and addressing common concerns in real-time. A study on an AI chatbot for a diabetes management trial showed a 15% improvement in medication adherence and a 10% reduction in participant dropout compared to a control group receiving standard reminders (48). NLP-based algorithms scanning EHRs for trial eligibility are transforming the recruitment process, making it faster and more precise (49). Retention is particularly important, as patients dropping out can significantly impact trial results and statistical power. By using AI to monitor patient engagement and predict dropout risk, researchers can intervene earlier with targeted support if patients are at risk of discontinuing participation.

Monitoring and data collection

AI technologies, including wearable devices, biosensors, and mobile health apps, enable continuous and real-time data collection, improving patient monitoring outside of traditional clinical settings. This rich stream of data is especially valuable in decentralized clinical trials (DCTs), where patients may not need to visit clinical sites as frequently. For example, remote monitoring of vital signs (heart rate, blood pressure, oxygen saturation, activity levels) and patient-reported symptoms through AI-enabled wearables allows for early identification of potential adverse events or changes in health status, providing real-time alerts to healthcare providers or trial coordinators (50). The Apple Heart Study, which enrolled over 400,000 participants, utilized AI algorithms on Apple Watch data to detect atrial fibrillation, demonstrating the feasibility of large-scale, AI-powered remote monitoring in a trial-like setting (51). More recent studies build on this, using AI to analyze complex patterns from multiple sensors to detect subtle changes indicative of disease progression or treatment response in conditions like Parkinson’s disease (52).

AI systems also enable the aggregation and intelligent integration of diverse data sources, such as medical imaging, laboratory tests, genomic information, and RWD, to create a comprehensive, longitudinal view of each participant’s health. This facilitates more accurate and timely decision-making during the trial process, such as adaptive trial designs where treatment arms can be modified based on interim AI-driven analyses.

Analysis and outcomes prediction

AI plays a pivotal role in the analysis phase by using advanced algorithms to process and interpret vast amounts of complex clinical data. Predictive models based on ML and DL can help forecast individual patient outcomes, such as the likelihood of treatment response, the risk of adverse events, or disease progression, allowing researchers to adapt the trial or develop more personalized treatment strategies.

For example, in oncology, AI models are increasingly used to predict which patients are more likely to benefit from a particular treatment (e.g., immunotherapy, chemotherapy) based on a combination of characteristics, such as specific genetic mutations (e.g., KRAS status in colorectal cancer), transcriptomic signatures, imaging biomarkers (radiomics), demographic factors (e.g., age, performance status), and clinical data (e.g., tumor stage, prior treatment responses) (14,53). These models improve the precision of clinical trials and help identify the most promising treatment pathways for patient subgroups. AI can also identify subtle trends, correlations, and patterns that would otherwise go unnoticed in large, complex datasets. For instance, AI-driven analysis of high-dimensional cytometry data (CyTOF) from an immunotherapy trial identified a novel T-cell subpopulation significantly correlated with positive treatment response, an insight missed by conventional statistical analysis (54). This enables a deeper understanding of treatment mechanisms and can improve the overall success rate of clinical trials.

Post-market surveillance

Once a drug or medical device reaches the market, AI continues to play an essential role in monitoring its long-term safety and effectiveness in broader, more diverse populations. AI systems are used to analyze RWD from EHRs, insurance claims databases, patient-reported outcome platforms, and even social media to track the safety and effectiveness of treatments in routine clinical practice.

RWE generated through these AI tools can help identify rare side effects, drug-drug interactions, and long-term treatment outcomes that may not have been apparent during the more controlled environment of pre-market clinical trials. For example, the FDA’s Sentinel Initiative utilizes AI and ML tools to screen vast databases of insurance claims and EHRs for potential adverse event signals. Sophisticated NLP algorithms can parse clinician notes or patient forum discussions to detect early warnings of unexpected side effects (55). By continuously monitoring a drug’s performance in the real world, AI enhances the ability to identify safety or effectiveness issues early, leading to timely regulatory actions, label updates, or improved clinical guidance, ultimately enhancing patient safety.

Regulatory landscape and ethical considerations

As AI tools become embedded in clinical research, robust oversight and ethical stewardship are essential to ensure these technologies deliver on their promise without compromising safety, fairness, or patient privacy. Regulators worldwide are crafting frameworks to govern AI applications, while researchers must proactively address issues of transparency, bias, and data protection to maintain public trust and scientific integrity.

FDA and EMA guidance

The regulatory landscape surrounding AI in clinical trials is evolving rapidly. The U.S. FDA, through its AI/ML Action Plan and subsequent guidance documents like the “Marketing Submission Recommendations for Predetermined Change Control Plans for Artificial Intelligence/Machine Learning (AI/ML)-Enabled Device Software Functions”, emphasizes the importance of lifecycle management, including Good Machine Learning Practice (GMLP), clear documentation, version tracking, robust validation, and post-market monitoring (13,56). These regulations aim to ensure that AI models used in clinical settings, especially adaptive algorithms or Software as a Medical Device (SaMD), remain safe and effective over time, addressing concerns about model drift and performance degradation as data landscapes or clinical practices change. The FDA also encourages the use of explainable AI (XAI) methods where feasible to enhance transparency by making the decision-making process of AI models more interpretable, particularly for high-risk applications.

Similarly, the EMA, in its “Reflection paper on artificial intelligence in the medicinal product lifecycle” and ongoing AI work plan, emphasizes transparency in AI decision-making, the need for detailed validation procedures, and robust data governance (11,57). EMA’s guidelines also focus on the importance of ensuring that AI models are ethically sound, free from bias, and appropriately validated for diverse populations, particularly when deployed in multinational trials. Both agencies are grappling with how to regulate complex AI systems, including LLMs and generative models, considering challenges in their validation, ensuring unbiased outputs, and managing their potential for emergent behaviors. There is a growing recognition of the need for international harmonization of AI regulations in healthcare, with bodies like the International Medical Device Regulators Forum (IMDRF) working on common principles.

Ethical challenges and bias

One of the most pressing ethical concerns in the use of AI in clinical trials is the potential for algorithmic bias. AI models are trained on data, and if this data reflects existing societal biases or underrepresents certain demographic groups (e.g., based on race, ethnicity, gender, socioeconomic status), the models can perpetuate or even amplify these biases, leading to unfair or inaccurate predictions and exacerbating health disparities (58). For instance, an AI model for patient recruitment trained predominantly on data from one ethnic group may perform poorly or unfairly disadvantage patients from other backgrounds when selecting for trial eligibility. This is particularly critical in clinical trials, where the goal is to generate evidence that is generalizable and equitable.

To address these issues, researchers and regulatory bodies advocate for curating diverse and representative datasets, implementing fairness-aware ML techniques, and conducting rigorous bias audits of AI models before and during their deployment in clinical settings. Furthermore, guidelines on AI explainability [e.g., using techniques like Shapley Additive Explanations (SHAP) or Local Interpretable Model-agnostic Explanations (LIME) to understand feature importance] are crucial to ensuring that AI-driven decisions in clinical trials are understandable, justifiable, and contestable (59,60).

Data privacy and security

With the increasing use of AI in clinical trials, which often involves processing large volumes of sensitive patient data, data privacy and security have become paramount concerns. The use of such data for AI model training, validation, and deployment raises the risk of data breaches, unauthorized access, and misuse, potentially eroding patient trust.

To address these concerns, robust data governance and adherence to regulatory frameworks like the GDPR in the European Union and the Health Insurance Portability and Accountability Act (HIPAA) in the U.S. are essential. These regulations mandate strict controls on how patient data is collected, processed, stored, and shared. Furthermore, privacy-enhancing technologies (PETs) are being explored and implemented. Federated learning, as discussed in section “Federated learning”, allows AI models to be trained collaboratively without centralizing raw data (21,37). Other PETs include differential privacy, homomorphic encryption, and secure multi-party computation, which aim to safeguard individual privacy while still enabling valuable insights to be derived from collective data (61). The responsible stewardship of data is fundamental to the ethical application of AI in clinical research.

Strengths and limitations of AI in clinical trials

As AI technologies become more deeply embedded in clinical research workflows, it is important to recognize both their transformative benefits and the practical challenges they bring.

Strengths

• Acceleration of trial timelines: AI-driven automation (e.g., in data extraction, patient matching) and predictive models (e.g., for site selection, outcome prediction) can significantly shorten various phases of clinical trials (62).
• Enhanced patient targeting and recruitment: sophisticated algorithms leveraging EHR data and NLP can identify eligible patient cohorts more efficiently and accurately than manual methods, improving enrollment rates and diversity if managed carefully (7).
• Improved data quality, integration, and interpretation: AI can manage, integrate, and analyze vast, heterogeneous datasets (genomics, imaging, RWD, wearables), uncovering complex patterns and generating richer insights than traditional methods (63).
• Optimization of trial design and execution: AI can help in designing more efficient protocols, selecting appropriate endpoints, and enabling adaptive trial designs based on real-time data analysis.
• Personalized medicine advancement: AI can identify patient subgroups most likely to benefit from specific interventions or those at higher risk of adverse events, paving the way for more tailored treatments in trials.
• Facilitation of decentralized trials and remote monitoring: AI powers tools for remote data capture, continuous patient monitoring, and real-time event detection, making DCTs more feasible and patient-centric (51).

Limitations

• Overfitting and lack of generalizability: models trained on limited, homogeneous, or site-specific datasets often perform poorly when applied to new, diverse populations or different clinical settings, a major hurdle for widespread adoption (64).
• Data scarcity, quality, and accessibility issues: high-quality, large-scale, diverse, and well-annotated datasets are essential for training robust AI models but are often difficult to obtain, particularly for rare diseases or underrepresented populations. Data silos and interoperability issues further compound this challenge.
• Interpretability and the “Black Box” problem: especially with complex DL models, understanding precisely how decisions are derived remains a significant challenge. This lack of transparency can hinder trust, clinical adoption, and the ability to identify and rectify errors or biases (60). While XAI techniques are emerging, they are not yet universally applicable or fully satisfactory.
• Implementation costs, infrastructure, and expertise: developing, validating, deploying, and maintaining AI systems requires substantial investment in specialized expertise (data scientists, AI engineers), computing resources, and robust data infrastructure, which can be prohibitive for some research institutions.
• Integration with existing clinical workflows: seamlessly integrating AI tools into established clinical research processes and EHR systems can be technically challenging and may face resistance from clinicians accustomed to traditional methods, requiring careful change management and user-centered design.
• Risk of perpetuating or amplifying bias: as discussed in section “Ethical challenges and bias”, if AI models are trained on biased data, they can perpetuate or even worsen existing health disparities. Continuous monitoring and mitigation strategies are crucial.
• Regulatory uncertainty and standardization: the regulatory landscape for AI in medicine is still evolving, creating uncertainty for developers and implementers. Lack of standardized validation methodologies and reporting guidelines further complicates cross-study comparisons and regulatory approval.
• Analysis of underperforming or failed AI implementations: while outright “failures” are less frequently published, analyses of AI projects that did not meet initial expectations highlight critical lessons. For instance, IBM Watson for Oncology faced significant challenges in real-world clinical adoption due to issues with data integration complexities, model generalizability beyond its training sites, and difficulties in aligning with intricate clinical decision-making workflows, underscoring the substantial gap that can exist between algorithmic performance in controlled settings and practical clinical utility in diverse, real-world environments (65,66). Learning from these experiences is vital for future success.

Future perspectives

As the integration of AI continues to reshape the clinical research landscape, the next frontier lies in exploring how these technologies will influence the future trajectory of clinical trials. Innovations in AI are not only streamlining traditional trial processes but are also enabling entirely new paradigms of study design, execution, and data interpretation.

AI-powered decentralized trials

One of the most exciting developments is the increasing adoption of DCTs, significantly facilitated by AI technologies. DCTs allow for patient data to be collected remotely, often via wearable devices, mobile apps, and other digital tools, reducing the burden on participants and potentially increasing trial diversity and reach. AI will play a key role in these trials by enabling real-time, intelligent monitoring of patients, analyzing continuous streams of multimodal data (e.g., physiological signals, patient-reported outcomes, environmental data), and dynamically adjusting interventions or alerts based on individual patient needs and incoming data (67). In DCTs, AI-driven platforms can help manage patient engagement through personalized communication, optimize data collection schedules, and predict and manage adverse events without requiring frequent in-person visits.

AI-powered tools can further enhance DCTs by ensuring greater inclusivity, enabling broader participant diversity across geographic regions and demographics without compromising data quality. Federated learning algorithms will be particularly important, as they allow healthcare institutions to collaborate on research by training models on local data without direct data sharing, which is crucial for large-scale DCTs spanning multiple jurisdictions and for research in conditions with limited patient populations, such as rare diseases (37,68). Additionally, AI algorithms capable of analyzing diverse data types, from genomics to EHRs to wearable sensor data, can create highly personalized treatment regimens for participants in real-time, thus enabling truly individualized treatment approaches within the framework of a DCT (67,69).

RWE and post-market surveillance

Another key area where AI will have a profound future impact is in the generation, analysis, and application of RWE. RWD, gathered from sources like EHRs, claims databases, wearable devices, patient registries, and social media, has the potential to complement traditional clinical trial data by providing critical insights into treatment effectiveness, safety, and value in broader, more heterogeneous patient populations and routine clinical settings. While RWE is currently utilized, future AI applications will focus on more sophisticated analytical methods, such as causal inference from observational data, the automated integration of highly heterogeneous RWD sources (e.g., patient-generated health data, genomics, social determinants of health), and the robust validation of RWE-derived endpoints.

The integration of advanced AI with RWE will transform post-market surveillance, enabling continuous, near real-time monitoring of product safety and effectiveness after approval. By applying sophisticated ML and NLP techniques to analyze large-scale RWD, AI can detect subtle patterns of adverse events or differential treatment effects much earlier and more efficiently than traditional methods (70,71). Future advancements will also likely involve AI-driven generation of ’synthetic’ or ‘hybrid’ control arms using RWE for clinical trials, potentially reducing the need for placebo groups in certain contexts and accelerating trial timelines (72). As regulatory agencies like the FDA and EMA increasingly recognize the value of RWE, AI will be instrumental in establishing robust frameworks for its validation, interpretation, and use in regulatory decision-making across the product lifecycle (73,74).

Ethical considerations and AI governance in the future

As AI continues to shape the landscape of clinical research, future ethical considerations and governance needs will become more complex. Beyond current concerns, future ethical governance will need to address the complexities of increasingly autonomous AI systems in clinical decision-making, such as AI agents that independently adjust trial protocols or treatment dosages. The long-term societal implications of AI in healthcare equity, including the potential for AI to widen or narrow existing disparities depending on its design and deployment, will require ongoing scrutiny (75,76). Developing agile, internationally harmonized regulatory and ethical frameworks that can keep pace with rapid technological advancements in AI will be paramount.

This includes proactive strategies for “ethics by design” and “responsible AI” principles embedded throughout the AI development lifecycle, from initial conception and data collection to model training, deployment, and post-market monitoring in clinical research (77,78). Continuous ethical auditing, dynamic consent models for data use in evolving AI research, and mechanisms for accountability when AI systems err or cause harm will be critical components of future AI governance in this domain. Ensuring public trust will require unprecedented transparency and multi-stakeholder engagement, including patients, clinicians, researchers, developers, and policymakers.

Future challenges and opportunities

While the future of AI in clinical research is promising, several evolving challenges must be addressed for its full potential to be realized. These include:

• Scalability of AI validation: developing efficient and reliable methods for validating increasingly complex and adaptive AI models, particularly those based on LLMs or generative techniques, across diverse settings and populations.
• The digital divide: ensuring that AI-driven innovations in clinical trials do not exacerbate health disparities due to unequal access to digital technologies, AI literacy, or data representation.
• Cybersecurity and data integrity: protecting sensitive patient data and ensuring the integrity of AI models against adversarial attacks or data poisoning as AI systems become more interconnected and critical to trial operations.
• AI literacy and workforce development: equipping clinical researchers, healthcare professionals, and regulatory staff with the necessary AI literacy and skills to develop, evaluate, and responsibly implement AI tools.
• Interoperability and data standards: overcoming persistent challenges in data interoperability and establishing common data models and standards to facilitate seamless data sharing and aggregation for AI development and multi-site trials.

At the same time, AI presents unprecedented opportunities to transform clinical research. As AI models become more sophisticated, they will be able to process increasingly complex and multimodal data, uncover novel biological insights, personalize treatments with greater precision, and predict clinical outcomes with higher accuracy. The convergence of AI with other emerging technologies like quantum computing, advanced omics, and digital twins could unlock entirely new frontiers in medical discovery and evidence generation. Collaboration between AI researchers, clinicians, bioethicists, regulators, and patient advocacy groups will be key to navigating the challenges and harnessing the immense power of AI to accelerate the development of safe and effective therapies.

Conclusions

In summary, AI is rapidly emerging as a transformative force in clinical research, with the potential to revolutionize every phase from protocol design and patient recruitment to data analysis and post-market surveillance. Recent advancements demonstrate significant progress in multimodal data integration, synthetic data generation for privacy preservation and data augmentation, and the development of sophisticated predictive models. AI-driven tools are already enhancing the efficiency, speed, and precision of clinical trials, offering new avenues for personalized medicine and improved patient outcomes.

However, the widespread adoption of AI in this critical domain is not without substantial challenges. Ethical concerns surrounding algorithmic bias, data privacy, and the interpretability of “black box” models demand careful consideration and proactive mitigation strategies. The evolving regulatory landscape requires ongoing adaptation and the development of robust frameworks for AI validation, oversight, and lifecycle management. Future success will depend on fostering responsible AI governance, promoting interdisciplinary collaboration, ensuring data diversity and quality, and committing to continuous innovation and critical evaluation. As AI technologies mature and their integration into clinical trial workflows deepens, they will undoubtedly reshape evidence generation, paving the way for more efficient, inclusive, and impactful medical research that ultimately advances human health.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-114/rc

Peer Review File: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-114/prf

Funding: None.

Conflicts of Interest: Both authors have completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-114/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Eisenstein EL, Collins R, Cracknell BS, et al. Sensible approaches for reducing clinical trial costs. Clin Trials 2008;5:75-84. [Crossref] [PubMed]
2. Lu X, Yang C, Liang L, et al. Artificial intelligence for optimizing recruitment and retention in clinical trials: a scoping review. J Am Med Inform Assoc 2024;31:2749-59. [Crossref] [PubMed]
3. Askin S, Burkhalter D, Calado G, et al. Artificial Intelligence Applied to clinical trials: opportunities and challenges. Health Technol (Berl) 2023;13:203-13. [Crossref] [PubMed]
4. Fleming N. How artificial intelligence is changing drug discovery. Nature 2018;557:S55-7. [Crossref] [PubMed]
5. Nguyen T, Le H, Quinn TP, et al. GraphDTA: predicting drug-target binding affinity with graph neural networks. Bioinformatics 2021;37:1140-7. [Crossref] [PubMed]
6. Esteva A, Kuprel B, Novoa RA, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature 2017;542:115-8. [Crossref] [PubMed]
7. Kantor K, Morzy M. Machine learning and natural language processing in clinical trial eligibility criteria parsing: a scoping review. Drug Discov Today 2024;29:104139. [Crossref] [PubMed]
8. Ibrahim M, Khalil YA, Amirrajab S, et al. Generative AI for synthetic data across multiple medical modalities: A systematic review of recent developments and challenges. Comput Biol Med 2025;189:109834. [Crossref] [PubMed]
9. Chen RJ, Lu MY, Chen TY, et al. Synthetic data in machine learning for medicine and healthcare. Nat Biomed Eng 2021;5:493-7. [Crossref] [PubMed]
10. Acosta JN, Falcone GJ, Rajpurkar P, et al. Multimodal biomedical AI. Nat Med 2022;28:1773-84. [Crossref] [PubMed]
11. European Medicines Agency. Reflection paper on artificial intelligence in the medicinal product lifecycle. 2023. Available online: https://www.ema.europa.eu/system/files/documents/scientific-guideline/reflection-paper-use-artificial-intelligence-ai-medicinal-product-lifecycle-en.pdf
12. Medicines and Healthcare products Regulatory Agency (MHRA). Software and AI as a Medical Device Change Programme Roadmap. 2023. Available online: https://www.gov.uk/government/publications/software-and-ai-as-a-medical-device-change-programme/software-and-ai-as-a-medical-device-change-programme-roadmap
13. U.S. Food and Drug Administration. Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) Action Plan. 2021. Available online: https://www.fda.gov/files/medical%20devices/published/US-FDA-Artificial-Intelligence-and-Machine-Learning-Discussion-Paper.pdf
14. Meti N, Saednia K, Lagree A, et al. Machine Learning Frameworks to Predict Neoadjuvant Chemotherapy Response in Breast Cancer Using Clinical and Pathological Features. JCO Clin Cancer Inform 2021;5:66-80. [Crossref] [PubMed]
15. Bennemann B, Schwartz B, Giesemann J, et al. Predicting patients who will drop out of out-patient psychotherapy using machine learning algorithms. Br J Psychiatry 2022;220:192-201. [Crossref] [PubMed]
16. Idnay B, Dreisbach C, Weng C, et al. A systematic review on natural language processing systems for eligibility prescreening in clinical research. J Am Med Inform Assoc 2021;29:197-206. [Crossref] [PubMed]
17. Dong F, Guo W, Liu J, et al. BERT-based language model for accurate drug adverse event extraction from social media: implementation, evaluation, and contributions to pharmacovigilance practices. Front Public Health 2024;12:1392180. [Crossref] [PubMed]
18. De Fauw J, Ledsam JR, Romera-Paredes B, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med 2018;24:1342-50. [Crossref] [PubMed]
19. Coudray N, Ocampo PS, Sakellaropoulos T, et al. Classification and mutation prediction from non-small cell lung cancer histopathology images using deep learning. Nat Med 2018;24:1559-67. [Crossref] [PubMed]
20. Dayan I, Roth HR, Zhong A, et al. Federated learning for predicting clinical outcomes in patients with COVID-19. Nat Med 2021;27:1735-43. [Crossref] [PubMed]
21. Heyndrickx W, Mervin L, Morawietz T, et al. MELLODDY: Cross-pharma Federated Learning at Unprecedented Scale Unlocks Benefits in QSAR without Compromising Proprietary Information. J Chem Inf Model 2024;64:2331-44. [Crossref] [PubMed]
22. Skandarani Y, Jodoin PM, Lalande A. GANs for Medical Image Synthesis: An Empirical Study. J Imaging 2023;9:69. [Crossref] [PubMed]
23. Rajkomar A, Dean J, Kohane I. Machine Learning in Medicine. N Engl J Med 2019;380:1347-58. [Crossref] [PubMed]
24. Geraci J, Rao P, Grandinetti C, et al. Current opportunities for the integration and use of artificial intelligence and machine learning in clinical trials: Good Clinical Practice perspectives. J Soc Clin Data Manag 2025; [Crossref]
25. Lee K, Mai Y, Liu Z, et al. CriteriaMapper: establishing the automatic identification of clinical trial cohorts from electronic health records by matching normalized eligibility criteria and patient clinical characteristics. Sci Rep 2024;14:25387. [Crossref] [PubMed]
26. Toni E, Ayatollahi H, Abbaszadeh R, et al. Machine Learning Techniques for Predicting Drug-Related Side Effects: A Scoping Review. Pharmaceuticals (Basel) 2024;17:795. [Crossref] [PubMed]
27. Zakeri M, Sansgiry SS, Abughosh SM. Application of machine learning in predicting medication adherence of patients with cardiovascular diseases: a systematic review of the literature. J Med Artif Intell 2022;5:5.
28. Helder I. NZhang X, Jonassen I, et al. Machine Learning Approaches for Biomarker Discovery Using Gene Expression Data. 2021.
29. Al-Kershi L, Altamimi A, Azzeh M. Predicting lung cancer TNM staging from radiological reports using natural language processing and machine learning. Network Modeling Analysis in Health Informatics and Bioinformatics 2025;14:45.
30. England BR, Roul P, Yang Y, et al. Extracting forced vital capacity from the electronic health record through natural language processing in rheumatoid arthritis-associated interstitial lung disease. Pharmacoepidemiol Drug Saf 2024;33:e5744. [Crossref] [PubMed]
31. Lee JY, Lee YS, Kim DH, et al. The Use of Social Media in Detecting Drug Safety-Related New Black Box Warnings, Labeling Changes, or Withdrawals: Scoping Review. JMIR Public Health Surveill 2021;7:e30137. [Crossref] [PubMed]
32. Hoang DT, Dinstag G, Shulman ED, et al. A deep-learning framework to predict cancer treatment response from histopathology images through imputed transcriptomics. Nat Cancer 2024;5:1305-17. [Crossref] [PubMed]
33. Ardila D, Kiraly AP, Bharadwaj S, et al. End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. Nat Med 2019;25:954-61. [Crossref] [PubMed]
34. McKinney SM, Sieniek M, Godbole V, et al. International evaluation of an AI system for breast cancer screening. Nature 2020;577:89-94. [Crossref] [PubMed]
35. Nazarian S, Glover B, Ashrafian H, et al. Diagnostic Accuracy of Artificial Intelligence and Computer-Aided Diagnosis for the Detection and Characterization of Colorectal Polyps: Systematic Review and Meta-analysis. J Med Internet Res 2021;23:e27370. [Crossref] [PubMed]
36. Lee G, Nho K, Kang B, et al. Predicting Alzheimer’s disease progression using multi-modal deep learning approach. Sci Rep 2019;9:1952. [Crossref] [PubMed]
37. Rieke N, Hancox J, Li W, et al. The future of digital health with federated learning. NPJ Digit Med 2020;3:119. [Crossref] [PubMed]
38. Sheller MJ, Edwards B, Reina GA, et al. Federated learning in medicine: facilitating multi-institutional collaborations without sharing patient data. Sci Rep 2020;10:12598. [Crossref] [PubMed]
39. Goncalves A, Ray P, Soper B, et al. Generation and evaluation of synthetic patient data. BMC Med Res Methodol 2020;20:108. [Crossref] [PubMed]
40. Gonzales A, Guruswamy G, Smith SR. Synthetic data in health care: A narrative review. PLOS Digit Health 2023;2:e0000082. [Crossref] [PubMed]
41. Murtaza H, Ahmed M, Khan NF. Synthetic data generation: State of the art in health care domain. Comput Sci Rev 2023;48:100546.
42. U.S. Food and Drug Administration. Using Artificial Intelligence and Machine Learning in the Development of Drug and Biological Products. Discussion Paper. May 2023. Available online: https://www.fda.gov/media/167973/download
43. Sharma P, Pathak L, Doke R, et al. Artificial intelligence in clinical trials: The present scenario and future prospects. In: AI Innovations in Drug Delivery and Pharmaceutical Sciences. 2024. Chapter 10.
44. Katsoulakis E, Wang Q, Wu H, et al. Digital twins for health: a scoping review. NPJ Digit Med 2024;7:77. [Crossref] [PubMed]
45. Laubenbacher R, Niarakis A, Helikar T, et al. Building digital twins of the human immune system: toward a roadmap. NPJ Digit Med 2022;5:64. [Crossref] [PubMed]
46. Unlearn.AI. TwinRCTs™ White Paper. Available online: https://www.unlearn.ai/ai-research
47. Antidote Technologies. AI-Driven Patient Recruitment: A Case Study in Oncology. Available online: https://www.antidote.me/hubfs/docs/case%20studies/Antidote%20Recruitment%20Case%20Studies%20-%20Download%20(1).pdf
48. Chawla R, Jaggi S, Gupta A, et al. Clinical Utility of a Digital Therapeutic Intervention in Indian Patients With Type 2 Diabetes Mellitus: 12-Week Prospective Single-Arm Intervention Study. JMIR Diabetes 2022;7:e41401. [Crossref] [PubMed]
49. Kalankesh LR, Monaghesh E. Utilization of EHRs for clinical trials: a systematic review. BMC Med Res Methodol 2024;24:70. [Crossref] [PubMed]
50. Izmailova ES, Wagner JA, Perakslis ED. Wearable devices in clinical trials: hype and hope. Clin Pharmacol Ther 2018;104:42-52. [Crossref] [PubMed]
51. Perez MV, Mahaffey KW, Hedlin H, et al. Large-Scale Assessment of a Smartwatch to Identify Atrial Fibrillation. N Engl J Med 2019;381:1909-17. [Crossref] [PubMed]
52. Li H, Zecca M, Huang J. Evaluating the Utility of Wearable Sensors for the Early Diagnosis of Parkinson Disease: Systematic Review. J Med Internet Res 2025;27:e69422. [Crossref] [PubMed]
53. Sun R, Limkin EJ, Vakalopoulou M, et al. A radiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study. Lancet Oncol 2018;19:1180-91. [Crossref] [PubMed]
54. Spitzer MH, Gherardini PF, Fragiadakis GK, et al. Systematic discovery of antigen-specific T cell signaling dependencies. Science 2015;349:1259425.
55. Pilipiec P, Liwicki M, Bota A. Using Machine Learning for Pharmacovigilance A Systematic Review. Pharmaceutics 2022;14:266. [Crossref] [PubMed]
56. U.S. Food and Drug Administration. Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence/Machine Learning. Federal Register. 2024. Available online: https://www.federalregister.gov/documents/2024/12/04/2024-28361/marketing-submission-recommendations-for-a-predetermined-change-control-plan-for-artificial
57. European Medicines Agency. HMA-EMA joint Big Data Taskforce Phase II report: ‘Evolving Data-Driven Regulation’. 2019. Available online: https://www.ema.europa.eu/en/documents/other/hma-ema-joint-big-data-taskforce-phase-ii-report-evolving-data-driven-regulation_en.pdf
58. Obermeyer Z, Powers B, Vogeli C, et al. Dissecting racial bias in an algorithm used to manage the health of populations. Science 2019;366:447-53. [Crossref] [PubMed]
59. Nzenwata UJ, Ilori OO, Tai-Ojuolape EO, et al. Explainable AI: A systematic literature review focusing on healthcare. J Comput Sci Appl 2024;12:10-6.
60. Samek W, Montavon G, Vedaldi A, et al. Explainable AI: interpreting, explaining and visualizing deep learning. 1st edition. Springer Cham; 2019.
61. Kaissis GA, Makowski MR, Rückert D, et al. Secure, privacy-preserving and federated machine learning in medical imaging. Nat Mach Intell 2020;2:305-11.
62. Cascini F, Beccia F, Causio FA, et al. Scoping review of the current landscape of AI-based applications in clinical trials. Front Public Health 2022;10:949377. [Crossref] [PubMed]
63. Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med 2019;25:44-56. [Crossref] [PubMed]
64. Vollmer S, Mateen BA, Bohner G, et al. Machine learning and artificial intelligence research for patient benefit: 20 critical questions on transparency, replicability, ethics, and effectiveness. BMJ 2020;368:l6927. [Crossref] [PubMed]
65. Strickland E. IBM Watson, heal thyself: How IBM overpromised and underdelivered on AI in health care. IEEE Spectrum 2019;56:24-31.
66. He J, Baxter SL, Xu J, et al. The practical implementation of artificial intelligence technologies in medicine. Nat Med 2019;25:30-6. [Crossref] [PubMed]
67. Inan OT, Tenaerts P, Prindiville SA, et al. Digitizing clinical trials. NPJ Digit Med 2020;3:101. [Crossref] [PubMed]
68. Xu J, Glicksberg BS, Su C, et al. Federated Learning for Healthcare Informatics. J Healthc Inform Res 2021;5:1-19. [Crossref] [PubMed]
69. Nguyen DC, Ding M, Pathirana PN, et al. Federated learning for smart healthcare: A survey. ACM Computing Surveys 2022;55:1-40.
70. Salas M, Petracek J, Yalamanchili P, et al. The Use of Artificial Intelligence in Pharmacovigilance: A Systematic Review of the Literature. Pharmaceut Med 2022;36:295-306. [Crossref] [PubMed]
71. Zhang L. Causal machine learning for reliable real-world evidence generation in healthcare. [dissertation]. Columbia University; 2023. Available online: https://academiccommons.columbia.edu/doi/10.7916/9mj1-z342
72. Thorlund K, Dron L, Park JJH, et al. Synthetic and external controls in clinical trials: A primer for researchers. Clin Epidemiol 2020;12:457-67. [Crossref] [PubMed]
73. European Medicines Agency. Guideline on registry-based studies. EMA/426390/2021. 2022. Available online: https://www.ema.europa.eu/en/documents/scientific-guideline/guideline-registry-based-studies_en.pdf
74. U.S. Food and Drug Administration. Framework for FDA’s Real-World Evidence Program. 2018. Available online: https://www.fda.gov/media/120060/download
75. Spector-Bagdady K, Rahimzadeh V, Jaffe K, et al. Promoting Ethical Deployment of Artificial Intelligence and Machine Learning in Healthcare. Am J Bioeth 2022;22:4-7. [Crossref] [PubMed]
76. Char DS, Shah NH, Magnus D. Implementing Machine Learning in Health Care - Addressing Ethical Challenges. N Engl J Med 2018;378:981-3. [Crossref] [PubMed]
77. Reddy S, Allan S, Coghlan S, et al. A governance model for the application of AI in health care. J Am Med Inform Assoc 2020;27:491-7. [Crossref] [PubMed]
78. Morley J, Machado CCV, Burr C, et al. The ethics of AI in health care: A mapping review. Soc Sci Med 2020;260:113172. [Crossref] [PubMed]