Clinical trials remain the cornerstone of evidence generation in drug development, but todays environment is more complex, costly, and data intensive than ever. Sponsors need smarter ways to extract more value from every data point and AI is rapidly becoming one of the most effective tools to do it. While AI is often associated with large, global studies, its most transformative impact may actually lie in small and medium sized trials. When applied responsibly and within regulatory frameworks, AI can help sponsors strengthen study design, enhance analytical depth, and build more defensible evidence packages without increasing sample size or cost.
In this joint perspective from Thera Business and 3S Pharma, we explore how AI can help sponsors turn structural constraints into competitive advantage through smarter design, deeper analytics, and fit for purpose governance.
Clinical Trials: The Cornerstone of Evidence-Based Health Research
The first controlled clinical trial in history is often ascribed to James Lind, who, in 1747 aboard HMS Salisbury, divided 12 sailors with symptoms of scurvy into pairs and gave each pair different dietary supplements, ultimately identifying citrus as the most effective treatment.
Clinical trials have evolved immensely since then, now operating under increasingly defined ethical, scientific, and regulatory expectations. Yet their core purpose remains unchanged, generating reliable evidence on whether a new intervention is safe and effective based on the results of testing a defined sample of study participants.
The current clinical trial landscape, however, is facing significant challenges. The number of registered trials has increased dramatically over recent decades, and trials are more complex, costly, and data-intensive than ever before. Artificial intelligence (AI) is increasingly being applied to improve efficiency, data quality, and monitoring, particularly in large, complex studies. However, small and medium-sized trials face a distinct set of structural challenges that require a different strategic lens. It is here that AI may offer some of its most meaningful impact.
Limited Sample Size: Low Statistical Power and Uncertain Conclusions
A fundamental issue for all clinical trials is sample size. An insufficient sample size affects a trial’s ability to detect true effects. This is particularly critical for small trials, which are more susceptible to random variation, lower precision, and underpowered results. These limitations increase the risk of inconclusive or false-negative findings and can make it harder to draw meaningful conclusions.
In rare diseases or niche populations, increasing sample size may be challenging, if not impossible. Rather than simply accepting these limitations, innovative AI-enabled approaches are being explored to support deeper analysis of the limited data that are available.
One promising strategy is the use of AI-enabled synthetic data and synthetic comparator approaches. By using historical clinical trial data, real-world evidence (RWE), and disease registries, AI modeling techniques may help develop external comparator groups drawn from real patients when enrolling a traditional control arm is not feasible. While data validation, source-data quality, and transparency with regulators are essential, these approaches can help generate additional context from limited datasets without necessarily expanding enrollment. It should be noted that fully synthetic (model-generated) patient data are not accepted as confirmatory comparator evidence; their role is in design simulation, method testing, and privacy-preserving data sharing, with external comparator data remaining traceable to real participants in line with ALCOA+ expectations.
Another approach is the use of Bayesian and AI-supported adaptive trial designs. Unlike traditional studies that follow a fixed plan from start to finish, these designs allow researchers to incorporate existing evidence and adjust certain decisions as new data are collected. AI tools can help refine dosing strategies, inform treatment allocation, or support early stopping decisions based on emerging results, provided that these adaptations are prespecified.
Finally, AI-driven analysis of complex data is helping researchers gain deeper insight from each enrolled participant. Advanced machine learning methods can analyze high-dimensional data sources such as imaging, genomics, wearable devices, and digital biomarkers to detect patterns that conventional statistical approaches may overlook. By improving endpoint sensitivity and helping to reduce variability, these tools can make small studies more informative and increase the likelihood of identifying meaningful treatment effects.
In small clinical trials, making the most of each patient’s data is not just beneficial; it is essential.
Recruitment and Retention Barriers: Protecting Study Completion
While statistical power is often the most visible constraint in small clinical trials, operational issues can be just as limiting. Even a well-designed study can struggle if eligible patients are difficult to identify or if participants withdraw before completion.
Recruitment and retention barriers are especially challenging in small or rare disease trials, where eligible patients may be geographically dispersed or where narrow inclusion criteria further reduce the pool of potential participants. Delays in enrollment extend timelines and increase costs, while even a small number of dropouts can meaningfully affect study outcomes.
AI-driven tools are beginning to address some of these challenges. Natural language processing applied to electronic health records can help identify potentially eligible patients more efficiently.Predictive models can also flag participants at higher risk of screening failure or dropout so that retention support (visit reminders, travel assistance, remote visit options) can be targeted where it is most needed. When combined with decentralized trial technologies and digital engagement tools, these approaches can reduce participant burden and help protect study completion rates, an essential safeguard when every enrolled patient can affect the analysis.
Bias, Variability, and Limited Generalizability: Strengthening Confidence in Results
Beyond recruitment challenges, small trials are naturally more vulnerable to bias, variability, and limited generalizability. With fewer participants, minor imbalances in baseline characteristics can have a greater influence on results, and differences across trial sites may introduce additional variability. In addition, findings from narrowly defined populations may not easily translate to broader real-world settings.
Advanced analytics can help mitigate these risks. Machine learning-based covariate adjustment methods can help account for clinically relevant baseline characteristics and participant imbalances, while improving the precision of treatment-effect estimates. AI-powered trial simulations allow researchers to test and refine study design assumptions before enrollment begins. Integration of RWE can further contextualize findings, helping to assess how results may apply beyond the study population and strengthening confidence in the conclusions for regulators and stakeholders.
Governing the Use of AI in Drug Development: Turning Innovation into Trustworthy Evidence
Across each of these applications, from synthetic comparators and adaptive designs to AI-enabled recruitment, retention, and real-world evidence integration, AI should not be treated as a plug-in solution. Its value in small clinical trials depends on being used deliberately, transparently, and within a fit-for-purpose governance framework.
The joint EMA–FDA Guiding Principles of Good AI Practice in Drug Development and ICH E6(R3) Good Clinical Practice point to the same central expectation: a risk-based, human-centric, and GxP-aligned approach to AI used for regulatory scope. In practical terms, this means that each AI application should have a clearly defined context of use. Sponsors should be able to explain what the tool is being used for, what decision it informs, what it does not do, and how human oversight is maintained.
For small trials, this is particularly important because each data point carries greater weight. AI-supported synthetic comparators, for example, require a predefined context of use, fit-for-purpose source data, documented provenance, transparent model development, and risk-based performance assessment. Adaptive designs and AI-supported decision rules should be prespecified in the protocol and statistical analysis plan, and trial blinding must be protected. Any decision on dosing, allocation, or stopping a trial based on unblinded interim data must be taken through an independent Data Monitoring Committee rather than by the sponsor study team. Where AI methods are used to analyze imaging and genomics, endpoints and digital biomarkers must be validated and prespecified in the protocol before unblinding. If AI is used to support recruitment or retention, the tool should be validated for the intended population, assessed for bias and equitable access. Its use must be explained clearly to participants and investigators, and be approved by the ethics committee/Institutional Review Board and regulators. Where an AI tool processes participant-level data, this must be described in the informed consent materials and conducted under a lawful basis for processing personal health data. AI models used to flag participants at higher risk of screening failure or dropout must never be used to decide who is enrolled: eligibility is determined solely by the protocol criteria.
The same principles apply to analytical methods that aim to reduce bias, variability, or uncertainty. Machine learning-based covariate adjustment, endpoint refinement, trial simulation, and RWE integration should be prespecified in the protocol and statistical analysis plan, interpretable enough to support scientific review, robust enough for the intended use, and documented in a way that allows regulators and stakeholders to understand the underlying data, assumptions, limitations, and performance. Data integrity expectations, including traceability, auditability, and ALCOA+ principles, remain central.
AI governance also extends beyond initial validation. Models should be monitored across their lifecycle for performance, data drift, and changes in the clinical or operational context. Computerized systems should be validated, version-controlled, access-controlled, and supported by audit trails. Sponsors should maintain oversight of vendors and technology partners, particularly where AI tools rely on external datasets, cloud infrastructure, or third-party algorithms.
Finally, responsible AI use requires strong data protection and transparency. Patient-level data, real-world data, and digital biomarker data must be handled under appropriate GDPR, HIPAA, and other applicable privacy and cybersecurity safeguards. Information provided to investigators, ethics committees, participants, and regulators should be clear, proportionate, and accessible, including the purpose of the AI tool, its limitations, and the role of human judgment.
Used in this way, governance does not slow innovation; it enables it. For small clinical trials, where confidence must often be built from limited evidence, responsible AI governance is what turns promising analytical tools into credible, reviewable, and trustworthy support for decision-making.
From Constraint to Competitive Advantage
Taken together, these innovations reflect a broader shift in clinical research: moving from simply attempting to increase sample size toward maximizing the value of every data point collected. In small clinical trials, AI does not replace methodological rigor. Instead, when used within a clear governance framework, it can augment good trial design, strengthen data quality, improve efficiency, reduce bias, and increase the reliability of conclusions drawn from limited but critically important patient populations.
This is where AI can become strategically powerful for small and medium-sized trials. Synthetic data, adaptive designs, advanced analytics, AI-enabled recruitment, and real-world evidence integration are not shortcuts around scientific standards. They are tools that, when appropriately designed, validated, monitored, and explained, can help sponsors generate more meaningful evidence from constrained settings.
Small trials may be structurally constrained, but with the right design, analytical strategy, and governance approach, their impact does not have to be. Responsible AI use can help turn small size from a limitation into a strategic advantage.
Global Insight. Diverse Expertise. Regulatory-Grade Evidence.
Thera Business and 3S Pharma support sponsors in designing, validating, and governing AI enabled clinical trial strategies that strengthen evidence, reduce risk, and accelerate decision-making. If you’re planning a small or early phase study, our team can help strengthen your evidence package through responsible, high-impact AI use.
Ready to move forward?