arnessing the parallel powers of real-world evidence (RWE) and artificial intelligence (AI) will be particularly beneficial for the development of medicines for underrepresented and underserved populations, such as women of childbearing age (WoCBA) and vulnerable people—especially children and elderly—with rare diseases. More broadly, the use and application of fit-for-purpose RWE and AI will expedite the clinical development process in all disease areas and deliver value to all patients and well-being of society in general. This was the prevailing view at the DIA Europe 2025 Data Science, Real-World Evidence and Artificial Intelligence track.

Regulators’ Vision: Clinical Evidence Generation 2030 and the Birth of a New Data Source

In February 2025, an expert team of 14 regulatory authors plus an author from academia collaboratively published a new vision for Clinical Evidence 2030, which included six key principles (see sidebox for complete list).

Principle 1 from Clinical Evidence 2030 calls for patients to be at the center and guide every step of evidence generation. Three areas stand out in this regard:

1. Rare disease patient perspective
   The use and acceptance of RWE in healthcare decision-making is relevant to all medicine development but has particular resonance with regard to treatments for rare diseases. For example, small patient numbers mean that conducting a typical randomized clinical trial might not be feasible, so clinical development that is augmented with RWE is likely to be of particular relevance in rare diseases. Patient groups such as EURORDIS (Rare Diseases Europe) have emphasized the needs:
   • To call for increased representation in clinical studies, which could be facilitated through better designed trials, including the inclusion of pragmatic elements from and deeper consultation with patients.
   • To red-flag the need for the continued inclusion and voting rights of patients in regulatory decision-making at the highest level (e.g., at CHMP) within the European regulatory network.
2. Randomized pragmatic clinical trials: Linking clinical research to clinical practice
   Pragmatic clinical trials can generate high-quality RWE for regulatory decision-making and lead to more patient-centric trials by broadening the patient population in alignment with clinical practice and increasing external validation of results; introducing more patient-relevant primary outcomes; reducing visits and follow-up; and increasing participation in trials through community engagement, with better access to and leveraging of real-world data sources.

   The PRagmatic Explanatory Continuum Indicator Summary PRECIS-2 is one tool that helps trialists make design decisions consistent with their trial’s intended purpose.

   One persuasive argument to adopt pragmatic approaches to clinical trials is the unique value they bring during the initial approval, when there is generally limited data available about the safety and efficacy of a medicine in the broader population. Well-designed randomized pragmatic clinical trials bridge that gap and ensure generation of data about the impact of a medicine in the real world at the earliest opportunity.

3. Including women of childbearing age from early to late development
   Another emerging call to change the current drug development dynamic is to include WoCBA early and throughout the value chain. The current unmet medical need: 90% of women take medications during pregnancy and breastfeeding, and 1 in 3 choose to discontinue treatment because of safety concerns, with potentially serious consequences to their health. There is a dearth of data for this patient population; it takes approximately 10-15 years after marketing approval for reliable information for this population to be included in the product information (PI) documents. Ideally, reliable information in the PI would be available within five years of initial marketing authorization.

   This could start at the pre-clinical stage with the development of better in silico models of women, thus derisking pharmacokinetic (PK) studies so that women could be included in pre-licensure studies. If the PK data on use of a medicine during pregnancy could then be included in the prescribing information (PI) at the time of initial approval, it would create a virtuous cycle enabling the generation of RWE in WoCBA as early as possible after the launch of the product.

   Artificial intelligence could help innovate nonclinical data generation, optimize in silico experiments, and integrate data from various sources to obtain reliable evidence of maternal and fetal outcomes and (potentially) link them.

Principle 4 from Clinical Evidence 2030 calls for the inclusion and embracing the full spectrum of data and methods during evidence generation, including the quickly emerging potential of machine learning (ML).

Machine Learning Can Potentially Enable Predictive Treatment Effect Modeling from RWD

Even though policies have been developed which encourage the use of retrospective studies and real-world evidence—for example, US FDA’s RWE projects under the auspices of the 21st Century Cures Act—challenges such as the high standard for demonstrating efficacy and validating biomarkers and diagnostic tests remain for precision medicine. Precision medicine requires the ability to describe the size of the biomarker population, to establish natural history of the disease stratified per biomarker status, and to demonstrate effectiveness of standard of care (SoC) in different populations.

The graphic below illustrates how ML and RWD were used to establish a model to predict drug response in ulcerative colitis. Deep learning, knowledge graphs, and RWD were used to identify patient-level drug response. The model was also capable of predicting efficacy in different subpopulations.

Addressing the challenges of trust, verification, and governance, while leveraging opportunities for efficiency and innovation, is essential in using AI and ML in model-informed drug development and precision medicine. The added value of integrating AI/ML into drug-disease modeling and discussed emerging opportunities will continue to grow, fueled by the increasing availability of high-dimensional and multimodal data, novel biomarkers, and digital real-world data. Regulatory frameworks governing the use of AI in the medicines lifecycle should be tailored, fit-for-purpose, risk-based, non-duplicative, and globally aligned, to foster innovation and support development of safe and efficacious medicines.

Emerging technologies such as wearables, genomics, AI, and ML keep the RWD and RWE landscapes evolving rapidly; if we are to ensure that these technologies are used responsibly, it’s critical we take both a pragmatic and agile approach to harmonization and convergence to ensure it is future-proofed in anticipation of potential new technological advances we will all be facing. It is also likely that regulators in different countries have different appetites for risk particularly when it comes to assessing innovation, which may lead to inconsistent assessments. However, as long as the level of risk taking is transparent, not only to sponsors but to the public, this will help to foster international trust and learnings around adoption of RWD and RWE.

The future is likely to see the integration of these emerging technology programs and RWE alongside more traditional clinical trial data and thereby provide more robust data to facilitate better healthcare decision-making by regulators and reimbursement bodies.

Principle 5 from Clinical Evidence 2030 highlights that clinical evidence generation should be planned earlier and collaboratively across healthcare stakeholders, allowing them to fully leverage the totality of evidence.

Real-world evidence in healthcare decision-making: pre- and post-approval
In general, the contribution of RWE to safety monitoring and disease epidemiology is well accepted, while using RWE to demonstrate efficacy requires further discussion. RWE can play a supportive role in evaluating the benefits and risks of medicines to support regulatory decision-making. An example of where the evidence was subsequently accepted as supportive was in the initial regulatory approval of abaloparatide, which was based on evidence from clinical trials augmented with RWE.

To facilitate the acceptance and use of these data, companies should interact early with regulators when considering inclusion of RWE in submissions, and the regulatory environment would benefit from further guidance on the regulators’ expectations of RWE.

Harmonization and convergence to enable global dossiers
Aligning regulatory definitions such as “RWD” and “RWE” should not unduly delay progress on other issues, such as defining data quality frameworks, which have a practical impact on sponsors’ ability to generate RWE that meets regulatory standards.

For example, the understanding of what certain words mean in the current discussion on data quality can differ: When we talk about “data completeness,” are we addressing how granular the data is or how much data is missing? It is important to reach agreement so that sponsors understand what regulators mean when they inform sponsors that there is a lack of completeness in their data.

Convergence is foundational. Regulators can work toward harmonization while acknowledging that they might not achieve uniformity in all areas, especially in the early stages of collaboration. In this context, industry representatives very much welcomed the announcement that the ICH management board has accepted a proposal to develop guidelines relating to RWE and asked how this work can be accelerated.

Convergence is also the right path forward to finding common ground on such foundational issues as scientific rigor, data quality, and interpretation of evidence. The key focus of harmonization and convergence should be on aligning core concepts and principles while accommodating national and regional nuances, while recognizing that uniformity may not always be feasible. There is a need to move beyond definitions and keep moving forward with practical solutions in terms of how RWE is used in national, as well as international, healthcare decision-making.

MHRA in the UK has shared its vision for RWD and RWE playing an integral role in their regulatory decision-making, and noted that the need for collaboration on this topic was factored in during the development of their guidance.

RWD and RWE are already contributing to regulatory decision-making in Germany, where health data labs (HDL) have been created and will soon be accessible to researchers, including industry. The main focus of HDL is to provide insurance claims and electronic health records (EHR) data for secondary use to optimize the research process, enable RWE research, and support regulatory and advisory activities. It emphasizes the importance of deep knowledge about data-generating processes and the requirements of researchers. Alongside this effort in Germany, other ongoing pilots are helping to pave the way, as all EU countries prepare for implementation of the European Health Data Space (EHDS).

EMA’s workplan for enabling the safe and responsible use of AI is shown below.

Lessons Learned from ICH M14

One recent industry exercise assessed and contrasted EMA, FDA, and ICH M14 guidelines (“General Principles on Plan, Design and Analysis of Pharmacoepidemiological Studies That Utilize Real-World Data for Safety Assessment of Medicines”) in terms of data accuracy, completeness, reliability, representativeness, and other relevant criteria, and generated a call to address the impact of inconsistent data quality frameworks and key considerations for harmonizing them. A complementary presentation identified use of RWE for efficacy and effectiveness in public assessment reports from regulatory and HTA bodies (86 submissions across 11 authorities) to identify research methods that most closely correlated with RWE acceptability.

One member of the ICH M14 working group has noted that the original authors heavily based their initial work on North American and European guidelines and were unintentionally excluding other regions; more countries were involved as the process moved forward. Flexibility and focus are essential: flexible enough to incorporate diverse requirements across different jurisdictions while developing guidance; and focused on the overall framework and principles, but not on too many details, such as specific study designs. Pairing up reviewers from different agencies in subgroups has proven successful in adjudicating comments to this guideline.

Conclusions

These common themes are emerging across Europe in the use of RWE and AI in clinical research and drug development:

• Enabling the use and acceptance of both AI and RWD/E requires high-quality data, transparency, and traceability, which in turn will lead to improved predictability and trust in the data.
• Major advances in AI and RWE alongside increasingly adaptive evidence-generation pathways will be particularly beneficial to underrepresented populations, such as WoCBA, and will allow them to participate earlier in the development cycle of new medicines.
• As AI and RWE become more embedded in drug development, regulators and drug developers continue to upskill their workforce, while it is expected that the content and structure of dossiers will remain the same (e.g., aligning with the CTD structure). Progress toward globally aligned regulatory frameworks for RWE and AI is very much welcomed by all stakeholders.