How Does AI Support Real-World Evidence Generation?

By Rasit Dinc

Introduction

In the ever-evolving landscape of healthcare, the generation and utilization of Real-World Evidence (RWE) has emerged as a cornerstone for informed decision-making. RWE, derived from the analysis of Real-World Data (RWD) collected from various sources such as electronic health records (EHRs), insurance claims, and patient registries, provides invaluable insights into the effectiveness, safety, and value of medical interventions in routine clinical practice [1]. The integration of Artificial Intelligence (AI) is further revolutionizing this field, offering unprecedented opportunities to unlock the full potential of RWD and accelerate the generation of robust and reliable evidence.

The Role of AI in RWE Generation

AI, particularly its subfields of machine learning (ML) and natural language processing (NLP), is transforming the way we approach RWE generation. These technologies enable the processing and analysis of vast and complex datasets that were previously inaccessible or too labor-intensive to handle. A systematic review of AI applications in RWE generation presented at ISPOR 2024 highlighted the diverse use of AI methodologies, including ML algorithms, NLP, and deep learning techniques, across a wide range of therapeutic areas [1].

One of the most significant contributions of AI is its ability to extract meaningful information from unstructured data sources. A large portion of clinical information is locked away in free-text clinical notes, radiology reports, and other narrative documents. NLP algorithms can be trained to understand and interpret this human language, allowing for the automated extraction of critical data points such as patient symptoms, disease progression, and treatment outcomes. This not only enhances the richness and completeness of RWD but also significantly reduces the time and effort required for manual data abstraction [1].

Furthermore, ML models can identify complex patterns and relationships within large datasets that may not be apparent through traditional statistical methods. These models can be used to predict patient outcomes, identify patient subpopulations that are most likely to benefit from a particular treatment, and even simulate clinical trials using RWD. The DIA Global Forum has emphasized the potential of ML to enable predictive treatment effect modeling from RWD, which could be particularly beneficial for precision medicine and for underrepresented populations where conducting traditional randomized clinical trials may not be feasible [3].

Navigating the Challenges: Towards Responsible AI

Despite the immense potential of AI in RWE generation, it is crucial to acknowledge and address the associated challenges. The use of AI in healthcare is not without its risks, and a responsible and ethical approach is paramount to ensure the development of safe, effective, and equitable solutions. A multi-stakeholder expert panel, as reported in the Journal of the American Medical Informatics Association (JAMIA), has identified several critical issues that need to be addressed, including data quality and documentation, bias, privacy, and accountability [2].

Real-world data, by its very nature, can be messy, incomplete, and heterogeneous. AI models are only as good as the data they are trained on, and biases present in the data can be amplified by the algorithms, leading to inaccurate and inequitable outcomes. For instance, if a dataset underrepresents a particular demographic group, an AI model trained on that data may not perform well for that group. Therefore, it is essential to implement robust data governance frameworks, including metadata standards and data quality assessments, to ensure that RWD is fit for purpose [2].

Transparency and interpretability are also key considerations. It is often difficult to understand how complex AI models, such as deep learning networks, arrive at their conclusions. This “black box” nature can be a major barrier to adoption in healthcare, where clinicians and regulators need to be able to trust and validate the evidence generated by these systems. The development of “nutrition labels” for AI systems, which would document the data sources, algorithms, and performance of the models, has been proposed as a way to enhance transparency and build trust [2].

Conclusion

The integration of AI into RWE generation holds the promise of a new era in healthcare, one where we can learn from every patient experience and continuously improve the quality and value of care. By leveraging the power of AI to process and analyze vast amounts of real-world data, we can generate evidence more efficiently, answer questions that were previously unanswerable, and ultimately, accelerate the translation of research into practice. However, to realize this vision, we must proceed with caution and a steadfast commitment to responsible AI. By addressing the challenges of data quality, bias, transparency, and accountability, we can ensure that AI-powered RWE generation is not only innovative but also ethical, equitable, and trustworthy.

References

[1] Verma, A., Patel, N., Krishna, A., Pruthi, J., Rai, MK., & Prasanna, R. (2024). Artificial Intelligence in Real-World Evidence Generation: Systematic Review of Applications and Challenges. ISPOR.

[2] Koski, E., Das, A., Hsueh, P. S., Solomonides, A., Joseph, A. L., Srivastava, G., ... & Quintana, Y. (2025). Towards responsible artificial intelligence in healthcare—getting real about real-world data and evidence. Journal of the American Medical Informatics Association, 32(11), 1746-1755.

[3] Michael, E., & Crane, G. (2025). Parallel Progress in Artificial Intelligence and Real-World Evidence Creates a New Vision for Clinical Evidence. Global Forum.