Supportive frameworks for generation of real-world evidence from real-world data

The importance of real-world evidence (RWE) derived from different real-world data (RWD) has been increasingly recognised, including its use in regulatory decision-making. Electronic health records, national registers and claims data are commonly utilised data sources, providing excellent opportunities for capturing the routine delivery of healthcare together with medication use and associated outcomes in representative real-life populations. As these data have typically been originally collected for other purposes than research, their application to RWE generation should involve fit-for-purpose assessment and appropriate formulation of study design. We highlight two frameworks that can support these crucial steps on the path from RWD to RWE. One of them, the Data Quality Framework is still evolving and thus not yet readily applicable for quality assessment, while the Target Trial Emulation Framework is already applicable and widely used.

Data Quality Framework

Before conducting the study, it is important to evaluate whether the data quality is appropriate for the planned analyses. At present, the European Medicines Agency (EMA) Data Quality Framework provides general considerations on data quality relevant for regulatory decision making. It defines the data quality dimensions (extensiveness, reliability, coherence, relevance and timeliness) and provides example metrics for each. In future, the framework aims to support data quality assessment procedures and selection of data sources for studies assessing, for example utilisation, safety and benefits of medications in real-world settings. At present, practical recommendations for assessing the quality of real-world data are under development (public consultation ended 28.2.2025).

Target Trial Emulation Framework

Derivation of regulatory-grade pharmacoepidemiological evidence requires appropriate control of biases and confounding in the observational setting where treatment allocation is rarely random. Target trial emulation provides a framework for causal inference from RWD. Briefly, target trial emulation is a two-step process. In the first step, the causal question is specified, and the protocol of a hypothetical randomized trial is written, which would answer that question. The protocol defines key elements such as inclusion and exclusion criteria, treatment strategies, treatment assignment, start and end of follow-up, outcomes, causal contrasts and includes an analysis plan. In the second step, the protocol components are emulated using RWD.

In Real4Reg, we apply this framework in the post-authorisation use case to investigate the risk of adverse outcomes of fluoroquinolone use (use case 3) and cardiovascular benefits of SGLT-2 inhibitors (use case 4) in work package 2 in close collaboration with work package 3. Our inclusion criteria (new users, presence of study participants in the applied data sources for adequate period before treatment initiation) aids in avoiding time-related biases and correct assignment of treatment initiation (so-called time zero) in our study. In randomised controlled trials, treatment is allocated at random, and in observational studies this is mimicked by applying methods that balance the distribution of observed confounders between the comparison groups. We utilise propensity scores, that is, the likelihood of treatment given the distribution of observed confounders, and develop, evaluate and implement different methods for deriving them.

Conclusion

The increasing accumulation of RWD into different data sources, together with advances in computing generates exciting opportunities for research and RWE generation. However, appropriate planning of such studies is of utmost importance. Existing frameworks, such as those highlighted in this post support the planning and thereby conduction of studies with better quality.