The European Medicines Agency (EMA) is seeking feedback on non-interventional studies (NIS) that use real-world data (RWD) for regulatory purposes.
In the draft reflection paper, EMA defines a NIS as a human study that does not fulfill any of the conditions defining a clinical trial. NIS collect data in normal clinical practice, often to assess safety after a medicine is authorized. The uncontrolled conditions, non-standardized treatments, absence of randomization, and uncertainties regarding data quality and completeness hinder the assessment of efficacy in NIS.
Electronic health data and registries have created new opportunities to use RWD to generate real-world evidence that reflects clinical practice, for example to describe patterns of drug use and validate outcome measures. Because NIS using RWD can generate a large amount of information for regulatory purposes, EMA wants people to understand the limitations of the studies and how to overcome them.
“A critical aspect when assessing the suitability of RWD for a regulatory purpose is the data quality, including data reliability and relevance … and, depending on the research question, the extent to which RWD truly reflects routine clinical practice. In this context, the data quality frameworks discussed in this document should be considered,” EMA wrote.
EMA addresses data quality in one of the four main chapters of the draft reflection paper. The agency said RWD “are often used in NIS without the ability to influence the way they were collected, coded and recorded,” making it essential to “adequately characterize” the dataset. EMA recommends assessing the quality of the RWD and presenting it in the feasibility analysis.
The agency wants researchers to address at least the reliability and relevance of the data in the protocol. Reliability determines whether the data are complete, trustworthy, and credible. Relevance shows whether the size and characteristics of the data and study design are capable of fully answering the research question. EMA recommends using a data quality framework to determine if RWD is suitable.
Other sections of the reflection paper discuss study design, governance and transparency, and statistical analyses. In the section on study design, the longest of the four core chapters, EMA addresses topics such as feasibility assessment, the differences between studies with descriptive and causal objectives, and how to manage bias and confounding.
“The non-experimental nature of NIS may lead to bias that distorts the measure of association due to processes of selection, misclassification and confounding. Confounders may be unknown or inadequately measured,” EMA said. “These sources of bias should be clearly identified at the design stage. They are not easily controlled in the analysis and the design should attempt to minimize their impact on the results.”
EMA is accepting feedback on the draft until 31 August.