It’s widely recognised that a lack of diversity in biomedical datasets can undermine the fairness, accuracy and applicability of research, and contribute to inequities in scientific discovery and health outcomes.

Increasing attention is therefore being paid to the representativeness of populations included in research studies, with funders, researchers and institutions being asked to consider equality, diversity and inclusion throughout research design and delivery.

This increased focus is welcome. However, some confusion has arisen about what diversity and representativeness mean in research practice. Inclusive research can be conflated with a requirement for outcomes to be analysed according to every demographic characteristic collected. In doing so, there is a risk that analyses may be undertaken within underpowered studies leading to spurious results.

Diversity, representativeness, and intersectionality

Diversity and representativeness are often used interchangeably, but they describe different concepts.

Diversity describes the observed variation within a study population, whereas representativeness considers how well that population reflects the people to whom the research findings are intended to apply – i.e. generalisability. A study population may therefore be diverse without being fully representative.

Intersectionality adds yet another dimension to be considered. Participant characteristics such as ethnicity, age, sex, socioeconomic circumstances, disability, and geography are not independent of each other. An intersectional lens is needed to understand how these factors interact to influence health outcomes, access to healthcare, and participation in research. However, recognition of intersectionality should not automatically result in a desire to divide and analyse datasets by increasingly smaller subgroups.

Research studies should be powered to answer a specific research question. Multiple subgroup analyses reduce statistical power – that is to say, it increases the risk of false negative findings. When multiple comparisons are undertaken without appropriate statistical corrections, the possibility of apparently significant findings occurring by chance is also increased.

Both approaches may widen existing inequalities by either masking genuine differences between population groups or suggesting differences where none exist, potentially leading to inappropriate conclusions about who benefits from research and healthcare interventions.

Why public and patient involvement is so important

If study populations are going to be representative, consideration needs to be given as to how to achieve representativeness when studies are designed, and not after recruitment has been completed.

The importance of robust patient and public involvement at all stages of the research lifecycle is often under appreciated. The design stage needs to consider who needs to be represented, which populations may face barriers to participation and whether features of the proposed study design could inadvertently exclude particular groups of interest.