Data Science: unpicking data diversity and inclusive research design

Toral Gathani and Brieuc Lehmann write that more precise definitions of inclusive research design and data diversity are needed, while still preserving scientific robustness. They note a lack of diversity in biomedical datasets can undermine fairness, accuracy and applicability of research, contributing to inequities in health outcomes.

Greater attention is being paid to the representativeness of study populations, but confusion exists about how to define diversity and representativeness in research practice. Diversity refers to variation within a study population, while representativeness considers how well the population reflects the target group. A study can be diverse without being fully representative.

Intersectionality must be considered, but dividing data into many subgroups is not always necessary and can reduce statistical power. Study populations must be representative from the start, not just after recruitment. Involving patients and the public throughout the research process is crucial, especially in identifying barriers faced by underrepresented groups.

Reporting study populations in detail, regardless of whether the characteristics are analyzed, helps improve data reuse. Detailed reporting allows for more targeted data sharing and pooling, increasing statistical power and enabling more robust research findings. Balancing inclusivity with scientific rigor is essential to maximize the value of diverse datasets and public investment in research.