We audited all single-site, NIHR-supported and Greater Manchester-CRN-managed studies that had closed to recruitment between 2016 and 2021, describing the key characteristics (ethnicity, age, sex, and social class/smoking status) of recruited subjects. We report that 6% of samples included participants from ethnic minority backgrounds compared to the proportion of ethnic minorities in the Greater Manchester population (16%). This underrepresentation was replicated across clinical speciality and when studies which specifically selected on participant characteristics were excluded. Overall, study samples provided a good representation of the Greater Manchester population in terms of sex. After removing single sex studies, 50% of study participants were women. As we anticipated, the participant population was older than the mean Greater Manchester population. We also report that most of the participant recruitment took place in centrally located Greater Manchester NHS Trust sites where centres of clinical excellence and research tend to be located.

Other contemporary reviews of representation in research also reported that women were adequately represented, but other groups were underrepresented. In a review of 213 Pfizer randomised controlled trials in the USA, participation was at or higher than US census levels for women and for Black or African American groups, but Hispanic or Latino, Asian, Native Hawaiian and Pacific Islander, American Indian, Alaska Native participants were underrepresented [17]. A recent evaluation of a large US biomedical research programme “All of Us” reported that, compared to US and state referents, non-Hispanic Black or African American groups were overrepresented whilst Hispanic and Latino, non-Hispanic Asian and multiracial groups were underrepresented. “All of Us” aims to recruit inclusively and equitably, to that end the authors insisted more should be done to over-recruit in order to redress the historic underrepresentation of certain groups in US health research[18]. In a study of the representativeness of 11 perinatal mental health studies conducted during the COVID pandemic, Black, Indigenous and other races and ethnicities were underrepresented despite the known racial and ethnic health disparities in pregnancy. Compared to our audit, there was detailed reporting of race and ethnicity in these studies. UK research may have a particular problem. Authors of a recent systematic review of the representativeness of 30 UK COVID vaccine trials reported limited and opaque reporting of participant ethnicity, similar to that found here, meaning it was difficult to be certain to what extent Asian, Black, Mixed and other ethnic minority groups were underrepresented in these trials [19].

There are several approaches to exploring the representativeness of patient recruitment. Some studies focus on a specific disease of interest and compare study participants with the characteristics of the general population with the same disease of interest. For example, a recent systematic review of 224 perioperative medicine trials reported that age exclusions and sampling bias meant the studied population were younger than the clinical population [20]. This approach addresses representativeness within a clinical speciality where age differences are easier to interpret. The approach we have taken here is complementary. Comparing the characteristics of patients recruited to research within the wider Greater Manchester population is complex when characteristics are clearly health related — such as age. However, there are characteristics where comparison is easier to interpret. For example, although there are some examples where recruitment of specific ethnic groups is indicated (e.g. sickle cell disease), this is not generally true. Our finding, that rates of ethnic minority participation are lower than the general population average, is of note. Although there are challenges in interpretation, our approach allows an evaluation of the totality of health research being conducted within a specific region, rather than a specific (and more limited) subset of research activity. Both are likely to be important to patients, those who commission health care, policy makers and funders.

Conducting this audit in Greater Manchester is a strength because, outside of Birmingham, this is the most diverse region in England. Access to the well-maintained CRN research dashboard improved the quality and content of the audit, particularly the ability to map recruitment. However, significant limitations remain. We were unable to obtain participant information for 53 out of 145 studies, accounting for ~ 4000 participants; therefore, it is possible there were studies with more or less diverse populations and their exclusion might change our overall findings. Ethnicity and social class measures (deprivation and smoking status) were less well reported and we had insufficient data to report high-level ethnic minority categories with confidence or index of deprivation at the patient level. Using recruitment site rather than participant address, means the data are not truly representative of where patients live. Social class and deprivation are important indicators of inequity because morbidity and severity of disease are overrepresented, and health research participation is underrepresented, in the most deprived areas and lowest social class. We were only able to use the IMD at the level of the recruitment centre. Although this gave us some measure of the aggregate deprivation of an area where the recruitment site was located, it did not tell us that people living in poverty in these areas of deprivation were taking part in that research. Therefore, to try to overcome this problem, we used smoking status as a patient-level indicator of social class. Not only is smoking often reported in health research, but in recent years, since smoking in the population has become less common, it acts as a relatively good indicator of social class [21]: rates of smoking increase with level of deprivation (31% in the poorest decile versus 9% in wealthiest) [22]. We report that most of the recruitment occurred within NHS hospital institutions located in central city locations which tend to sit in more deprived areas, and 33.6% reported smoking. However, the most common type of clinical research was cancer, meaning we should expect elevated levels of smoking in the sample population; we might interpret the high proportion of non-smoking in this cohort as evidence that most of the sample population are from wealthier areas than where they were recruited. Ideally, we would have used patient-level area code data to calculate IMD and describe the wealth distribution of the sample.

These data, and the recent NIHR audit aggregating data across 140 RCTs conducted between 2019 and 2021 [12], provide valuable insight into the representativeness of funded clinical studies in the UK. Research participation provides advantages both to patients and healthcare services [23]. However, if clinicians do not consider the evidence to be generalisable to their clinical patient population, they might not offer an intervention [6, 10].

This study highlights significant gaps in the reporting of basic information about participants in clinical studies — especially social class, smoking status and ethnicity. These elements will be important if we are to address the challenges of inclusion and equity in research that are prioritised in the new CRN contract from 2024 [24].

The CRN supports participants to access and take part in research; and healthcare institutions and researchers to recruit participants for studies. A key feature of ‘Open for Business’ (UK Government strategies post Brexit) is to encourage inward investment from industry to undertake their research (Pharma and Medtech) in the UK market [25, 26]. High-quality science, rich NHS data and patient resources are key aspects of the offer. Delivering inclusion in studies is vital if findings are to generalise to the wider population and deliver improved outcomes. Individual-level data on sample diversity and levels of deprivation, therefore, must be improved in these datasets; geo-mapping could be used to identify and monitor the representativeness and equity of samples. Different healthcare levels, and non-centrally located institutions, should be adequately supported to exploit their potential to address inequity by undertaking research in areas of high disease burden [10].

It was challenging to reach investigators and obtain even the most basic summary information needed to ascertain representativeness. Equitable research should start with monitoring and transparent reporting of participant characteristics during recruitment. Obtaining data from studies once closed is challenging. We recommend that CRN support should be predicated on a minimum reporting of participant characteristics from the study outset. To make this as easy and secure as possible, CRNs could request non-identifiable participant data in a specific format, to be converted automatically to metrics that allow fair comparison and monitoring over time.

Digital tools can support equity monitoring of clinical studies. We are currently scoping the development of an in-study digital tool to monitor participant equity within the CRN’s open data platform through the MRC DATAMIND mental health informatics programme [27]. The DATAMIND equity audit tool will enable routine monitoring and reporting of recruitment patterns of clinical studies. The digital tool can support research teams to plan for more equitable recruitment by clearly identifying specific underserved groups. Other digital approaches have used machine learning to quantify and visualise gaps in the representativeness of clinical studies [28].

We think our approach is potentially translatable to other regions or countries where health research is undertaken, although it is dependent on good data collection on research activity at a regional level, as our ability to conduct this audit was dependent on online access to the majority of health research conducted in England. This is a unique advantage, and we are not sure if there are other equivalent examples. All health systems will have to deal with these issues as they arise within their own regions and administrative authorities. We believe representation cannot be tackled if it is not monitored over time. We hope the recommendations made serve as a model for how representation in research can be addressed.