Exploring ethnic representativeness in diabetes clinical trial enrolment from 2000 to 2020: a chronological survey.

AIMS/HYPOTHESIS: Ethnic representativeness of participant enrolment in diabetes RCTs involving multiple ethnicities remains unknown. The aims of this study were to evaluate the status and temporal trend of ethnic representativeness in enrolment to diabetes RCTs, and to assess under-enrolment of non-white ethnic groups and explore trial characteristics associated with under-enrolment.
METHODS: We conducted a chronological survey by systematically searching the literature to include eligible RCTs published between January 2000 and December 2020. We assessed temporal trends in enrolment of ethnic groups in the included trials. Univariable logistic regression was used to explore the association between trial characteristics and under-enrolment of non-white groups, using a participant to prevalence ratio of <0.8 to define under-enrolment. This study was registered in PROSPERO (CRD42021229100).
RESULTS: We included 405 RCTs for analysis (327 multi-country trials, 69 conducted in the USA and nine conducted in the UK). The median enrolment rate of all non-white groups was 24.0% in the overall RCTs. Trials conducted in the USA and the UK had median enrolment rates of 29.0% and 12.0% for all non-white groups, respectively. There was a temporal trend towards increased participation of non-white ethnic groups in the overall RCTs; however, no significant improvement over time was found in the US or UK trials. Non-white groups were under-enrolled in most included trials: 62.3% (43/69) in US trials and 77.8% (7/9) in UK trials. The US trials with a high female proportion were associated with lower odds of under-enrolment of all non-white groups (OR 0.22; 95% CI 0.07, 0.65), while trials receiving funding from industry showed increased odds of under-enrolment (OR 4.64; 95% CI 1.50, 14.35). Outpatient enrolment and intervention duration were significantly associated with under-enrolment of Black participants. Only a small proportion of trials reported subgroup results or explored the effect modification by ethnicity. CONCLUSIONS/
INTERPRETATION: A temporal trend towards increased non-white ethnic enrolment was found in diabetes RCTs globally, but not in the USA or the UK. Non-white ethnic groups were under-represented in the majority of diabetes trials conducted in the USA and the UK. Some trial characteristics may be associated with non-white under-enrolment in diabetes trials. These findings provide some evidence for non-white ethnic representativeness in diabetes trials over the past two decades, and highlight the need for more effective strategies and endeavours to alleviate under-enrolment of non-white ethnic groups.

Introduction

In light of the increasing prevalence of diabetes mellitus worldwide, various RCTs have been performed to provide high-quality evidence for the prevention, treatment and management of diabetes. Nevertheless, concerns have been raised about whether participants are enrolled into RCTs proportionately to disease prevalence in the population. For instance, in a systematic review, women were found to be under-represented in cardiovascular outcome trials on heart failure [1]. Ethnic under-representativeness is also a concern in health research because health inequities have been reported among non-white populations in terms of the management of digestive diseases, coronavirus disease-2019 (COVID-19)-related deaths, cardiovascular mortality and respiratory outcomes, and some of the under-representativeness may be related to socioeconomic factors and structural racism [2-5].

It has been reported that participants in RCTs have typically reached a higher educational level, demonstrate greater adherence to therapy, and have a higher socioeconomic status than patients in clinical practice [6]. Fewer adverse effects and reduced mortality rates are also observed in trial participants compared with those who are eligible for inclusion but are not enrolled in trials [7]. However it is unclear whether participant enrolment in diabetes RCTs is representative of disease distribution among those living with diabetes. Many non-white ethnic groups have a higher diabetes prevalence than white groups worldwide, and so would stand to benefit more from results of diabetes trials. For instance, one recent systematic review found that the glucose-lowering efficacy of sodium–glucose cotransporter 2 inhibitors and dipeptidyl peptidase 4 inhibitors was more potent in trials comprising participants of predominantly Asian ethnicity than in trials comprising participants of predominantly white ethnicity [8]. As inadequate proportional representativeness of ethnic groups in trials would lead to an evidence gap in terms of treatment or management recommendations tailored to a specific non-white group [9], improving the representativeness of non-white ethnic groups in diabetes RCTs may be an important route to addressing the significant inter-ethnic health inequities. Therefore, the aims of this study were to evaluate the status and temporal trend of ethnic representativeness in diabetes RCT enrolment (objective I), and to assess under-enrolment of non-white ethnic groups and explore trial characteristics associated with under-enrolment (objective II). This study was registered on PROSPERO (Prospective Register of Ongoing Systematic Reviews; identifier: CRD42021229100).

Methods

Search strategy

We conducted a chronological survey by systematically searching MEDLINE, EMBASE, CINAHL and the Cochrane Library for eligible RCTs from 1 January 2000 to 31 December 2020. Descriptors including synonyms for trials and diabetes were used for the search (see electronic supplementary material [ESM] Table 1 for the search strategy used in MEDLINE). Only trials published in English were included due to lack of language translation resources for non-English studies. The reference lists of included trials, reviews, commentaries or editorials were also checked for potentially relevant trials.

Trial selection

Four reviewers working in two pairs independently screened titles and abstracts. We included full-text RCTs involving multiple ethnicity that assessed the effects of interventions on treatment or management of diabetes and prevention of complications compared with standard care or placebo. We limited participants to those aged ≥18 years old and with type 1 or type 2 diabetes. Trials that did not have an RCT design, did not exclusively include diabetic patients at baseline, did not include two or more ethnicities, or did not aim to explore the effect of interventions were excluded. Trials with a predefined focus on a specific non-white group were excluded; for instance, a trial targeting the Black population only was not eligible. We also excluded trials that focused on gestational diabetes, which is a temporary form of diabetes in most cases. To ensure that we included only RCTs with the potential to impact clinical practice, a threshold for trial sample size of ≥400 patients was used for inclusion, on the assumption that trials with smaller sample sizes were mostly early-phase and single-centre studies [10, 11]. We only used trial data from original publications reporting primary outcomes and main results, i.e. secondary, exploratory or subgroup analyses were not included for analyses. However, some trials do not present information on non-white enrolment in the original publications. Instead, they publish ethnicity-related data in subsequent publications; these studies with data on ethnicity were eligible for inclusion. Over one-third of the studies considered were excluded during the full-text assessment process due to lack of ethnicity data in both original and subsequent publications (Fig. 1). Studies with insufficient information for extraction were also excluded, including protocols, short reports, commentaries and letters to editors.

Fig. 1

Flow diagram showing the trial selection process. PPR, participation to prevalence ratio

Outcomes

The primary outcomes were the enrolment percentages of non-white groups extracted from the included RCTs (for objective I) and the under-enrolment as defined below (for objective II). Secondary outcomes included under-enrolment of Black, Asian, Hispanic and/or other non-white groups. We extracted data about percentages of enrolment of non-white groups from both multi-country and single-country trials, but could only evaluate under-enrolment in single-country trials or when data on multi-country trials were available by ethnicity and study country. Some trials only report data for all non-white groups, rather than data for each individual non-white group; therefore, the number of included trials for analyses of primary outcomes differed from the number of included trials for secondary outcomes.

To define under-enrolment in individual countries/regions, we used the participation to prevalence ratio from cardiovascular trials, in which a participation to prevalence ratio <0.8 indicated under-enrolment [1, 12]. The participation to prevalence ratio is calculated as the percentage of non-white people among trial participants divided by the percentage of non-white people in the diabetic population. The percentages of non-white groups in the diabetes population were obtained from national census reports available in the literature or calculated by dividing the estimated number of participants with diabetes among the non-white group by the total estimated number of all the participants with diabetes (regardless of their race/ethnicity) (ESM Table 2). For instance, the Black group was considered as under-enrolled in the USA if the trial included <12.2% Black patients with diabetes, given the percentage of Black adults with diabetes among the US diabetes population is 15.2%. However, if no data on non-white enrolment for individual countries/regions could be extracted from multi-country trials, we could not define under-enrolment for them due to no global participation to prevalence ratio being available for the global diabetes population stratified by ethnicity.

Data extraction

Two reviewers (JZ and YW) independently extracted data from the included RCTs. Information collected included first, senior and corresponding authors, year of publication, sample size, mean age of the participants, sex proportion, inclusion and exclusion criteria related to ethnicity, enrolment by ethnicity, country/region of the trial coordination office (for multi-country trials), enrolment location, type of intervention, type of randomisation, trial reimbursement for participants, type, frequency and duration of follow-up, treatment effect modification by ethnicity, and funding source. The two reviewers discussed and resolved disagreements in data extraction, and consulted a third reviewer (GL) for a decision if no consensus was reached.

We compared the enrolment performed in a specific country/region with the corresponding ethnic distribution. If the trial was conducted in multiple countries/regions, we tried to extract enrolment proportions by ethnicity in each specific country/region. Many of the multi-country trials did not report detailed non-white enrolment proportions stratified by study country. For these trials, we contacted the corresponding authors via email to seek detailed data.

Statistical analyses

Continuous variables are described using median and upper and lower quartiles (Q1 and Q3), and categorical variables are described using counts and percentages. Results are shown separately for the overall included trials (for objective I) and single-country trials (for objective II) because data from single-country trials could be used to assess the relationship between non-white ethnic under-enrolment and trial characteristics. The non-parametric Kolmogorov–Smirnov test was used as a test for goodness of fit to compare an observed sample distribution with a reference probability distribution to determine whether the enrolment rates by ethnicity significantly differed from the ethnic distribution in the diabetes population. We used the kernel-weighted local polynomial smoothing curve to display enrolment of non-white groups by ascending year of publication. The Jonckheere–Terpstra proportion trend test was used to test whether there was a temporal trend in enrolment of non-white ethnic groups.

For objective II, we used univariable logistic regression models to perform exploratory analyses of the association between trial characteristics and under-enrolment of non-white groups. Associations were quantified as ORs and corresponding 95% CIs. An OR >1 indicated an association between the trial characteristic and increased odds of under-enrolling non-white groups. No multivariable logistic regression analysis was performed due to the small number of included trials and insufficient statistical power. We did perform a post hoc exploratory analysis using a multivariable linear regression model to assess the relationship between trial characteristics and non-white enrolment percentages in the US trials, where the model included all trial characteristics except for those with a variance inflation factor >4. A negative β coefficient implied a relationship between the trial characteristic and a lower enrolment rate of non-white ethnic groups. No regression analysis was performed for the UK trials due to the limited number of trials; instead, we performed a post hoc analysis comparing the proportions of trial characteristics between the UK trials with and without under-enrolment.

All statistical tests were two-sided, with a p value <0.05 taken to indicate statistical significance. We performed all data analyses using STATA software, version 13.0.

Results

A total of 18,006 records were included for title and abstract screening. After assessing a total of 1463 full-text articles for eligibility, we included 405 RCTs for analysis. There were 78 single-country trials (69 in the USA and nine in the UK) and 327 multi-country trials for which data on non-white ethnic groups by specific country could not be extracted (Fig. 1).

Status and temporal trend of non-white ethnic representativeness in diabetes RCTs

Among the overall 405 included trials, the median enrolment rate of non-white ethnic groups was 24.0% (6.4% for Black, 11.2% for Hispanic and 8.5% for Asian groups, respectively). Only 26 of the trials (6.4%) provided subgroup results or explored effect modification by ethnicity, of which seven trials (10.1%) were from the USA and one trial (11.1%) was from the UK. Only one trial specifically mentioned a significant effect modification by ethnicity (p<0.01) [13], and all the remaining trials showed no significant differences in efficacy, effectiveness and/or safety outcomes between ethnicities. However, it is unknown whether these trials did not show significant subgroup effects because of insufficient statistical power or the true absence of effect modification, because most of the trials (20/26, 76.9%) performed exploratory post hoc subgroup analyses by ethnicity without a predefined hypothesis or sample size consideration.

The temporal change in non-white ethnic enrolment is shown in Fig. 2 and ESM Fig. 1. A significant trend towards increased non-white group enrolment rate was found chronologically across the last two decades for the overall trials (p=0.04).

Fig. 2

Enrolment percentages of all non-white ethnic groups in all included diabetes trials published between 2000 and 2020 (n=405), and in the US trials (n=69) and UK trials (n=9) trials

Under-enrolment of non-white ethnic groups in the US and UK trials, and trial characteristics related to under-enrolment

Table 1 summarises the trial characteristics for the US and UK trials. Approximately half of the trials were published after 2010 (50.7% for the USA and 44.4% for the UK). The trials had a median female proportion and patient age of 46% and 58 years, respectively, for the USA and 41% and 63 years, respectively, for the UK. The trials mainly focused on type 2 diabetes (63.8% for the USA and 55.6% for the UK) and enrolled outpatients (47.8% for the USA and 55.6% for the UK). No trials reported inclusion criteria specifically related to ethnicity or whether patients received reimbursement for compensation. Approximately half of the US trials (47.8%) and three-quarters of the UK trials (77.8%) were funded by industry. The multi-country trials generally had similar characteristics to the US and UK trials (ESM Table 3). As shown in Table 2, the US trials had a median enrolment rate of 29.0% (Q1–Q3: 22.5–37.9%) for all non-white ethnicities (n=69 trials): 13.5% for Black participants (n=62), 11.6% for Hispanic participants (n=38), 2.6% for Asian participants (n=34) and 3.3% for other non-white ethnicities (n=63). The UK trials had a median enrolment rate of 12.0% (Q1–Q3: 9.4–21.0%) for all non-white ethnicities (n=9 trials): 5.0% for Black participants (n=5), 12.0% for Asian participants (n=4) and 3.1% for other non-white ethnicities (n=6).

Table 1

Summary of trial characteristics for the included US and UK trials published between 2000 and 2020

Trial characteristic	RCTs performed in the USA (n=69)	RCTs performed in the UK (n=9)
Year of publication
2000–2004	11 (15.9)	2 (22.2)
2005–2009	23 (33.3)	3 (33.3)
2010–2014	20 (29.0)	0
2015–2020	15 (21.7)	4 (44.4)
Sample size	685 (538, 1191)	2721 (826, 9597)
Patient age (years)	58 (55.0, 60.3)	63 (52.4, 64.0)
Female proportion	46 (40.9, 52.6)	41 (40.9, 43.0)
Trial primary objective
Glycaemic control	22 (31.9)	4 (44.4)
Management	28 (40.6)	2 (22.2)
Complications	19 (27.5)	3 (33.3)
Mixed	0	0
Had criteria related to ethnicity	0	0
Trial reimbursement for patients	0	0
Outpatient enrolment	33 (47.8)	5 (55.6)
Type of diabetes
Type 1	0	0
Type 2	44 (63.8)	5 (55.6)
Unspecified	25 (36.2)	4 (44.4)
Type of randomisation
Individual	67 (97.1)	6 (66.7)
Cluster	2 (2.9)	3 (33.3)
Type of intervention
Medication	42 (60.9)	2 (22.2)
Lifestyle or education	10 (14.5)	2 (22.2)
Device	5 (7.2)	2 (22.2)
Other	12 (17.4)	3 (33.3)
Frequency of intervention
>1 time/week	45 (65.2)	4 (44.4)
1–4 times/month	3 (4.3)	1 (11.1)
Other	21 (30.4)	4 (44.4)
Duration of intervention (weeks)	8.6 (6.0, 14.0)	15.0 (6.0, 72.0)
Type of follow-up
Face to face	23 (33.3)	4 (44.4)
Telephone	2 (2.9)	0
Other	44 (63.8)	5 (55.6)
Frequency of follow-up
Weekly	1 (1.4)	0
Monthly	4 (5.8)	1 (11.1)
Yearly	4 (5.8)	1 (11.1)
Unknown	60 (87.0)	7 (77.8)
Duration of follow-up (months)	12 (6.0, 24.0)	18 (6.0, 84.0)
Subgroup analysis by ethnicity reported	7 (10.1)	1 (11.1)
Source of funding
Public or institute	27 (39.1)	0
Industry	33 (47.8)	7 (77.8)
Combination	9 (13.0)	2 (22.2)

Values are n (%) or median (Q1, Q3)

Table 2

Non-white ethnic enrolment percentages in the diabetes trials conducted in the USA and the UK

	All non-white groups	Black group	Hispanic group	Asian group	Other non-white groups
RCTs conducted in the USA
Number of included trials	69	62	38	34	63
Trial enrolment percentage	29.0 (22.5, 37.9)	13.5 (9.6, 16.4)	11.6 (10.0, 19.5)	2.6 (1.4, 3.8)	3.3 (1.3, 6.5)
RCTs conducted in the UK
Number of included trials	9	5	0	4	6
Trial enrolment percentage	12.0 (9.4, 21.0)	5.0 (3.4, 5.0)		12.0 (6.5, 38.9)	3.1 (1.1, 9.6)

Values are medians (Q1, Q3)

All distributions of non-white enrolment significantly differed from the ethnic distribution in the diabetes population in the US trials (p values <0.01) (ESM Table 4). Similarly, there were significant differences between non-white enrolment in the UK trials and the population distribution (p values <0.01). As shown in Fig. 2, no significant temporal change was observed in trials performed in the USA (p=0.17) or the UK (p=0.34). Figure 3 shows the trend in US trials by specific ethnicity. A temporal trend was observed for increased enrolment of Black and Asian participants; however, the trend was not significant (ESM Table 4). Inclusion of Hispanic participants fluctuated over time, with a non-significant temporal trend. Likewise, no significant trend regarding enrolment of Black and Asian participants was observed in UK trials.

Fig. 3

Temporal trend of enrolment by specific non-white ethnic group in the US trials published between 2000 and 2020. (a) Black participants; (b) Asian participants; (c) Hispanic participants; (d) other non-white participants

The under-enrolment rate for all non-white groups in the US trials was 62.3% (43/69), while for a specific non-white subgroup, the under-enrolment rate ranged from 33.3% for other non-white groups (21/63) to 82.4% for Asian participants (28/34) (ESM Table 4). The results from univariable logistic regression models are shown in Table 3 and ESM Table 5. For all non-white ethnic groups, trials with a female proportion ≥46% showed 78% lower odds of under-enrolment (OR 0.22; 95% CI 0.07, 0.65; p<0.01). Funding from industry was significantly related to increased odds of under-enrolment in all non-white groups when compared with funding from a public body or institute (OR 4.64; 95% CI 1.50, 14.35; p<0.01). The results from the multivariable linear regression model also showed that funding from industry was significantly related to reduced non-white enrolment (β=−15.44%, p<0.01) (ESM Table 6). For Black participants, trials recruiting patients from outpatient settings were more likely to have under-enrolment compared with inpatient settings (OR 2.92; 95% CI 1.02, 8.37), while a duration of intervention >8 weeks showed 87% lower odds of under-enrolment (OR 0.13; 95% CI 0.04, 0.40). No significant relationship was found between year of publication and under-enrolment of any specific non-white ethnic group.

Table 3

Relationship between trial characteristics and under-enrolment of non-white ethnic groups based on a univariable logistic regression model in the US trials published between 2000 and 2020

	Under-enrolment of non-white ethnic groupsa
Trial characteristic	OR (95% CI)	p value
All non-white groups (n=69)
Year of publication ≥2010	0.38 (0.14, 1.05)	0.06
Female proportion ≥46%b	0.22 (0.07, 0.65)	<0.01
Source of funding
Public or institute	Reference	-
Industry	4.64 (1.50, 14.35)	<0.01
Combination	1.56 (0.34, 7.13)	0.57
Black group (n=62)
Year of publication ≥2010	0.37 (0.11, 1.19)	0.10
Female proportion ≥46%b	0.36 (0.13, 1.03)	0.06
Outpatient enrolment	2.92 (1.02, 8.37)	0.04
Duration of intervention >8 weeksb	0.13 (0.04, 0.40)	<0.01
Hispanic group (n=38)
Year of publication ≥2010	1.83 (0.48, 7.07)	0.38
Duration of intervention >8 weeksb	3.43 (0.88, 13.39)	0.08
Asian group (n=34)
Year of publication ≥2010	0.20 (0.03, 1.32)	0.09
Duration of follow-up >12 monthsb	5.00 (0.76, 32.93)	0.09
Other non-white group (n=63)
Year of publication ≥2010	0.92 (0.83, 1.02)	0.12

Results shown for year of publication and other factors with a p value <0.1; an OR >1 indicates that the trial characteristic was related to increased odds of under-enrolment; results for the full list of factors are shown in ESM Table 5

aNumber (percentage) of included trials with under-enrolment: 43 (62.3%) for all non-white groups, 25 (40.3%) for Black people, 23 (60.5%) for Hispanic people, 28 (82.4%) for Asian people, 21 (33.3%) for other minority groups

bThe cut-off point used was the median value

Of the UK trials, 77.8% (7/9), 80.0% (4/5) and 75.0% (3/4) showed under-enrolment of all non-white groups, Black participants and Asian participants, respectively. No significant differences in trial characteristics were found between the UK trials with and without under-enrolment of all non-white groups (ESM Table 7).

Discussion

In this chronological survey, we assessed the enrolment of non-white ethnic groups in diabetes RCTs involving multiple ethnicities in the past two decades. There was a trend towards increased participation of non-white ethnic groups in the overall RCTs chronologically, but not in the US or UK trials. There were significant differences in the non-white group distributions between the trial and diabetes populations, with non-white under-enrolment of 62.3% and 77.8% for the US and UK trials, respectively. The US trials with a high female proportion were significantly associated with lower odds of under-enrolment of all non-white groups, while trials receiving funding from industry had increased odds of under-enrolment.

Although non-white ethnic enrolment appeared to improve over time in the multi-country RCTs, no further explorations could be specifically performed by their study country. Only 12 trial authors indicated that they could not provide the data we requested or had no more data access in our email communications. By contrast, the US and UK trials failed to show a significant temporal trend. Moreover, while all non-white groups accounted for 42.8% and 35.4% of the US and UK diabetes populations, respectively (ESM Table 2), the included trials only had a median percentage of 29.1% in the USA and 12.0% in the UK for non-white enrolment. The non-white under-representativeness of patient enrolment in diabetes trials thus remained an issue of concern in current practice, in line with a guideline indicating that reductions in health disparities among ethnic groups requires more urgent attention and endeavours [14]. Whether the lack of a temporal trend in non-white enrolment in the USA and the UK would remain robust after incorporating information from the multi-country trials is unknown due to the limited data on specific countries extracted from the multi-country RCTs.

Some authorities or guidelines have advocated the incorporation of ethnicity information into diabetes research [15, 16]. In our study, the enrolment of non-white participants indicated some improvement, but such participants remain largely under-represented in diabetes trials. Previous evidence suggested reasons for non-white under-enrolment in clinical trials, e.g. patients’ lack of transportation and childcare, limited access to healthcare and adequate insurance, not being aware of their eligibility, and mistrust of physicians and research [17-19]. More importantly, implicit biased perception from physicians or research of non-white candidates deterred effective communication and clinical interactions with patients, thereby discouraging candidates from participating [20, 21]. Nevertheless, research has found that non-white groups are at least as likely as white groups to participate in trials if they are offered a trial with adequate details provided [22, 23]. Female patients were reported to consent less frequently to participation in trials than men, partly due to fear of adverse health events and negative experiences of research engagement [24, 25]. Enhanced female enrolment may therefore reflect a low level of implicit bias against sex and a better interaction between research personnel and trial candidates. Thus, a high female proportion in trials may be related to increased enrolment of non-white groups for similar reasons during the trial recruitment process. While US federal law and National Institutes of Health guidelines require proportional ethnic representativeness in research funded by the National Institutes of Health [26], industry-funded trials were less likely to recruit non-white groups, yielding an OR for under-enrolment of 4.64 for all non-white groups. Implicit bias against non-white groups may lead research personnel to believe that non-white groups would be poor candidates, with compromised mutual trust and low compliance [27, 28]. The impulse to identify favourable effects of intervention and/or potential conflicts of interest for researchers in industry-funded trials may also consciously or unconsciously discourage them from effectively communicating with non-white candidates. Potential factors related to specific under-enrolment of Black participants included outpatient settings and long intervention durations. Referral bias and/or logistical difficulties in outpatient settings may be important barriers for Black participants in particular [29]. The inverse association between long intervention duration and under-enrolment of Black participants merits further investigation, with potential improvements such as enhanced community engagement, strategic and cultural motivations for minority involvement, and positive physician–patient relationships [30, 31]. However, these results should be interpreted with caution given the exploratory nature of our analyses, the small number of included trials, with potentially insufficient statistical power and wide imprecision, and potential confounding factors that we were unable to capture.

Non-white ethnic enrolment has been systematically explored in trials relating to cancer [32, 33], vaccines [34, 35], behaviour [36, 37] and cardiology [38, 39]. These analyses either focused on specific high-profile journals [39] or a specific country [35, 36], targeted trials of drugs for approval by the US Food and Drug Administration [38], or covered a short time span [32–34, 37]. However, similar findings were reported in these different areas, i.e. non-white groups were less likely to be enrolled and were under-represented in clinical trials relative to the disease distribution, highlighting the need for more effective strategies to enhance non-white group enrolment.

There are also some epidemiological studies exploring the differences in engagement, management and healthcare service usage between ethnicity in routine practice or some diabetes-related programmes [40-43]. However, we are not aware of any previous study systematically assessing the temporal trend of non-white enrolment in diabetes trials. A recent study has reported non-white ethnic under-representativeness in eight US trials of type 1 diabetes from 2015 and 2020, focusing on technologies approved by the Food and Drug Administration [44]. Unlike this previous work, our study comprehensively investigated non-white enrolment in diabetes RCTs in the past two decades. Another difference is our exploratory results regarding trial characteristics in relation to non-white enrolment, which may provide insights into current trial practices and potentially generate targeted recommendations for enhancing non-white ethnic enrolment.

Despite endeavours and policies aimed at overcoming the social and structural causes of healthcare inequities among ethnic groups, under-representativeness in trial enrolment undermines this goal. Under-enrolment of non-white participants deprives them of the benefits of trial participation [45]. More importantly, results from trials that lack ethnic diversity and representativeness may not be generalisable to non-white groups in clinical practice [46]. Thus, despite the disproportionately higher rates of diabetes in non-white ethnic populations, ethnic differences in effects of genes or gene–environment interactions, patients’ responses to interventions, and pathophysiological processes remain largely unknown. Therefore, outcomes stratified by ethnicity are recommended to be reported in Phase III clinical trials, in order to explore the heterogeneity of intervention effects and improve evidence-based decision-making. Unfortunately, we found that only a small proportion of trials (6.4%) published results stratified by ethnicity, and almost none of them showed a significant ethnic effect modification. Thus, it is not possible to assess the heterogeneity of intervention effects in trials without ethnic subgroup reporting or to evaluate whether selective outcome reporting exists [47]. Equitable selection and enrolment of eligible patients, in combination with predefined ethnicity-related hypotheses and transparent subgroup result reporting, are needed to enhance the credibility and generalisability of trial findings and to mitigate the current practice of ethnic under-representativeness. Furthermore, post hoc analyses of trials and meta-analyses on effect estimates among non-white groups [8, 48], real-world evidence studies [49] and transport studies [50] may be potential solutions in progressing towards improved ethnic representativeness in RCTs.

Strengths and limitations

Our assessment may offer insights into improving non-white engagement in future diabetes trials. Another strength was the duplication of study processes, including literature search, screening, data extraction and statistical analyses, to increase the credibility of the study findings. However, some limitations exist in this study. We were unable to obtain data on non-white enrolment by specific country in multi-country trials; therefore, it was unknown to what extent the data from multi-country trials would impact our findings about trial enrolment in the UK and USA. Although we wished to assess non-white representativeness in specific countries, we could only evaluate under-enrolment in US and UK trials due to the lack of other single-country trials and lack of such data from multi-country trials. Likewise, under-enrolment could not be assessed in multi-country trials because no global participation to prevalence ratio is available for the global diabetes population stratified by ethnicity. Similar to previous studies, we assumed that the non-white ethnic distributions in the diabetes population were constant over time. It was not possible to collect data on detailed enrolment strategies, whether the research personnel were members of minority groups, or whether they had a conflict of interest with non-white groups, even though these data may be important factors influencing non-white representativeness. We only included trials in English language, which may omit some eligible studies from analyses and thus impair our study findings. Furthermore, only data on the US trials were used for regression analyses, and these associations between trial characteristics and non-white ethnic enrolment may not be generalisable to other countries or other research areas. Given the small number of included US trials and the occurrence of under-enrolment, no multivariable logistic regression analysis could be performed to adjust for other covariates in the model. Therefore, our exploratory results regarding trial characteristics in relation to non-white under-representativeness were hypothesis-generating in nature and should be interpreted with caution. In addition, we only included trials of ≥400 patients for analysis, which limited further exploration and may compromise our results. Future research is recommended to increase the trial sample size for further analyses by including small-scale trials.

Conclusions

In this chronological survey, we observed a temporal trend towards increased non-white ethnic enrolment in diabetes RCTs involving multiple ethnicities globally, but not in the USA or the UK specifically. Non-white ethnic groups were under-represented in most trials performed in the USA and the UK. Some trial characteristics were found to be associated with non-white enrolment, such as the proportion of women, and whether the study was industry-sponsored. These findings provide some evidence regarding non-white ethnic representativeness in diabetes trials over the past two decades, and show that more strategies and endeavours are required to alleviate the under-enrolment of non-white ethnic groups.

(PDF 227 kb)