Advancing diversity in clinical trials requires joint efforts, novel strategies
Click Here to Manage Email Alerts
Due to a complex interplay of genetic, lifestyle and environmental factors, as well as access to health care, the burden for advanced eye disease is highest among racial and ethnic minority groups.
But disparities in clinical trial participation persist, with an overrepresentation of white participants in most trials.
Source: Mike McElwaine, Wilmer Eye Institute
“In many of the pivotal trials for which our products have been approved, either industry sponsored or non-industry sponsored such as the DRCR Retina Network protocols, certain groups are underrepresented. And the challenge is that these are the same groups that unfortunately bear most of the burden of advanced eye disease,” Adrienne W. Scott, MD, professor of ophthalmology at Wilmer Eye Institute, Johns Hopkins University School of Medicine, said.
A cross-sectional study at Cole Eye Institute, Cleveland Clinic, compared U.S. census data and the representation of racial and ethnic minority groups in randomized clinical trials (RCTs) of diabetic macular edema and retinal vein occlusion (RVO). The analysis found that the demographic distribution diverged significantly from that of the U.S. population in 22 of the 23 RCTs included.
“This implies that trials don’t give us the answers we are looking for because the patients we see in the clinic and treat every day are not the ones that were studied,” Aleksandra Rachitskaya, MD, associate professor of ophthalmology at Cleveland Clinic Cole Eye Institute and principal investigator of the study, said. “However, significant efforts have been made in recent years to address this issue and increase diversity in clinical trial patient recruitment, and it’s a continuous work that involves a lot of stakeholders.”
A profound gap
In the study that Rachitskaya co-authored with Abdul-Hadi Kaakour, MD, and Hong-Uyen Hua, MD, the racial and ethnic demographic characteristics of U.S.-based phase 3 RCTs on DME and RVO published between 2004 and 2023 were compared with 2010 U.S. census data. Of the 169 screened trials, only 23 were based in the U.S. and provided information on race and ethnicity of participants.
“Summing up the participants of all selected trials, we had over 9,000 people, and we found that most of the clinical trials did not reflect the U.S. population. Participants identifying as white tended to be overrepresented, with an overall presence of 80.4% as compared with the 63.7% indicated by census data,” Rachitskaya said. “Hispanic groups were underrepresented in 15 trials, those who identified as Black were underrepresented in nine, and Asians were underrepresented in 10 trials.”
RCTs are the source of evidence-based information on which therapeutic decisions are made, and a disproportionate enrollment of specific racial and ethnic groups limits the generalizability of trial findings. It also creates a disconnect with the emerging needs in health care because Black and Hispanic people experience higher rates of type 2 diabetes, leading in turn to higher rates of diabetic retinopathy among patients who belong to those groups.
“Patients from African and Hispanic minorities also tend to have worse vision at presentation as compared to patients who identify as white, and if we have this inadequate representation in research, it’s really hard to judge how safe or effective a therapy might be in those individuals. There is also a small study suggesting that the response to anti-VEGF might be different among different patient populations. We need more extensive evidence on this, but this possibility should be considered,” Rachitskaya said.
Studies that are more diverse and inclusive would help shed light on the reasons why certain populations bear the brunt of visual impairment and vision loss from potentially treatable eye conditions, Scott said.
“They could tell us more about the biological factors but also the behavioral, social and environmental contributors to eye health, including the degree of health literacy, access to eye care and the barriers to access. They could potentially clarify why some patients with diabetic macular edema are more responsive to anti-VEGFs and some are more responsive to steroid medications,” she said.
She said that in the literature, over the past 3 to 4 years, there has been an increasing number of studies evaluating disparities in eye care.
“It’s quite revealing that much of what we know about the treatments that we use as standard of care are derived from an overall white male population,” she said.
The quest for health equity
Health disparities have gained increasing attention from policymakers and federal health agencies.
In the 2021-2022 session, Congress introduced the Health Equity and Accountability Act, a comprehensive set of strategic policy solutions designed to eliminate racial and ethnic health disparities. In the same session, the Diversifying Investigations Via Equitable Research Studies for Everyone Trials Act and the Diverse and Equitable Participation in Clinical Trials Act specifically addressed the issues related to demographic diversity and equitable access to clinical trials, setting directions for the FDA to require recruitment from underrepresented groups and conduct decentralized clinical trials to improve demographic diversity.
“In clinical trials for FDA-regulated medical products, we are now required to clearly report race and ethnicity data, and there are studies, such as a recent phase 4 study, that specifically enrolled underrepresented minorities as their main study population,” Rachitskaya said.
Companies in the health care space have also realized that taking tangible steps to integrate diversity and inclusion in their strategies is now mandatory to be in line with societal expectations and regulations, thereby ultimately being the key to commercial success.
“A large chunk of the road from drug discovery and development to commercialization is in the clinical research,” Pierre R. Theodore, MD, executive director of health equity and patient inclusion at Genentech, said.
The involvement of underserved populations in clinical trials happens because of a comprehensive commitment to the needs of the population, which is a key part of a company’s mission strategy, he said.
“That means and involves everything from patient advocacy efforts, where we’re listening to the voices of the patients, to screening efforts in which we don’t have a direct commercial stake but that is important to bring more patients into treatment pathways, and then all the way to education of clinical trial specialists so that they know how to interact directly with patients and improve their enrollment in clinical trials,” he said.
Elevatum, a blueprint for trials to follow
Genentech has undertaken a few projects to advance health equity and inclusive research in ophthalmology.
“One of them is our relationship with the American Diabetes Association. We treat diabetic eye disease with our products, but diabetic eye disease is a small subsection of a much larger group of patients who have diabetes. This relationship allows us to listen to community leaders, participate with boots on the ground screening activities for diabetic eye disease, and engage in health literacy campaigns,” Theodore said.
The second project is the Elevatum phase 4 trial looking at faricimab in the treatment of diabetic macular edema, specifically focusing on Black/African American, Hispanic/Latino and Indigenous populations.
“No regulatory agency forced us to do this study, but because it’s meaningful to us for our long-term strategy, we did the trial, and we have really taken significant lessons back around what it means to reach underserved patients in clinical research. I think that’s a unique example of where Genentech invests its resources,” Theodore said.
Scott, who is one of the trial investigators, said that to facilitate enrollment in Elevatum, Genentech specifically addressed the well-known barriers that affect the willingness of underrepresented patients to participate in trials. Transportation to and from the clinic is provided, as well as meal stipends and compensation.
“Patients can also receive treatment in the fellow eye if needed, and eligibility criteria have been broadened to include up to 20% of patients with HbA1c up to 12%. Sometimes patients in these groups get excluded because of the stricter HbA1c cutoff,” Scott said.
In this way, the study is more representative of the patient population that clinicians encounter in real-world practice.
“Not necessarily the patients with the best systemic glycemic control, but more of a typical patient you see in your practice,” she said.
Inclusive enrollment starts with outreach to investigators to discuss best practices in clinical trials and unmet clinical needs who may see a larger number of patients from traditionally underrepresented groups. This is another aspect that differentiates Elevatum from other trials that are usually conducted in large academic and tertiary centers where the population tends to be homogeneously white.
“Elevatum intentionally targeted investigators from the same underrepresented backgrounds, practicing in the places where patients have a high burden of eye disease, and engaged them from the start in the initial discussions about trial design and patient recruitment strategies,” Scott said.
This proved successful, and the study enrolled ahead of expectations.
“Enrollment at my institution was slower in that eligible patients were required to be treatment naive without any past treatment for their diabetic eye disease, but for the overall multicenter clinical trial, enrollment overall completed ahead of expectations. I think Elevatum could be a blueprint for other clinical trials to follow to increase diversity among patients who participate,” she said.
A collaborative effort to overcome barriers
Another initiative from Genentech was the creation of the U.S. Advancing Inclusive Research Site Alliance, a coalition of clinical research sites that have a track record of reaching traditionally underserved populations, including six oncology centers, three ophthalmology centers and now growing into additional disease indications.
“What we do is create a dialogue around how to reach underserved patients, what are the barriers and how to overcome them, as well as issues related to cultural sensitivity, language and translation, simplification of inclusion criteria and protocols. By creating this community of practice, the sites are able to train one another and then generate data that we can share more broadly. We believe this will help to drive inclusive clinical research,” Theodore said.
Rachitskaya agreed that broadening clinical trial participation requires a collaborative effort, time investment and a mindset change.
“It’s going to be a process because there are multiple barriers we’ll have to address, and it’s easier said than done,” she said.
Access for underserved populations is often limited by the fact that most trials are conducted in large academic centers and big private practices.
“We might have to go to our county hospitals or community health centers and enroll patients who are there. Additionally, there have been studies showing that patients respond best to caregivers who mirror their personal characteristics, and therefore increasing diversity among our medical and research personnel is also important. There are initiatives that aim to increase diversity in trainees, and several national organizations such as American Academy of Ophthalmology, American Society of Retina Specialists and Vit-Buckle Society have mentoring programs to encourage participation in clinical trials by diverse investigators. It’s a multitiered approach to improve the situation,” she said.
Adequate compensation for patients who participate in trials is also important because research is time-consuming and people with diabetes or RVO are usually working adults who may have multiple jobs.
“To spend hours without being compensated is cost prohibitive,” Rachitskaya said.
And again, she said that inclusion and exclusion criteria should be modified because people from medically underserved populations often present with worse vision and might be excluded because they have more advanced disease, more comorbidities or worse control of diabetes.
“Finally, patients need to be properly informed and educated to overcome diffidence and reluctance based on the unfortunate history we have of research abuse and discrimination,” Rachitskaya said.
AI systems inherit human biases
The lack of diversity and representation in trials leads to the formation of datasets that are misrepresentative of the population variability. AI algorithms that are trained on these data are biased by design because they mirror these inequalities. When applied to underrepresented minorities, they can lead to misinterpretation and diagnostic error, further amplifying disparities in health care.
“Most of the data we use to train our AI models is from public resources that often do not contain information about ethnicity, gender and age. Therefore, we don’t even know what we are training our models on,” Amitha Domalpally, MD, PhD, research director of the Wisconsin Reading Center, University of Wisconsin-Madison, said.
Other datasets are imbalanced toward an almost exclusively white population.
“There are lots of macular degeneration AI algorithms out there, and most of them are trained on AREDS data because those are big and publicly available. However, the AREDS population is 96% white, which is far from reflecting the current U.S. demography. As a consequence, these algorithms will fail because they don’t know how to recognize disease in the underrepresented groups,” Domalpally said.
The risk of failure is bound to increase when AI algorithms tested and validated in an almost exclusively white population are used across the world.
“We know that the way AMD presents in Asia is different from the way AMD presents in the U.S. and Europe. Therefore, there’s absolutely no way that those algorithms work in Japan or China,” Domalpally said.
Simple biological differences may also hinder the performance of an AI algorithm that is not properly trained on diversity. Skin pigmentation, for instance, is also represented in the retina, specifically in the retinal pigment epithelium, and makes a retina in a white person different from a retina in an Asian or Black person, Domalpally said. Because most of the clinical trials have a 99% white population, human graders have been trained for many years on retinal images that belong to the white population and may not have the same confidence in detecting disease features in pigmented retinas.
“We tested graders’ confidence on detecting severity of diabetic retinopathy on white, Black and Hispanic retinal images. With curated good-quality images from a public dataset, there wasn’t a difference, but when we looked at all-comers, reflecting whatever was submitted for teleophthalmology, ungradable rates were around 20% for Black and Hispanic compared to 5% for whites. And in another clinical trial dataset, it was about 18% for whites and close to 40% for Blacks and Hispanics because of that pigmentation. Ungradable means that the graders cannot tell whether there is disease or not, and this is of course a major problem that we then transfer to our AI models,” she said.
The problem of the biased nature of large datasets has come to light with AI, which mirrored human bias and highlighted the need for equitable inclusion in trials as a prerequisite for equitable care.
“We’ve always had bias, but when you develop AI, there are thousands, millions of human labels that go into it, and bias is amplified,” she said. “I think it will take another 5 years before we start seeing bias go away from our datasets. We could meanwhile generate synthetic data, but that doesn’t seem to be the solution because we would still be excluding the actual populations, perpetuating existing inequalities in the real world. It could be a temporary measure but is not the real solution.”
Overcoming bias
Recently, significant efforts have been made by several groups to overcome the problem of bias in AI models. LumineticsCore (formerly IDx-DR, Digital Diagnostics), an FDA-cleared AI-based screening system for diabetic retinopathy, was developed by a group of researchers in the U.S. who made an active effort to form a dataset that is more representative of the diverse U.S. population.
“Whites are about 60%, but the remaining 40% are distributed,” Domalpally said.
Another data generation project specifically for diabetes and diabetic retinopathy, called AI-READI, is funded by the NIH as part of the Bridge to Artificial Intelligence program.
“They are actively enrolling about 4,000 participants balanced for self-reported race/ethnicity, gender and diabetes disease stage. It is an about $20 million project setting the stage for widespread adoption of AI,” she said.
The Collaborative Community on Ophthalmic Imaging is a large group that gathers members from academic institutions, government agencies, leading professionals and subspecialty organizations worldwide. Its aim is to advance innovation in the ophthalmic imaging space, and through an open forum, annual conferences and publications, it has widely discussed the issue of bias.
“In their recently published [article] ‘Considerations for addressing bias in artificial intelligence,’ they rightly remarked that AI-driven digital health technologies have a great potential to improve equitable access to diagnosis and treatment but may also exacerbate disparities if bias is not properly addressed. We need, first of all, to be aware that there is bias, define specific solutions for mitigating it and incorporate them into the framework of the AI development life cycle,” Domalpally said.
“We are in the stage where we are aware of the problem, and we are starting to implement changes. We all need to work together — individual investigators, the institutions, their research teams, the sponsors, the people who develop the protocols. Everybody has to be aware and make it a priority,” Rachitskaya said.
- References:
- Abràmoff MD, et al. NPJ Digit Med. 2023;doi:10.1038/s41746-023-00913-9.
- Aguwa UT, et al. Am J Ophthalmol. 2022;doi:10.1016/j.ajo.2022.03.006.
- Aguwa UT, et al. Curr Opin Ophthalmol. 2023;doi:10.1097/ICU.0000000000000970.
- Allison K, et al. Cureus. 2022;doi:10.7759/cureus.24424.
- Artificial Intelligence Ready and Equitable Atlas for Diabetes Insights. https://aireadi.org/. Accessed May 15, 2024.
- Bains A, et al. JAMA Ophthalmol. 2023;doi:10.1001/jamaophthalmol.2023.4638.
- Baxter SL. JAMA Ophthalmol. 2022;doi:10.1001/jamaophthalmol.2022.3930.
- Berkowitz ST, et al. JAMA Ophthalmol. 2021;doi:10.1001/jamaophthalmol.2021.0857.
- Bowe T, et al. Ophthalmol Retina. 2022;doi:10.1016/j.oret.2022.01.018.
- Brinson J, et al. Ophthalmic Epidemiol. 2023;doi:10.1080/09286586.2023.2249540.
- Coney JM, et al. J Natl Med Assoc. 2022;doi:10.1016/j.jnma.2021.12.011.
- Domalpally A, et al. Lancet Digit Health. 2021;doi:10.1016/S2589-7500(21)00140-0.
- Gallifant J, et al. Lancet Digit Health. 2023;doi:10.1016/S2589-7500(23)00150-4.
- H.R.6584 – DEPICT Act. https://www.congress.gov/bill/117th-congress/house-bill/6584/text. Accessed April 29, 2024.
- H.R.7585 – Health Equity and Accountability Act of 2022. https://www.congress.gov/bill/117th-congress/house-bill/7585/text. Accessed April 29, 2024.
- Kaakour AH, et al. JAMA Ophthalmol. 2022;doi:10.1001/jamaophthalmol.2022.3929.
- Maloca PM, et al. J Biophotonics. 2024;doi:10.1002/jbio.202300274.
- Montazeri F, et al. Ophthalmol Sci. 2023;doi:10.1016/j.xops.2023.100402.
- Moore DB. JAMA Ophthalmol. 2020;doi:10.1001/jamaophthalmol.2020.2107.
- Nakayama LF, et al. BMJ Open Ophthalmol. 2023;doi:10.1136/bmjophth-2022-001175.
- Nwanyanwu K, et al. JAMA Ophthalmol. 2021;doi:10.1001/jamaophthalmol.2021.4203.
- Osathanugrah P, et al. Am J Ophthalmol. 2021;doi:10.1016/j.ajo.2020.09.042.
- Paik KE, et al. PLOS Digit Health. 2023;doi:10.1371/journal.pdig.0000313.
- Peters U, et al. Ther Innov Regul Sci. 2023;doi:10.1007/s43441-022-00464-3.
- Sanjiv N, et al. J Natl Med Assoc. 2022;doi:10.1016/j.jnma.2021.12.016.
- S.2706 – DIVERSE Trials Act. https://www.congress.gov/bill/117th-congress/senate-bill/2706. Accessed April 29, 2024.
- Scott AW, et al. JAMA Ophthalmol. 2021;doi:10.1001/jamaophthalmol.2021.2053.
- Vidal GA, et al. Contemp Clin Trials. 2024;doi:10.1016/j.cct.2023.107416.
- For more information:
- Amitha Domalpally, MD, PhD, of Wisconsin Reading Center in Madison, Wisconsin, can be reached at domalpally@wisc.edu.
- Aleksandra Rachitskaya, MD, of Cole Eye Institute, Cleveland Clinic, can be reached at rachita@ccf.org.
- Adrienne W. Scott, MD, of Johns Hopkins University School of Medicine, can be reached at ascott28@jhmi.edu.
- Pierre R. Theodore, MD, of Genentech, can be reached at theodop1@gene.com.
Click here to read the Point/Counter to this Cover Story.
Collapse
Click Here to Manage Email Alerts