Some actions to consider for your next trial or submission.
Yes, you read that headline right. Intrigued? Good, but I have misled you just a little, as this paper is about Clinical Trial Diversity, but is not so much “politically incorrect” as “politically agnostic”. What I hope to accomplish in this paper is to briefly define the problem, answer the question “why should I care?” (Hint—it has nothing to do with political correctness), and some actions you may wish to consider for your next trial or submission.
Recently, I was asked to sit on a panel at the Octane MedTech Innovation Forum in sunny Irvine California October 2021. I was certainly happy to escape the 40-degree Minnesota Autumn, but perhaps not so overjoyed at the topic—Diversity in Clinical Trials. In my current role, I increasingly see a push for a diverse population in In Vitro Diagnostic (IVD) trials, but little, if any action in Medical Device trials. With respect to drugs and biologics, my company at least has major initiatives, training, and task forces on the topic. Perhaps the difference between IVDs and devices may be a function of the size of the IVD trials (the last one we were involved in was 10,000 patients), and Device trials tend to be a small fraction of that—in the low hundreds, and sometimes less than 100—but then again, perhaps the IVD people have cottoned on to something device has yet to fully realize, let alone embrace. What’s that? I contend, as I did on the panel, that it is not about being politically correct or “woke”, but rather it should be all about the science. If we are admonished to “follow the science” in other aspects of life, why not in our clinical trials, as after all, is that not what we’re supposed to be doing? Besides that, there are compelling longstanding ethical reasons to pursue representative diversity in clinical trials.
This is not a new concept. In 2002, nearly 20 years ago, The New England Journal of Medicine published an editorial by T.E. King MD whose opening lines read
”The rational use of a new drug or treatment should be based on the results of controlled clinical trials that are well designed, avoid bias, and include subjects representing the full range of patients who are likely to receive the treatment once it is marketed. In addition to age, sex, diet, underlying disease, and the concomitant use of other medications, race and genetic factors may play pivotal parts in the variability of subjects' responses to a medication.”1
As King observed two decades ago, despite the efforts of various government agencies “it is uncertain…whether the participation of minority groups in clinical trials has increased”. As we will see below, the situation remains unacceptable, but there are concrete measures that can be taken to effect meaningful change.
Fundamentally, our clinical trials to date have not been representative of the populations in which the therapy is used. This state of affairs is fundamentally bad science—you cannot expect to see the same results in a diverse heterogenous population that were seen in a relatively homogenous population. It is very simple: if you wish to extrapolate the results of a trial to the entire population who will use the drug or device (or IVD for that matter), then the sample (clinical trial participants) must be representative of that population. To fail to do this is a fundamental sampling error that may lead manufacturers to conclude their product is safe and efficacious when in fact it may not be for the intended population. Much has been made in the past about how good results seen in approval trials are not mirrored when the therapy is released “into the wild” or the general population—and whilst some of that may be related to difference in performance at research sites, off-label use, patient non-compliance, etc., perhaps some of the disparity might just result from initial sampling error.2 Why? Simply because the inclusion and exclusion criteria (I/E) are constructed in such a fashion as to control variables and ensure a degree of homogeneity in the trial population. From a purely scientific perspective, there’s nothing wrong with this—it’s actually good scientific methodology—but it does mean the population studied is not the same as the broader population who may meet one or more of the exclusion criteria: from the perspective of generalizability to the entire population, many approvals trial I/E criteria actually create a sampling error situation.
If we consider published data from Moderna on the COVE COVID-19 study in the US (and this is one of the more diverse studies where a conscious effort was made to represent population racial makeup appropriately)), we can see that 63% of patients were Caucasian, 20% Hispanic, 10% Black, 4% Asian and 3% other.3 US Census Bureau 2019 estimates for the same population breakout nationally are 60.3% Caucasian, 18.5% Hispanic, 13.4% Black and 5.9% Asian—other, including Alaskan Natives, Pacific Islanders etc. account for 4.3%4 We can see in this simple example that Caucasians were over-represented in the COVE study, whilst Blacks and Asians were quite under-represented (by 34% and 47.5% respectively). US FDA’s numbers for new drugs or biologics seeking approval in 2019 puts the Caucasian participation rate at 73%, with most other groups under-represented.5 For specific therapy types, the situation is even more unbalanced – Nazha et al report on nine advanced-solid-tumor phase III trials comprising 1711 patients—Caucasian participation ranged from 77.2% to 97.5% whilst Black participation ranged from 0-4%.6 These disturbing figures are in the face of a known 28% higher cancer-related mortality rate in Blacks compared to Caucasians.7 In another type of cancer, Caucasian females have a 131.8/100,000 incidence of breast cancer and Blacks 124.7, yet Blacks are 39.7% more likely to die from the disease, despite the lower incidence rate.8 It would seem that appropriate representation in clinical trials should be a priority to ensure therapies developed are effective in the populations that need them the most.
Other examples abound in a variety of therapeutic areas. Just recently a meta-analysis published in May 2021 of over 33,400 open-angle glaucoma clinical trial participant data showed 70.7% of the study population was Caucasian, 16.8% was Black, 3.4% was Hispanic/Latino, and 9.1% consisted of other races/ethnicities, including Asian, Native Hawaiian or Pacific Islander, American Indian or Alaska Native, and unreported.9 For the sake of brevity, however, given the wide play this issue has been receiving in various news media in recent times, perhaps we can just take it as read that there are significant problems in the makeup of many clinical trials in terms of representation of the population in whom the therapy will be employed.
Think about it for a moment—if you release a therapy into a population in whom it has not been adequately tested, doesn’t that make it unproven? Is that ethical? The absolute bedrock of everything we do as researchers must be founded in ethical behavior. Must. Surely there is no debate on this topic? Scientifically, we know there are problems with sampling error—this is especially true in genomics: as part of the research for the panel discussion I was involved in recently I discovered that overall, according to The Cancer Genome Atlas, of 5729 samples sequenced to develop therapies for conditions such as prostate and squamous cell lung cancer, 77% (n = 4389) were white, 12% (n = 660) were black, 3% (n = 173) were Asian, 3% (n = 149) were Hispanic, and less than 0.5% combined were from patients of Native Hawaiian, Pacific Islander, Alaskan Native, or American Indian decent.10 How can a manufacturer possibly assure the unsuspecting public that their proposed treatment will work adequately in Asians, Hispanics, or Pacific Islanders for example? The simple answer is they can’t – and that borders on the unethical. How can this be? As I reported in 2014—seven years ago! —12.9% of households speak Spanish at home in the US.11 You could argue that many Hispanics are actually racially Caucasian, and that has some degree of merit, but there are compelling other reasons to actively represent them in clinical trials, as will be discussed later.
The other reason you should care, if questionable ethics are insufficient to sway you (they can, and are, always debated to a degree) is that the US FDA cares, as do other Regulatory Agencies around the world. US FDA issued final guidance on the topic entitled Enhancing the Diversity of Clinical Trial Populations — Eligibility Criteria, Enrollment Practices, and Trial Designs Guidance for Industry November 2020,12 so you know they’re serious about it, especially as guidances rapidly turn into expectations with FDA. Unfortunately, FDA’s guidance refers explicitly to drugs and biologics, and makes no mention of my own particular specialty, Medical Devices. Despite this shortcoming, the guidance has broad application in the Device field.
On that topic, devices, I have often heard manufacturers observe “I get clinical trial diversity when it comes to drugs, but how does that apply to devices or combination drug/device products? Anatomy and Physiology (A&P) is changed that much by race or ethnicity, is it?” To this position I respond as follows – 1) in order to make that claim (lack of meaningful difference in A&P) you need to have studied it, and per the basic sampling error problem described above that hasn’t happened, 2) Racial and cultural differences can make a huge difference to how a patient interacts with Software as a Medical Device or even cares for an implanted device, 3) cultural activities and foods can vary how a device functions (think for example about long periods of fasting and an insulin pump (Ramadan), and 4) because of differential access to healthcare related to income, mistrust, lack of physical clinics etc. some minorities, especially Blacks and Hispanics present, if they present at all, with much more advanced disease than may be allowed in a clinical trial. On the panel which prompted this paper, one of my co-panelists observed that Blacks presented with far more advanced aortic stenosis than Caucasians, and this made use of their cardiac valves much more difficult – the problem is that in order to get the product approved in the first place, inclusion/exclusion criteria in pivotal studies often excludes really advanced disease, so you really don’t know how it will work in that population. That particular problem is worthy of its own brief discussion.
FDA’s guidance on enhancing the diversity of clinical trial populations has a lot to say about broadening inclusion criteria and easing up on exclusionary criteria to increase minority participation. There remains, however, a major problem in terms of establishing performance goals for a therapy – relevant priors. We’ve all had the experience of going to US FDA and proposing a study design with a particular performance goal in mind, and having FDA come back with “paper X shows this outcome – that should be your goal”. Paper X however, and most likely the research papers you used to develop your own performance goals in the first place, is almost certainly derived from a non-diverse population – that is, after all, the fundamental problem under discussion. Bayesian statistics rely on appropriate relevant priors to be worthwhile, so, if your (or FDA’s) relevant priors are fundamentally flawed by sampling error, how can you then rely on them to develop meaningful performance goals for our product? If you enroll minorities who present with much more advanced disease and/or considerably more co-morbidities due to lack of access to healthcare would you not expect to see poorer outcomes than in a much healthier population? Of course!
Manufacturers want to get their therapy approved—many will benefit from it. FDA wants to ensure it is safe and effective. Relying on flawed non-representative historical data to set performance goals is not going to achieve either outcome in a diverse population. Rather, it exacerbates the situation as manufacturers push to get their therapy into the population where it is most likely to be “successful”—if the relevant prior was in a predominantly Caucasian population, then a new study demonstrating comparative performance should be done in the same population: that is pretty basic science. FDA pushes for a very high-performance bar without considering performance may be lower in a diverse population. And so, it goes on—what to do?
Three simple principles can dramatically change the quality of your next study. There are more things you can do to be sure but trying to “boil the ocean” all at once is usually frustrating and unproductive. Consider these elements and you’ll be well on your way - Design, Sites, and Engagement
Hands up who wants to do questionable and flawed science! No takers? OK, then as researchers, let’s take representative population sampling seriously and employ good scientific methods instead. Is it easy? Well, it’s different, and change is often not easy. Is it necessary? Only if you actually take pride in your work and make people’s lives better with proven therapies.
Timothy Pratt, PhD is the Executive Director of Clinical Development for Syneos Health