This is the third of a three-part series on issues affecting informed consent for participation in clinical research. To read the first installment, click here; to read the second installment, click here.
Clinicians, researchers, scholars, and patient advocates generally agree that obtaining informed consent (IC) for participation in clinical research should consist of a process that informs the potential participant about a research study’s aims and design, the study’s risks and potential benefits, alternative options available to the participant, and the participant’s right to withdraw from the study at any time. It should also help the potential research participant to understand the study protocol and the individual’s rights by presenting the information in an appropriate, easy-to-assimilate format tailored to their individual needs.
But despite a near-unanimous agreement about what IC should be and how it should be implemented, we continue to fall short of this ideal. And in particular, there’s a perception that the regulatory community, including the Department of Health and Human Services’ Office of Human Research Protections (OHRP), the FDA, and the system of institutional review boards (IRBs), has stifled innovation with the potential to improve different aspects of the consent process.
In this third installment of posts on informed consent, I’ll suggest ways that digital technology and AI could be applied to make IC better and more efficient, based on my experience at FDA.
During my time at FDA, sincere efforts were made to overcome the perception that IC could not be reformed and encourage the research community to engage in innovative research and demonstration projects capable of challenging an ossified system. These included FDA guidance promoting digital consent methods and explicit calls for innovation in the literature. In addition, an internal program, led by the Office of Clinical Policy in the Office of the Chief Medical Officer, highlighted the benefits of innovation in IC, helped review teams get comfortable with the need for improvement, and actively sought their input on how to improve the process.
Technology with the potential to improve the consent process is evolving rapidly. For several decades now, it has been possible to use the digital environment to improve the information base available for a potential research participant to interact with, given the typical limitations on time for direct human interaction in clinical settings. User interfaces have improved, with clunky canned videos giving way to interactive chatbots that can provide detailed information about the study protocol and address multiple issues in consent. Here’s a short list of newer approaches made possible (or made easier) with technology:
Asynchronous Learning
A patient or potential research participant may find it difficult to focus on and absorb unfamiliar or abstruse information amid the distractions of the clinic. But simply sending a patient home with reams of paper for “homework” is no solution. Instead, digital tools can enable the prospective participant to engage with this information in his or her spare time, with interactive formats that allow them to learn at a pace that suits them and to delve as deeply into a topic as they wish.
Matching Study Materials to the Individual’s Level of Understanding
The United States comprises over 320 million people representing a wide array of ages, levels of education, cultural backgrounds, languages, and learning preferences. Having only a single version of a form to convey information about something as complex and nuanced as IC only makes sense if there is no way practicable to adapt the message to the specific needs of the potential participant. Of course, the adaptive human interaction with the research participant that takes place in the clinic is a critical skill for the clinical research staff, but the time and energy required can be substantial. In the near future, adaptive generative AI chatbots should, with appropriate safeguards, be able to answer a participant’s questions in ways best suited to their individual needs and preferences. In addition, being able to use video formats also helps overcome literacy issues and some vision limitation.
A patient or potential research participant may find it difficult to focus on and absorb unfamiliar or abstruse information amid the distractions of the clinic. But simply sending a patient home with reams of paper for “homework” is no solution.
One important potential advantage of such technologies is that they can do the “heavy lifting” of presenting essential background information, freeing site personnel to focus efforts on more valuable interactions, such as discussing areas of uncertainty and helping the potential participant weigh the information presented according to the their values and with the additional context of the opinions of the clinicians and researchers involved.
Language Issues
Within the U.S., a significant part of the population does not speak English as first language. Moreover, many clinical trials enroll participants in other countries. The need for translation and translators in clinical research is a moral imperative. But providing timely access to accurate translation services—particularly for languages less frequently encountered in the United States—is also a long-standing challenge. However, recent technological progress may offer solutions. During my time working at Google’s parent company, Alphabet, I was amazed to see the evolution of Google’s translation function. The company went from armies of people doing basic translation work to fewer people doing higher-level functions. While work will be needed to reach adequate quality in the medical context, using readily available digital translation capabilities could become a routine function and with so much investment in AI in health care delivery, translation back and forth from “medical-ese” to lay language in all languages could soon be routine.
Involving Families
All too often, the decision about whether or not to participate in research is influenced by the opinions of a family member or close friend who is not able to be there in person for the IC process. Enabling “group chats” via social media opens up the possibility of including family members and trusted friends and companions in the discussions about research participation.
Matching the Presentation of Information to the Level of Interest
Anyone who has been part of IC discussions in prospective clinical research knows that some people just want to know what the clinician recommends and are uninterested in further details while others may want to know everything possible. Digital media combined with generative AI is perfect for this function. Especially in the context of asynchronous learning and follow-up, people can engage at their own pace and according to their own level of interest, reserving direct human interactions for the most important issues.
Understanding the Problem and the Counterfactual of Participation
As I note in the second installment of this three-part series on IC, clinicians are typically not prepared to engage in a frank discussion of the risks and uncertainties of routine clinical practice. There is ample literature to document that variation in practice and cognitive errors place people at risk. Further, if there are several treatment alternatives and no one knows which is better, it’s obvious that if one option actually is better, then those patients who are treated with the inferior alternative will be harmed. In this scenario, not only is there harm to the individual patient but also to the larger community because there has been no learning through research that could eventually have yielded benefit to everyone. This is not the fault of the clinicians, but it is a problem that can be addressed through a commitment to recommending research in situations where uncertainty is significant.
One question that often emerges around clinical research is whether participants in trials do better than people with the same clinical situation who do not participate? My interpretation of the several methodologically difficult studies that have addressed this issue is that trial participants fare at least as well and probably better. Whether trial participation is causal, the kinds of people who volunteer would have done better anyway, or overall care in the context of a trial is better (my favorite explanation), I can’t say for sure; but at least we can say that there’s no credible evidence that participation in trials in the modern regulatory environment is harmful overall.
Testing Understanding
There are relatively few data about the retention of information among research participants, but what data there are point to significant comprehension issues at the time of consent and significant loss of information over time. For example, the concept of randomization and “therapeutic misconception” (in which people assume a benefit from participation even when the consent form says that benefits are unknown) present challenges to the ideal of consent. Although no specific process will be 100% effective, the availability of sessions that are not time-limited (thus allowing for asynchronous information and interaction) holds great promise.
Creating a robust system of actual forms and processes could enable a rapid learning approach to consent. What approaches have been most successful? What approaches, when exposed to sunlight, were counterproductive?
One caveat is that the areas that would benefit the most from the availability of digital resources, namely underserved and rural areas, have the least penetration of high-speed internet connectivity. The failure to implement high-speed internet despite the availability of funds demands that we resuscitate this idea, not just for research but also for clinical care. “Clinician deserts” and “internet deserts” overlap considerably.
When federal policies governing research in humans (known as the “Common Rule”) were updated in 2016, it included an update in the ClinicalTrials.gov registry that enabled registration of consent forms, but such registration was not required. While I’m in favor of requiring this information about IC, we could at least take advantage of the registry to improve the IC process in the absence of a mandate. Creating a robust system of actual forms and processes could enable a rapid learning approach to consent. What approaches have been most successful? What approaches, when exposed to sunlight, were counterproductive? The application of LLMs to a large corpus of consent information could open up a significant avenue for the research ethics and research operations communities to identify best methods.
So, What’s Holding Us Back?
In this series I’ve tried to make the case that while IC is only one component of a clinical research enterprise in the US that is unnecessarily slow and expensive, it’s nonetheless ripe for improvement along with other elements of clinical research operations. If we fail to make multiple improvements in each component of the enterprise, we will face many more unanswered questions about health and healthcare than should be the case.