7

Real-World Evidence, Accessibility, and Health Equity

Serena Jingchuan Guo began by providing the Food and Drug Administration (FDA) definitions of “real-world data” and “real-world evidence.” Real-world data, she said, are data relating to patient health status or delivery of health care that are routinely collected from a variety of sources. Real-world evidence is clinical evidence derived from real-world data about the use and potential benefits or risks of a medical product. Such evidence might concern the effectiveness and safety of a drug or treatment, for example, or provide details about heterogeneity in treatment effect.

However, she continued, real-world clinical data are not collected for the purpose of providing evidence about the effectiveness of drugs or treatments, and there are many structural and methodological challenges with transforming such data into real-world clinical evidence. The challenges include missing clinical data, the often-poor organization of such data, and the fact that such data often do not provide the details researchers need to generate clinical evidence.

Despite those limitations, Guo continued, real-world evidence is uniquely positioned to address a number of important questions concerning the use of glucagon-like peptide-1 receptor (GLP-1R) agonists in the treatment of various disorders. These questions include the long-term effects of GLP-1R agonists on central nervous system (CNS) disorders, the safety profile of GLP-1R agonists in special populations (e.g., pregnant women, children, and aging populations), individualized treatment effects and precision dosing of GLP-1R agonists (e.g., for users with clinically high or low benefit or the balance of intended effect versus off-target effect), and head-to-head comparisons of the benefit-risk profile for different drugs (e.g., semaglutide versus tirzepatide).

Finally, Guo said, recent research has shown significant geographic and racial disparities in the use of GLP-1R agonists. For example, one study of the use of various glucose-lowering drugs, including GLP-1R agonist class drugs, for use in the treatment of type 2 diabetes found that non-Hispanic Black patients who were newly diagnosed with type 2 diabetes were only two-thirds as likely as newly diagnosed White patients to begin these drugs, even when clinical details were accounted for (Chen et al., 2024). The same study also found large geographic variation in how likely newly diagnosed patients were to use these drugs in the treatment of their diabetes.

Guo concluded her introduction by saying that the session would focus on three somewhat interrelated topics: real-world evidence, accessibility, and health equity.

USING REAL-WORLD DATA FOR TRIAL EMULATION

Jiang Bian, a professor of biomedical informatics at the University of Florida (UF) College of Medicine and the chief data scientist at UF Health,

spoke about ways to use real-world data to carry out what he called “synthetic trials,” which were very similar to the simulated trials that Rong Xu had discussed in the session on substance use disorders. Bian carries out his synthetic trials using a technique called trial emulation, which he described as a framework for dealing with the biases in observational data and achieving the sort of rigor one looks for in real trials.

As background, he noted that there is nothing new about using real-world data to draw conclusions about drug efficacy and safety (Concato and Corrigan-Curay, 2022; Sherman et al., 2016). For instance, the FDA has been doing postmarket surveillance with patients’ electronic health records for years. “We need to figure out new ways to use this,” he said. Other sorts of real-world data include health care claims data, tumor registry data, linked mother–baby data (i.e., birth records), and many others, and the various types of data are most powerful when they are linked to paint a more complete picture of patients. Importantly, Bian continued, there is an existing data infrastructure, both nationally and internationally, for dealing with real-world patient data. For example, PCORnet, which was founded by the Patient-Centered Outcomes Research Institute (PCORI), has data on about 100 million patients held in eight clinical data research networks. Trial emulation is one way to take advantage of those data, he added.

As an example of what is possible, Bian pointed to a study carried out by his team that looked at data from two large real-world datasets for more than 170 million patients over 10 years. The purpose of the study was to identify FDA-approved drugs that might be useful in treating Alzheimer’s disease. Using machine learning, the researchers emulated trials for thousands of medications and found five medications who use was associated with a decreased use of developing Alzheimer’s disease among patients with mild cognitive impairment (Zang et al., 2023).

In emulating a target trial, Bian said, one needs to specify seven key components of the target trial: the eligibility criteria, the treatment strategies being compared (including their start and end times), assignment procedures, the follow-up period, the outcome of interest, the causal contrasts of interest, and an analysis plan. Two components that are particularly important, he said, are the eligibility criteria and how outcomes in real-world data are defined.

To illustrate another way trial emulations can be used, Bian described a target trial emulation study that looked at how GLP-1R agonists might be used to treat Alzheimer’s disease and related dementias. Using electronic health records from the OneFlorida+ dataset, they looked at a total of 33,858 patients with type 2 diabetes who were given various glucose-lowering drugs. Those who had received GLP-1R agonists had a reduced risk of developing Alzheimer’s disease and related dementias, he said, but

he emphasized that the trial emulation also made it possible to examine the drug’s heterogeneous treatment effects across different subpopulations. For instance, if a patient had cardiovascular disease and some type of cerebrovascular disease, the treatment effect of the GLP-1R agonist was much bigger. “It does point [to] a way where you maybe design the drug for [a] specific subpopulation and where the real-world data can help you find that subpopulation,” he said.

Bian then briefly described current work he is carrying out with Fei Wang of Weill Cornell Medicine on the use of real-world data in the design of trial eligibility criteria. The research, funded by the National Institute on Aging, is motivated by the issue of trial generalizability. Since trials are conducted in a constrained environment and do not match the target population in the real world, the goal of designing trial eligibility criteria is to minimize the gap between the trial subjects and the real-world target population to maximize the trial’s generalizability as well as predicting potential outcomes and safety signals. Bian and Wang are working to apply explainable artificial intelligence methods to real-world data in order to assess the quantitative impacts of various eligibility criteria on clinical outcomes. One specific goal is to develop a prototype toolbox for eligibility design criteria.

Bian concluded by discussing some of the challenges in using real-world data. Misclassification issues are a major problem, for instance, as many clinicians are not accurate in coding diagnoses. One approach to solving this would be to develop more accurate computable phenotypes using informatics, though this would not completely prevent misclassification errors. Another major issue is that much of the information in electronic health records does not exist as structured data; some 80 percent of clinical information is in the form of free-text narratives, Bian said. Natural language processing may offer a solution to this problem. And there is a long list of other data types that are not readily accessible to researchers, such as imaging, genomics, and microbiome data. Ultimately, making these sorts of data easily available for research purposes will require a better data infrastructure, he said.

SHORTAGES OF GLP-1R AGONISTS

Robert Kosko, senior program management officer for the drug shortage staff at FDA and a commissioned officer in the U.S. Public Health Service, described FDA’s role in ensuring the supply of drugs in the United States and, specifically, what is being done to deal with current shortages of GLP-1 class drugs.

He began by describing the FDA’s role in ensuring the availability of safe and effective drugs in the United States. FDA staff works with manufacturers on both short-term and long-term strategies to address actual or

potential supply interruptions, he said. “We don’t want to just address the immediate issue, but we also want to ensure there are not continued issues that could impact future supply.” The staff also works to develop a risk-benefit analysis for each drug shortage situation as well as to distribute information related to drug shortages by posting public information on the FDA website and reaching out to professional organizations and patient advocacy groups.

The staff has various tools to deal with a drug shortage situation, Kosko said. For instance, if a manufacturer notifies the FDA of a potential or actual shortage of concern, FDA staff can reach out to other manufacturers of the product to see if they can ramp up their own production. They can use regulatory flexibility to release a product to the market that does not meet the current FDA-approved specifications but with added safety controls or additional testing. They can also expedite the review of proposals and regulatory submissions. “And as a last resort,” he said, “we will explore the potential importation of product to assist with a shortage situation. This requires careful evaluation of the product attributes, manufacturing facilities, and labeling.”

Then Kosko turned to the current GLP-1R agonist supply situation in the United States, as of September 2024, when the workshop took place. The first GLP-1R agonist product shortage was posted on the FDA website at the end of March 2022, he said, and additional GLP-1R agonist products were added to the shortage website in August 2022, December 2022, July 2023, and April 2024. To date only the injectable formulations and not the oral formulations have experienced shortages in the United States, he said, and it appears that supply is now improving, with four of the injectable GLP-1R agonist products currently available in all of their approved forms, while most versions of the remaining products have limited availability.

He closed by talking about what the FDA is doing to assist with the supply of GLP-1R agonists. “First, we communicate on a regular basis with the sponsors regarding their current supply and demand, as well as projected demand and potential ways to increase supply,” he said. “We also update availability information on our drug shortage website at least every 2 weeks and more frequently if additional supply information is provided by a sponsor.” The staff responds to inquiries from patients and health care providers regarding these products with the most up-to-date availability information. Concerning the FDA’s efforts to increase the supply of GLP-1R agonists, he said that the agency’s main tactic is to expedite the review of proposals and regulatory submissions. “We have worked and continue to work with all sponsors of these products to provide feedback on any proposals as well as expedite the review of drug application supplements that can assist with increased supply.” Finally, he said, FDA staff members participate in various working groups, attend workshops, and meet with their

international regulatory counterparts to discuss current supply situations in their jurisdictions as well as best practices to address these shortages.

POTENTIAL BARRIERS AND SOLUTIONS TO WIDENING ACCESS TO GLP-1R AGONISTS FOR THE TREATMENT OF OBESITY

Fatima Cody Stanford discussed barriers to accessing GLP-1R agonists for the treating obesity and what might be done to overcome those barriers. Racial and ethnic minorities are the groups most likely to be affected by those barriers, she said, which is particularly important because those minorities are also often more affected by obesity. For instance, obesity affects more than 60 percent of Black women. And many minority patients with obesity find it difficult or impossible to gain access to GLP-1R agonists such as semaglutide or tirzepatide to treat their obesity. “I happen to work at one of the best-resourced hospitals in the world,” she said, “and I can’t get these to patients today.” The situation is even worse for Medicaid and dual-eligible patients, that is, those who are eligible for both Medicare and Medicaid.

Stanford said the barriers to and solutions for widening access to GLP1R agonists to treat obesity vary according to the actors involved. And, following an article she and colleagues recently published in Nature Medicine, she identified five groups of actors to consider: health care professionals, patients, insurance companies, governmental regulatory agencies, and drug manufacturers (Waldrop et al., 2024).

For health care professionals, she said, a major barrier is that a large percentage of them are not educated about obesity and perceive obesity as a consequence of personal choices, rather than as a disease. Stanford said potential solutions would include improved training within medical schools and residencies, cultural competency training, and training in implicit bias, weight bias, and the effects of weight stigmatization.

Patient-related barriers include the high cost of the medications, the need for long-term use, and safety concerns. The solutions Stanford suggested included helping patients to understand obesity as a disease requiring long-term therapy, improving how health care providers educate patients on the risks and benefits of starting and maintaining anti-obesity medications, and using better insurance coverage and third-party payer discounts to improve patients’ access to these medications, regardless of their ability to pay.

Moving on to barriers related to insurance companies, Stanford pointed to the high cost of the medications and the need for long-term use. Potential solutions include facilitating greater access to the medications to improve cost sharing and removing drug therapies for obesity that have less efficacy and undesirable side effects, such as orlistat.

Barriers facing government regulatory agencies include both safety concerns and addressing stigma and bias related to obesity. According to Stanford, solutions include having these agencies follow medical society and drug manufacturer guidelines and evaluating more clinical data on safety and efficacy endpoints to support long-term use.

Finally, the barriers that drug manufacturers face include safety concerns and the fact that these treatments require long-term use, Stanford said. She emphasized the need for more studies specifically analyzing long-term use to provide clinical data on long-term safety and efficacy endpoints.

DISCUSSION

Challenges for Data Collection

Ellen Mowry asked how to overcome the confounding that will inevitably arise in any clinical study of GLP-1R agonists because people with higher socioeconomic status—who have access to better health care overall—will be far more likely than people with less money to have used these drugs. This could skew the results of any research on the benefits and risks of the drugs. Stanford replied that a group at Massachusetts General Hospital is planning to develop a data repository using data from individuals—typically with higher socioeconomic status—who are taking GLP-1R agonists. The group plans to start accumulating data from the point that individuals begin treatment with these drugs and working with collaborators in neurology, rheumatology, and other areas, record the results, analyze those results, and publish them. “I think this is going to be important for us to do across the board,” she said. However, the data repository is likely to suffer from the same problem Mowry identified since most patients of lesser means will not be able to afford the drugs.

Bian commented that as long as some patients of limited means are receiving the drugs, researchers can use real-world data and data science methods to correct the bias. Also, he added, researchers might get funding to recruit specific minority cohorts to take these drugs to counteract some of the bias in the cohort.

Brian Fiske noted that GLP-1R agonists are starting to be prescribed for things other than type 2 diabetes and obesity, so data will start being accumulated in other places, such as by doctors treating CNS disorders. How can data be brought together from all these different specialists? Bian said that the problem is not so much collecting data, since the data can be found in electronic health records and prescription-dispensing records, and nationally a great deal of effort is going into creating integrated data systems where all these different kinds of data are linked together. The real problems arise from data quality issues, he continued. “Yes, if you ask the

physician to enter more data, you may get more accurate data, but that’s going to take the physician away from treating the patient, so that’s not necessarily the right approach.”

Karen Glanz commented that people go to many different pharmacies, which complicates the collection of data. “I’ve done studies on glaucoma where we couldn’t even get [data on] whether prescriptions were filled,” she said, and she suggested that it would be valuable to set up registries to collect this sort of data. It is also very useful, she added, to collect data on patients’ social determinants of health—such things as socioeconomic levels and whether patients have housing instability or food insecurity. “It’s probably one of the greatest variables where we see a lot of missing data in electronic medical records,” she said, “because it’s voluntary. It’s whether people want to answer, and it’s also at a pretty crude level.”

Elizabeth Mietlicki-Baase asked if anyone is tracking the outcomes or adverse effects of compounded versions of GLP-1R agonists, which some people have turned to because of the shortages of the manufactured versions. Kosko said that a team at FDA is tracking adverse effects for the compounded versions. Glanz referred to a recently published article (Ashraf et al., 2024) that studied compounding pharmacies that were advertising GLP-1R agonists online and found that most of them did not satisfy all the criteria for compounding pharmacies and at least one was providing drugs that were fake and potentially dangerous.

Peter Park, a clinical scientist at Eli Lilly, asked if there are countries with national health systems that collect good, well-structured data for use in real-world evidence analyses. Bian said that he believes the lack of such structured data is universal “because the documentation burden is huge.” Indeed, some Asian countries do not even have electronic health record systems. Thus, it will be important to find ways, perhaps based on artificial intelligence, to extract real-world information from physicians’ notes and turn that information into structured data.

Glanz added that another part of the challenges is that physicians’ records generally do not even contain much of the information needed for analyses. For instance, many doctors prescribe treatments for patients and ask the patients to let them know if there are any problems, so there may be no record of concerns or successful results. In her use of a GLP-1R agonist, it was the pharmacist who asked her to report any side effects, so those would not be listed in her medical record at the doctor’s office.

Prior Authorization and Supply

Stanford commented that one hurdle for patients seeking GLP-1R agonists is prior authorization from insurance companies, which can be quite time-consuming to obtain. Massachusetts General Hospital has a

central prior authorization team, and on average each prior authorization for one patient takes 1 hour. “We’ve seen this year already 51,000 patients at the Mass General Hospital weight center,” she said, translating into a tremendous amount of effort spent just on prior authorizations, and it will only get worse once GLP-1R agonists are commonly prescribed for other diseases.

Nicole Boschi, the director of regulatory affairs at the National Multiple Sclerosis Society, pointed out that enabling Medicare and Medicaid to pay for weight loss drugs will require a change in current laws. Stanford said that there is a proposed bill, the Treat and Reduce Obesity Act, which would make that change, but after 12 years of effort, it has still not been passed.¹ Concerning the supply of these drugs, Kosko said that both of the major manufacturers of GLP-1R agonists are working to increase supply by adding manufacturing lines at existing facilities or bringing new facilities online, and the FDA is working to assist them with these expansions. “But we know the demand just continues to increase for these products and will with these new indications,” Kosko said.

Health Equity

Guo asked the panelists for strategies for avoiding the growing disparities in the availability of GLP-1R agonists and improving health equity. Stanford suggested that there are lessons to be learned from how the costs of HIV drug regimens were brought down after effective HIV drugs were developed in the late 1990s. This is particularly relevant because people with HIV need to stay on their drugs for life, and it appears that people using GLP-1R agonists for obesity may also need to maintain their usage indefinitely. Bian agreed, noting the PrEP and HIV antiviral drugs are covered by Medicaid.

A workshop participant asked Stanford how the systemic challenges of access to GLP-1R agonists compare between the United States and other countries, such as Canada, that have public health systems. Stanford answered that one difference can be found in the costs of the drugs; for instance, the average monthly price for semaglutide in the United States is about $1,400, while it is $278 in Canada. In other high-income countries around the world, she added, government-negotiated drug prices tend to be significantly lower than in the United States.

In conclusion, one of the themes shared by several workshop participants is that real-world data can play an important role in answering various questions about the use of GLP-1R agonists to treat various disorders;

___________________

¹ For more information on the Treat and Reduce Obesity Act, see https://www.congress.gov/bill/118th-congress/house-bill/4818 (accessed November 27, 2024).

one valuable way to use the data is in trial emulation. Maximizing the usefulness of such real-world data will, however, require overcoming a number of challenges, such as misclassification issues and varying standards and formats. Workshop participants also highlighted gaps in health care access and quality as another challenge in the field, given that systemically marginalized and under-resourced populations are already finding it more difficult than members of other groups to gain access to GLP-1R agonists to treat obesity.