Measuring Clinical Trial Representation in the Era of Precision Medicine

As cancer treatments become more targeted in the age of precision medicine, the methods used to measure whether a clinical trial is truly representative of the patient population struggle to keep up with the pace. Recently, June Cha, PhD, a health policy professional, and Andrew Friedson, PhD, a health economist, were co-authors of a report titled Measuring Representation in Clinical Trials: Simulations Demonstrating How Current Methods Fail in the Context of Precision Medicine. In this conversation with Oncology Data Advisor, they further explore efforts to improve trial accuracy through more granular registry data, AI-driven cohort building, and institutional and federal collaboration.

Defining Representation in the Era of Precision Medicine

The foundational goal of clinical trial representation is to ensure that a study population mirrors the real-world population intended to receive the treatment. Dr. Friedson notes that without this alignment, results cannot be safely generalized. "If I’m going to be looking at disease population X in the study and I want to know how it’s going to generalize to the overall disease population, I want these things to line up. Otherwise, what I study in the lab or in the clinical trial is not going to tell me what I need to know about the study when I take the drug into the actual field," he explains.

As oncology moves toward narrower indications, the definition of a "representative" population must also evolve. Dr. Cha emphasizes that traditional metrics like age, sex, and ethnicity are no longer sufficient on their own. "Increasingly, genetics and molecular profiles are becoming very important. Without that knowledge, we're essentially flying blind when it comes to knowing whether this therapy will work well in the real world compared to what's in the study," she says. "In a precision medicine space, it makes this conversation even more urgent and more complicated because the indications and the study target populations are becoming more narrowed to molecular subtypes."

Limitations of Current Methodologies

Currently, representativeness is measured by comparing a trial’s demographic mix to data from a broad cancer registry. If the trial population deviates significantly from the registry, it is deemed non-representative; however, Dr. Friedson argues that this “urn model” of probability fails when the underlying genetics of a subpopulation differ from the general registry. 

"The issue that we're running into here is that under precision medicine, the urn that you're drawing these balls from is not what you see in the registry. It's a subpopulation that could be genetically distinct," he observes. Using triple-negative breast cancer (TNBC) as an example, he points out that the TNBC population is demographically different from the general breast cancer population. "The test you’re going to do that’s saying, ‘If we draw from this urn, how likely are we to get this?' is not going to be validating. In fact, our paper shows how frequently that test is going to be completely wrong."

Study Findings

The researchers conducted simulations to determine how sensitive these representation tests are to small "drifts" in the underlying population data. By varying the registry demographics by as little as 1%, they found that the results of the statistical tests frequently flipped from representative to non-representative. 

"All we're doing is asking, ‘How often are these tests to our registry going to be different from our true underlying population of this specific genotype?" Dr. Friedson explains. "The answer that the paper shows is that very often, these representation tests are very sensitive to this type of error." This suggests that many trials currently labeled as representative may be failing to capture the specific demographic nuances of the genomic cohort they aim to treat.

System-Level Approaches and Community Engagement

Addressing these gaps requires a multi-stakeholder effort involving the National Institutes of Health (NIH), FDA, and Centers for Medicare & Medicaid Services (CMS). Dr. Cha suggests that the NIH should prioritize funding for biomarker validation, while the FDA’s diversity action plans must account for molecular subtypes. Furthermore, she highlights the need for a robust care infrastructure to ensure that underrepresented patients are even visible to researchers. 

"If patients can't really access the healthcare, then their information isn't going to be accurately represented and captured in the medical records, so they're invisible to any kind of sophisticated AI cohort-building tools," Dr. Cha notes. She advocates for "on-the-ground efforts" such as community health workers, trusted local organizations, and faith-based partnerships to build trust and address social determinants of health like transportation and affordability.

Clinical Implications for Oncologists

For clinicians evaluating trial data for use in practice, the primary concern should be external validity. Dr. Friedson advises oncologists to look beyond the "representative" label. "The more precisely targeted a specific study is, the more you really need to make sure that your patient population is lining up with the patient population of the study," he says.

Dr. Cha adds that clinicians must ask critical questions about the comparator groups used in research. "If the study claims representativeness, the first question should be, ‘representative of who?’ Ask whether the comparator population aligns with the clinical indications, paying attention to the specific biomarkers and mutations that are being targeted."

Key Takeaways

• The Precision Gap: Traditional representation tests often compare trial cohorts to broad cancer registries, which may not reflect the demographic realities of specific molecular subtypes.

• High Sensitivity to Error: Current statistical methods for measuring representation are highly sensitive; even a 1% difference in the reference population can lead to incorrect conclusions about a trial's validity.

• Beyond Demographics: Representation in precision medicine must include genetic and molecular profiles alongside age, sex, and ethnicity to ensure real-world efficacy.

• Ecosystem-Wide Action: Improving trial accuracy requires more granular registry data, AI-driven cohort building, and federal collaboration between the NIH, FDA, and CMS.

• Clinical Scrutiny: Oncologists should verify that the enrolled trial population—not just the general disease population—matches the specific biomarkers and demographics of the patients they are treating in practice. 

About the Speakers

June Cha, PhD, is a health policy professional with two decades of leadership experience spanning infectious disease research, non-profit and multilateral organizations, the government, and the pharmaceutical industry. Her research focuses on connecting biomedical innovation, regulatory frameworks, and reimbursement for innovative medical products throughout the ecosystem. Dr. Cha leads policy efforts aimed at accelerating biomedical innovation and access, as well as promoting a prevention-first approach to healthcare.

Andrew Friedson, PhD, is a health economist with expertise in healthcare sectors and insurance. He has published peer-reviewed research on health behaviors, markets, and policy in premier journals spanning economics, public policy, and medicine. Dr. Friedson is the author of Economics of Healthcare: A Brief Introduction, a textbook published by Cambridge University Press and used in classrooms around the country.