How AI-Powered Clinical Trials Could Expand Access — and Raise Equity Questions — for American Patients

For decades, participating in a clinical trial has required proximity to major research hospitals, the ability to take time off work for repeated visits, and often just knowing such trials exist—barriers that have systematically excluded rural Americans, working-class families, and communities of color from the frontier of medical research.

The numbers tell the story: minorities comprise over 40% of the U.S. population but represent only 2% to 16% of clinical trial participants, with Hispanic and Latino patients at approximately 1% median enrollment and Black and African American patients at less than 5%.

Now artificial intelligence is rapidly reshaping how patients are found, screened, and enrolled—a transformation fueled by a market projected to grow from $1.92 billion in 2025 to over $10 billion by 2035. The technology promises to finally solve longstanding access problems, but it also risks automating the very inequities that have plagued medical research for generations.

The Promise: Technology That Could Finally Find the Overlooked

AI recruitment systems use natural language processing and machine learning to automatically analyze electronic health records, evaluating hundreds of data points per patient instantly—work that previously required manual chart review. These systems can reduce screening time by 42.6% and identify protocol-eligible patients three times faster with 93% accuracy.

The NIH-developed TrialGPT algorithm processes patient summaries, retrieves relevant trials from ClinicalTrials.gov, and explains how patients meet enrollment criteria, achieving 87.3% accuracy while reducing clinician screening time by 40%. Dr. Zhiyong Lu, NLM Senior Investigator, explains that TrialGPT "could help clinicians connect their patients to clinical trial opportunities more efficiently and save precious time that can be better spent on harder tasks that require human expertise".

Crucially, AI can analyze social determinants of health including income, housing, and education to identify eligible patients in underserved populations—people algorithms might actually find when human recruiters wouldn't think to look.

Meanwhile, decentralized clinical trials enable patients to participate virtually from home or visit local community locations, saving nearly four hours per visit and eliminating costs for gas, parking, and lodging—burdens that disproportionately affect rural patients and those without flexible work schedules. AI-powered remote monitoring platforms use FDA-cleared wearables, biosensors, and digital biomarkers to collect continuous real-world data without requiring clinic visits. Data collected at home is often more accurate than clinic-based measurements because it captures daily behaviors without the white coat effect.

Decentralized trials have already demonstrated success: they recruited three times more patients three times faster than traditional models in a 2018 study and improved recruitment of older adults, non-White participants, and those with lower education levels or limited mobility. Participant satisfaction runs high, with 90% reporting satisfaction, 80% more willing to join similar trials in the future, and 85% feeling more able to participate when remote options exist.

The Historical Weight: Why Scrutiny Is Essential

The promise of AI-driven inclusion must be weighed against clinical research's well-documented history of excluding the very populations these tools claim to help reach. Women's underrepresentation stems from biological factors, caregiving responsibilities, limited access to trial sites, and lack of awareness. For racial and ethnic minorities, decades of systemic injustices, medical mistrust rooted in abuses like the Tuskegee study, stringent eligibility criteria, and financial and transportation barriers have restricted access. More than 50% of U.S. trials listed on ClinicalTrials.gov between 2000 and 2020 did not even report enrollment data by race and ethnicity, obscuring the full extent of the problem.

This exclusion produces medical knowledge based primarily on certain bodies, meaning treatments may not work as well for everyone and side effects may go unnoticed in underrepresented groups until drugs reach the broader market. Three-quarters of those surveyed believe that greater diversity in clinical trial recruitment will lead to more effective and suitable drugs.

The Peril: How Efficiency Doesn't Guarantee Fairness

Machine learning models trained on incomplete or historically skewed data—such as electronic health records that underrepresent minority populations—may systematically deprioritize those same groups, reinforcing exclusion through algorithmic bias. Biases compound throughout the AI lifecycle, from imbalanced data samples and implicit provider biases embedded in medical record annotations to developer naivety and poor real-world generalizability when systems encounter populations different from training data.

The risks aren't theoretical: one widely used AI system prioritized healthier white patients over sicker Black patients for care management because it was trained on healthcare cost data rather than actual medical need—a reminder that optimization for the wrong metric can embed discrimination at scale.

Black box AI systems lack explainability, making it impossible to verify fair recruitment practices or understand why certain patients were selected or excluded—a particular problem in communities where trust depends on transparency. And while integrating AI may increase efficiency in trial operations, there is currently a lack of robust evidence that AI tools actually increase demographic diversity in enrollment.

Even when AI identifies diverse candidates, decentralized trials create their own barriers: reliable high-speed internet, smartphones, and connected health devices remain inaccessible in many underserved areas. The high cost of sophisticated AI tools creates disparities between well-funded health systems and those serving under-resourced communities, potentially leaving providers and patients in marginalized areas without access to these technologies' benefits.

The Fundamental Shift: From Geographic to Digital Gatekeeping

For generations, proximity to major research hospitals determined who could participate in clinical trials, concentrating opportunity in urban areas and affluent communities. AI and decentralized trials promise to dissolve those geographic barriers—but risk replacing them with digital ones, where participation depends on the richness of your electronic health record, the sophistication of your local healthcare system's technology infrastructure, and your personal access to reliable internet and smart devices.

In this new landscape, your data profile becomes your ticket: whether algorithms trained on historical patterns recognize you as a candidate, whether your medical records are digitized in a format AI systems can parse, whether your doctor works in a health system that has invested in these recruitment tools.

Seventy-six percent of industry leaders are actively investing in AI for clinical development, signaling that this transformation is already underway whether or not the equity questions have been answered.

Why This Matters Beyond Abstract Fairness

Who participates in clinical trials today determines whose bodies inform the next generation of treatments—which populations' side effects get documented, which genetic variations get studied, whose responses to medication establish the standard of care. Trial participation also determines who gets first access to potentially life-saving treatments before they reach the general market.

The equity implications extend to trust itself: communities that see themselves systematically excluded or only included through opaque algorithmic processes have reason to doubt whether medical institutions serve their interests. Without intentional intervention, these powerful efficiency tools will accelerate existing patterns of exclusion at unprecedented scale.

The Path Forward: What Intentional Design Looks Like

The difference between AI that expands access and AI that entrenches exclusion comes down to deliberate design choices being decided right now.

The FDA has issued draft guidance on mandatory Diversity Action Plans, requiring sponsors to specify enrollment goals by race, ethnicity, sex, and age so that participants reflect U.S. disease prevalence, with compliance becoming binding 180 days after final guidance is issued. FDA Commissioner Robert M. Califf characterized the guidance as "an important step—and one of many ongoing efforts—to address the participation of underrepresented populations in clinical trials to help improve the data we have about patients who will use the medical products if approved".

The MRCT Center published an ethical framework in May 2026 examining AI use across five stages of the recruitment process, proposing safeguards to address risks including discrimination, selection bias, and erosion of public trust. The Council of Medical Specialty Societies and the Doris Duke Foundation launched the Encoding Equity alliance in 2024 to identify incorrect use of race in clinical algorithms and design accurate, equitable decision tools.

Technical solutions exist to build fairness into AI from the start: the FRAMM framework, a fairness-aware recommender system, improved minority enrollment by up to 60% in simulation studies by explicitly correcting for bias in patient matching algorithms.

Human-centered approaches have proven remarkably effective alongside technology: targeted interventions raised Black patient participation from 13% to 41% in some trials. Institutions like the MCW Cancer Center have launched patient navigation programs to guide patients through eligibility, education, and decision-making, while Johns Hopkins is conducting randomized studies comparing high-intensity versus low-intensity navigation to identify the most effective approaches.

AI can reduce clinical research costs by an estimated $28 billion per year, meaning resources exist to invest in equity safeguards—the question is whether efficiency savings get reinvested in access or simply increase profit margins. Addressing digital infrastructure gaps is equally critical—ensuring that decentralized trials provide internet access, devices, and technical support rather than simply assuming participants already have them.

The technology to expand access exists; what remains is the will to deploy it in ways that genuinely democratize participation—building transparency into algorithms, auditing for bias, investing in digital infrastructure for underserved communities, combining AI efficiency with human navigation support, and holding trial sponsors accountable for diversity outcomes.