How AI Can Help Satisfy FDA’s Drug, Device Diversity Requirements

Wendy Cheng addresses the potential role artificial intelligence and machine learning can play in satisfying the diversity requirements as part of the Food and Drug Omnibus Reform Act of 2022.

This article was originally published at STAT on April 19, 2024.

 

The Food and Drug Omnibus Reform Act of 2022 (FDORA), which was passed as part of the Consolidated Appropriations Act of 2023, will encourage greater diversity in clinical trials and help ensure new medications and treatments are developed with more representative populations in mind. But FDORA represents a paradigm shift for life sciences companies in the way individuals will be identified for clinical trials and enrolled in them, requiring companies to look closely at and rethink their current trial strategies.

FDORA will require trial sponsors to include diversity action plans for Phase 3 and other pivotal studies of new drugs and devices to contribute to more generalizable clinical data, which can improve both regulatory decisions and public health.

While the date these regulations will go into effect remains to be determined (at the time this article was published, the FDA had not set the expected date of its updated guidance), the mandates for change are clear: Trials that start enrolling participants after the effective date will be required to include a diverse patient population, with a focus on racial and ethnic diversity. Other demographic and non-demographic factors — including sex, gender identity, age, pregnancy status, lactation status, and multiple comorbidities — will also need to be considered.

What may not be clear at this point are the potential implications of FDORA on operational and analytical logistics, including the need for sufficient sample sizes and stratified enrollment and analyses by patient subgroup. Even before these new regulations were being developed, life sciences organizations often struggled to recruit enough patients for clinical trials, and many trials did not reach statistical significance. This begs the question: What can life sciences companies do to increase efficiency in clinical trials to maximize the likelihood of success in drug development, considering the need to include diversity? I propose some potential answers that may involve artificial intelligence (AI) and machine learning (ML).

Focus on health equity in clinical trials

Health equity has come under increased scrutiny with the rising recognition that medicine is not one-size-fits-all, and that personalized medicine is essential for quality care. The populations enrolled in clinical trials have historically not reflected the patient population for the condition at hand. Instead, they have relied heavily on white male participants, rendering some trial findings not generalizable to non-white male patients who may have different responses to therapeutic products due to variability in intrinsic factors like genetics and disease presentations and extrinsic factors like environmental influence and socioeconomic factors.

By requiring life sciences companies to include more diverse populations, the FDA is forcing the industry to prioritize broader racial and ethnic representation and rethink what future clinical trials and studies might look like.

The need for diversity in clinical trials and studies is not new. The FDA has previously issued guidance documents, such as the 2020 “Enhancing the Diversity of Clinical Trial Populations,” to recommend approaches to improve the inclusion of underrepresented populations in clinical trials. Approaches that have been promoted include broadening trial eligibility criteria using enrichment strategies, which target inclusion of populations with certain characteristics that increase the likelihood of demonstrating the treatment effect, if one exists. The three broad categories of enrichment strategies, as outlined by the FDA in 2019, include:

• strategies to decrease intra- and interpatient variability, such as enrolling patients more likely to provide consistent baseline values or adhere to treatment
• prognostic enrichment strategies such as enrolling high-risk patients more likely to experience the disease endpoint
• predictive enrichment strategies such as enrolling patients more likely to respond to the treatment

Life sciences companies can leverage these as a basis for developing strategies that fulfill diversity requirements.

Leveraging AI and ML to implement diversity action plans

The FDA faces a complex balancing act in its mandate to ensure both broader representation in clinical trials and expedite the delivery of life-saving treatments. Effectively implementing a diversity action plan will require addressing potential bottlenecks that could inadvertently slow the drug approval process. Notably, requiring additional subpopulations to ensure diversity may necessitate expanded sample sizes to allow for sufficient statistical power for stratified analyses. Enrichment strategies, with their aim to reduce nontreatment-related heterogeneity or maximize the treatment effect size to allow for the use of a smaller study population, may offer the key to containing trial sample sizes in face of the diversity requirement.

AI and ML have seen immense growth and adoption in recent years. With the continued proliferation of more mature technologies, these tools offer methodologic solutions to enrichment strategies via patient profiling, predictive modeling, or counterfactual simulation to identify individuals who are more likely to be at high risk, or adherent or responsive to treatment.

Life sciences companies could undertake these efforts now using data from prior trials and real-world data sources that offer good generalizability to inform future trial designs, well before FDORA’s effective date. In fact, as summarized in a May 2023 FDA discussion paper on using AI and ML in drug and biologic development, adoption of these tools in new drug application submissions has risen notably in the past five years. The number of NDA submissions with AI and/or ML components increased nearly 10-fold in just one year, from 14 in 2020 to 132 in 2021.

While AI and ML offer great potential for patient selection, the FDA has emphasized that rigorous validation processes are essential to ensure the accuracy of the underlying predictive models or algorithms. To avoid bias, it is critical to ensure high sensitivity, high specificity, and high predictive values, as well as external validity, when using AI and ML tools. Otherwise, if validity is mediocre or questionable, enrichment strategies fail and the need to expand trial sample size to include all patients would be merited, thereby negating the benefits of enrichment.

As the mandate for instituting diversity action plans approaches, proactive investment in rigorous studies to develop and validate robust AI and ML tools for patient selection is important. Emphasis should also be placed on identifying suitable and representative data source(s) to develop and validate AI/ML tools, ideally in real-world settings where patients may be more reflective of the patient population likely to take the therapeutic product upon approval.

The road ahead

As life sciences companies await the FDA’s final guidance on the requirement for diversity action plans, now is an opportune time to lay the groundwork for enrichment strategies in anticipation of future trials. Specifically, once the diversity composition of a trial population is defined, a company should decide the best enrichment strategy to maximize trial efficiency based on the expected effect of the therapeutic product and the patient population for whom the product is intended. Part of this process involves determining whether the strategy is to minimize variability or to target patients with certain prognoses or likelihood of response.

Once the goal is set, a company must carefully select the appropriate AI/ML methodology and data source(s) and implement them to maximize internal and external validity. Companies can then apply AI and ML solutions and tools in patient recruitment and enrollment strategies to maximize the success of future trials. If implemented properly, AI/ML techniques offer a promising solution to diversifying clinical research, making possible a better understanding of medicine and improved health for people that is the norm rather than the exception.

 

The views and opinions expressed in this article are those of the author and do not necessarily reflect the opinions, position, or policy of Berkeley Research Group or its other employees and affiliates.

To view the original article or to learn more, visit STAT →