Once the patient population (biology, genetics, demographics) with the disease is understood, Principle 2 anchors on “Where” these patients are (Supplemental Fig. 2). The FDA issued draft guidance for the creation of DAPs by sponsors of pivotal studies to increase enrollment of historically underrepresented populations to improve the strength of the evidence for the intended-use population [11]. Roche has submitted over 30 DAPs to the FDA for 23 therapies across oncology, ophthalmology, neurology, autoimmune, and infectious diseases. Processes, procedures, and platforms have also been developed to monitor and track diverse enrollment.

Not all countries have adequate infrastructure and expertise to conduct clinical trials. Therefore, established regional frameworks can be used to advance diverse trial enrollment globally. The International Council for Harmonisation (ICH) Guideline on General Principles for Planning and Design of Multi-Regional Clinical Trials (E17) provides a framework for multiregional clinical trials to support global development and regulatory approvals. It highlights that enrollment should be proportional to the geographical burden of disease, to maximize regional representativeness within operational constraints [12].

Preferred sources, such as the US Cancer Statistics (USCS) and the Institute for Health Metrics and Evaluation (IHME) Global Burden of Disease study, should be identified based on data quality, representativeness, and disease area coverage to provide epidemiology estimates at the indication or disease area level [6, 8, 9, 13]. These sources should provide incidence or prevalence of the disease by region to determine the global burden of disease, as well as age and sex distribution to set overall global targets, and race and ethnicity distribution in the USA to set the US targets. A limitation of these sources is that they provide information about epidemiology estimates at the indication level with limited information about subpopulations, e.g., biomarker-driven subsets, stage, and evolution beyond the initial diagnosis. Different epidemiology estimates may be needed if the diversity metric varies by these factors. These estimates can come from real-world data sources or literature reviews. Therefore, augmenting multiple data sources may be required to generate epidemiologic estimates that closely approach the trial population, or the development plan may require triangulation, using two disparate data sources to generate an epidemiology estimate.

To measure representativeness, Roche uses the participation-to-prevalence ratio by dividing the percentage of a particular population among the study participants (e.g., percentage of women among participants in a lung cancer trial) by the percentage of the particular population in the disease population (e.g., percentage of women in the USA with lung cancer). A range of 80–120% has been suggested as adequate representation, where each study should aim for zero error between these proportions [14, 15].

Country-specific enrollment goals should adhere to standardized local requirements and frameworks. Diversity outside the USA cannot be captured or measured by US-specific demographic frameworks alone, such as the Office of Management and Budget’s (OMB’s) standards on race and ethnicity [16], as ex-US patients may not identify to the OMB categories and/or concepts of race/ethnicities are not globally relevant. The use of regional and/or country-specific enrollment by demographic variables ensures that trial results are reflective of countries’ diversity, as well as how these concepts are recorded per national statistical offices or census bureaus. For the USA, population health data licensed from the IHME or USCS can be used to set within-country enrollment targets for demographic groups [6, 13]. Assessments of accessibility should be formalized to ensure efficient and diverse patient enrollment as well as codifying data-informed tactics for enrolling representative populations. Outside of the USA, the site selection process is guided by country-specific disease prevalence estimates. For example, phase 3 studies for chronic obstructive pulmonary disease (COPD) replicated the COPD global disease burden and epidemiology data, and 35 countries were selected as study sites accordingly (Fig. 3). This included low- and middle-income countries (LMICs), such as China, Kenya, and the Philippines, to enroll patients according to unmet needs and with the greatest burden (NCT05595642).

Country-specific needs are also very different in LMICs versus high-income countries, and this framework may not be clearly defined. To have a global impact/implications, metrics need to be defined to assess gaps between epidemiological data and operational feasibility amongst countries. Building capability before clinical trials in partnership with local stakeholders (such as patient networks, non-governmental organizations, ministries of health, and community navigators) can help enable future clinical trial access and inclusive research, as was done with the Ghana Women’s Health Project [17]. When considering these countries, not for profit, but for health equity, a regulatory path for drug approval and sustainable access must also exist.

There are data-informed opportunities to ensure clinical trials are representative of real-world patient populations. Data sources such as the US Census, USCS, the Surveillance, Epidemiology, and End Results Program, IHME, electronic health record-derived real-world data (e.g., TrinetX and Flatiron), and genomic databases are available for epidemiological estimates of US and global populations [18,19,20,21,22]. The representativeness index (R-index) has been proposed as a standardized metric for reporting representativeness of race, ethnicity, and other diversity-related attributes of clinical trial participants, regardless of therapeutic area and trial phase [23]. While no one data source is perfect, key factors to consider are data access, population and geographic coverage, demographic characteristics, and availability of clinical and genomic details.

Study site placements should also be based on operational capabilities, regulatory considerations, demographic data, and disease burden. Methods include using health systems, small area demographics, disease burden, and operational performance data to optimize representativeness and efficiency. Pre-site activation surveys can be coded with questions that assess the baseline state of sites’ advancing inclusive research (AIR) commitments and involvements and guide site selection decisions. This approach allows discrepancies in enrollment metrics to drive local insights and hypothesis generation. Following data-driven processes for site placement ensures that representative trials are not managed at the expense of efficiency and allows discrepancies in enrollment metrics to drive local insights and hypothesis generation. For instance, all else being equal in terms of population and disease burden data, a study site that enrolls slowly compared with a peer must have factors that may merit investigation, such as socioeconomic barriers that have yet to be identified, organizational limitations, or otherwise. If data suggest that a site takes more time to start up or is slow to enroll, one must uncover if it relates to site-based inefficiencies, such as less infrastructural support, or enrollment challenges with historically under-represented patients because of justified mistrust due to systemic racism or limited English proficiency. The former might be a site that would not be selected, the latter should be chosen and provided with appropriate resources, such as the Roche Site Academy, a sponsor program that offers core training to sites that are new to research and provides access to clinical trial equipment and inclusive research training for established sites [24, 25]. Sixteen global sites have utilized this program, including in Africa and India. Data-driven site placement also provides quantitative evidence of efforts by a sponsor if study targets are not met at the time of the last patient enrolled and can lead to the activation of additional resources to address barriers and trial support.

Even if not all countries have adequate infrastructure and expertise to conduct clinical trials, the aim is for enrollment to be proportional to the geographical burden of disease and to maximize regional representativeness within operational constraints. There is also the ethical obligation to only conduct clinical trials in countries with a path to regulatory approvals. To increase the operational capacity of LMIC countries to participate in clinical trials, and accounting for this later policy, Roche-Genentech developed a program called AIR Site Alliance in sub-Saharan Africa and India, intending to expand to other geographies. The AIR Site Alliance program enabled a significant increase in the number of clinical trials conducted in selected LMIC countries [26].