Expanding the use of real‐world data through the development of user‐friendly tools to support drug development: Example of an organ impairment dashboard

Real-world data (RWD), and more specifically electronic health record (EHR) data, have gained increasing interest and utilization in the pharmaceutical industry over the past few years, especially given their ability to support decision-making in drug development. From research and development to post-approval, EHR data can help inform various stages and components of the drug development process.1 Although randomized controlled clinical trials have been the gold standard for decades, the use of RWD to complement these studies paves the way for more inclusive clinical trials, since RWD can provide a more representative composition of patients with a certain disease. Additionally, leveraging RWD when designing clinical trials allows researchers to plan for and design studies with some initial insights on the patient population of interest, enabling scientists to curate inclusion/exclusion criteria that are more directed and focused in order to get meaningful results. Recent advances in making RWD/EHR data more accessible include the OHDSI ecosystem, which aims to bring large-scale analytics and access to health data to the broader community in an open-source FAIR format.2 In particular, there have been European efforts through the EHDEN/DARWIN projects to make EHR data more accessible.3 We believe that by putting forward a standardized framework (particularly for a use case specific to clinical pharmacology), we might complement the efforts by the DARWIN project to showcase how EHR data can have an impact in a particular field. Overall, the use of RWD can provide qualified evidence when considering an appropriate scientific question and taking into account meaningful steps to standardize and generalize the data.4 However, utilization of EHR data is often limited and constrained due to various factors such as restrictions on data access (which may be out of scope at an individual scientist level) and the need for coding skills. As an example, to answer more clinically based questions, gaining insights from EHR data can be particularly challenging since it requires individuals to have both clinical expertise (to formulate the question, identify the right patient population, and interpret the results) as well as technical expertise (to understand the data, acknowledge their limitations, and wrangle the data into a more usable format). This can be addressed by developing user-friendly dashboards which enable clinical scientists to interact with complex data and standardize common analysis practices for processing and interpreting EHR data. By removing the need to be proficient in coding or wrangling big data, scientists can query RWD to help address drug development-related questions and support decision-making. Such efforts have led to the development of analytical methods to simulate and assess the impact of clinical and demographic variables on trial enrollment and outcomes, such as in the Trial Pathfinder study.5 We present an analysis workflow that was internally developed as a dashboard; this tool was created to analyze the prevalence of hepatic and renal impairment in real-world patient cohorts as a use case example to demonstrate how to increase accessibility and utilization of RWD to support drug development. Since dedicated organ impairment studies can cost approximately 3–5 million dollars, leveraging RWD to (1) inform study design and (2) understand the need, relevance, and impact a dedicated organ impairment study may have (i.e., by providing insights on the proportion of the target population affected by these comorbidities), can be extremely beneficial. Renal and hepatic impairment are common comorbidities for patients diagnosed with cancer. Since organ impairment can directly affect the elimination of a drug, and thus its exposure, it is important to understand the severity and extent of changes in clearance when renal and/or hepatic are the primary route(s) of elimination. However, these patients are often excluded from pivotal clinical trials, and conducting dedicated studies to evaluate the impact of organ impairment on drug exposure can be challenging due to the paucity of patients. Increasing accessibility to RWD for the evaluation of prevalence and incidence of organ impairment can greatly help in early decision-making, waiving, and/or better designing of clinical pharmacology studies for certain diseases, as seen for both ipatasertib and polatuzumab vedotin; more specifically, for both molecules, RWD was used to understand the prevalence and severity of organ impairment in their respective indications and patient populations, which in turn aided in understanding the need and feasibility to conduct dedicated organ impairment studies.1, 6, 7 In addition, we also want our clinical trials to mirror the real-world population and not exclude patients from trials that could likely benefit from our medicines (advancing our diversity and inclusion goals for trials). Hence, knowing the incidence of organ impairment in the real-world setting, together with the knowledge of the drug's metabolism/elimination pathways, may also present opportunities to include a larger patient population than initially planned. Using the nationwide Flatiron Health EHR-derived de-identified database, a longitudinal database comprising of de-identified patient-level structured (i.e., analysis dataset similar to clinical trial data) and unstructured (e.g., physician notes) data curated via technology-enabled abstraction, we can evaluate the real-world incidence of renal and/or hepatic impairment in oncology patient populations of interest, that is, a target population for a particular drug.8, 9 We developed an easy-to-use dashboard to assess the prevalence of organ impairment. Users are able to create virtual cohorts that most accurately represent their clinical trial population of interest, following which the dashboard categorizes patients based on a predefined renal (Cockcroft https://rstudio.github.io/shinydashboard/), it was easy to implement the backend functions with input toggles that could easily be manipulated by a user in an outward facing interface. Lastly, we incorporated intuitive visuals using "ggplot" (https://ggplot2.tidyverse.org/) and interactive plots using "plotly" (https://plotly.com/) so that the user would easily be able to convey results to their team and interact with the data in real time. We believe that standardizing the analysis workflow through an easy-to-use application reduces the risk of misinterpretation while creating more reproducible results. Overall, this organ impairment dashboard puts RWD and analyses at the fingertips of scientists without requiring any coding skills, allowing for widespread use and adoption of RWD as well as assessing the feasibility and optimizing clinical trial design, informing drug development decisions, and supporting health authority interactions. We believe that developing additional dashboards with multiple RWD sources (i.e., electronic health records and claims data) to answer various drug development questions can be informative and helpful for the scientific community to not only complement randomized clinical trials but also to promote more representative, diverse studies and reduce patient burden. The use of standardized dashboards may also increase the regulatory acceptance of the data generated. We hope this dashboard can serve as a template and be a springboard for other institutions to create their own dashboards, expanding the use of RWD among the scientific community to help supplement drug development. All authors participated in research design; Bianca Vora, Erick Velasquez, Rong Zhang, and Krystian Igras contributed to or performed the data analysis; all authors wrote or contributed to the writing of the manuscript. The authors acknowledge Dale Miles, Benjamin Wu, Rui Zhu, Rucha Sane, Scott Pivirotto, Chunze Li, and Amita Joshi for their helpful feedback and discussions during the development of this dashboard and writing of this manuscript. No funding was received for this work. All authors are employees and stockholders of Roche/Genentech, Inc. None. The data that support the findings of this study were originated by and are the property of Flatiron Health, Inc., which has restrictions prohibiting the authors from making the data set publicly available. Requests for data sharing by license or by permission for the specific purpose of replicating results in this manuscript can be submitted to email protected. Data S1: Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.