The thoughtful integration of new and powerful emerging technologies, including digital health tools, artificial intelligence (AI) and machine learning, new modalities, throughout the drug R&D process offers opportunities to address critical barriers and streamline clinical trials for prevalent chronic diseases.
Qi Liu, senior science advisor in the Office of Clinical Pharmacology and Translational Sciences at FDA, offered her perspective on how innovations—new modalities, analytic tools, and new sources of data—that have enabled drug R&D and led to new treatment options for cancer patients could be applied to other chronic disease areas. Grace Colón, CEO of InCarda Therapeutics, provided three case studies that illustrated the value of thinking about technology in a more holistic way. John Ngai, director of NIH’s Brain Research through Advancing Innovative Neurotechnologies (BRAIN) Initiative, discussed developing a collection of new tools for exploring and understanding the brain, many of which could be applied to prevalent chronic diseases involving the brain. Andrew Radin, co-founder and CEO of twoXAR Pharmaceuticals (renamed Aria Pharmaceuticals since the time of the workshop) spoke about AI applications to enable drug discovery for chronic diseases. Elizabeth Kunkoski, health science policy analyst with FDA’s Office of Medical Policy clinical methodology team, offered a regulatory perspective on the development and use of health technologies in clinical research.
FDA focuses primarily on drug development and regulation and does not play a major role in drug discovery. However, evaluating drug products has given Liu a unique perspective on how innovations in drug discovery
have led to new treatment options for improving public health. She shared a few examples from oncology that may offer applicable lessons for other prevalent chronic diseases. Liu categorized new innovations into three categories: (1) therapeutic modalities, (2) types of data, and (3) analytical tools.
New Therapeutic Modalities
Over the past decade, there has been significant growth in the development and use of targeted therapies, which Liu described as new molecular entities that are intended for a subset of patients who are identified through molecular testing. Biologics, particularly monoclonal antibodies, represent a fast-growing class of targeted therapy. As an example, Liu pointed to immune checkpoint inhibitors, which she said have been transformative in the treatment of patients with cancer and have become an increasingly important part of cancer treatment. She noted that James Allison and Tasuku Honjo won the 2018 Nobel Prize in Physiology or Medicine for their pioneering work in this field.
Liu also highlighted bispecific antibodies and antibody–drug conjugates as important innovations for the treatment of cancer. In 2009, only one antibody–drug conjugate had been approved by FDA (Joubert et al., 2020). In 2020, nine products had approval, and several others were in the pipeline. The first FDA approval of a bispecific antibody came in 2014—blinatumomab was approved for use in the treatment of acute B cell lymphoblastic leukemia1—and many others are under clinical development.
Liu mentioned several other therapeutic modalities, including cell-based therapy, oligonucleotide-based therapy, microbiome-based therapy, and viral therapy. She emphasized that “today’s breakthroughs in oncology are the result of decades of investment in cancer research and drug development.” In 1971, President Nixon signed the National Cancer Act,2 and in 2016, Congress passed the 21st Century Cures Act,3 which authorized $1.8 billion in funding for the Cancer Moonshot4 over 7 years. Industry has also invested heavily in oncology, which is now the largest pharmaceutical therapeutic area. Liu suggested a few reasons for this success: (1) basic research in cancer has led to a number of promising drug candidates; (2) there are strong financial incentives for investment; and (3) FDA’s regulatory strategy for cancer drugs has been flexible and innovative.
2 National Cancer Act of 1971, Public Law 92-218, 92nd Cong. (December 23, 1971).
3 21st Century Cures Act of 2016, Public Law 114-255, 114th Cong. (December 13, 2016).
4 For more information, see https://www.cancer.gov/research/key-initiatives/moonshot-cancer-initiative (accessed July 16, 2021).
New Types of Data
Integrating multi-omics data, which broadly cover genome, epigenome, transcriptome, proteome, metabolome, and microbiome data, can be a powerful way to study complex diseases, said Liu. She highlighted a new type of -omics data, radiomics data, which she described as information extracted from medical images using algorithms. Radiomics has the potential to improve disease detection, characterization, and assessment, as well as prediction of treatment response.
Liu then discussed the opportunity and obligation to derive real-world evidence from real-world data5—information relating to patient health status and/or the delivery of health care routinely collected from a variety of sources, including, but not limited to, electronic health records, claims and billing activities, product and disease registries, and patient-generated data—such as data from smartphones and wearables. She noted that in 2018, FDA issued a framework for the use of real-world evidence in regulatory decision making (FDA, 2018).
New Analytical Tools
Given the increased volume and types of data now available, Liu stated, “New analytical tools are needed to transform big data into smart decisions.” She described a few examples of analytical tools that can play important roles in drug discovery and development:
- Model-informed drug development (MIDD),6 which Liu described as the “application of exposure-based, biological, and/or statistical models, derived from preclinical and clinical data sources to address drug development and/or regulatory issues.” She listed a few modeling approaches, such as physiologically-based pharmacokinetic models, quantitative systems pharmacology models, and quantitative structure–activity relationship models. As an example, Liu pointed to an application of MIDD to examine the in vitro antiviral activity of hydroxychloroquine to in vivo concentrations in order to predict how effective the drug might be in treating COVID-19 in humans. The study, which employed physiologically-based pharmacokinetic modeling and simulation, concluded that at normal doses, hydroxychloroquine was unlikely to achieve a high enough concentration to have an antiviral effect in humans
5 For more information, see https://www.fda.gov/science-research/science-and-researchspecial-topics/real-world-evidence (accessed July 16, 2021).
6 For more information, see https://www.fda.gov/drugs/development-resources/model-informed-drug-development-pilot-program (accessed July 16, 2021).
- AI and machine learning offer numerous applications for better understanding diseases and drug targets, generating and evaluating drug candidates and combination therapy, improving clinical trial design, and advancing precision medicine by improving diagnosis and treatment.7 Liu shared a few examples of submissions with machine learning components received by FDA’s Center for Drug Evaluation and Research, which included applications for predicting drug response based on baseline factors, identifying predicated biomarkers for drug response, and identifying drug abuse–related problems in postmarket settings.
(Fan et al., 2020). The paper was published in May 2020, and on June 15, FDA revoked its emergency use authorization for both hydroxychloroquine and chloroquine, referring to this MIDD analysis as one of the reasons for the decision (FDA, 2020a).
Data sharing will be a key factor in encouraging innovation in prevalent chronic diseases, Liu said. Despite the vast amount of data generated, scientific breakthroughs have been relatively limited for most prevalent chronic diseases. The use of new analytic tools can help translate those data into smart decisions. “Investments from government and industry are needed for scientific breakthroughs. Data sharing and precompetitive collaboration can potentially benefit everyone, especially the patients. In the pursuit of innovation, if we want to go far, we need to go together,” she said.
Colón offered three case studies illustrating ways innovation could be considered in a more holistic way, integrating various approaches in a value-based way to improve outcomes for patients: atrial fibrillation (AF), idiopathic pulmonary fibrosis (IPF), and COPD.
AF, one of the most commonly diagnosed types of heart arrhythmias, impacts an estimated 6 million people in the United States alone (Morillo et al., 2017). It is a progressive chronic disease that is associated with increased morbidity and reduced quality of life in some patients. AF is esti-
7 A more in-depth discussion on the applications of AI and machine learning in drug development is presented later in this chapter.
mated to cost the U.S. health care system more than $50 billion per year,8 she said, if all expenses are included—therapeutics, health care usage, and invasive surgery.
The more a patient experiences AF, the more likely and the more extensive further episodes become. Yet, there is no good treatment option to stop an episode quickly, Colón explained. A number of antiarrhythmic drugs are available, but these have limitations: Some must be administered intravenously and others may take hours before they take effect. Some drugs taken orally are available for suppression, but may only work at high levels, so patient compliance is poor, and there are associated safety and tolerability issues.
InCarda Therapeutics is developing an inhaled version of a well-known drug, flecainide, which has the potential of stopping AF within a few minutes after an 8-minute inhalation.9 The treatment is being studied in a medically supervised setting to verify flecainide response, but Colón suggested that if this medical product is approved for broader settings (e.g., work, home, travel), it could offer patients more options for determining their course of treatment. She added that the effectiveness of this intervention could be increased by combining it with the use of digital health technologies for disease monitoring. For example, a device that detects abnormal arrhythmia, coupled with confirmation from a patient’s physician, could help facilitate increased awareness of AF and give the patient more control over their own health.
Idiopathic Pulmonary Fibrosis
IPF is a progressive, debilitating, and fatal lung disease. Although groundbreaking treatments have been approved in the past few years, most notably pirfenidone and nintedanib, these drugs have significant side effects, which can limit patient adherence. Colón pointed to the work of Avalyn Pharma, which is working on an inhaled version of pirfenidone.10 Early studies in animal models have shown promising results, suggesting that small inhaled doses of pirfenidone can deliver therapeutic levels of the drug in lung tissue. Although the results are preliminary, she said this example “highlights the innovative use of an existing drug with a new modality to change the paradigm and potentially help these patients significantly with a much lower systemic dose and even improved compliance.”
9 For more information, see https://incardatherapeutics.com/2021/01/19/incarda-therapeutics-announces-enrollment-of-first-u-s-patient-in-phase-2-instant-trial-of-inrhythm-for-treatment-ofatrial-fibrillation (accessed July 16, 2021).
Chronic Obstructive Pulmonary Disease
COPD is a complex, multimodal disease associated with many comorbidities. There are no good single biomarkers that can help diagnose or manage COPD. Yet, diagnosis and management are important given that numerous triggers (e.g., viral infection, allergies) can lead to hospitalization. To address this gap, ProterixBio—of which Colón is the executive chair—has developed a disease activity score that uses an algorithm that takes into account a variety of inflammatory and immune response biomarkers measured against known patient populations.11 Colón suggested that this approach has the potential to enable clinical trials for COPD and help physicians and health care systems better understand and monitor their patients over time.
Colón envisioned a future in which partnerships among different companies could integrate technologies for measurement, treatment, patient coaching, and interventions and apply these approaches toward prevalent chronic conditions, such as respiratory disease. New technologies offer opportunities to consider the overall patient journey and treatment paradigms, she said. Alternative routes of delivery and new modalities for existing drugs are two areas of promise. The use of novel biomarkers and integrating multiple biomarkers could lead to new endpoints and perhaps combination endpoints. These approaches could be integrated with digital health tools that provide better patient-reported outcomes and patient quality metrics. Holistic solutions “will require an integration of science, engineering, and medicine to address these problems together,” Colón said. She concluded that these holistic approaches must take into account disparities in access to care and clinical trials by working to remove financial and transportation barriers so participants can enroll and stay in clinical trials.
The BRAIN Initiative seeks to revolutionize the understanding of the human brain by accelerating the development and application of innovative technologies.12 This collaborative public–private partnership includes federal agencies, private-sector companies, nonprofit organizations, foundations, and academic institutions. Ngai laid out seven key areas of focus for the initiative:
- Discovering the diversity of cell types in the mammalian brains and other brains in order to “create a parts list”;
- Creating maps at multiple scales, or “wiring diagrams”;
- Developing technology to monitor neural activity;
- Developing technology to modulate neural activity;
- Identifying fundamental principles in order to create a solid grounding for the work in theory and data;
- Creating human brain research networks; and
- Integrating the various approaches.
Ngai emphasized that the tools and resources developed by the BRAIN Initiative can serve as a foundation for basic discovery, leading to clinical research and, ultimately, clinical use. Work by the BRAIN Initiative is already helping to identify molecules, cells, and circuits affected in neurologic and neuropsychiatric disorders, he said, and pointing the way to interventions such as deep brain stimulators, sensory and motor neural prostheses, and targeted molecular and gene therapies.
Ngai suggested a few resources that could help support innovation in other fields, such as prevalent chronic diseases. For example, the BRAIN Initiative Cell Census Network13 aims to create comprehensive brain cell atlases that integrate molecular analyses, connectivity, physiology, and other data for the mouse and other mammalian species, including humans and non-human primates. He suggested that the Cell Census Network may serve as a useful source of information for researchers seeking new cures for human brain disorders and other conditions.
The initiative already offers a large variety of resources that can be used in brain research. Ngai mentioned three brain cell data repositories: the Neuroscience Multi-Omic (NeMO) data archive for transcriptomic, epigenomic, and other omic information14; the Brain Image Library, with microscopy and other imaging information15; and the Distributed Archives for Neurophysiology Data Integration (DANDI), which offers neurophysiology information.16 There are also repositories of analytical tools that allow the general research community to use the information from the data archives, including for disease-specific projects.17
Finally, Ngai spoke about the importance of carefully considering the ethical, legal, and societal implications for the work of the BRAIN Initiative. The Initiative’s Neuroethics Working Group18 has developed a set
13 For more information, see https://braininitiative.nih.gov/brain-programs/cell-census-network-biccn (accessed July 16, 2021).
18 For more information, see https://braininitiative.nih.gov/about/neuroethics-working-group (accessed July 16, 2021).
of neuroethics guiding principles, which deal with issues around safety and privacy as well as agency, malign, and dual use (Greely et al., 2018). Ultimately, he said, it is important to be mindful, responsive, and engaged with the public. He emphasized the need to make technologies more accessible to all people and engage communities that traditionally have been underserved and underrepresented as true partners so that the work of the BRAIN Initiative can benefit all.
Looking forward, Ngai said it has been challenging to find ways to share work generated from the BRAIN Initiative in ways that are broadly accessible and useful to researchers, but there are efforts to make data more accessible to researchers, including those working in drug R&D. Additionally, he mentioned that the BRAIN Initiative is developing disease-agnostic tools that could be applied more generally to other disease areas. “What we are hoping to provide is a framework upon which we can hang the disease-specific projects,” he said.
While AI may seem new and exciting, Radin reminded workshop participants, this technology has been around for decades. In recent years, computing power and data storage have improved dramatically and been available at lower costs. Algorithms, such as deep learning and neural networks, have enabled more accurate predictions based on real-world data (e.g., weather predictions). Radin suggested thinking about AI as a tool one can use to predict an event using real-world data. Just as AI might be used to make weather predictions based on temperature, pressure, humidity, and cloud cover, this technology can be applied to many other domains, including drug discovery. Similar to weather prediction, Radin said, the process of finding a drug to treat a specific condition involves a number of uncertainties: Will the molecule reach the desired target in the body? Will the molecule interact with other targets that could lead to side effects? Does the animal model characterize well what will happen in a human? AI has potential applications for all of these questions and many more.
Radin’s company, which is now known as Aria Pharmaceuticals, uses AI applications to identify candidate first-in-class small molecule treatments—drugs that use novel mechanisms of action—for complex diseases. There is an opportunity, Radin argued, to address unmet medical need by better understanding the biology of disease with the use of computational modeling systems.
As an example, Radin shared work the company has done on lupus, a disease that affects between 161,000 and 322,000 people in the United States, according to a 2008 estimate (CDC, 2018). Given that lupus is an
inflammatory disease, the typical initial treatment for lupus is the administration of non-steroidal anti-inflammatory drugs, which are safe, but not particularly effective for most lupus patients. Radin said many of the more effective drugs are also more toxic. For example, cyclophosphamide, which may be given to lupus patients with severe kidney inflammation, has a number of severe side effects, including infertility, birth defects, and blood clotting. He pointed to the need for a drug that is effective and well tolerated—likely a drug that has a different mechanism of action than drug treatments that are currently available.
Radin’s company used AI applications to screen for small molecules with new mechanisms of action (see Figure 3-1). The first step, AI-driven discovery, was computationally intensive. The researchers characterized key biological features of lupus and built an in-silico model of the disease, based on 25 different orthogonal data sources (e.g., gene expression, protein expression, clinical data, and phenotypical data). These data were then processed by more than 60 methods to identify key features of the disease that could be mapped onto a small molecule library. Out of a small molecule library of 2 million compounds, 50,000 molecules were preliminarily screened as appropriate for the project, and the AI algorithm identified 3,000 molecules with the highest predicted efficacy out of that subset. Using AI-assisted review, researchers then screened the 3,000 molecules, examining the mechanism of action, safety profile, and other properties. This left 80 candidate molecules, which were then examined based on likelihood
to be effective against the disease or have serious side effects. Intellectual property issues were also taken into consideration. This left nine molecules for testing in standard preclinical efficacy models. Radin said two of the molecules, TXR-711 and TXR-712, showed improvements in organ function and decreased inflammation in standard in vivo mouse models.
Radin highlighted the power of AI for advancing drug discovery. From project launch to the identification of nine candidate small molecules for screening took about 4 weeks. The preclinical testing took about 3 months. He pointed out that without the use of AI applications, this work could take years. He added that this approach was able to identify more candidate molecules than traditional approaches, which typically require far more in vivo testing to find one or two candidates.
Radin’s company has had similar promising results across its portfolio of 18 diseases, including fibrotic diseases, immunoinflammatory conditions, and oncology. He emphasized that this work may not only lead to the discovery of new treatments, but it can also help the scientific community gain new insights into the biology of these disease areas.
Kunkoski offered a regulatory perspective on the use of digital health technologies in clinical research. Digital health technologies use computing platforms, connectivity, software, and sensors for health care and related uses (FDA, 2020b). Technologies include passive measuring devices, such as accelerometers, glucometers, and electrocardiograms, as well as more interactive tools, such as mobile phone apps and smart watches. She suggested that the possibilities for data collection through the use of digital health technologies are endless and can make it possible to find the right drug at the right time for a patient.
One of the first steps for using a digital health technology in a clinical trial is verification and validation of the technology, which is intended to ensure that the data collected during the study are reliable. The verification step examines how accurate or precise the technology is in collecting data. In the case of an accelerometer, for instance, it might be important to assess how accurately it measures acceleration, if it is reliable in different environments, and whether readings are affected by the placement of the device. The validation step checks that the technology works as intended in the field when used by real people.
Speaking on the use of digital health technologies in clinical trials, Kunkoski listed a few potential applications. For example, digital tools could be used to remotely monitor a patient’s response to a drug over a dosing cycle, which could be particularly relevant for chronic diseases. Capturing
continuous or frequent measurements over time can provide valuable longitudinal information compared to cross-sectional data that clinicians might gather during a periodic clinic visit. “All of this means potentially fewer in-person visits for [research participants] enrolled in clinical trials,” she said, “which means less time taking off of work, traveling less to research centers, which will enable more participation and greater retention in clinical trials.”
For researchers planning to use a digital health technology in a clinical trial, Kunkoski said, the technology does not necessarily need to be approved by FDA or cleared for marketing. Instead, the main focus is whether the technology can provide sufficient high-quality evidence for FDA to be able to draw conclusions about the safety and effectiveness of the therapeutic intervention being studied.
After verification and validation, another important step in preparing a digital health technology for use in a clinical trial is formulating an endpoint. The researcher must specify the clinical characteristic of the event being measured for each research participant and how the measurements will be used to assess the endpoint. Will that endpoint be, for instance, a single measurement, such as blood pressure, or the result of repeated measurements over a certain cycle, such as 24-hour ambulatory blood pressure monitoring? There can also be composite endpoints, she said, such as a combination of patient-reported outcomes and actigraphy measurements. Part of formulating the endpoint is determining the timeframe. Should it be steps in 1 hour? One day? One week? Or should it be a weekly progression over time? How should the response variable be reported—change from baseline, mean value, peak value, number of events, time to event?
Next, Kunkoski listed a number of factors that should be taken into account when selecting a digital health technology for a research study. The study population makes a difference, she said, observing that for many chronic diseases, there are elderly populations that may find it challenging to use the technology effectively. Other considerations include the study design, safety monitoring, technical support, and available training. “If you do not properly train the study personnel, investigators, and patients upfront,” Kunkoski said, “you are not going to be able to get the data to make your trials successful.”
As a case example, Kunkoski pointed to the use of digital health technologies with an accelerometer, gyroscope, and GPS to monitor how much a patient is moving after a hip replacement surgery (Bini et al., 2020). She suggested that the potential for real-time feedback may have the most promise given that these types of digital health technologies can provide the ability to engage patients in their recovery and allow them to track their own progress day by day.
Kunkoski reiterated that digital health technologies make it possible to collect many types of data remotely. They allow for broader participation
and retention in clinical trials, and for more continuous data collection. “Now is the time for innovative thinking,” she said. “We are excited about the potential for incorporating these technologies in clinical research moving forward.”
Bini, S. A., P. L. Schillling, S. P. Patel, N. V. Kalore, M. P. Ast, J. D. Maratt, D. J. Schuett, C. M. Lawrie, C. C. Chung, and G. D. Steele. 2020. Digital orthopaedics: A glimpse into the future in the midst of a pandemic. Journal of Arthroplasty 35(75):S68–S73.
CDC (Centers for Disease Control and Prevention). 2018. Systemic lupus erythematosus. https://www.cdc.gov/lupus/facts/detailed.html (accessed July 16, 2021).
Fan, J., X. Zhang, J. Liu, Y. Yang, N. Zheng, Q. Liu, K. Bergman, K. Reynolds, S. M. Huang, H. Zhu, and Y. Wang. 2020. Connecting hydroxychloroquine in vitro antiviral activity to in vivo concentration for prediction of antiviral effect: A critical step in treating patients with coronavirus disease 2019. Clinical Infectious Disease 71(12):3232–3236.
FDA (U.S. Food and Drug Administration). 2018. Framework for FDA’s real-world evidence program. https://www.fda.gov/media/120060/download (accessed July 16, 2021).
FDA. 2020a. Coronavirus (COVID-19) update: FDA revokes emergency use authorization for chloroquine and hydroxychloroquine. https://www.fda.gov/news-events/press-announcements/coronavirus-covid-19-update-fda-revokes-emergencyuse-authorization-chloroquine-and (accessed July 16, 2021).
FDA. 2020b. What is digital health? https://www.fda.gov/medical-devices/digital-healthcenter-excellence/what-digital-health (accessed July 16, 2021).
Greely, H. T., C. Grady, K. M. Ramos, W. Chiong, J. Eberwine, N. A. Farahany, L. S. M. Johnson, B. T. Hyman, S. E. Hyman, K. S. Rommelfanger, and E. E. Serrano. 2018. Neuroethics guiding principles for the NIH brain initiative. The Journal of Neuroscience 38(50):10586–10588.
Joubert, N., A. Beck, C. Dumontet, and C. Denevault-Sabourin. 2020. Antibody-drug conjugates: The last decade. Pharmaceuticals (Basel) 13(9):245.
Kim, M. H., S. S. Johnston, B.-C. Chu, M. R. Dalal, and K. L. Schulman. 2011. Estimation of total incremental health care costs in patients with atrial fibrillation in the United States. Circulation: Cardiovascular Quality and Outcomes 4(3):313–320.
Morillo, C. A., A. Banerjee, P. Perel, D. Wood, and X. Jouven. 2017. Atrial fibrillation: The current epidemic. Journal of Geriatric Cardiology 14(3):195–203.
Mullard, A. 2015. FDA approves first bispecific. Nature Reviews Drug Discovery 14(1):7.
This page intentionally left blank.