Clinical Researcher—June 2023 (Volume 37, Issue 3)

PEER REVIEWED

Brian Mai; Andrea Roman, MS; Alondra Suarez, MS

 

Streamlining pain points of the clinical trial process to reduce costs and improve patient results can be accomplished with artificial intelligence (AI). For example, AI can support patient recruitment and retention by seeking potential participants and foreseeing the probability of subject withdrawal. Patient monitoring can be improved by collecting real-time data using wearable devices and sensors on the patient’s physiological parameters. These data can be analyzed using AI to identify patterns indicating the start of potential adverse events or complications. Furthermore, AI can help data quality by lessening the effects of confounders and expanding the scope of use for AI wearable technology.

Further, deep learning models can be trained to continuously analyze and interpret patient data for individual patients and across cohorts. This allows researchers to uncover patterns that may otherwise have been unrecognizable.

In this paper, we examine how, although there are challenges and drawbacks to implementing AI in clinical trials, it has potential benefits that make it a viable tool for the pharmaceutical industry.

Background

AI has become increasingly prevalent in the healthcare industry, with one area of application being in clinical trials for pharmaceutical products, which are time consuming, expensive, and labor intensive. Indeed, clinical trials require a large investment of resources, and it takes years to bring a drug to market.

Part of the issue is due to the frequent failure of clinical trials, which, when they occur, come extremely late in the overall development cycle for a drug. Only around 10% of drugs entering the clinical trial stage eventually receive U.S. Food and Drug Administration (FDA) approval. The failures of the other 90% are often ascribed to poor patient cohort selection, recruiting tactics, and insufficient infrastructure to support complex clinical trials.{1}

In such a challenging environment, AI can be leveraged to streamline various aspects of the clinical trial process, such as patient recruitment, data analysis, pattern recognition, and identification of potential adverse events. Researchers could expedite the drug development process, potentially reducing costs and improving patient outcomes. However, as with any new technology, there are hurdles to overcome and drawbacks to be faced when implementing AI in clinical trials.

An Overview of AI

AI is a branch of computer science that aims to create intelligent machines that can perform tasks that typically require human intelligence, such as visual perception, speech recognition, and decision making. One of the critical components of AI is machine learning, a subset of AI that involves algorithms and statistical models to enable machines to learn from data and improve their performance over time.

Deep learning is a subfield of machine learning that uses artificial neural networks with multiple layers to learn complex patterns in data. Neural networks are a set of algorithms that are modeled after the structure and function of the human brain. They consist of layers of interconnected nodes or neurons that process information and make predictions based on input data. Neural networks are commonly used in deep learning applications, and can be trained to perform various tasks, including image and speech recognition, natural language processing, and decision making.{1}

Current Uses of AI in Clinical Trials

AI can improve patient recruitment by identifying and screening potential participants based on inclusion and exclusion criteria. This can help reduce the time and cost associated with patient recruitment, which is a common clinical research issue.

Once the trial is under way, patient monitoring can be improved by incorporating the use of AI, which in turn can help improve patient safety and reduce the risk of adverse events. For example, AI algorithms can detect and predict adverse events by analyzing various data types, such as vital signs and patient-reported outcomes. This can help researchers identify potential safety concerns more quickly and take appropriate action, such as modifying the study protocol or adjusting the dosage of the drug being tested.

Additionally, wearable devices and other sensors can collect real-time data on patients’ physiological parameters, which can be analyzed using AI to identify patterns indicating the onset of a potential adverse event or complication. In short, AI can help improve patient monitoring during clinical trials, leading to better patient safety and more efficient drug development.

Meanwhile, the majority of clinical trials experience some level of subject dropout. There are various ways that AI could improve subject retention. This could involve using AI to identify factors associated with a high risk of patient dropout. AI models could be used to predict the probability of subject dropout, which would allow researchers to be proactive in subject outreach. This could significantly decrease the resources and time restraints associated with clinical trials. Furthermore, wearable technologies, such as an Apple Watch or Fitbit, combined with AI, could improve both subject retention and monitoring.

It is important to note that clinical trial participants are not in a completely controlled environment. Studies must be planned with factors such as loss to follow-up and variability of self-reported patient data in mind. AI allows researchers to minimize the effects of such confounders and improve data quality.

Researchers could combine AI with wearable technologies to simplify self-reporting data collection, as patients would only have to wear the technology that gathers the appropriate biological data rather than manually collecting the data. This could expand the scope of use for wearable technologies in clinical trials as AI could be trained to analyze this data in real-time and improve adherence to the study protocol. Offering another way to mitigate issues related to data reliability will help ensure the collection of complete datasets. Of the various AI methods, deep learning can be used to analyze data collected by wearable technologies and diagnostic devices.

AI Software and Tools

Deep learning is a class of machine learning methods based on artificial neural networks that mimic information processing and distributed communicated nodes in humans. The neural networks use multiple layers to extract higher level features from input progressively.{2} Deep learning models can be trained to continuously analyze and interpret patient data for an individual patient and across cohorts, while automatically adjusting to disease expression and treatment response changes. This tool will allow researchers to learn from complex datasets and uncover patterns that may otherwise have been unrecognizable.

In a 2015 study, Shah, et al. evaluated the efficacy of clinical outcomes generated from technology-enabled non-invasive diagnostic screening (TES) using smartphones and other point-of-care medical sensors versus conventional vital signs examination. TES synergistically identified clinically significant abnormalities in subjects who presented as usual in routine health screenings. Physicians verified TES findings and used routine health screening data and medical history responses for comprehensive diagnoses for at-risk patients. The researchers concluded that, while routine health screening continues to be necessary, the emerging techniques of TES can play an essential supporting role in the early detection of disease, continuous monitoring throughout clinical trials for adverse events, and providing personalized screening and care to support clinical trials.{3}

A second study evaluated an AI system that continuously analyzes arterial pressure waveform during surgery and warns if hypotensive events are expected within the next 15 minutes. The researchers concluded that this study demonstrated that using AI compared to standard care resulted in less intraoperative hypotension.{4}

These two studies demonstrate the applicability of AI methods for real-time analysis of clinical data and for early detection of adverse events in clinical trials. While physicians must still provide standard patient care, AI can help physicians and researchers detect the onset of abnormalities and adverse events much sooner. Thus, AI methods offer a cost-effective and rapid solution to improve patient outcomes in clinical trials.

AI’s language compatibility is an additional tool for scouring data across the web quickly thanks to natural language processing (NLP). Like human comprehension, a device can be programmed to understand written or spoken words. Within the medical practice, this can be used to read through physicians’ comments and pathology reports to determine if the patient meets the eligibility requirements to enter a specific clinical trial.

With the help of NLP, researchers devised an AI tool called Criteria2Query, which standardizes inclusion and exclusion criteria within databases and allows professionals to gather information simplistically without needing extensive context. Another AI creation, made by the same researchers, involves searching from the patient’s perspective.{5} ClinicalTrials.gov can be a daunting database for those unfamiliar with what to look for in a trial. Hence, DQuest, another AI NLP tool, generates a series of dynamic questions for patients to answer and then filters their options based on the responses. While the accuracy could improve, an initial study showed that it could exclude 60% to 80% of trials for which the patient was not eligible. Allowing patients to select their trials could increase satisfaction and retention rates.

HIPAA, Patient Privacy, and Patient Rights

With the Health Insurance Portability and Accountability Act (HIPAA), patients have the right to privacy and control over their personal health information (PHI), including the right to access their personal health files, petition for corrections, and be informed about how their PHI is disclosed and used.{6} Researchers and healthcare organizations must comply with HIPAA regulations and as well as other privacy laws to ensure patients’ PHI is protected.

Utilizing AI technology in patient recruitment can lead to more challenges in ensuring compliance with these regulations. For example, AI algorithms may need to access a large amount of patient data to identify appropriate clinical trials. However, these data must be de-identified and protected to ensure patient privacy and prevent unauthorized access to PHI. Additionally, third-party organizations in AI-driven patient recruitment have the potential to create other risks to patient privacy and compliance with privacy regulations.

Healthcare professionals and researchers must prioritize compliance with HIPAA and other privacy regulations to ensure that patient data are de-identified and protected, as well as put appropriate safeguards in place to prevent unwanted access to PHI. Lastly, they must ensure that third-party organizations involved in AI-driven patient recruitment are in compliance with HIPAA and other privacy regulations.{7}

To address these concerns, patients and subjects should be provided with clear and concise information about how AI technology is used in patient recruitment. This includes how their PHI will be used and protected. Researchers must work to build trust with patients and subjects by highlighting the potential benefits of using AI technology in clinical trials while also being transparent about potential concerns and risks.

By prioritizing compliance with privacy regulations and patient and subject education, researchers and organizations can maximize the potential benefits of AI technology while ensuring that patients and subjects have confidence and are informed about participating in clinical trials.

FDA Regulations and Revisions

The FDA has been following an action plan for AI use in medical devices. It involves a series of proposed measures the agency would take in response to stakeholder feedback from an initial outline of regulatory modifications.{8} Most recently, the FDA has issued an updated revision of its guidance document for the “Predetermined Change Control Plan for AI/[Machine Learning]-Enabled Devices,” which highlights and defines for device manufacturers “what” changes are being made to the device using machine learning and “how” the changes in algorithms will be redeveloped to still maintain safety and efficacy.{9}

In addition, the FDA is taking action to ensure the harmonization of Good Machine Learning Practices (GMLP), which will focus on removing bias from AI algorithms and enforce a standardized system for scouring patient health records. To maintain transparency with patients, the FDA hosts public workshops with the Patient Engagement Advisory Committee on how device labeling can foster transparency between manufacturers and users while also building trust in AI medical devices. Stakeholders were concerned about bias within AI algorithms and suggested better methods to ensure validity.

The FDA will back regulatory scientists in their methodology and research on exploring, identifying, and eliminating bias within AI. The FDA was also asked to clarify what “real-world performance” would look like for AI devices. In response to that, they will be working together with stakeholders that are piloting this application.

Conclusion

The future of the use of AI in clinical trials is promising. While personal AI hardware could cost upwards of $10,000 to $4 million for building complex language processing systems, there is still zero cost to use open-source options that can give worthwhile results. There are still safety and efficacy concerns over AI, but the new hope is that clinical trial data interpretation will become a faster, more efficient, and reliable system with the help of technological advancements. The progress of patient care will open new doors for personalized medicine and allow inclusivity and transparency for patients of all backgrounds and medical needs to be involved in clinical trials. AI is not meant to replace clinical trial professionals, but rather to supplement the work that is being done to support the development of groundbreaking medical products, which will positively impact countless lives.

References

1. Harrer S, Shah P, Antony B, Hu J. 2019. Artificial Intelligence for Clinical Trial Design. Trends in Pharmacological Sciences. https://pubmed.ncbi.nlm.nih.gov/31326235/
2. Zhou Q, Chen Z, Cao Y, Peng S. 2021. Clinical Impact and Quality of Randomized Controlled Trials Involving Interventions Evaluating Artificial Intelligence Prediction Tools: A Systematic Review. NPJ Digital Medicine. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8553754/
3. Shah P, Yauney G, Gupta O, Patalano II V, Mohit M, Merchant R, Subramanian SV. 2018. Technology-Enabled Examinations of Cardiac Rhythm, Optic Nerve, Oral Health, Tympanic Membrane, Gait and Coordination Evaluated Jointly with Routine Health Screenings: An Observational Study at the 2015 Kumbh Mela in India. BMJ Open. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5914894/
4. Wijnberge M, Geerts BF, Hol L, Lemmers N, Mulder MP, Berge P, Schenk J, et al. 2020. Effect of a Machine Learning-Derived Early Warning System for Intraoperative Hypotension vs Standard Care on Depth and Duration of Intraoperative Hypotension during Elective Noncardiac Surgery: The Hype Randomized Clinical Trial. JAMA. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7078808/
5. Woo M. 2017. An AI boost for clinical trials. Nature. https://pubmed.ncbi.nlm.nih.gov/31554996/
6. Health Insurance Portability and Accountability Act of 1996. Centers for Disease Control and Prevention. https://www.cdc.gov/phlp/publications/topic/hipaa.html
7. Sharing and Utilizing Health Data for AI Applications (Roundtable Report). Center for Open Data Enterprise. 2019. https://www.hhs.gov/sites/default/files/sharing-and-utilizing-health-data-for-ai-applications.pdf
8. Artificial Intelligence/Machine Learning (AI/ML)–Based Software as a Medical Device (SaMD) Action Plan. 2021. FDA Center for Devices and Radiological Health. https://www.fda.gov/media/145022/download
9. Predetermined Change Control Plans for AI/ML–Enabled Devices. 2023. FDA Center for Devices and Radiological Health. https://www.fda.gov/medical-devices/medical-devices-news-and-events/cdrh-issues-draft-guidance-predetermined-change-control-plans-artificial-intelligencemachine

Brian Mai is an Imaging Analyst at Exelixis and a graduate student in the MS program for Medical Product Development Management at San Jose State University.

Andrea Roman, MS, is Senior Histology Laboratory Technician with Exact Sciences and a recent graduate in Medical Product Development Management at San Jose State University.

Alondra Suarez, MS, is a Regulatory Affairs Specialist with Silk Road Medical, Inc. and a recent graduate in Medical Product Development Management at San Jose State University.