Patient selection and recruiting techniques, paired with the inability to monitor and coach patients effectively during clinical trials, are two leading causes of high trial failure rates, according to Trends in Pharmacological Sciences. However, while patient recruitment is vital at the beginning of a study, patient retention is paramount for ensuring the trial moves through all key phases. If a comprehensive and robust retention strategy is not implemented, it will cost stakeholders valuable time and money and put the trial’s viability at risk and sponsors in the hot seat.
Artificial Intelligence (AI) has become a key player in today’s 52 billion clinical trials market. However, it also has limitations that can significantly impact a trial’s pipeline – specifically retention. Below we examine the current financial and strategic challenges of today’s clinical trial and discuss the benefits and limitations of using predictive AI technology. We address these critical limitations by demonstrating how Projective Health AI may be the solution to more advanced patient retention and “smart” clinical trials.
The Urgent Challenges Facing Today’s Clinical Trials
Common trial failures can arise from safety issues and lack of funding, failing to follow FDA guidance, or patient recruitment, enrollment, and retention problems. However, retention can mean the difference between a successful trial and total study failure. So what’s at stake? The staggering costs incurred during an average clinical trial. According to the U.S. Health and Human Services report, the average expenditure of phase 1, 2, and 3 clinical trials across therapeutic areas are around 4, 13, and 20 million, respectively. To narrow these costs, even more, phase 3 studies for new drugs approved by the FDA cost an average of $41,117 per patient. Further, according to a recent report in JAMA, U.S. biopharmaceutical companies spent over 1 billion between 2009 and 2018 on R&D alone. And guess where a significant chunk of this spending fell? Recruitment and retention.
Unfortunately, despite all this spending, approximately 30 percent of patients drop out of clinical trials, resulting in hefty financial losses. Case in point, the average cost to recruit a new patient if one drops out is $19,533. Additionally, studies show that 80 percent of trials are delayed by at least a month due to issues such as retention, causing losses of approximately $600,000 and potentially as high as 8 million per day. With all these costs at stake, ensuring that the retention protocol is comprehensive and robust is paramount.
Barriers to Predictive AI
Regulators are now turning to AI technologies to address these costly clinical trial issues. According to Clinical Trials Arena, predictive AI offers advantages such as examining large amounts of data to detect patient subgroups that may benefit a clinical trial. However, while predictive AI is valuable in making clinical trial stages more efficient, it still faces many unknowns. These limitations include issues such as data interoperability and inherent bias. Interoperability and access to data are vital to developing an AI algorithm. Yet, according to Visceral Medicine, the current data security and privacy challenges, resistance to sharing data, and managing inconsistent information across multiple sources, make it hard to create a clear pathway to develop this technology. Additionally, many datasets are used to train an algorithm. Unfortunately, this process can lead to insufficient data and leave out significant parts of the historically underrepresented population – leading to bias risk. And the limitations of predictive AI do not end there.
While AI can monitor participant adherence during the trial, it is often done using wearable sensors, digital biomarkers, or video. According to an article in Clinical Trials Arena, AI-driven technology is designed to pick up specific patient behavioral patterns, such as avoidance of taking the medication, and to predict possible non-compliance during a trial. Unfortunately, according to Digital Health CRC chief innovation officer Stefan Harrer, PhD., “Video data is the most sensitive type of data and must be handled with utmost privacy. There is a high bar in video privacy as trust needs to be earned from the user,” he explains to Clinical Trials Arena. These privacy issues may lead to patient distrust of tracking devices like video monitoring – deterring them from staying in the study. Additionally, trial developers increasingly need to add staff and patient training to their budgets to ensure these devices are used correctly. And while many patients have smartphones and iPads, many devices aren’t familiar to them. These barriers can lead to delays, wasting valuable time and money. Using less invasive, simple, and cost-saving measures to understand patient behavior may be the key to better retention and “smart” clinical trials. Enter projective AI technology.
Predictive v. Projective AI
To understand the benefits of projectable health AI technology, it is necessary to highlight the fundamental differences between predictive and projective AI. Predictive AI is based on probability. This model is often used as a predictive tool by healthcare professionals to gather data based on symptoms and the progress of a disease or disorder. However, according to IoT for all, there are limitations in using predictive AI, including but not limited to:
- It requires a considerable amount of data. Leading developers to experience challenges in obtaining the high-quality data needed to create effective AI tools.
- Predictive AI can suffer from a lack of diversity of participants, limiting its data and skewing results during clinical trials.
Predictive AI also requires the patient to be symptomatic, excluding healthy population data. Endominance’s Projective Health AI (P.H.A.I.) tool is based on a patient’s current environment and their current pattern of behavior. Therefore, the focus is on behavior, not symptoms, and includes healthy population data. And when we say behavior – we mean behavior as an interaction between a person’s genotype and phenotype and their environment. For example, someone with asthma in an environment with low air quality will become more sedentary and stay indoors even if this is not indicative of their behavior in a more asthma-friendly environment.
Yes, predictive AI can also be used to anticipate the dropout risk for patients during clinical trials; however, it comes with the above challenges, risks, and barriers. But what if a clinical research coordinator could gain personalized insight into their participant’s behavior? What if a contract research organization could better understand a patient’s compliance and compatibility throughout each trial phase? Our P.H.A.I. tool has been developed to address these retention challenges.
How P.H.A.I. Maximizes Participant Retention
P.H.A.I. is entirely remote and accessible online and works regardless of ethnicity, gender, or age. However, our tool’s strength lies in its ability to address the biggest pain point for sponsors and CROs – retention. Our technology can help reduce patient dropout rates from 6 to 15% by addressing possible issues such as:
- Is the patient responsive and able to communicate their needs?
- How resilient is the participant to stress? Will they be persistent or too sensitive to stressors?
- Will the participant comply in areas such as following trial protocol or accurately reporting their symptoms?
- Is the participant and clinical staff compatible?
Unlike predictive AI, P.H.A.I. is built using our proprietary behavioral processing model (PCB Model), which offers an evidence-based explanation of how a person’s behavioral output is triggered by their environment and traits, combined with how that person perceives/interprets these stimuli. In short, the PHAI model measure’s the root of behavior at the subconscious level and how it influences one’s conscious decision-making process.
And unlike predictive AI, which requires vast datasets to train an algorithm (often leaving out significant parts of underrepresented populations), P.H.A.I. does not rely on population-driven statistical data. Instead, it formulates individual results derived from the responses of each user with an embedded algorithm. Since our algorithm is built from our own holistic and proprietary data sets, we can tailor results to each user based on their responses. We continue to feed P.H.A.I. with additional data sets from our research while also connecting with other health metadata to improve the algorithm’s accuracy. Unlike predictive AI, which relies on a patient’s symptoms, our goal is to project the future of a person’s wellness, regardless of their current health condition, by integrating data from healthy and at-risk populations into our already-robust projectable AI.
The Future of “Smart” Clinical Trials Starts With Projectable AI
Every clinical trial aims to keep participants engaged in the study from start to finish. However, if retention strategies and processes are not robust and comprehensive, it can lead to massive delays, high dropout rates, or a total trial failure. These issues can result in wasted time and resources and the possible loss of millions – or even billions of research dollars. Predictive AI has its place in the clinical trial market – but it’s time to go beyond its boundaries. By using our P.H.A.I tool CROs and sponsors can feel confident about gaining personalized insights of participants – leading to higher retention rates, reduced financial risks, and overall clinical trial success.
