Applying artificial intelligence technology to parts of the clinical trial process could increase trial success rates.
Artificial intelligence could improve key parts of the clinical trial process, including selection and recruitment and patient monitoring, according to a study published in Trends in Pharmacological Sciences.
It takes between 10 and 15 years and costs between $1.5 and $2.0 billion to bring a new drug to market, researchers noted, and about half of this time and capital is dedicated to clinical trials.
But despite significant investments, clinical trials still have high failure rates, the team stated.
Clinical trial failures are mainly due to poor recruiting and selecting techniques, as well as an inability to effectively monitor patients. Artificial intelligence tools have emerged as a viable way to improve these processes and increase clinical trial success rates, researchers said.
“AI is not a magic bullet and is very much a work in progress, yet it holds much promise for the future of healthcare and drug development,” said lead author and computer scientist Stefan Harrer, a researcher at IBM Research-Australia.
Researchers and sponsors are already using AI to streamline selection and recruitment processes, the team stated. Selecting and recruiting patients to participate in clinical trials is often the main cause of trial delays, with 86 percent of all trials failing to meet enrollment deadlines.
“Every clinical trial poses individual requirements on participating patients with regards to eligibility, suitability, motivation, and empowerment to enroll,” researchers said.
“An eligible patient might not be at the stage of the disease, or belong to a specific sub-phenotype, that is targeted by the drug to be tested, thus making that patient unsuitable. Eligible and suitable patients might not be properly incentivized to participate, and, even if they are, they might not be aware of a matching trial or find the recruitment process too complex and cumbersome to navigate.”
AI tools can help enhance patient selection by reducing population heterogeneity, choosing patients who are more likely to have a measurable clinical endpoint, and identifying a population more capable of responding to treatment.
Technologies like natural language processing (NLP) and machine learning (ML) could improve processes like electronic phenotyping, a method focused on reducing population heterogeneity. Electronic phenotyping can be a complex task and requires sophisticated approaches to identifying heterogeneity across multiple data types and patient records.
“Although early methods relying on hand-crafted rules were effective for simple cases, they proved to be insufficient for more complex and more nuanced cases,” researchers said.
“In recent years there have been increasing efforts to design a diverse range of machine learning methods, ranging from natural language processing to association rule mining to deep learning, that have shown great progress towards being able to handle complex real-world situations.”
In addition to improving electronic phenotyping, AI methods can help patients understand complex clinical trial eligibility criteria.
“Assistive systems using AI techniques, like NLP and ML, can be used to automatically analyze EHR and clinical trial eligibility databases, find matches between specific patients and recruiting trials, and recommend these matches to doctors and patients,” researchers said. “Such AI-based clinical trial matching systems have successfully been demonstrated and have proved their value in real-life use cases.”
AI tools can also help sponsors monitor patient behavior and response to drugs throughout clinical trials, thus helping them keep track of potential patient dropouts.
The research team noted that the average dropout rate across clinical trials is 30 percent, and just 15 percent of clinical trials don’t experience patient dropout.
If patients drop out of clinical trials because they don’t adhere to trial protocols, sponsors have to recruit additional participants, leading to substantial costs and delays.
“To comply with adherence criteria, patients are required to keep detailed records of their medication intake and of a variety of other data-points related to their bodily functions, response to medication, and daily protocols,” researchers said.
“This can be an overwhelming and cumbersome task, leading to on average 40 percent of patients becoming non-adherent after 150 days into a clinical trial.”
Wearable devices and video monitoring can automatically collect this data, making the process easier for patients, researchers said. Machine learning and deep learning models can then analyze this data in real time, logging any relevant events for clinical trial investigators.
What’s more, clinical trial administrators could even use AI techniques to predict the dropout risk for some patients.
“Picking up early warning signs for non-adherence allows proactive engagement with individual patients and permits the root causes of problematic behavior to be addressed: for example, severe side effects or incompatibility of study and personal routines could be detected and remedied before they lead to dropout,” researchers said.
“The choice of sensors and analytical models is highly disease-specific and will need to be part of the clinical study design.”
While AI, machine learning, and other analytics tools have great potential to streamline clinical trial processes and reduce drug development time, researchers noted that these technologies shouldn’t be applied to real-world trials just yet.
“Major further work is necessary before the AI demonstrated in pilot studies can be integrated in clinical trial design,” Harrer concluded. “Any breach of research protocol or premature setting of unreasonable expectations may lead to an undermining of trust – and ultimately the success – of AI in the clinical sector.”
Source : Healthit Analytics