Feature

It’s a match! Connecting patients to clinical trials with AI

AI trial matching platforms are overcoming the obstacles associated with clinical trial recruitment, but human oversight and data standardisation are crucial. By Urte Fultinaviciute.

Credit: Shutterstock/olya osyunina

One of the biggest pain points in clinical trials is patient recruitment. Clinical trials often have strict eligibility criteria that look for suitable subjects who are healthy enough to participate. However, finding athletic patients is time consuming and costly.

Dr Daniel Vorobiof, chief medical director at patient-centric network platform Belong.Life, says that for many years clinical trials have been in a problematic situation with only a small percentage of patients being accrued into clinical trials. “Many patients have never heard about clinical trials and their own doctors have never talked to them about it,” he says. 

A data analysis by Clinical Trials Arena revealed that the most common reason for trial termination is a low accrual rate. A different analysis noted that 86% of all trials do not meet enrolment timelines and almost one-third of Phase III trials fail because of slow enrolment. 

However, the prevalence of trial terminations due to low accrual is decreasing, possibly due to an extended use of technology-aided solutions, including artificial intelligence (AI). Experts shared their thoughts with Clinical Trials Arena on how AI-powered trial matching can accelerate patient identification. Still, as with many technology solutions, AI is not perfect and certain limitations can slow down the process. 

Benefiting all stakeholders 

People who stop responding to existing therapies always look for the next best thing, and sometimes the only option left is a clinical trial, says Belong.Life’s chief technology officer Irad Deutsch. Belong.Life offers patient support across various therapy areas, but there is a high demand for clinical trials among oncology patients. 

“We understood that in order to scale up, we cannot use a 100% human-based effort to provide the support and answers to all these demands,” Deutsch notes. Over the past seven years, Deutsch and his team developed an AI-powered technology that automates most of the trial matching process. 

AI trial matching platforms also increase awareness among physicians, says Dr Michel van Harten, CEO at myTomorrows, which recently released a physician-focused clinical trial search tool TrialSearch AI. The search tool streamlines the matching process and gives physicians the opportunity to refer a patient to a clinical trial, he explains. 

myTomorrows investigators conducted a test run on how the TrialSearch AI works by creating 10 fictitious patient profiles across 10 different diseases. The tool, which operates on a Large Language Model (LLM), was able to reduce pre-screening checking time for physicians by 90%. 

Other use cases also demonstrated the efficiencies of AI-based clinical trial matching. A 102-patient Australian study demonstrated 95.7% accuracy for exclusion and 91.6% accuracy for overall eligibility assessment among cancer patients. 

Unsurprisingly, this technology benefits both pharma companies and clinical trial sponsors. van Harten explains that if trial enrolment is faster, sponsors can reach the finish line more quickly. “If a trial was earlier enrolled and completed, the faster they can get to commercialisation,” he adds. 

Limitations in sight 

For AI tools to operate efficiently, the data input in public clinical trial registries must be up to a certain standard. However, because data and information are submitted by humans, data standardisation is a crucial process. Both TrialSearch AI and Belong.Life AI engines require standardised and structured databases to operate. 

As with humans, AI also has language barriers. Currently, the Belong.Life engine supports English, French, German, Spanish, and Hebrew, while TrialSearch AI platform uses data from ClinicalTrials.gov and the European Clinical Trials Register. 

Data privacy is another hot topic in the world of AI. “You need to adhere to the highest standards when it comes to protecting their data and privacy,” van Harten says. 

However, as with the wider application of AI, AI-powered trial matching platforms are not regulated. Recently, European Parliament released a mandate draft to ensure the human-centric and ethical development of AI in Europe. While it does not have healthcare-specific rules, it outlines the possible future of AI. 

While waiting for healthcare-specific regulations, van Harten recommends establishing partnerships with rigorous data safety standards and expertise in patient safety. 

Using nanomaterials to deliver drugs to treat TB and infectious lung diseases can provide numerous advantages over traditional drug delivery methods.

While NPs have been developed for TB over the past decade, the therapeutic systems have become prominent using diagnostic and therapeutic methods (theranostic). Theranostic approaches to TB management were designed to conduct nuclear imaging, optical imaging, ultrasound, imaging with magnetic resonance, and computed tomography. 

Problems with resistance to conventional TB drugs mean therapeutic methods require high doses of numerous medications over a longer time. Issues with the practical capabilities of traditional drugs for TB also exist. Solubility, stability, and penetration impact the drugs’ effectiveness. Traditional drugs may also create resistance over time, a relapse, and extend to other body parts, leading to secondary TB.  

The future of AI 

AI has been moving into every possible industry, including healthcare and pharma. Earlier this year, Microsoft’s BioGPT tool demonstrated “human parity” in analysing biomedical research. Deutsch says that AI was a very generic term six months ago, but since the dawn of ChatGPT, people can feel its presence in their everyday lives. 

As such, people are more open minded about AI than before, and both patients and physicians are more welcoming of the technology. The TrialSearch AI tool was launched a couple of weeks ago and myTomorrows has 30 physicians on board to test the beta version. “We are getting some qualitative feedback that this AI application is very practical and it’s saving time,” van Harten says. 

While AI matching platforms are efficient, they are far from being perfect, and human oversight is still a crucial element. Deutsch explains that Belong.Life’s Natural Language Processing engine has been trained for seven years, and while most of the suggested trials are appropriate for patients, sometimes there is still a need for fine tuning. 

According to Deutsch, the adoption of AI is at the same development stage as autonomous cars. “It can drive autonomously, but you have to be there and take control of the wheel if it is required,” he adds. 

van Harten agrees that physicians are at the centre of decision making and that they can correct AI when needed. “It’s just a clinical decision-making tool and not a clinical decision maker,” he adds. Vorobiof notes that there is no replacement for the human side in this process, but instead AI should be viewed as sharing the workload. 

Various AI applications are being tested in several stages of conducting clinical trials. From using predictive analytics and AI to potentially bypass animal testing, to digital twins allowing researchers to digitally replicate trial participants and observe them in two scenarios in real time, AI is here to stay.