Tech & Trials: Recruiting is hard, AI can make it easier
Recruiting for clinical trials is hard… and expensive. And a large component of that time involves people sifting through medical data, by hand, to see if someone qualifies for a study or not.
In fact, according to one of the latest Tufts studies on the topic, about a third of studies under-enrolled and about 14% didn’t enrol a single patient.1 This all has a real impact, as recruitment issues lead to trial delays, which lead to fewer medicines coming quickly to the public.
As is a theme in this series, artificial intelligence (AI) is beginning to address some of these challenges, making recruiting faster, more diverse, and daresay… easier?
Finding patients
The first challenge in a clinical trial is finding the right patients to participate. This involves matching a patient’s medical history (from their charts, labs, etc.) to the inclusion and exclusion criteria of a trial. This is important because many patients will not agree to participate, so the goal is to find as many potential patients as possible to see who might be willing to learn more.
Most of the challenge here involves understanding enough about the medical history to know if an inclusion or exclusion applies. For instance, let’s say an exclusion criterion is to exclude alcoholics in a Phase 1 trial. Sometimes the medical record makes it easy – maybe that diagnosis is in a note or a formal field. But other times, maybe it must be inferred, say by the answer to the question, “How many drinks per week, on average, do you have?”
Even more challenging is finding patients for rare disease trials. Traditional methods may miss certain patients simply because relevant information is buried in electronic medical records and is not commonly seen, but AI can analyse vast amounts of such data to uncover potential participants who might otherwise go unnoticed.
One way to do this is via AI that has a medical understanding (see previous instalments in this series on data), or via linguistic matching that can be trained on records of patients that met certain criteria (remember what was discussed on supervised learning?) There are many examples where AI has helped screen patients with high accuracy2 and included agreement with humans3 - so, this technology is not a pipe dream.
Optimising trial design
A common area for AI is in trial design – optimising the design to make it less likely that adverse events might occur or figuring out which outcomes to measure that will be most effective. But in recruiting a big pain point is patient retention – if trials are too hard on patients (too many things to comply with, too far away, etc.) patients will leave and new ones need to be recruited. This is also a problem that AI can help with.
There is work on using AI to identify unnecessary steps in a protocol (those that don’t seem to add much clinical outcome benefit, but are hard on patients). For instance, maybe your trial doesn’t really need to gather labs twice a week – that’s a big patient burden and maybe the results don’t make a huge difference compared to gathering those labs once a week. Less burden means happier participants and therefore a shorter trial. And AI is great at identifying those cases, especially since it would be hard for a human to pick up on that.
Promoting diversity and inclusivity
It’s well known that clinical trials have historically lacked diversity (in the US, clinical trials weren’t required to include women until 1993). One driver of the continued lack of diversity in clinical trials is the reach of recruitment efforts. Here, AI can play an important role in two capacities. First, algorithms can be designed to actively seek out underrepresented populations by analysing broader data sets and highlighting candidates based on a diverse array of demographic and socioeconomic indicators. This approach can address the lack of diversity in clinical trials by ensuring that studies include patients from a wider variety of backgrounds and health conditions.
Second, just like with any messaging, AI can be leveraged to personalise that message, so that it’s most meaningful. At the simplest level, since many messages require legal review, a recruitment process might generate many messages around different benefits of a trial, and let an AI manage which messages get sent to whom, to maximise outreach.
Things AI can’t help with
In a trial, AI can help write better copy for brochures, help ease the burden on patients, and help find patients that may be a good fit.
But many participants come to a trial via the tried-and-true method: human communication. Patients talk with their doctors and, if the doctor knows about a trial, they will discuss it with a patient and make a recommendation. Talking with your doctor can make everyone comfortable with the decision, and honestly, I don’t know if an AI will ever provide that level of comfort and empathy.
But, as we’ve seen, there are positive ways that AI can impact patient recruiting to find, keep, and include patients for clinical trials.
