Interview
AI in oncology: trials, tribulations and future opportunities
AI has the potential to improve several aspects of oncology trials, but sponsors must find ways to harness the technology’s true potential, writes Annabel Kartal‑Allen.
Main image: Anthony Costello, CEO of Medidata
Main video credit: Infi Studio/Shutterstock.com
I n recent times, artificial intelligence (AI) has begun to dominate the conversation in life sciences, with its use rapidly expanding into new areas, including drug discovery and development, as well as drug manufacturing.
This trend can also be strongly observed across the clinical trials landscape, as sponsors increasingly use the technology in their day-to-day workflow to overcome bottlenecks with AI’s use, which often spans across areas like predictive protocol management and site feasibility evaluations, as well as patient enrolment, recruitment and retention.
As GlobalData documents oncology’s continued dominance in terms of new trial initiations throughout 2020-2025, sponsors are continuing to seek out new ways to stay ahead in the highly competitive landscape, with many turning to the promise of AI to drive efficiencies, enhance proactivity and potentially improve patient outcomes by bringing drugs to the market faster.
While some are heralding AI as a transformative tool for clinical trials, challenges to implementation remain, and some sponsors are not attaining the results they would hope for with the technology.
In conversation with Pharma Technology Focus, Anthony Costello, CEO of Medidata, notes that AI holds the potential to revolutionise oncology clinical trials, but implementation hurdles must be overcome for the technology to meet industry expectations.
This interview has been edited for length and clarity.
Annabel Kartal‑Allen: How can companies effectively implement AI into their site feasibility evaluation protocol, and what are the key hurdles they should be aware of when looking to do this?
Anthony Costello: Currently, clinical trial sites often have vastly different abilities to manage the types of protocol-driven requirements needed. There’s a lot of fantastic sites out there, but they’re very burdened – often they’re running multiple studies that all focus on the same disease area, so feasibility doesn’t necessarily mean a site is a good pick.
A lot of our work with AI and site feasibility has focused on historical work, which provides a good ‘virtual twin’ of the site’s conduct and capability, which helps evaluate, based on a current protocol design, whether that site will perform at a high level on that trial.
By having this information, sponsors can make a decision about whether they want to interact with that site around those changes, or whether they want to pass that site over for a particular study.
Annabel Kartal‑Allen: Are there any particular benefits that AI can offer to oncology trial operators?
Anthony Costello: AI can certainly help in a lot of clinical trial settings, including oncology, as studies in this disease area tend to be among the more complex to get running off the ground to enrol patients, despite their impactful and important nature.
Where we really see AI touching the process is by taking complex trials and building them much faster, as well as looking for areas of the study where there’s likely to be a high burden for patients and sites due to complex protocols with a lot of requirements, travel and site interaction.
By using AI, operators can simplify trials, reduce the burden to only the most important data requirements that are necessary to prove a hypothesis, and then build studies more quickly. This can reduce what used to be weeks or months or human involvement to initiate a study down to hours or days.
Annabel Kartal‑Allen: How can sponsors use AI to enhance patient enrolment and retention, and are there any challenges to be cognizant of when implementing the technology?
Anthony Costello: Before we get dropout or retention challenges, there are recruitment problems, and bringing patients into studies that are overly burdensome is a challenge. If you’re lucky enough to enrol patients, retention could prove problematic, as the burden begins to, in many therapeutic areas, outweigh the promise of the new therapy; AI can help with all of these challenges.
For example, you can use AI to choose when and how to communicate with patients, so you can precisely contact them in a way they respond to, at the right time of day when they’re able to perform a clinical trial-related activity.
By approaching patients in a personalised way, you can impact retention rates, as well as adherence rates, which measures how much a patient is engaging with the clinical trial activities in the timeline the protocol requires. Staying on top of these subtleties can make the difference between a patient staying in a trial or dropping out, and AI can allow us to measure those kinds of effects across large populations more readily than we ever could before.
Annabel Kartal‑Allen: While many are trialling AI in their daily workflow, not all are getting what they hoped out of the technology. Why do you think some AI pilots are failing, and how can operators go about implementing AI to maximise its impact?
Anthony Costello: This is a question without a simple answer. The life sciences sector in general is a tough place to implement new technology because it’s a heavily regulated industry, and regulatory bodies oversee the data capture and the permissions to utilise that data.
The other factor is that we’re dealing in private health information for people, and that makes it a very challenging place to implement technology and expect that it’s going to be readily available for use across the industry.
Over the last 20 or 30 years, we have seen adoption hurdles and barriers to technology that have taken some time from pilot to full implementation. AI takes that all to a new level, because you’ve got the machine doing the learning and making suggestions; there’s an extra layer of risk aversion that I think comes to play in much of this industry.
The other important thing when using AI in this highly specialised industry is where the data are coming from that train the AI. This is why we have trained our AI on historical clinical trials, which provide the cleanest datasets in any sector, as they’ve been prepared for regulatory bodies to review and potentially approve new pharmaceuticals. This is the cleanest training ground you’re ever going to get for AI, and technologies trained on a bedrock of data that is inherently reliable, clean and trustworthy is an important consideration.
It’s early days still, but within a year or two from now, we will see how many of the early-stage pilot exploration projects have made it to full-blown implementation.
