Key Challenges in Clinical Trials and How AI Can Help Overcome Them

Across multiple surveys on challenges facing modern clinical trials, a few consistent themes arise, which lend themselves naturally to support from Artificial Intelligence (AI).[1],[2]

Here we’ll focus on three of them.

Diversity

Diversity, righty, has become an important focus within clinical trials. Recruiting and retaining more diverse participants is crucial to more realistically reflect and address therapy outcomes and safety in the real world. However, it’s challenging to recruit and retain trial participants, especially with the added requirement of ensuring appropriate levels of diversity.

One way AI can help is by providing users with near-instant insight into the demographic makeup of their recruits. This can be done in a blinded fashion (e.g., the cohort overall) or unblinded in some manner.

At ReadoutAI, we automatically generate ICH E3 compliant Baseline Demographic reports, complete with narratives and tables, directly from the data. A toy example is shown below (with participants split based on whether they drink coffee (the “Y” column) or don’t (the “N” column).

Baseline Screen

The takeaway is that you can’t target what you can’t measure. So, if it takes you weeks or months to understand the demographics of your participants, rather than doing so in an instant, as participants are randomizing, then you can’t course correct quickly and adaptively.

Data Gathering and Quality

Another top challenge in clinical trials is data – how to get it and how to make sure it’s of sufficient quality. A trial accumulates data from various sources – labs, sites, even patient reported outcomes, and all of this data must be checked for consistency and errors. For instance, a site may accidentally enter a temperature in Farenheit of 9.87 instead of 98.7 or a lab value might have been cut off in the file transfer, making it 100X smaller than would be possible.

Currently, many of these edit and data checks are done manually. While some systems, such as EDCs, can have edit checks in them – they are programmed by hand, on a per trial basis. And further, those only cover the data in the EDC (likely from the sites) and not from other sources like a lab or PRO capture.

Again, this is an area ripe for AI. Fundamentally, if an AI can understand the type of data it’s looking at, it can determine its quality.

With ReadoutAI, we do this in two steps. First, our AI makes an assessment of the type of data its looking at – is it an Age, a Date or a measure of White Blood Cells? Second, it checks the data – an Age is a number, and it can’t be negative or more than 140, for instance. Our Knowledge Base knows about data from labs to vitals to demographics, and it applies this knowledge to not just flush out unusual data, but to find data that would be inconsistent with human life – since that must be wrong (a human is never -10 years old…)

Intro To Readout Mds Short

Above is an example showing demographic data from onboarding (in yellow), combined with lab data (in green), and data from a site visit (in ZZZZ). Going from left-to-right, we see the steps where first the data is interrogated to determine its type, and then the data errors are flagged.

These flagged errors can then generate a “Query Resolution” process for the data to be corrected in a regulated fashion. But the key is that, among volumes of data across many sources, the AI can help the team identify data errors.

Talent Shortages

Finally, a key issue in trials is the massive shortage of smart people who can do hard jobs. The number of trial starts continues to outpace the number of PIs and sites, but even more specific jobs, such as biostatisticians and medical writers, are in significant demand.

Unfortunately, ReadoutAI can’t hire people. But it can make existing humans more productive. We’ve focused very deliberately on automating many of the tasks that take away from high-level, productive time, for some of these specialized experts. For instance, we free biostatisticians and medical writers from Adverse Event and Demographic reporting. We do the statistics and the writing (and even the MedDRA coding), in compliant formats, to free them to focus on the challenging tasks of efficacy analysis. As described above, rather than combing through data by hand, we free Data Managers to focus on the query resolution components and database design. The key is that while AI doesn’t replace humans and can’t help you hire, it can help make your existing teams that much more productive (and possibly even happier, as they have help).

These are just three of the major issues in clinical trials today – while not comprehensive, we do hope it sheds some light on the ways in which AI, such as ReadoutAI, can help to mitigate them. If you want to chat further, you can always drop us a line!

[1] https://www.ppd.com/blog/challenges-opportunities-in-clinical-trials/

[2] https://www.six-degrees.com/the-biggest-perceived-challenges-facing-clinical-trials-in-2023/