Site Selection done by AI - NO WAY

NO, NO, NO - this is exactly how I react every time I hear about the newest solution offering AI-powered site selection. And I will definitely tell you why I feel like that. But first I want to make one amendment.

There is a great place for AI in improving site selection and I will give you a few examples. Here are a few applications already contributing to better and easier identification of sites thanks to Artificial Intelligence:

• Most site and investigator databases are created with data coming from clinical trial registries. These datasets though usually have the issue of duplicates making users have to manually decide if the 3 sites are the same or if they are actually one site. AI is being used to deduplicate these sites and investigators with very high accuracy leading to a much cleaner and more accurate database. (I know that firsthand as my team did this a few years ago at TrialHub and I witnessed the results.)??

• NLP (a type of AI) is used to understand much better the trial sites and investigators have been involved in, narrowing down things like endpoints, procedures, eligible patients, trial performance etc. etc. This also helps to identify the right sites and investigators with the right capacity and experience. (Here also speaking from experience)

• AI is used also to support the filling in of site questionnaires, making the burden on sites much lower and the questionnaires themselves much more adequate and accurate. Here I know there are plenty of solutions that are applying this one way or another (On top of my head are Inato and Yendou though both have slightly different angles)

I am sure there are even more applications of AI in site selection (if you know of such please add in the comments so you can also help me learn more :)?

Yet, I am less positive about AI making the site selection decisions for clinical trials. I won’t name any brand names who are doing this because I don’t want to be mean. I’d rather focus on why I believe AI can NOT select the right sites for us today. The keyword is “today” because in the ideal world, which may be in a few years from know things may change. Here’s a list of what needs to change in order for AI to be able to do the selection for us:

1. Data
2. Data
3. Data

Yes, DATA. AI is as powerful as the data behind it. And today we don’t have the detailed and accurate data we all need so that the AI can be accurate in his/her predictions of which site will perform the best. Today we leverage data from:

• Clinical trial registries are not good enough. First, not all sites and investigators are listed, unfortunately. Second, you don’t have site-level performance parameters you can take from there.

• Claims/EHR data only shows us an approximate number of eligible patients sites that might have access to. Again no insights into how well they can potentially recruit patients. Let’s not forget that this is very geographically limited - maybe just the USA and a few hospitals in a few other countries and that’s about it.?

• Datasets like Sunrise are again just in the USA. It is quite interesting to see based on this data how many patients have been recruited in the established sites and investigators. Then again, the moment you are planning a Phase II and III clinical trial, exactly when you need more sites, this means you are left alone.

• Private databases like Medidata and QDS are great, but again limited with the number of sites and the number of trials they have been collecting information from.?

When you add to that the other parameters like Competition, Diversity, and Standard of Care to the equation of what makes sites and investigators a good fit for the trial, you will quickly realize we simply don’t have all the details we need so that we can “teach” the AI to perform site selection for us. Not yet!

I want to encourage all of you to read when thinking about AI or any other technology, please don’t forget that at the end of the day this is just one instrument helping us to achieve a goal (the Jobs-to-be-done). When looking at AI solutions think first, can they help me achieve better productivity, and better results and how much can I trust what’s behind.

Speaking of the ROI of AI in Life Sciences, here’s the latest report from 德勤 that featured a few success stories (including TrialHub ) on how AI can indeed help and what’s next.?