What's in a name? Our first clinical trial

Jaime Domingues

Recently we welcomed the news that we have officially selected a study name for BenevolentAI's first clinical trial. I thought I would share the story of how, using our tech and data, we were able to quickly and comprehensively choose an acronym for this trial.

Clinical trials that appear on the clinical trial registry, available to view at ClinicalTrials.gov, are required by the FDA to have a brief title. Brief is a bit misleading however, as shown by this example: 

"Comparison of Volinanserin and Lormetazepam in the Treatment of Insomnia Characterized by Sleep Maintenance Difficulties".

So, often acronyms which help describe these clinical trials are included in the registry to make things more manageable, waste less paper, reduce wrist pain and spend less time typing out tweet-length titles.

In order to provide our clinical trial with an appropriate (and appropriately sexy) acronym, Our clinical trial project manager managed to generate a list of around 20 potential acronyms based off the brief title for the trial. In staying with the ethos of BenevolentAI we aimed to ensure each of these potential acronyms satisfied 2 conditions before a final shortlist was put forward to a vote by the company:

  1. It should not already be in use by another clinical trial in the same or similar indication

  2. It should not suggest or imply any effect or results to patients or observers (e.g. CURE)

If one person were to do this manually for each of the 20 potential candidates, it would take them weeks to comprehensively search the registry (which have no means of searching on acronyms) and to the best of our knowledge there is no easily accessible database that would help with this. So instead, with the help of a member of our bioinformatics department, all 235 000+ registered clinical trials from ClinicalTrials.gov we have stored internally (and continuously updated) were able to be searched against. With this list, the team were able to search across the entire registry and, impressively, return trials with acronyms that had exact matches or one letter changes specifically in the same or similar indication as our proposed clinical trial, all within half a day. We were also able to check whether any of the proposed acronyms were found in the brief title of existing trials as well to ensure any search with the acronym was unique to our trial.

From one generated list and half a day's work, we were able to confidently provide an acronym for our first clinical trial where traditional methods would have taken weeks and been less comprehensive. It may be a little taste of the value of what we do, but it is something BenevolentAI can feel proud about and celebrate. Well done guys!

Jaime Domingues