How it works
TrialSeek is the swiss army knife for discovering, monitoring, and analyzing clinical trials. It empowers researchers, clinicians, and business professionals to find exactly what they need by combining standard search tools with advanced AI models in an elegant interface.
Search
The Problem:
It is difficult to search for complex criteria using the standard search fields on clinicaltrials.gov. For example, you might want to do a search for trials that use add-on Ezetimibe as an alternative for increasing statin dosage. Your best approach with the standard search fields is something like:
Intervention: Ezetimibe AND StatinThis search returns 52 trials that you'll then have to sift through to find the exact studies you're looking for.
Another option would be to spend time to learn the complex search options. The how-to pages are very well done, but require significant practice if you want anything beyond the basic fields — and still may not get you to the correct results.
The above is just one example of an avalanche of edge cases that make it near impossible to craft the perfect search. Language is messy. Traditional search methods often include studies they shouldn't, and more importantly, omit studies you need.
Fortunately, clinicaltrials.gov has a comprehensive and open API that allows anyone to build on top of their data repository and tools.
The Solution:
With TrialSeek, you can just describe exactly what you want:
Intervention: Ezetimibe
Deep Search Criteria: Interventions section specifically compares add-on Ezetimibe vs increasing statin dosageThis search returns the exact 13 trials that meet your criteria, saving you from manually sifting through the 39 that don't.
In general, the workflow is:
Step 1 - Standard Search:
Start with very simple Standard Search criteria. The purpose of this step is simply to narrow down the number of studies passed to step 2, not to query for specific details.
Step 2 - Deep Search:
Enter a list of Deep Search criteria. AI will read each study in-full and assess whether it meets your criteria. You can specify any criteria about any part of the study in any amount of detail, far beyond the capabilities of the Standard Search.
This framework completely bypasses traditional classification, keywording, etc., and relies entirely on the knowledge of state-of-the-art AI models to get you exactly what you want. TrialSeek currently offers GPT 4.1 Mini and GPT 4.1. The capabilities are impressive, and will only improve as OpenAI and others continue to scale their systems.
Monitoring
The Problem:
- The traditional way to keep an eye on a set of studies is through RSS feeds. While loved by a niche audience, RSS is no longer widespread, and likely doesn't get used to its full capability.
- There is no seamless way to see incremental changes to a study as they happen. The current set of tools requires much navigation, scrolling, and comparison to decipher what exactly changed with each study update.
The Solution:
TrialSeek compiles all data for all versions of each study monitored into an easy-to-use interface. All study updates have a clean presentation of the exact changes, along with a brief AI-generated summary describing them.
FAQ
Why are some searches so fast?
Whenever TrialSeek asks an AI model to assess a deep search criterion, it saves the result for future use. If you need a criterion to be re-assessed for a given study, TrialSeek will automatically bypass the AI model and use the cached result. This is much faster and doesn't use any credits.
How is credit usage estimated?
When you queue up a set of search criteria, an estimate will appear underneath the search button before you click search.
- For searches with only one deep search criterion, this estimate is exact.
- For searches with more than one deep search criterion, this estimate is a conservative calculation of how many requests will be needed per criterion from historical data. For example, if a study doesn't pass the first criterion, the second criterion will not be assessed to prevent unnecessary usage of the model.
Additionally, the estimate does not account for previously assessed criteria. The actual usage will be much lower than the estimate for searches that contain criteria you (or anyone else) have already run before.