OpenEvidence: Free AI-Powered Medical Search Engine

Physicians make multiple decisions every day. According to one study, a group of 10 pediatric cardiologists made on average 158 decisions per day over 7.5 days. Physicians do not always follow the best medical evidence when it comes to decision-making. This is due to many reasons. They may not have the time to perform a literature search. They may lack the appropriate tools due to expense or complexity. Moreover, there may not be a satisfactory answer to every clinical question.

As a result, decisions are often made that are not evidence-based or are based purely on their anecdotal experience. That reminds me of the quote by W. Edwards Deming “Without data, you are just another person with an opinion.” We must strive to find evidence based on high-quality data to validate our decisions.

Fortunately, there are excellent medical educational programs out there to help answer our frequent medical questions. Newer tools are much faster and more current than relying on textbooks that grow old quickly. Programs such as UpToDate are outstanding but cost $559 per physician per year which may be cost-prohibitive for some.

Large language models such as Perplexity ai have added a new educational choice but they are not medically oriented nor are they medically curated for accuracy. The results may be hit or miss.

I only recently became aware of OpenEvidence which is an AI-powered medical search engine developed by Harvard and MIT. They have a large team of clinicians and data scientists to support their efforts.

OpenEvidence scored a 90 on the standard USMLE. It is the engine behind Elsevier’s Clinical Key option. They state on their website that OpenEvidence is “Trusted by medical professionals at 7,000+ care centers across the US. “ Despite these successes, many clinicians are not aware of this educational option.

The GUI is simple (below) without any options or filters. The question you submit can be a single simple sentence, such as “What causes long COVID?” or a more complicated patient scenario can be inputted, asking for a differential diagnosis.

I will include the results of two searches but I want to list several very helpful features first:

1. To the right of each reference is a “Details” option which opens a shortened summary of each reference.
2. Users can rate each response as helpful or not helpful
3. There is a share option that copies the URL to the clipboard.
4. There are follow-on questions after the references so you dig deeper.

Below is a screenshot of a search “What patients with obesity should receive GLP-1 agonists?”

For the second search, I inputted “What is the differential diagnosis of high- frequency hearing loss in children” and the results are below

Comments

I performed about twenty medical searches using OpenEvidence and found the results to be extremely helpful. The results are quickly obtained and well-referenced. These questions were primarily clinical. When I asked the engine about generative AI and medicine it returned some 2024 references but missed several I was aware of. To the best of my knowledge, the references are current based on seeing references from this year.

Several important questions remain. When would you use this tool, instead of high-quality resources such as UpToDate? How should this resource be validated or benchmarked? Is this engine adequate for an admission differential diagnosis by housestaff? Would it be more useful as a phone app?

Happy Searching!