News about AI in healthcare is everywhere — as a scribe to help with billing and the EHR, as a way to get a second opinion on difficult cases, and innumerable other uses. Physicians have, rightfully, been concerned about claims that AI can safely give medical advice.
A new study from a clinical AI research team led by physician researchers from Stanford University School of Medicine and Harvard Medical School called NOHARM (Numerous Options Harm Assessment for Risk in Medicine) looked at just that: How safe is the medical advice AI gives?
NOHARM evaluated how often recommendations from multiple AIs could result in patient harm. The study looked at 1,100 physician-derived clinical scenarios spanning 10 medical specialties. It broadly concluded that clinically specialized AI systems — like Doximity Ask — consistently outperformed more general-purpose models like ChatGPT, Gemini, or Claude. This reinforces what we already know to be true: Rather than forcing doctors to adapt to platforms that work for a wide variety of scenarios, tools built specifically for medicine belong to medical applications and its end users. Indeed, imagine if the EHR had been built with us in mind.
The results from NOHARM show that healthcare AI should no longer simply be measured on if it can answer medical questions correctly, but whether those recommendations are actually consistently safe for patients. After all, knowing medicine and practicing medicine are different.
That’s why Doximity Ask came out ahead of so many, including OpenEvidence. Beyond just sharing an answer, Ask is powered by PeerCheck™, where practicing physicians themselves continuously review AI-generated answers for clinical accuracy, evidence quality, completeness, and potential bias. This user experience with built-in agentic reasoning along with the key element of human feedback not only delivers faster, reliable clinical responses, but also builds trust, the same kind of trust that we strive to provide for our patients.
As it continues to evolve to represent the needs of real-world clinical practice, Doximity Ask was built to anticipate the needs of real-world practice, reflecting the roles of actual physicians and Doximity end-users who played a direct role in its development and deployment. Indeed, each stage in its algorithm (Retrieval, Ranking, Generation, Verification, Physician Review) was designed to anticipate and address potential gaps, assumptions, and biases that we encounter in everyday face-to-face interactions with our patients.
Medicine is constantly evolving, as new studies or guidelines are published and new therapies and treatments are developed. We all want to do what’s best for the patient in front of us. That’s why it’s so useful to have a product like Doximity Ask, which uses the most relevant literature from a constantly updated index of millions of peer-reviewed publications and specialty society guidelines, refreshed daily.
We all remember what it’s like to be a new medical student and be overwhelmed by the firehose of medical knowledge to learn. Having Doximity Ask in your pocket is like having the world’s most knowledgeable medical team always at your side. The correct medical answer is one thing, but the safest is another — and when it comes down to it, safety is what matters most to our patients.
