The Pulse
Google’s AMIE Included the Final Diagnosis in 90% of Clinic Cases
Google says its AMIE medical AI included a patient’s final diagnosis among its seven leading possibilities in 90% of cases in a supervised study at Beth Israel Deaconess Medical Center. The single-center trial involved 100 patients who chat
AI.info Team · Written with AI ·
AMIE included the eventual diagnosis among its seven leading possibilities in 90% of cases in a real-clinic study involving patients at Beth Israel Deaconess Medical Center, Google said October 8. The result comes from a small, single-center feasibility study, not a trial showing that the chatbot diagnoses patients as well as doctors or improves health outcomes.
One hundred adults completed a text conversation with AMIE before an urgent-care appointment; 98 went on to see a primary care provider. In 75% of cases, the final diagnosis appeared among AMIE’s top three possibilities. It was AMIE’s single most likely diagnosis in 56% of cases, according to Google Research’s account of the study.
AMIE gathered histories before the appointment
AMIE, short for Articulate Medical Intelligence Explorer, asked patients about their symptoms and medical history through a secure web link. It generated a transcript and summary for the clinician to review before the visit. A physician watched each conversation live and could stop it if a pre-set safety condition arose; the supervisors stopped none.
The researchers compared AMIE’s differential diagnoses and management plans with those from primary care providers. A blinded panel of clinical evaluators rated the overall quality of the AI’s and clinicians’ differential diagnoses similarly, and found no significant differences in the appropriateness or safety of their management plans. Clinicians rated their own plans higher on practicality and cost effectiveness.
“While further research is needed, this study demonstrates the initial feasibility, safety, and user acceptance of conversational AI in a real-world setting, representing crucial steps towards clinical translation.”
Peter Brodeur et al., A prospective clinical feasibility study of a conversational diagnostic AI in an ambulatory primary care clinic
Supervision shaped the result
The study tested AMIE as a pre-visit assistant, not as a substitute for a clinician. The AI did not have access to patients’ electronic health records, could not examine them, and did not process visual or other multimodal input. Doctors, by contrast, saw patients and worked within the clinic’s care setting—advantages that may help explain their higher ratings for practical and cost-effective plans.
Google reported that clinicians found AMIE’s summaries useful in preparing for visits in 75% of cases and said the summaries influenced their approach to care in more than half. The company also said patients’ attitudes toward AI became more positive after interacting with the system. Those findings describe participants’ experience in this study; they do not establish that the tool would have the same effect in other clinics or among patients who chose not to participate.
A feasibility study, not an outcomes trial
The researchers call the work a prospective, single-center feasibility study. It had no controlled comparison with the clinic’s ordinary intake process, so it cannot establish whether adding AMIE improves diagnosis, saves clinicians time, or changes patient outcomes. The chat was text-only, and live physician supervision was part of the tested workflow.
The paper’s 90% figure also needs its precise definition: the final diagnosis appeared somewhere in AMIE’s first seven diagnostic possibilities. It does not mean the AI independently reached the correct diagnosis 90% of the time, nor that its top answer matched the clinician’s in that share of cases. Whether supervised pre-visit chats can improve care in a controlled, larger study remains an open question.