Artificial Intelligence Safe to use in Primary Care Setting

AN ARTIFICIAL INTELLIGENCE (AI) system has demonstrated safety, specificity, and sensitivity for speciality-level diagnosis in the primary care setting. Designed by researchers at the University of Iowa, Iowa City, Iowa, USA, the autonomous system was successfully used to detect diabetic patients who were likely to have more than mild diabetic retinopathy that required further care, which is a severe complication of diabetes and the leading cause of vision loss in adults. This trial led to the device becoming the first autonomous AI diagnostic system to be authorised by the U.S. Food and Drug Administration (FDA) across any sphere of medicine.

Nine hundred adult diabetes patients with no history of diabetic retinopathy were enrolled from 10 American primary care settings. The IDx-DR AI device was used to assist the operator in taking retinal images of the patients with a robotic camera and then to make a clinical diagnosis, detecting characteristics such as microaneurysms, haemorrhages, and lipoprotein exudates. As a control, the trial participants also had retinal images taken using specialised equipment without AI operated by retinal photographers certified by the Wisconsin Fundus Photograph Reading Center (FPRC), which is the gold standard diagnostic method for diabetic retinopathy. “This was much more than just a study testing an algorithm on an image. We wanted to test it in the places where it will be used, by the people who will use it, and we compared it to the highest standard in the world,” stated principal investigator Dr Michael Abramoff, Department of Ophthalmology and Visual Sciences, University of Iowa.

With data available from 819 of the 900 study participants, it was clear that IDx-DR exceeded all prespecified superiority endpoints of the trial (>85% sensitivity and >82.5% specificity). The AI system correctly identified 173 participants with diabetic retinopathy out of the 198 identified by the FPRC readers, equating to a sensitivity of 87%. In addition, the specificity and imageability rate were 90% and 96%, respectively, showing the system was also able to effectively diagnose patients and detect those who were disease-free to a similar level to the current gold standard technique.

With the first IDx-DR systems beginning to be used in the clinic, these results are of great significance for diabetes patients, for whom early detection and treatment of diabetic retinopathy can reduce the risk of blindness by 95%. The team noted that these autonomous AI systems may also be effective for detecting other diseases, such as glaucoma and macular degeneration, providing the primary care setting with speciality diagnostic techniques.