Machine learning models based on electronic health record data can predict whether glaucoma patients will need surgery. A study presented at the 2023 annual meeting of the American Academy of Ophthalmology used a 5-center consortium, with 1 site serving as an external set, to test the machine learning models.

Sophia Ying Wang, MD, of Stanford University in California, and colleagues from the Sight Outcomes Research Collaborative randomly selected 3,000 patients from each center to be in development and test sets and 34,747 patients for model training. Among 43,321 patients, 5,544 (12.8%) underwent glaucoma surgery. Model area under the receiver operating characteristic curve ranged from 0.651 to 0.675 on the random test set and 0.623 to 0.673 on the external test site.

Although the results were generalizable across different independent academic eye centers, Dr. Wang and colleagues noted caution should be taken when deploying these models to populations outside of the original training set.