Personalized model learns patients’ glaucoma-progression dynamics

By Cheryl Guttman Krader

Las Vegas-A novel glaucoma monitoring system is being developed to guide providers in establishing personalized monitoring schedules that would help them avoid missing significant progression in glaucoma suspects or patients with open-angle glaucoma, Joshua D. Stein, MD, MS.

In an initial validation study using data from patients in the Advanced Glaucoma Intervention Study and the Collaborative Initial Glaucoma Treatment Study, the approach detected open-angle glaucoma progression 51% earlier, with 33% better accuracy, and 37% fewer tests than use of a fixed annual monitoring interval, said Dr. Stein, at Glaucoma 2015 during the annual meeting of the American Academy of Ophthalmology.

“We are now applying for an NIH RO1 grant to validate the model on a sample from the Ocular Hypertension Treatment Study and to enhance its inputs and outputs,” said Dr. Stein, associate professor of ophthalmology and visual sciences, University of Michigan, Ann Arbor.

The project is a collaboration between Dr. Stein and colleagues at the University of Michigan together with researchers at the University of Iowa and University of Pittsburgh, along with Ocular Hypertension Treatment Study investigators.

The system is designed to integrate information on IOP, visual fields, and changes in OCT to determine disease stability. And, it dynamically incorporates new information at successive visits with the historical information to help determine whether the treatment regimen should be changed.

“This is a personalized model that learns the patients’ unique disease-progression dynamics over time,” Dr. Stein said. “In addition it is flexible so that providers can tailor their monitoring based on characteristics of individual patients such as age and disease severity, and it can be easily integrated into the busy clinical setting.”

The system uses Kalman filtering-a forecasting and noise reduction technique useful for modeling complex, large systems, which was first used by NASA engineers to guide space flights to the moon. It combines a population-based understanding of disease evolution with the individual’s characteristics to predict values of key clinical parameters in the future.

“If we see a patient in our clinic who is a glaucoma suspect, we may be asked to be a soothsayer to figure out if years later, the visual field will continue to look normal or be changed,” he said. “Wouldn’t it be great to have some insider information.”

Unlike traditional approaches for identifying progression that compare the individual to a normative database, the Kalman filter generates personalized information and progressively learns about how the state changes over time.

In addition, it is able to extract measurement noise from the estimates.

“We all know how noisy visual field and IOP data can be,” Dr. Stein said.

Results of the initial validation study were published in 2014 [Schell GJ, et al. Ophthalmology. 2014;121:1539-1546].

Dr. Stein demonstrated the performance of the forecasting technique with examples from 3 patients that showed the visual field mean deviation predicted by the model was very close to the observed values.