Development and Evaluation of an Artificial Intelligence Model to Set Target IOP for Glaucoma

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Glaucoma management relies on achieving target intraocular pressure (IOP) to prevent vision loss.
• Current methods for setting target IOP include clinician judgment and societal guidelines, with varying effectiveness.
• Personalized target IOP prediction may improve glaucoma treatment outcomes.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI) model for predicting personalized target intraocular pressure (IOP) in glaucoma patients.
• To compare the impact of AI-predicted, clinician-defined, and guideline-based target IOP on the rate of visual field (VF) worsening.

Main Methods:

• A machine-learning model was trained on a dataset of 14,871 eyes with baseline structural, functional, and clinical data.
• The model predicted target IOP for a progression dataset of 10,559 eyes with longitudinal VF testing.
• Linear models assessed the effect of target IOP differences on VF worsening (mean deviation slope).

Main Results:

• The AI model achieved a mean absolute error of 2.28 mmHg in predicting target IOP.
• AI and clinician target differences showed similar effects on VF outcomes, with a 0.032 dB/year and 0.026 dB/year faster rate of mean deviation worsening per 1 mmHg increase, respectively.
• Both AI and clinician target differences were superior to Canadian Ophthalmological Society (COS) guidelines and mean absolute IOP in mitigating VF worsening.

Conclusions:

• AI models demonstrate comparable performance to glaucoma specialists in setting target IOP.
• AI-based target IOP prediction is superior to using mean IOP or societal guidelines for preventing glaucoma progression.
• Further research is warranted to evaluate the clinical utility of AI-guided target IOP in broader patient populations.