Table 4 Summary of artificial intelligence-based glaucoma risk prediction models that do not use visual fields and imaging data.

Baxter et al.²⁸

Predicting need for surgical intervention within 6 months for patients (N = 385) with open angle glaucoma

48 features that can be broadly categorized into vital signs, body mass index, smoking status, comorbidities, hospitalization status, medications, and lab values

Logistic regression

Accuracy: 62%

Sensitivity: 78%

Specificity: 50%

Mehta et al.²⁹

Predicting self-report of open angle glaucoma in a population (N = 1689) without a clinical diagnosis at the time of testing

Age, gender, ethnicity, body mass index, forced vital capacity, peak expiratory flow, heart rate, diastolic and systolic blood pressure, diabetes, recent nicotine and caffeine intake, intraocular pressure, corneal hysteresis, and corneal resistance factor

Extreme gradient boosting (XGBoost)

Accuracy: 75%

Tielsch et al.³⁰

Predicting glaucoma in a normal population (N = 5308)

Age, race, intraocular pressure, family history of glaucoma, and diabetes

Logistic regression

Predicted probability threshold ≥ 0.025

Sensitivity: 86%

Specificity: 66%

Current study

Predicting self-report of glaucoma in a population without a clinical diagnosis at the time of testing

Age, gender, race, BMI, systolic and diastolic blood pressures, and comorbidities

Logistic regression, support vector machine, and adaptive boosting

Accuracy: 68%–74%

Sensitivity: 52%–57%

Specificity: 69%–77%