Development of a machine learning model using systemic and ophthalmic parameters to detect sleep-disordered breathing in glaucoma patients.

PURPOSE: To develop and validate a machine-learning model using systemic and ophthalmic parameters that predicts sleep-disordered breathing (SDB) in patients with open-angle glaucoma (OAG). STUDY DESIGN: Retrospective cross-sectional study METHODS: We analyzed 513 patients with OAG (955 eyes) treated at Seiryo Eye Clinic. All the participants underwent comprehensive ophthalmic examinations, including Humphrey visual field (HVF) testing, and home sleep apnea testing (HSAT) to obtain the 4% oxygen desaturation index (ODI). SDB was operationally defined as ODI-SAS, ie, sleep apnea syndrome (SAS) defined by HSAT-derived 4% ODI ≥15 events per hour. Sixteen algorithms were trained to predict ODI-SAS; model performance was assessed by the area under the receiver operating characteristic curve (ROC-AUC), precision, recall, the F1 score, and the Matthews correlation coefficient (MCC). RESULTS: One hundred fifty-eight patients (30.80%) met the ODI-SAS criterion. ODI-SAS status was associated with a history of hypertension, higher body mass index, family-reported apnea, shorter axial length, fewer antiglaucoma medications, and worse inferocentral total deviation on HVF (all P < 0.05), but not with self-awareness of snoring or previous SAS diagnosis. A gradient-boosted decision-tree model (CatBoost) achieved the best performance (ROC-AUC 0.86), with precision 0.692, recall 0.711, F1 score 0.701, and MCC 0.543. CONCLUSION: Machine-learning models can predict ODI-SAS in glaucoma using systemic risk factors together with ophthalmic features, including inferocentral visual-field defects. Such models may help identify OAG patients who warrant formal sleep evaluation.