Discrimination of Glaucoma Patients From Healthy Individuals Using Combined Parameters From Spectral-domain Optical Coherence Tomography in an African American Population.

PURPOSE

To create a multivariable predictive model for glaucoma in an exclusively African American population and to compare the performance of the model with individual structural parameters derived from SD-OCT.

PATIENTS AND METHODS

A total of 103 healthy eyes and 118 glaucomatous eyes of African American patients underwent SD-OCT optic disc and macular scanning. Twenty-seven optic nerve head, retinal nerve fiber layer (RNFL), and ganglion cell parameters were collected. A multivariable model was derived using a backward elimination variable selection procedure. Areas under the curve were used to measure the diagnostic performance of the individual parameters and the multivariable model.

RESULTS

The best performing parameters for glaucoma patients included inferior quadrant thickness (AUC=0.9239), average RNFL thickness (AUC=0.9209), sup2 RNFL thickness (AUC=0.9157), superior quadrant thickness (AUC=0.8906), and vertical CDR (AUC=0.8640). The best performing parameters for early glaucoma patients were sup2 RNFL thickness (AUC=0.8680), inferior quadrant thickness (AUC=0.8571), average RNFL thickness (AUC=0.8550), superior quadrant thickness (AUC=0.8420), and inf2 RNFL thickness (AUC=0.8420). The AUC of the multivariable model was 0.8918 for early glaucoma and 0.9744 for moderate/advanced glaucoma. There was some variability in the performance of the model based on disc size.

CONCLUSIONS

These findings confirm that several individual RNFL, ONH, and GCA parameters have excellent diagnostic performance in differentiating glaucomatous patients from healthy patients in African American population. A multivariable model was developed and validated with high diagnostic accuracy.