Two Machine Learning Models to Economize Glaucoma Screening Programs: An Approach Based on Neuronal Nets

Background: In glaucoma screening programs, a large proportion of patients remain free of open angle glaucoma (OAG) or have no need of IOP lowering therapy within 10 years of fol-low-up. Is it possible to identify a large proportion of patients already at the initial exam and thus to safely exclude them already at the initial exam? Methods: A total of 6889 subjects re-ceived a complete ophthalmological examination, including objective optic nerve head and quantitative disc measurements at the initial exam and after an average follow-up period of 11.1 years, complete data were available of 585 individuals. Two neuronal network models were trained and extensively tested. To allow the models to refuse to make a prediction in doubtful cases, a reject option was applied. Results: The first prediction model for the first endpoint ‘re-maining free of OAG within 10 years’ rejected to make a prediction in 46.4% of all subjects. In the remaining cases (53.6%), 271/271 (=100%) were correctly predicted. A prediction for the second endpoint ‘remaining free of OAG and no IOP lowering therapy within 10 years’ was rejected in 57% and in the remaining cases (43%), 253/253 (=100%) received a correct prediction. Conclusion: Most importantly, no eye was predicted false-negatively or false-positively. 43% of all eyes can safely be excluded from a glaucoma screening program for up to 10 years if one wants to be cer-tain that the eye remains free of OAG and will not have any need for a IOP lowering therapy. The corresponding model significantly reduces the screening amount and work load of oph-thalmologists. In the future, better predictors and models may enlarge the number of patients with a safe prediction to further economize time and health care budgets in glaucoma screening.