PURPOSE OF THE STUDY: We recently reported that it is beneficial to apply least absolute shrinkage and selection operator (Lasso) regression to predict future 24-2 visual field (VF) progression. The purpose of the current study was to investigate the usefulness of Lasso regression to predict VF progression in the central 10 degrees (10-2) in glaucoma patients. METHODS: Series of 10 VFs (Humphrey Field Analyzer 10-2 SITA-standard) from each of 149 eyes in 110 open angle glaucoma patients, obtained over 5.7±1.4 years (mean±SD) were investigated. Mean deviation values of the 10th VF were predicted using varying numbers of VFs (ranging from the first to third VFs to the first to ninth VFs), applying ordinary least square regression (OLSLR) and Lasso regression. Absolute prediction errors were then compared. RESULTS: With OLSLR, prediction error varied between 5.4±5.0 (using first to third VFs) and 1.1±1.6 dB (using first to ninth VFs). Significantly smaller prediction errors were obtained with Lasso regression, in particular with small numbers of VFs (from 2.1±2.8: first to third VFs, to 1.0±1.6 dB: first to ninth VFs). A large λ value, which is an index showing the degree of penalty in Lasso regression, was observed when a small number of VFs were used for prediction. CONCLUSION: Mean deviation prediction using OLSLR with a small number of VFs resulted in large prediction errors. It was useful to apply Lasso regression when predicting future progression of the central 10 degrees, compared to OLSLR.
PURPOSE OF THE STUDY: We recently reported that it is beneficial to apply least absolute shrinkage and selection operator (Lasso) regression to predict future 24-2 visual field (VF) progression. The purpose of the current study was to investigate the usefulness of Lasso regression to predict VF progression in the central 10 degrees (10-2) in glaucomapatients. METHODS: Series of 10 VFs (Humphrey Field Analyzer 10-2 SITA-standard) from each of 149 eyes in 110 open angle glaucomapatients, obtained over 5.7±1.4 years (mean±SD) were investigated. Mean deviation values of the 10th VF were predicted using varying numbers of VFs (ranging from the first to third VFs to the first to ninth VFs), applying ordinary least square regression (OLSLR) and Lasso regression. Absolute prediction errors were then compared. RESULTS: With OLSLR, prediction error varied between 5.4±5.0 (using first to third VFs) and 1.1±1.6 dB (using first to ninth VFs). Significantly smaller prediction errors were obtained with Lasso regression, in particular with small numbers of VFs (from 2.1±2.8: first to third VFs, to 1.0±1.6 dB: first to ninth VFs). A large λ value, which is an index showing the degree of penalty in Lasso regression, was observed when a small number of VFs were used for prediction. CONCLUSION: Mean deviation prediction using OLSLR with a small number of VFs resulted in large prediction errors. It was useful to apply Lasso regression when predicting future progression of the central 10 degrees, compared to OLSLR.