Table 1 Summary of main features of models combining OCT parameters for the diagnosis of early glaucoma
Model | Analytical Method | Proposed Combination | Predicted Probability Cutoff Points | AUC 95% CI width | Strength | Weakness | Validation |
|---|---|---|---|---|---|---|---|
GSDI [37] | Multivariable logistic regression | Average RNFL+GCC; focal loss volume RNFL+GCC; VCDR | Not provided | – | Improved diagnostic ability | Inter-variable collinearity | Internal |
OCT Glaucoma Diagnostic Calculator [39] | Multivariable logistic regression | Age, color code for SN, ST, and min GCIPL, CDR, and values of CDR, VCDR, IT GCIPL and inferior quadrant RNFL. | < 0.3 = low; 0.3–0.6 = intermediate; > 0.6 = high | 0.046 | Improved diagnostic ability | Inter-variable collinearity | Internal |
UNC OCT Index [36] | Exploratory Factor Analysis; multivariable logistic regression | Composite RNFL (average, superior and inferior quadrants), composite ONH (VCDR, CDR, rim area), composite GCIPL (all 8 parameters), age | ≤0.34 = low; > 0.34 = high | 0.011 | Minimized inter-variable collinearity, improved diagnostics ability | Some information in the original set of variables may be lost when running the factor analysis | Internal, external |
Baskaran et al. [32] | Classification And Regression Tree | RNFL (superior, inferior and nasal quadrant, symmetry); ONH (disc area, VCDR, cup volume) | Not provided | 0.01 | Low misclassification rate; improved diagnostic ability | Inter-variable collinearity | Internal |
Baskaran et al. [32] | Linear Discriminant Analysis | RNFL (average, IQ, SQ); ONH (VCDR, CDR, disc and rim area) | Not provided | 0.02 | Low misclassification rate; improved diagnostic ability | Inter-variable collinearity | Internal |
Blumberg et al. [31] | Logistic regression | Disc area; RNFL (superior quadrant, clock-hours 8, 12, 1, 6); age | Not provided | 0.247 | Improved diagnostic ability | Wide AUC 95% CI | Internal |
Fang et al. [30] | Logistic regression | Average RNFL, VCDR, rim area | Not provided | 0.109 | – | Variables chosen arbitrary. Wide AUC 95% CI. Inter-variable collinearity. | No |
Huang et al. [33] | Linear Discriminant Analysis | RNFL (ST1 and 2, NU2, IT1), disc area, standard deviation of superior and inferior hemispheric GCC | ≤0.131 | 0.045 | Improved diagnostic ability | Inter-variable collinearity | No |
Yoshida et al. [35] | Random Forest classification | ONH, RNFL, and GCIPL | Not provided | 0.028 | Improved diagnostic ability | Risk of overfitting | Internal |