Abstract
Purpose :
Currently, there is no presentation biomarker that is significantly correlated with treatment outcome in optic neuritis (ON). Visual fields (VF) reveal global and regional deficits, reflecting visual pathway dysfunction. Archetypal analysis (AA), a form of unsupervised machine learning, has been used in glaucoma to monitor focal VF defects quantitatively. We theorized that AA could detect quantifiable, disease-specific patterns of VF loss that correlate with treatment effect and visual outcome, as well as reveal residual deficits in ON.
Methods :
Using the R statistical environment, we performed AA on 3,892 VFs prospectively collected from 456 eyes in the ON Treatment Trial. We decomposed each study eye VF into a weighted sum of its component ATs (total weight=1.0). We defined a minimum AT weight change (7%) distinguishable from normal fluctuations by decomposing VFs of 61 control eyes into ON-specific ATs. To compare outcomes and recovery rates between treatment groups, we created a mathematical model to describe change in AT weight and MD over time.
Results :
We used a 10-fold cross-validation model to create 16 ON-specific ATs (Figure 1), which were distinct from control VFs. At presentation, AT2, a severe global dysfunction pattern, had the highest weight (0.33±0.40). Over 6 months, we observed an exponential increase in AT1, a normal AT (Figure 2), and an exponential decrease in AT2 weight. AT1 demonstrated the greatest relative weight increase due to intravenous methylprednisolone (IVMP: 0.57±0.24; placebo: 0.51±0.26; oral prednisone: 0.49±0.27, p=0.01). Eyes with AT1 weight ≥0.19 at baseline (≥1 SD above mean) had better visual outcomes for both MD (-1.13 vs. -3.86 dB, p<0.001) and AT1 weight (0.68 vs 0.50 p<0.0001); however, a treatment benefit for IVMP occurred only for eyes with AT1 weight <0.19 at baseline (p=0.0115). Patients receiving IVMP recovered fastest (p<0.001). At 6 months, 182/227 eyes with MD>-2.00 dB had at least one abnormal AT.
Conclusions :
For the first time, AA reveals specific features of vision loss associated with outcome and treatment effect in ON. AT1 is a quantifiable biomarker of VF function, identifying patients who may benefit most from treatment. AA shows residual deficits in eyes labeled as normal by MD at outcome.
This is a 2021 ARVO Annual Meeting abstract.