Predicting False-Positive Glaucoma Hemifield Test Results by Representative Glaucomatous Visual Field Patterns

Abstract

Purpose : Glaucoma Hemifield Test (GHT) is an important functional measure for glaucoma diagnosis. Assuming that glaucomatous visual field (VF) loss is irreversible, a reversal from 2 consecutive GHT outside normal limits (ONL) to within normal limits (WNL) is considered a false-positive result. In this retrospective cohort study, we investigate the relationships between such GHT reversals and the VF loss patterns of the baseline in early-stage glaucoma patients.

Methods : From 5 large glaucoma practices, eyes with reliable baseline and two annual follow-up Humphrey VFs (fixation loss≤33%, false negative/positive rates≤20%, SITA Standard 24-2), baseline mean deviation (MD) ≥-3dB and GHT ONL for the first 2 VFs were selected. For each baseline VF, the pattern coefficients are calculated for the 11 glaucomatous VF loss archetypes (ATs) and the normal AT, chosen from all 17 VF ATs identified by [1] (see Fig. 1; AT indices according to [1]). Logistic regression was applied to model the relationship between AT coefficients, MD, pattern standard deviation (PSD), and false-positive GHT. Stepwise regression was performed to select the best predictive model for false-positive GHT. 10 times repeated 10-fold cross validation was applied to evaluate the predictive model performance by the area under the receiver operating characteristic curve (AUC).

Results : 2,506 eyes were selected for data analysis. 12.1% of the GHTs reversed from 2 consecutive ONLs to WNL at follow-up. Fig. 2a shows the best predictive model (relevant parameters: 8 ATs, MD, and PSD): Higher values of MD and lower values of PSD and all ATs predict GHT false positives with an AUC of 0.717 (95% confidence interval: [0.715, 0.719]; Fig. 2b). A lower similarity to glaucomatous ATs as well as to the normal AT, reflected by negative coefficients, implies a higher probability of false positive GHTs.

Conclusions : 12.1% of eyes with baseline MD≥-3dB and two consecutive GHT ONL reverse to normality and may therefore be false-positives. Using a previously proposed computational VF decomposition method [1], we show that these false-positives can be statistically predicted by the lack of similarity to typical glaucomatous and typical normal VF patterns at baseline. Our model can be potentially used to recommend patients for VF retest.


[1] Elze et al., J. R. Soc. Interface 12:20141118, 2015

This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.

 

Fig. 1

 

Fig. 2