Standard automated perimetry, in common with other psychophysical tests, is profoundly influenced by reliability of performance. Reliability can be quantified using indices such as false negatives (FNs), false positives (FPs) and fixation losses (FLs), with cut-off values used to differentiate reliable and unreliable tests aiming to balance reliability with the risk of excluding too many tests.1 The current manufacturer guideline for the Humphrey Field Analyzer recommends a reliability limit of 20% FLs and 15% FPs.2
Tan and colleagues emphasize that setting reliability cut-offs is oversimplistic and introduces a false dichotomy between "reliable" and "unreliable". Reliability actually falls along a continuum and even unreliable visual fields may contain useful information. By discarding unreliable tests, one reduces the number of available tests, and potentially reduces power to detect progression. This study quantified the effects of reliability indices on mean deviation (MD) and pattern standard deviation (PSD).
A total of 1,828 visual field tests from 830 healthy participants were included in this cross-sectional study, with regression analysis used to examine the relationship between reliability indices and MD and PSD. A higher rate of FNs was associated with lower MD and higher PSD, whereas a higher rate of FPs was associated with higher MD and lower PSD. The relationship was non-linear, with a greater effect on global indices observed with higher rates of error. MD decreased by 1.15 dB per 5% increase in FNs when FNs were ?15% but decreased by only 0.71 dB when the FN rate was <15%. MD increased by 1.26 dB per 5% increase in FPs when FPs were ≥15%, compared a 0.65 dB increase when FPs were <15%. A similar relationship was observed between FPs and FNs and PSD. Although FLs are the most common reason for poor reliability3, they had no significant effect on MD and only a small effect on PSD, perhaps due to averaging of regions of falsely high or low sensitivity resulting in little overall effect.
The current recommendation to disregard FNs from reliability assessment does not mean that FNs should not be considered when estimating progression ratesThe study provides insight into the effect of unreliable responses on MD and PSD which may allow clinicians to better estimate rates of progression. Even low levels of FNs or FPs had a significant impact on MD and PSD meaning it may be important to consider the effect of reliability indices on MD when calculating rates of progression. Furthermore, the current recommendation to disregard FNs from reliability assessment2 does not mean that FNs should not be considered when estimating progression rates. FNs had a similar (though inverse) effect on global indices as FPs. Patients with glaucoma are well known to have higher FN rates due to inability to see the catch trial stimuli, however if FN rates affect MD and PSD this may need to be taken into account in progression analysis. Further research is needed as the study was limited to including only healthy subjects and it is possible that the relationship between reliability and global indices may differ in eyes with established visual field defects. For example, although FLs were not strongly associated with global indices in healthy subjects, others have reported higher FLs to decrease PSD and increase MD in patients with glaucoma.4