Abstract
Purpose:
To develop a rigorous method for the determination of a Visual Field “Event”, defined as “a statistically significant sudden change in the overall sensitivity of an automated visual field (VF)”.
Methods:
We previously developed a method for detection of VF progression (Richards et.al. ARVO 2011 Program # 4155 ) by accounting for the “noise” at every location within a VF, based on transformation of variable to approximate a Gaussian distribution. Once the pointwise noise has been so defined for a series of VF’s, a sudden change in overall sensitivity (“Event”) may be determined by treating an existing series of Zeiss-Humphrey Sita-Standard 24-2 VF’s as a 52-dimensional normal distribution and computing the 52-dimensional Chi-square value for a new VF, based on the established means and variances, and requiring p<0.05. We tested this method by analyzing 92 VF’s of 10 eyes of 10 patients and comparing results with modified Parrish-Hodapp-Anderson (PHA) criteria ( AAO BCSC 2011-2012 Section 10 p 77) for: Case 1, Deepening of existing scotoma; Case 2, Depression of a point adjacent to existing scotoma; Case 3, Depression of a previously normal point; and Case 4, Depression of two adjacent normal points.
Results:
Results for all 4 cases are significant at p<0.05 level by Mann Whitney U Test. Case 1: Criterion of 7 db is too weak; deepening by >10 db is needed. Case 2: Criterion of 9 db is too strong; deepening by 5 db is sufficient. Case 3: Criterion of 11 db is too strong; deepening by 7 db is sufficient. Case 4: Criterion of 5 db is too weak; deepening by 6 db is needed.
Conclusions:
Transformation of variable and accounting for pointwise noise of a series of VF’s permit, for the first time, computation of a precise statistical measure of an “Event” when a new VF is obtained. Previously used criteria, such as PHA, while reasonable estimates, may not always be reliable.
Keywords: 758 visual fields