Using the described technique, a well-formed S-cone ERG was
obtained at frequencies of 4 Hz or less. The absence of any response
after the blue–green transition indicates that there was no
contamination from L- or M-cones. The purity of the S-cone signal
obtained by this method could be demonstrated by changing the green
intensity away from the point of balance
(Fig. 1) . A small increase in green produced M-cone ON responses at the offset
of blue. A reduction in green produced M-cone OFF responses at the
offset of blue, and a further reduction in green reduced the intensity
below the rod saturation level, resulting in rod intrusion. To obtain a
large-amplitude late negative response (S-cone PhNR) the frequency was
reduced to 1.67 Hz.
The group-averaged ERG signals elicited by the three stimulation
techniques each had a characteristic waveform in which the small
initial negative wave was followed by a major positive-going potential
and a larger negative-going potential.
(Figs. 2 3) . The a-wave amplitude was measured from the baseline, and the leading
edge of the b-wave and PhNR were measured peak to peak. As in previous
studies,
2 the latencies were less significantly affected
in the POAG group than were the potential amplitudes. Therefore, only
amplitude data are presented in the present analysis. The data for each
electrophysiological measure, using the three criteria, are shown in
Table 2 . All three methods produced statistically significant results. The most
significant difference between the POAG and the control subjects was
achieved with the S-cone PhNR. (
P = 0.0001). Similarly,
the ROC analysis provided the most favorable ROC area score (0.86) with
a sensitivity of 89%, specificity of 74%, and general error rate of
19%, for a cutoff criterion of 5.38 mV. The S-cone PhNR therefore had
the best performance, closely followed by the PERG (positive P50–N95).
The PhNR (L&M-cone) scored less effectively but still provided a
significant indication of diffuse neural damage
(Table 2) . The only
significant correlations of electrophysiologic and perimetric data were
between the P50–N95 and the pattern standard deviation (PSD;
Spearman’s rho, −0.512,
P = 0.32), and the S-cone
PhNR and mean deviation (MD; Spearman’s rho, −0.679,
P = 0.002). These two forms of ERG were poorly
correlated with each other and may be regarded as relatively
independent criteria
(Fig. 4) .