A total of 39 animals were used for the portion of the study comparing the optic nerve grading to axonal survival with the TEM sampling method. Both experimental and control eyes were available in all but one animal, in which the optic nerve section from the experimental nerve was not usable. Axon counts with the TEM full-count method were obtained in both eyes of 12 (31%) animals. Two, 5, 22, and 4 animals were treated with 0.05, 0.1, 0.2, and 0.4 μg ET-1/d respectively, whereas 6 were treated with PSS. Four, 11, 9, 5, and 10 animals were killed after 10, 21, 42, 63, and 84 days, respectively.
The agreement rates among the three observers for the eight sets of nerves is shown in
Table 1 . The mean ± SD interobserver κ for the eight sets of nerves was 0.67 ± 0.09. There was no obvious trend in interobserver variability with grading session (Spearman’s ρ = 0.571;
P = 0.139). The interobserver κ for all 170 nerves over the eight sets was 0.66 (95% CI: 0.58–0.74). When the control eyes were removed from the analyses, the agreement rates were lower
(Table 1) . The mean ± SD interobserver κ for the eight sets of nerves was 0.55 ± 0.21, whereas the interobserver κ for all experimental nerves over the eight sets was 0.55 (95% CI: 0.43–0.67). The intraobserver κ for the three observers were 0.80, 0.81, and 0.80 for all eyes and 0.60, 0.64 and 0.71, respectively, for experimental eyes only. The interobserver agreement results are shown in
Table 2 .
The median optic nerve damage grade in the 38 experimental eyes used for the comparison study was 2, with a range from 0 to 10. Thirty-eight (97%) of the 39 control nerves had a damage grade of 0, whereas 1 (3%) had a grade of 1. There was a good relationship between the optic nerve damage grade and axonal survival with the TEM sampling method (Spearman’s ρ = −0.677;
P < 0.001;
Fig. 2 ); however, there was considerable dispersion in the damage grades with mild axonal loss, or there was considerable dispersion in axonal loss estimated with the sampling method for a given damage grade. The relationship between damage grade and axonal survival with the TEM full-count method is shown in
Figure 3 (Spearman’s ρ = −0.926;
P < 0.001). The association between damage grade and axonal survival using the full full-count method was significantly higher than that with the sampling method (
P = 0.032). When only the subset of 12 animals in whom axonal survival was obtained with both methods was analyzed, the association between damage grade and axonal survival with the full-count method (Spearman’s ρ = −0.926) was higher than that with the sampling method (Spearman’s ρ = −0.827,
P < 0.001), although the differences did not reach statistical significance (
P = 0.338).
In nerves that had axonal survival determined by both TEM methods, the sampling method sampled a mean ± SD of 10.1% ± 2.0% and 9.9% ± 1.4% of the total number of axons in the experimental and control eyes, respectively. In these animals, the full-count method yielded a mean ± SD of 68,330 ± 40,630 axons in the experimental eyes and 127,190 ± 11,400 axons in the control eyes. The regression equation describing the relationship between axonal survival estimates using the two methods
(Fig. 4)yielded a slope of 0.839 and an intercept of 0.126. A Bland-Altman plot showing the difference versus mean of the two methods
(Fig. 5)shows that when the mean ratio of axonal survival was less than 0.7, the sampling method always underestimated the extent of damage, approaching differences in survival ratio of 0.2 in some cases. When the extent of damage was slight, the sampling method yielded survival ratios that were generally closer to those of the full-count method.