An unbiased stereological method for corneal confocal microscopy in patients with diabetic polyneuropathy

An unbiased stereological method for corneal confocal microscopy in patients with diabetic polyneuropathy

A summary of the demographic data of the participants obtained by Chen et al. is presented here1: The mean age was significantly higher in the DSPN( +) group compared to both the control and the DSPN(-) group, p < 0.001 (59 ± 11 years vs 44 ± 15 years and 44 ± 13 years, respectively) and the duration of diabetes was significantly longer in the DSPN( +) group (39 ± 14 years vs 23 ± 15 years). Neuropathy disability score was significantly higher in the DSPN( +) group compared to controls. Vibration perception threshold was significantly higher in the DSPN( +) group compared to both DSPN(-) and control (25.2 ± 13.4 V (DSPN( +)) vs. 7.6 ± 5.5 V (DSPN(-)) and 6.0 ± 5.5 V (control)), p < 0.001). Similarly, the peroneal motor nerve conduction velocity was lower in both diabetic groups compared to control and lowest in the DSPN( +) group (31.0 ± 9.5 m/s (DSPN( +)) vs 43.9 ± 3.1 m/s (DSPN(-)) vs 49.1 ± 3.4 m/s (control), p < 0.001).CCM parameters with the new methodManual CCM analysisUsing the randomized and area adjusted method, CNFD and CNFL were significantly reduced in the DSPN( +) group compared to both the control and the DSPN(-) group. In contrast, the CNFL values were larger in the DSPN(-) group compared to the healthy controls. There were no differences in CNBD between the groups (see Supplementary Table S1 and Supplementary Fig. S1, A online).Automated CCM analysisFor all CCM parameters (CNFD, CNFL and, CNBD) the values were significantly reduced in the DSPN( +) and DSPN(-) group compared to the controls and were lowest in the DSPN( +) group (see Supplementary Table S2and Supplementary Fig. S1, B online).Unadjusted versus adjusted area, randomized sampling method The adjusted area increased the actual CCM values by 35–64%. There was a strong correlation between the unadjusted and the adjusted area CCM parameters with Pearson’s correlation coefficients between r = 0.87 to r = 0.93 (see Supplementary Table 2 and Supplementary Fig. S2, A and B online). The relative increase in CNFL (both automated and manual analysis) when using the adjusted area was significantly lower in the control group (Kruskal Wallis test, p < 0.05, unpaired t-test, p < 0.05) compared to both the DSPN(-) and the DSPN( +) group. Likewise, for the CNFD and CNBD measurements, when using the adjusted area, there were tendencies for a higher relative increase in the diabetic groups compared to the control group (Kruskal Wallis test, p values from p = 0.06 to p = 0.4). There were no differences between the diabetic groups (CNFL: p = 0.052, manual, p = 0.08 automatic, unpaired t-test, CNFD and CNDB p > 0.05, Kruskal Wallis test).New method compared to skin biopsyThere was no correlation between the randomized sampling method and area adjusted method and IENFD, Pearson’s correlation coefficients between r = 0.04 and r = 0.13, respectively (Fig. 1). Likewise, there was no correlation between IENFD and the CCM values originally reported by Chen et al. (manual r = 0.17 (CNFL), r = 0.13 (CNBD), r = 0.23 (CNFD) and automated: r = 0.17 (CNFL), r = 0.10 (CNBD), r = 0.14 (CNFD)). Figure 2 presents a representative example of the ROC curves (CNFD and IENFD values). The AUC’s for DSPN were between 0.63 and 0.75 with no statistically significant differences between the two methods (CNFD; chi2 = 1.16, p = 0.56, CNFL: chi2 = 5.64, p = 0.06 and CNBD: chi2 = 3.16, p = 0.21).Figure 1Scatter plot and regression line of IENFD vs CCM results using the area adjusted and randomized sampling method, grouped by method of analysis (A: manual, B: automated). The dots represent the mean values from the individual participants and the dashed lines indicate the regression line. The Pearson’s correlation coefficients were r = 0.12 (manual method) and r = 0.13 (automated method) for the CNFD values, r = 0.11 (manual method) and r = 0.10 (automatic method) for the CNFL values and r = 0.066 (manual method) and r = 0.038 (automated method) for the CNBD values.
Figure 2ROC curve. AUC between 0.68 and 0.75, no statistical difference between the three curves, chi2 = 1.16, p = 0.56.
New method compared to the standard methodTo secure an equal analysis method, only the automated analyses were compared. The randomized sampling and adjusted area method generated numerically (8–40%) larger CCM parameters (Table 1) compared to the standard procedure (p < 0.05, paired t-test), except for CNFD in the control group.Table 1 Randomized sampling and adjusted/unadjusted area versus standard method with automated analysis.
When comparing the randomized sampling method to the standard method,
using the unadjusted area the CCM values were reduced compared to the standard method (p < 0.05, paired t test), but since the absolute difference was small it was not considered clinically relevant (Table 1).Interobserver reliabilityThere was no significant difference in mean between the investigator and the blinded second observer for CNFL, CNFD, and CNBD (p = 0.14, p = 0.29, p = 0.49, respectively, paired t-test) and correlations of variance were (investigator vs. observer) 32% vs. 45%, 39% vs. 17% and 50% vs. 48% for CNFL, CNFD, and CNBD respectively.

Via Source link