Detection of diabetic sensorimotor polyneuropathy by corneal confocal microscopy in type 1 diabetes: a concurrent validity study

Abstract

Objective: We aimed to determine the corneal confocal microscopy (CCM) parameter that best identifies diabetic sensorimotor polyneuropathy (DSP) in type 1 diabetes and to describe its performance characteristics.

Research design and methods: Concurrent with clinical and electrophysiological examination for classification of DSP, CCM was performed on 89 type 1 diabetic and 64 healthy subjects to determine corneal nerve fiber length (CNFL), density, tortuosity, and branch density. Area under the curve (AUC) and optimal thresholds for DSP identification in those with diabetes were determined by receiver operating characteristic (ROC) curve analysis.

Results: DSP was present in 33 (37%) subjects. With the exception of tortuosity, CCM parameters were significantly lower in DSP case subjects. In ROC curve analysis, AUC was greatest for CNFL (0.88) compared with fiber density (0.84, P = 0.0001), branch density (0.73, P < 0.0001), and tortuosity (0.55, P < 0.0001). The threshold value that optimized sensitivity and specificity for ruling in DSP was a CNFL of ≤14.0 mm/mm(2) (sensitivity 85%, specificity 84%), associated with positive and negative likelihood ratios of 5.3 and 0.18. An alternate approach that used separate threshold values maximized sensitivity (threshold value ≥15.8 mm/mm(2), sensitivity 91%, negative likelihood ratio 0.16) and specificity (≤11.5 mm/mm(2), specificity 93%, positive likelihood ratio 8.5).

Conclusions: Among CCM parameters, CNFL best discriminated DSP cases from control subjects. A single threshold offers clinically acceptable operating characteristics, although a strategy that uses separate thresholds to respectively rule in and rule out DSP has excellent performance while minimizing unclassified subjects. We hypothesize that values between these thresholds indicate incipient nerve injury that represents those individuals at future neuropathy risk.

Figure 1

ROC curve for the identification of DSP by the four CCM parameters in the 89 subjects with type 1 diabetes. The curve representing the performance of CNFL is depicted by the solid black line. The AUC for CNFL was the largest among the CCM parameters (0.88). See text for estimates of AUC for each test and their statistical comparisons with CNFL. According to the first approach for determining threshold values for the diagnosis of DSP described in research design and methods, the single point on the CNFL curve with optimal operating characteristics had a sensitivity of 85% and a specificity of 84%, corresponding to a threshold value of 14.0 mm/mm² (indicated by *). However, according to the second approach in which two threshold values were sought, the value 15.8 mm/mm² (indicated by †) had a sensitivity of 91% for ruling out DSP with a negative likelihood ratio of 0.16, whereas the value 11.5 mm/mm² (indicated by ‡) was associated with a specificity of 93% and a positive likelihood ratio of 8.5.

Figure 2

Box-and-whisker plots demonstrating the distribution of CNFL in 64 healthy volunteers and 89 type 1 diabetic subjects according to neuropathy status. Type 1 diabetic control subjects without DSP were divided into two groups based on the presence or absence of sural nerve conduction study abnormalities. Type 1 diabetic case subjects with DSP were divided into mild, moderate, and severe groups based on the presence of less than four, four, or greater than four lower-limb nerve conduction study abnormalities, respectively.

Figure 3

Representative images of CCM according to membership in the following groups: healthy volunteers (A), diabetic control subjects without DSP (B), and diabetic case subjects with varying DSP severity (C–E). TCNS scores of 0–5 are considered to represent low likelihood of DSP, 6–8 likelihood of mild neuropathy, 9–12 likelihood of moderate neuropathy, and 13–19 likelihood of severe neuropathy.