Corneal nerve imaging in systemic neuropathy: A minireview of experimental and precision medicine applications

Table 2 Representative quantitative evidence and patient-level diagnostic performance of corneal confocal microscopy

Clinical setting	Representative evidence	Sample size	Reference standard/comparator	Key quantitative findings	Diagnostic performance and interpretation
Impaired glucose tolerance/prediabetes[1]	3-year longitudinal cohort	30 participants with IGT + 17 controls; 10 progressed to T2D, 15 remained with IGT, and 5 reverted to normal glucose tolerance	3-year glycemic trajectory (progression to T2D, persistent IGT, or reversion to normoglycemia); healthy controls	Progressors showed lower baseline CNFD (20.0 ± 2.2 fibers/mm2 vs 30.7 ± 1.5 fibers/mm2), CNBD (25.6 ± 5.2 branches/mm2 vs 37.0 ± 2.7 branches/mm2), and CNFL (13.7 ± 1.2 mm/mm2 vs 20.4 ± 3.2 mm/mm2) than controls	Exploratory association with metabolic trajectory. No externally validated prediction model or neuropathy-specific prognostic threshold was developed
Diabetic peripheral neuropathy[5]	Systematic review and meta-analysis	38 studies; approximately 4000 participants	Study-specific clinical and/or neurophysiological definitions of DPN across included studies	CNFL was lower in established neuropathy than in diabetes without clinically established neuropathy by a pooled MD of -3.08 mm/mm2 (95%CI: -3.58 to -2.58; 34 studies; n = 3868). IWL was lower by -4.11 mm/mm2 (95%CI: -5.10 to -3.12; 6 studies; n = 459)	Strong evidence for group-level differences, but the meta-analysis did not establish a pooled patient-level diagnostic threshold
Diabetic sensorimotor polyneuropathy[48]	Pooled multinational multicenter cross-sectional study	998 participants with diabetes: 516 with type 1 diabetes and 482 with type 2 diabetes	Toronto consensus criteria incorporating lower-limb electrophysiological abnormality	Automated CNFL showed an AUC of 0.77 in type 1 diabetes and 0.68 in type 2 diabetes	For the overall cohort, a CNFL threshold of 12.3 mm/mm2 yielded AUC 0.71, sensitivity 67%, specificity 66%, PPV 59%, and NPV 74%. A lower threshold of 8.6 mm/mm2 provided 88% specificity, whereas CNFL > 15.3 mm/mm2 provided 88% sensitivity for exclusion. A substantial intermediate range remained unclassified
Predominantly mild/recent type 2 diabetic polyneuropathy[47]	Comparative cohort	374 participants: 214 with DPN, 63 with diabetes without DPN, and 97 controls	Toronto criteria for DPN; IENFD and thermal thresholds were not included in the DPN case definition	CNFL was significantly lower in patients with DPN than in participants without DPN and controls	AUC 0.55, sensitivity 14.4%, and specificity 95.7%. IENFD showed higher sensitivity (51.1%). These findings demonstrate that significant group-level differences do not necessarily translate into useful individual screening performance
Small- or mixed-fiber neuropathy[51]	Prospective unselected neurological cohort	680 patients assessed; 244 with small- or mixed-fiber neuropathy were included in the primary sensitivity analysis, and 179 patients with established alternative diagnoses were used for specificity calculations	Predefined clinical criteria for SFN/MFN; comparison with IENFD and cold-detection threshold	Limited concordance was observed between CCM and skin biopsy. Among the 244 affected patients, only 41 were abnormal on both tests, whereas 66 had abnormal CCM alone and 63 had abnormal skin biopsy alone	CCM sensitivity 44% (95%CI: 38%-51%), specificity 75% (95%CI: 69%-81%), and AUC 0.63. Skin biopsy showed sensitivity 43%, specificity 99%, and AUC 0.74. CCM therefore cannot be considered a direct substitute for skin biopsy
Parkinson’s disease with autonomic involvement[8]	Cross-sectional phenotyping study	71 patients with PD and 30 healthy controls: 14 without autonomic symptoms, 14 with single-domain autonomic involvement, and 43 with multiple-domain autonomic involvement	SCOPA-AUT autonomic-domain classification; healthy controls	CNFD decreased from 30.88 ± 2.42 fibers/mm2 in patients without autonomic symptoms to 23.63 ± 3.93 fibers/mm2 in those with multiple-domain autonomic involvement. CNFL decreased from 17.54 ± 2.03 mm/mm2 to 12.85 ± 2.55 mm/mm2	The combination of CNFD, CNBD, and CNFL yielded an AUC of 0.872 for distinguishing single-domain autonomic involvement from no autonomic involvement (n = 14 vs n = 14; sensitivity 85.7%, specificity 92.9%) and an AUC of 0.915 for distinguishing multiple-domain from single-domain autonomic involvement (n = 43 vs n = 14; sensitivity 79.1%, specificity 92.9%). These values represent cross-sectional phenotype discrimination rather than diagnosis of Parkinson’s disease or prediction of future progression
Chemotherapy-induced peripheral neuropathy[11]	Prospective longitudinal study	95 patients recruited; 73 included in the post-treatment analysis, 32 completed paired clinical Total Neuropathy Score assessments, and 14 underwent paired skin-biopsy and CCM assessment	Longitudinal clinical Total Neuropathy Score; paired skin biopsy available in a subgroup	Longitudinal reductions in corneal nerve density and density-to-tortuosity measures were reported following neurotoxic chemotherapy, whereas CNFL did not uniformly decrease	No validated disease-specific AUC, sensitivity/specificity threshold, or diagnostic cutoff is currently available. Findings support potential sensitivity to longitudinal treatment-related nerve changes but remain exploratory

AUC: Area under the receiver operating characteristic curve; CCM: Corneal confocal microscopy; CNBD: Corneal nerve branch density; CNFD: Corneal nerve fiber density; CNFL: Corneal nerve fiber length; CNF: Corneal nerve fiber; DPN: Diabetic peripheral neuropathy; IGT: Impaired glucose tolerance; IWL: Inferior whorl length; MD: Mean difference; NPV: Negative predictive value; PPV: Positive predictive value.