Corneal confocal microscopy for the diagnosis of diabetic peripheral neuropathy: A systematic review and meta-analysis.

INTRODUCTION: Corneal confocal microscopy (CCM) is a rapid non-invasive ophthalmic imaging technique that identifies corneal nerve fiber damage. Small studies suggest that CCM could be used to assess patients with diabetic peripheral neuropathy (DPN). AIM: To undertake a systematic review and meta-analysis assessing the diagnostic utility of CCM for sub-clinical DPN (DPN- ) and established DPN (DPN+ ). DATA SOURCES: Databases (PubMed, Embase, Central, ProQuest) were searched for studies using CCM in patients with diabetes up to April 2020. STUDY SELECTION: Studies were included if they reported on at least one CCM parameter in patients with diabetes. DATA EXTRACTION: Corneal nerve fiber density (CNFD), corneal nerve branch density (CNBD), corneal nerve fiber length (CNFL), and inferior whorl length (IWL) were compared between patients with diabetes with and without DPN and controls. Meta-analysis was undertaken using RevMan V.5.3. DATA SYNTHESIS: Thirty-eight studies including ~4,000 participants were included in this meta-analysis. There were significant reductions in CNFD, CNBD, CNFL, and IWL in DPN- vs controls (P < 0.00001), DPN+ vs controls (P < 0.00001), and DPN+ vs DPN- (P < 0.00001).
CONCLUSION: This systematic review and meta-analysis shows that CCM detects small nerve fiber loss in subclinical and clinical DPN and concludes that CCM has good diagnostic utility in DPN.

Introduction

Diabetic peripheral neuropathy (DPN) affects ~50% of patients with diabetes and leads to significant morbidity including neuropathic pain, erectile dysfunction, and foot ulceration . Currently, the diagnosis of DPN in clinic relies on symptoms, loss of sensation to the 10 g monofilament, neurological examination, and occasionally electrophysiology . However, these methods do not reliably detect small nerve fiber damage which occurs in early DPN .

In 2003, we showed that the ophthalmic technique of corneal confocal microscopy (CCM) can identify corneal small nerve fiber loss in patients with early and established DPN . Subsequently we and others demonstrated good diagnostic utility for DPN , , , comparable to intra‐epidermal nerve fiber density (IENFD) , . CCM also predicts incident DPN , and identifies individuals at higher risk of developing DPN . However, some studies have failed to demonstrate corneal nerve fiber loss in patients with and without DPN , , which has been attributed to a small sample size and variances in image acquisition and analysis protocols .

We have undertaken a systematic review and meta‐analysis to generate a definitive single estimate for the diagnostic utility of CCM in sub‐clinical and clinical DPN.

Methods

Data sources and searches

This systematic review and meta‐analysis is reported in accordance with MOOSE guidelines . The protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO) on November 2020 (CRD42018093498). Four databases were chosen to search for this systematic review: PubMed, EMBASE (Ovid), CENTRAL, and web of science (WoS)‐ (1900‐present). In the PubMed and CENTRAL database both Mesh subject headings and keywords were searched; in Embase‐(1988‐present) Emtree subject headings and keywords were utilized. Numerous terms were tested for relevancy and the final search strings for the three databases can be found in Table S1 in the supplement. Article language was limited to English and no date restrictions were set. A segment of the grey literature was searched through the use of dissertation and theses (ProQuest) and Clinicaltrials.gov. The databases were searched from inception to April 2020.

We included observational studies that reported on at least one of the following CCM parameters: corneal nerve fiber density (CNFD), corneal nerve branch density (CNBD), corneal nerve fiber length (CNFL), or inferior whorl length (IWL) in any of the following three groups: patients with type 1 and/or type 2 diabetes with diabetic peripheral neuropathy (DPN+), without diabetic peripheral neuropathy (DPN−), and controls. Cross‐sectional and longitudinal observational studies were included in this systematic review and meta‐analysis. Narrative reviews, systematic reviews, correspondence, and case reports were excluded. Study country, age, diagnosis (DPN+, DPN−, control), duration of diabetes, HbA1c, software used for image analysis, CNFD, CNBD, CNFL, and IWL were extracted when available. Studies using CCMetrics, ACCMetrics, ImageJ, and other morphometric software to quantify CNFD, CNBD, and CNFL were included. IWL was quantified using CCMetrics and ACCmetrics only. Data presented as median (IQR) were converted into mean ± SD using an online calculator and data presented as mean ± SEM were converted into mean ± SD using the RevMan calculator . HbA1c presented in (%) was also converted into (mmol/mol) using the NGSP calculator, where NGSP % must be between 3 and 20 . Original studies that staged DPN as per the diabetic neuropathy study group in Japan (DNSGJ) were classified as: DPN− for stage I, DPN+ for stages II–V, for meta‐analysis reporting purpose , . Stage I was reported as DPN− and stages II–III were reported as DPN+ in this study . Patients classified according to the modified neuropathy disability score (NDS) were grouped as: scores between 0–2 (DPN−) and 3–10 (DPN+) , . No neuropathy was classified as DPN− and mild‐severe neuropathy was classified as DPN+ , , , . No differentiation was made for either painful or painless DPN and both were classified as DPN+ , . Where the vibration perception threshold (VPT) was used, <15V was classified as DPN− and ≥15V as DPN+ .

Study selection

After the removal of duplicates, all citations were screened for relevance using the full citation, abstract, and indexing terms, before excluding studies deemed as irrelevant. Where there was a lack of consensus a third (senior) author was consulted. Duplicates were removed and the most recent and complete versions of the studies were reviewed for eligibility. Relevant studies were assessed by two reviewers (HG and INP) to assess eligibility according to the pre‐specified inclusion and exclusion criteria. Full manuscripts of these potentially eligible citations were obtained. Two reviewers made the final inclusion and exclusion decisions independently and in the case of disagreement, a third reviewer was consulted to resolve any conflicts. A flow chart of search results was produced (Figure S1). A data collection tool was developed to extract the data from each study. Data verification was undertaken by two reviewers (HG and INP). In the event of missing data, the authors were emailed to obtain unpublished data.

Data extraction and quality assessment

The included studies were assessed using the Cochrane Collaborations tool for assessing the risk of bias (section 8.5) . The tool categorizes the risk of bias into high, moderate, low, or unclear risk. This tool assessed six domains: selection bias, performance bias, detection bias, attrition bias, reporting bias, and other bias, where applicable. Quality assessment was undertaken by two reviewers (AK and GP). If the risk of bias of a study was unclear, the effect of removing the study was checked and relevant outcomes were reported (Table S2).

Data synthesis and analysis

Meta‐analysis was performed in RevMan (version 5.3) . Random effects meta‐analysis was used in anticipation of heterogeneity due to differences in study population and type and duration of diabetes. The mean difference (MD) with a 95% confidence interval (CI) was calculated for CNFD, CNBD, CNFL, and IWL. The Chi‐squared (χ2) test was used to test for difference between subgroups. The I2 statistic was calculated, which is derived from Cochrane’s chi‐squared test Q and is used to describe the percentage of between‐study variations attributed to variability in the true exposure effect . An I 2 value of 0–40% was classified as not important, 30–60% moderate, 50–90% substantial, and 75–100% considerable .

Results

The search strategy identified 1,310 records (Figure S1). In total, 557 papers were screened on the basis of titles and abstracts, of which 508 were excluded, leaving 49 full text papers of which 38 were included in the meta‐analysis.

Study characteristics

The studies were conducted in Canada , , , , , United Kingdom , , , , , , , , , , , , , , , , Germany , Denmark , Australia , , , , , , , Japan , , , , and China , (Table 1).

Table 1

Characteristics of the included studies

Study	Country	Group	n	Age (years)	Duration of diabetes (years)	HbA1c% – mmol/mol	CCM Type	Software for image analysis	Assessment with CCM
									CNFD	CNBD	CNFL	IWL
Ahmed et al. 31	Canada	DPN+ DPN− Control	33 56 64	50 ± 14.3 34.9 ± 14.8 38.9 ± 17.6	31.4 ± 13.5 17.6 ± 14 N/A	8.7 ± 2.1–72 7.4 ± 1.3–57 NS	HRT‐II	CCMetrics	√	√	√
Ostrovski et al. 32	Canada	DPN+ DPN− Control	13 13 20	56.2 ± 8.7 30.3 ± 13.7 41.3 ± 17.3	34.8 ± 13 10.7 ± 6.2 N/A	8.5 ± 2.2–69 7.5 ± 1.3–58 5.5 ± 0.4–37	HRT‐III	CCMetrics ACCMetrics	√	√	√
Lovblom et al. 10	Canada	DPN+ DPN−	11 54	38 ± 16 34 ± 15	21 ± 9 17 ± 12	8.1 ± 1.6–65 7.6 ± 1.3–60	HRT‐III	CCMetrics	√	√	√
Sivaskandarajah et al. 33	Canada	DPN+ DPN− Control	33 63 64	48.5 ± 13.7 32.7 ± 13.6 38.3 ± 16.4	32.3 ± 13.1 17.3 ± 12.2 N/A	8.4 ± 1.6–68 7.5 ± 1.2–58 5.6 ± 0.4–38	HRT‐III	CCMetrics	√	√	√
Hertz et al. 26	Canada	DPN+ DPN− Control	14 12 20	NS NS 41.4 ± 17.3	NS NS N/A	NS NS 5.5 ± 0.4–37	HRT‐III	CCMetrics	√	√	√
Alam et al. 9	UK	DPN+ DPN− Control	31 30 27	53.3 ± 11.9 38.8 ± 12.5 41 ± 14.9	37.2 ± 13.1 17.2 ± 12 N/A	8.5 ± 1.5–69 8 ± 1.3–64 5.5 ± 0.3	HRT‐III	CCMetrics	√	√	√
Azmi et al. 34	UK	DPN+ Control	29 32	61.9 ± 12.3 47.7 ± 1.6	46 ± 13.9 N/A	8.3 ± 1.3 5.7 ± 0.6	HRT‐III	ACCMetrics	√	√	√
Chen et al. 35	UK	DPN+ DPN− Control	29 63 84	63 ± 12 44 ± 15 46 ± 15	19.9 ± 11.7 20 ± 11.1 N/A	8.6 ± 3.6–70.4 ± 16 8 ± 4.1–63.9 ± 21.2 5.6–37.4 ± 3.5	HRT‐III	CCMetrics ACCMetrics	√	√	√
Brines et al. 21	UK	DPN+ DPN− Control	60 21 48	35.3 ± 14.3 37.1 ± 16.5 46.2 ± 16.9	35.3 ± 14.3 17.9 ± 15.1 N/A	8.2 ± 1.3–66 7.9 ± 1.3–63 5.7 ± 0.3–39	HRT‐III	ACCMetrics	√	√	√
Petropoulos et al. 36	UK	DPN+ DPN− Control	25 28 15	60.1 ± 10.2 42.4 ± 14.7 NS	24.8 ± 19.5 16.2 ± 9.3 N/A	7.6 ± 1.5–60 NS 5.4 ± 0.5–36	HRT‐III	CCMetrics	√		√	√
Chen et al. 8	UK	DPN+ DPN− Control	17 46 26	59 ± 11 44 ± 13 44 ± 15	39 ± 14 23 ± 15 N/A	8.5 ± 1.3–69 8.2 ± 1.4–66 5.5 ± 0.3–37	HRT‐III	CCMetrics ACCMetrics	√	√	√
Petropoulos et al. 37	UK	DPN+ DPN− Control	61 50 47	56.5 ± 13.2 44.2 ± 15.6 52 ± 13.2	35.33 ± 14.3 23 ± 14 N/A	8.4 ± 1.8–68 7.9 ± 1.7–63 5.6 ± 0.3–38	HRT‐III	CCMetrics	√	√	√
Petropoulos et al. 38	UK	DPN+ DPN− Control	100 86 55	NS NS 51.7 ± 11.4	34.4 ± 17.3 24.2 ± 21.2 N/A	7.9 ± 1.6–63 7.7 ± 1.6–61 5.5 ± 0.3–37	HRT‐III	CCMetrics ACCMetrics	√	√	√
Ponirakis et al. 39	UK	DPN+ DPN−	46 64	60.75 ± 8.9 45.5 ± 14.4	36.5 ± 14.4 22.25 ± 13	8.6 ± 0.4–70 7.62 ± 0.48–60	HRT‐III	CCMetrics	√	√	√
Quattrini et al. 24	UK	DPN+ DPN− Control	44 10 15	59.3 ± 17.25 43.5 ± 10.2 55 ± 18.5	NS NS N/A	8.01 ± 2.32–64 7.16 ± 1.26–55 NS	Confoscan‐P4	Morphometric software	√	√	√
Tavakoli et al. 40	UK	DPN+ DPN− Control	67 34 17	59 ± 18.2 55 ± 11.1 55 ± 19.8	17.8 ± 29.55 10.7 ± 10.6 N/A	8.2 ± 2.70–66 8.1 ± 1.57–65 <6.5 < 48	Confoscan‐P4	Morphometric software	√	√	√
Tavakoli et al. 25	UK	DPN+ DPN− Control	96 42 26	59 ± 20 57 ± 13 53 ± 3	59 ± 20 57 ± 13 N/A	8.30 ± 30.14–67 7.88 ± 10.23–63 ~5.8–40	Confoscan‐P4	CCMetrics	√	√	√
Kalteniece et al. 41	UK	DPN+ DPN− Control	69 47 22	62.08 ± 11.6 46.9 ± 13.2 50.32 ± 13.7	20.78 ± 17.8 16.04 ± 12.2 N/A	7.19 ± 10.16–55 7.72 ± 2.06–61 5.48 ± 00.42–36	HRT‐III	CCMetrics	√	√	√	√
Kalteniece et al. 28	UK	DPN+ Control	140 30	65.09 ± 1.13 61.2 ± 1.33	21.8 ± 2.05 N/A	7.5 ± 0.17–58 5.63 ± 00.06–38	HRT‐III	CCMetrics	√	√	√	√
Malik et al. 4	UK	DPN+ DPN− Control	14 4 18	59.2 ± 9.9 53 ± 18.5 57.8 ± 11.5	23.4 ± 6.25 21.3 ± 3.6 N/A	8.15 ± 1.3–66 7.8 ± 0.8–62 <6.5–48	Confoscan‐P4	Morphometric software	√	√	√
Ponirakis et al. 42	UK	DPN+ DPN− Control	33 41 70	64.1 ± 1.79 44.3 ± 2.19 41.8 ± 1.63	37.6 ± 3.2 23.3 ± 2.03 N/A	7.9 ± 0.26–63 7.5 ± 0.18–58 5.29 ± 0.12–34	HRT‐III	ACCMetrics	√
Puttgen et al. 27	Germany	DPN+ Control	116 46	67.3 ± 9 66 ± 5.2	17.6 ± 13 N/A	7.41 ± 1.3–57 5.44 ± 0.23–36	HRT‐III	CCMetrics ACCMetrics	√	√	√
Andersen et al. 12	Denmark	DPN+ DPN− Control	27 117 25	71.4 ± 3.1 69.7 ± 2.7 71.2 ± 0.69	12.2 ± 1.23 11.67 ± 1.12 N/A	6.95 ± 0.48–52 6.6 ± 0.33–49 5.5 ± 0.22–37	HRT‐III	ACCMetrics	√	√	√
Tummanapalli et al. 44	Australia	DPN+ DPN− Control	28 35 34	NS	NS	8.45 ± 0.5–69 7.59 ± 0.6–59	HRT‐III	ACCMetrics	√	√	√	√
Dehghani et al. 47	Australia	DPN+ DPN− Control	13 20 17	NS	NS	NS	HRT‐III	CCMetrics ACCMetrics			√
Tummanapalli et al. 49	Australia	DPN+ DPN− Control	23 27 29	47 ± 15 32 ± 10 37 ± 11	22 ± 13 15 ± 9 N/A	8.89 ± 1.9–74 7.83 ± 1.02–62 NS	HRT‐III	ACCMetrics	√	√	√	√
Tummanapalli et al. 43	Australia	DPN+ DPN−	35 35	51 ± 9.5 44.5 ± 11	NS	8 ± 1.4–64 8 ± 2–64	HRT‐III	ACCMetrics	√	√	√	√
Pritchard et al. 48	Australia	DPN+ DPN− Control	25 82 80	NS NS 37.0 ± 17.8	NS	NS	HRT‐III	CCMetrics			√	√
Edwards et al. 46	Australia	DPN+ DPN− Control	88 143 61	58 ± 9 48 ± 16 52 ± 14	23 ± 14 14 ± 12 N/A	8.2 ± 1.7–66 7.8 ± 1.2–62 5.4 ± 0.3–36	HRT‐III	CCMetrics		√	√	√
Dehghani et al. 45	Australia	DPN+ DPN− Control	39 108 60	NS NS NS	NS NS N/A	NS	HRT‐III	ACCMetrics	√	√	√
Ishibashi et al. 18	Japan	DPN+ DPN− Control	55 23 28	56.4 ± 14.1 48.1 ± 10.6 50.2 ± 7.41	9.6 ± 16.3 5.8 ± 5.8 N/A	8.03 ± 3.0–64 7.7 ± 2.11–61 5.6 ± 0.26–38	HRT‐III	ImageJ	√	√	√
Ishibashi et al. 19	Japan	DPN+ DPN− Control	153 47 40	56.03 ± 10.3 53.4 ± 7.54 53.6 ± 12.65	12.4 ± 8.2 10.5 ± 14.8 N/A	8.3 ± 3.5–67 7.3 ± 1.4–56 5.7 ± 0.32–39	HRT‐III	ImageJ	√	√	√
Ishibashi et al. 22	Japan	DPN+| DPN− Control	115 47 45	54.4 ± 19.1 52.4 ± 9.6 52.8 ± 4.7	7.9 ± 11.4 5 ± 4.5 N/A	9.06 ± 4.4–76 8.5 ± 1.4–69 5.5 ± 0.03–37	HRT‐III	ImageJ	√	√	√
Ishibashi et al. 50	Japan	DPN+ DPN− Control	18 57 42	59.4 ± 8.1 54.4 ± 12.1 53.1 ± 11.7	13.6 ± 10.61 6.7 ± 6.34 N/A	9 ± 1.74–75 9.1 ± 2.4–76 5.7 ± 0.4–39	HRT‐III	ImageJ	√	√	√
Li et al. 51	China	DPN+ DPN− Control	79 49 24	70.15 ± 7.34 67.12 ± 6.01 68.3 ± 5.19	12.58 ± 7.28 9.79 ± 7.09 N/A	7.94 ± 1.86–63 7.07 ± 0.96–54 5.88 ± 0.82–41	HRT‐II	CCMetrics ACCMetrics	√	√	√
Xiong et al. 23	China	DPN+ DPN− Control	79 49 24	70.3 ± 10 67.12 ± 6.13 68.63 ± 5.2	12.57 ± 10.2 9.79 ± 7.14 N/A	7.95 ± 3.4–63 7.07 ± 1.68–54 5.88 ± 0.83–41	HRT‐II	ImageJ	√	√	√
Pritchard et al. 70	Australia, Canada, UK	DPN+ DPN−	16 74	51 ± 14 42 ± 16	29 ± 16 15 ± 12	8 ± 1.1–64 7.9 ± 1.2–63	HRT‐III	CCMetrics			√
Pritchard et al. 52	Australia, UK	DPN+ DPN− Control	48 100 60	57 ± 11 43 ± 16 46 ± 15	34 ± 16 20 ± 15 N/A	8.6 ± 1.8–70 8 ± 1.2–64 5.5 ± 0.3–37	HRT‐III	CCMetrics		√	√

Data are presented as mean ± SD. CNFD, corneal nerve fiber density; CNBD, corneal nerve branch density; CNFL, corneal nerve fiber length; IWL, inferior whorl length; NS, not stated, N/A, not applicable.

Corneal nerve fiber density

DPN+ vs DPN−

Twenty‐nine studies , , , , , , , , , , , , , , , , , , , , , , , , , , with 3,214 (1,677 DPN+ and 1,537 DPN−) participants were included in the meta‐analysis. The CNFD (fiber/mm2) was significantly lower in the DPN+ group compared with the DPN− group (MD = −7.01, 95% CI −7.45 to 6.57, P < 0.00001) (CCMetrics (MD = −6.83, 95% CI −7.82 to −5.84, P < 0.00001), ACCMetrics (MD = −7.77, 95% CI −8.32 to −7.22, P < 0.00001), ImageJ (MD = −3.48, 95% CI −4.64 to −2.33, P < 0.00001), and morphometric software (MD = −11.40, 95% CI −15.42 to −7.38, P < 0.00001)). There was a significant difference in the magnitude of the CNFD reduction in the DPN+ group between studies (χ2 = 19.32, P = 0.0002) (Figure 1a).

Figure 1

(a) Forest plots of corneal nerve fiber density (CNFD) in patients with diabetic peripheral neuropathy (DPN+) and without diabetic peripheral neuropathy (DNP−). (b) Forest plots of corneal nerve fiber density (CNFD) in patients with diabetic peripheral neuropathy (DPN+) and healthy control. (c) Forest plots of corneal nerve fiber density (CNFD) in patients without diabetic peripheral neuropathy (DNP−) and healthy control.

DPN+ vs control

Twenty‐nine studies , , , , , , , , , , , , , , , , , , , , , , , , , , , with 3377 (1994 DPN+ and 1383 control) participants were included in the meta‐analysis. The CNFD (fiber/mm2) was significantly lower in the DPN+ group compared with the controls (MD = −11.94, 95% CI −12.25 to −11.62, P < 0.00001) (CCMetrics (MD = −10.83, 95% CI −11.26 to −10.40, P < 0.00001), ACCMetrics (MD = −13.75, 95% CI −14.26 to −13.25, P < 0.00001), ImageJ (MD = −8.98, 95% CI −10.40 to −7.55, P < 0.00001), and morphometric software (MD = −22.26, 95% CI −27.67 to −16.85, P < 0.00001). There was a significant difference in the magnitude of the CNFD reduction in the DPN+ group between studies (χ2 = 15.50, P = 0.001) (Figure 1b).

DPN− vs control

Twenty‐seven studies , , , , , , , , , , , , , , , , , , , , , , , , , with 3,035 (1,620 DPN− and 1,415 control) participants were included in the meta‐analysis. The CNFD (fiber/mm2) was significantly lower in the DPN− group compared with the controls (MD = −5.85, 95% CI −6.12 to −5.57, P < 0.00001) (CCMetrics (MD = −5.76, 95% CI −6.15 to −5.37, P < 0.00001), ACCMetrics (MD = −5.91, 95% CI −6.32 to −5.50], P < 0.00001), ImageJ (MD = −5.89, 95% CI −7.13 to −4.65, P < 0.00001), and morphometric software (MD = −11.07, 95% CI −16.34 to −5.80, P < 0.0001). There was no significant difference in the magnitude of the CNFD reduction in the DPN− group between studies (χ2 = 4.01, P = 0.26) (Figure 1c).

Corneal nerve branch density

Thirty studies , , , , , , , , , , , , , , , , , , , , , , , , , , , , with 3,552 (1,763 DPN+ and 1,789 DPN−) participants were included in the meta‐analysis. The CNBD (branch/mm2) was significantly lower in the DPN+ group compared with the DPN− group (MD = −3.36, 95% CI −4.11 to −2.61, P < 0.00001) (CCMetrics (MD = −10.37, 95% CI −12.56 to −8.18, P < 0.00001), and ACCMetrics (MD = −8.20, 95% CI −10.20 to −6.20, P < 0.00001). There was a significant difference in the extent of the CNBD reduction in the DPN+ group between studies (χ2 = 30.97, P < 0.00001), (Figure 2a).

Figure 2

(a) Forest plots of corneal nerve branch density (CNBD) in patients with diabetic peripheral neuropathy (DPN+) and without diabetic peripheral neuropathy (DNP−). (b) Forest plots of corneal nerve branch density (CNBD) in patients with diabetic peripheral neuropathy (DPN+) and healthy control. (c) Forest plots of corneal nerve branch density (CNBD) in patients without diabetic peripheral neuropathy (DNP−) and healthy control.

Thirty studies , , , , , , , , , , , , , , , , , , , , , , , , , , , , with 3,460 (2,072 DPN+ and 1,388 control) participants were included in the meta‐analysis. The CNBD (branch/mm2) was significantly lower in the DPN+ group compared with the controls (MD = −11.00, 95% CI −11.65 to −10.35, P < 0.00001) (CCMetrics (MD = −20.87, 95% CI −22.05 to −19.68, P < 0.00001), ACCMetrics (MD = −7.34, 95% CI −8.35 to −6.32, P < 0.00001), ImageJ (MD = −4.79, 95% CI −6.05 to −3.53, P < 0.0001), and morphometric software (MD = −21.81, 95% CI −26.61 to −17.01, P = 0.0003)). There was a significant difference in the magnitude of the CNBD reduction in the DPN+ group between studies (χ2 = 30.98, P < 0.00001) (Figure 2b).

Twenty‐six studies , , , , , , , , , , , , , , , , , , , , , , , , with 2,813 (1,606 DPN− and 1,207 control) participants were included in the meta‐analysis. The CNBD (branch/mm2) was significantly lower in the DPN− group compared with the controls (MD = −6.37, 95% CI −7.31 to −5.44, P < 0.00001) (CCMetrics (MD = −11.08, 95% CI −13.40 to −8.75, P < 0.00001), ACCMetrics (MD = −11.17, 95% CI −13.46 to −8.88, P < 0.00001), ImageJ (MD = −3.34, 95% CI −4.52 to −2.17, P < 0.0001), and morphometric software (MD = −16.26, 95% CI −21.14 to −11.37, P = 0.007)). There was a significant difference in the magnitude of the CNBD reduction in the DPN− group between studies (χ2 = 33.32, P < 0.00001) (Figure 2c).

Corneal nerve fiber length

Thirty‐four studies , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , with 3,868 (1,855 DPN+ and 2,013 DPN−) participants were included in the meta‐analysis. The CNFL (mm/mm2) was significantly lower in the DPN+ group compared with the DPN− group (MD = −3.08, 95% CI −3.58 to −2.58, P < 0.00001) (CCMetrics (MD = −3.74, 95% CI −4.49 to −2.99, P < 0.00001), ACCMetrics (MD = −2.80, 95% CI −3.57 to −2.04, P < 0.00001), ImageJ (MD = −1.57, 95% CI −2.06 to −1.09, P < 0.00001), and morphometric software (MD = −3.49, 95% CI −5.63 to −1.35, P = 0.001). There was a significant difference in the magnitude of the CNFL reduction in the DPN+ group between studies (χ2 = 25.42, P < 0.00001) (Figure 3a).

Figure 3

(a) Forest plots of corneal nerve fiber length (CNFL) in patients with diabetic peripheral neuropathy (DPN+) and without diabetic peripheral neuropathy (DNP−). (b) Forest plots of corneal nerve fiber length (CNFL) in patients with diabetic peripheral neuropathy (DPN+) and healthy control. (c) Forest plots of corneal nerve fiber length (CNFL) in patients without diabetic peripheral neuropathy (DNP−) and healthy control.

Thirty‐two studies , , , , , , , , , , , , , , , , , , , , , , , , , , , , with 3,459 (2,036 DPN+ and 1,423 control) participants were included in the meta‐analysis. The CNFL (mm/mm2) was significantly lower in the DPN+ group compared with the controls (MD = −6.05, 95% CI −6.77 to −5.34, P < 0.00001) (CCMetrics (MD = −6.91, 95% CI −8.06 to −5.76, P < 0.00001), ACCMetrics (MD = −5.49, 95% CI −7.03 to −3.95, P < 0.00001), ImageJ (MD = −4.14, 95% CI −4.72 to −3.56, P < 0.00001), and morphometric software (MD = −6.07, 95% CI −8.64 to −3.50, P < 0.00001). There was a significant difference in the magnitude of CNFL reduction between studies (χ2 = 19.59, P = 0.0002) (Figure 3b).

Thirty studies , , , , , , , , , , , , , , , , , , , , , , , , , , , , with 3,149 (1,786 DPN− and 1,363 control) participants were included in the meta‐analysis. The CNFL (mm/mm2) was significantly lower in the DPN− group compared with the controls (MD = −2.87, 95% CI −3.34, −2.40, P < 0.00001) (CCMetrics (MD = −3.12, 95% CI −4.06 to −2.19, P < 0.00001), ACCMetrics (MD = –2.63, 95% CI −3.43 to −1.83, P < 0.00001), ImageJ (MD = −2.78, 95% CI −3.35 to −2.22, P < 0.00001), and morphometric software (MD = −2.68, 95% CI −3.48 to −1.88, P < 0.00001). There was no difference in the magnitude of the CNFL reduction in the DPN− group between studies (χ2 = 0.72, P = 0.87), (Figure 3c).

Inferior whorl length

Six studies , , , , with 459 (205 DPN+ and 254 DPN−) participants were included in the meta‐analysis. The IWL (mm/mm2) was significantly lower in the DPN+ group compared with the DPN− group (MD = −4.11, 95% CI −5.10 to −3.12, P < 0.00001) (CCMetrics (MD = −3.42, 95% CI −5.47 to −1.36, P = 0.001), and ACCMetrics (MD = −4.40, 95% CI −5.53 to −3.28, P < 0.00001). There was no significant difference in the magnitude of the CNFL reduction in the DPN+ group between studies (χ2 = 0.68, P = 0.41), (Figure 4a).

Figure 4

(a) Forest plots of inferior whorl length (IWL) in patients with diabetic peripheral neuropathy (DPN+) and without diabetic peripheral neuropathy (DNP−). (b) Forest plots of inferior whorl length (IWL) in patients with diabetic peripheral neuropathy (DPN+) and healthy control. (c) Forest plots of inferior whorl length (IWL) in patients without diabetic peripheral neuropathy (DNP−) and healthy control.

Six studies , , , , , with 520 (310 DPN+ and 210 control) participants were included in the meta‐analysis. The IWL (mm/mm2) was significantly lower in the DPN+ group compared with the controls (MD = −10.36, 95% CI −13.30 to −7.42, P < 0.00001) (CCMetrics (MD = −11.62, 95% CI −15.97 to −7.28, P < 0.00001), and ACCMetrics (MD = −8.32, 95% CI −9.40 to −7.24, P < 0.00001)). There was no significant difference in the extent of the IWL reduction in the DPN+ group between studies (χ2 = 2.08, P = 0.15), (Figure 4b).

Five studies , , , , with 399 (219 DPN− and 180 control) participants were included in the meta‐analysis. The IWL (mm/mm2) was significantly lower in the DPN− group compared with the controls (MD = −3.81, 95% CI −4.56 to −3.06, P < 0.00001) (CCMetrics (MD = −4.43, 95% CI −5.56 t0 −3.29, P = 0.003), and ACCMetrics (MD = −3.34, 95% CI −4.33 to −2.34, P < 0.00001). There was no significant difference in the extent of IWL reduction in the DPN− group between studies (χ2 = 2.11, P = 0.15), (Figure 4c).

Discussion

In this large systematic review and meta‐analysis of over 3,000 participants, CCM demonstrated a consistent reduction in four major corneal nerve parameters in patients with DPN compared with healthy controls and those without DPN. Furthermore, we demonstrate a lesser but significant reduction in all corneal nerve parameters in patients without DPN compared with controls, suggesting that CCM detects early sub‐clinical DPN. This is consistent with the demonstration of corneal nerve loss in subjects with impaired glucose tolerance , recently diagnosed type 2 diabetes and children with type 1 diabetes . The greater corneal nerve loss in patients with DPN compared with those without DPN is consistent with studies showing that corneal nerve loss is associated with the severity of DPN , , , , and has good sensitivity and specificity for diagnosing DPN , , . Both CNFD and IENFD have a comparable diagnostic performance for DPN , , , although in a study of patients with recently diagnosed type 2 diabetes there were differences in the extent of small nerve fiber damage between CCM and skin biopsy . Additionally, a reduction in corneal nerve parameters is associated with incident DPN , , and greater corneal nerve loss , and augmented nerve branching occurs in patients with painful diabetic neuropathy. CCM could act as a biomarker as defined by the NIH Biomarkers Definitions Working Group ; it is non‐invasive, easily measured, and produces rapid results with high sensitivity , , . It allows the detection of subclinical DPN, and there is minimal overlap in corneal nerve parameters between patients with and without DPN and healthy people. In addition, CCM identifies those at risk of developing DPN , , .

The outcomes of the current review extend considerably the findings of a previous systematic review and meta‐analysis showing a reduction in CNFD, CNBD, and CNFL in patients with and without DPN compared with controls from 13 studies with 1,680 participants and a more recent trial sequential meta‐analysis which showed a reduction in CNFD, CNBD, and CNFL in patients with and without DPN compared with controls in 13 studies with 1,830 participants .

In the present review we have included IWL which has the potential to detect earlier nerve damage , , , especially in patients with painful diabetic neuropathy , .

The reliability of establishing a single estimate for the effect size of corneal nerve outcome measures from all the published studies may be affected by the inclusion of the same subjects from several studies, type of CCM used to acquire the images, the mode of image acquisition, and the image analysis tool used to quantify corneal nerve parameters. Our analysis showed that the type of software used for image analysis had no significant influence on the heterogeneity of corneal nerve outcomes. Whilst the corneal nerve measure was lower when using automated (ACCMetrics) compared with manual (CCMetrics, ImageJ) software, the magnitude of difference in corneal nerve parameters between groups was comparable , .

Our sensitivity analysis shows no evidence of significant bias or heterogeneity (Doc S1). This was expected, given that there may be differences in corneal nerve parameters between patients with type 1 and type 2 diabetes , , and in relation to HbA1c and glycemic variability , presence of metabolic syndrome and hypertension or hyperlipidemia , .

Conclusions

Corneal confocal microscopy is a rapid, non‐invasive and reproducible imaging technique to quantify small nerve fiber damage. Our systematic review and meta‐analysis provides robust evidence that corneal confocal microscopy can be used to diagnose sub‐clinical and established DPN.

Disclosure

Approval of the research protocol: N/A.

Informed Consent: N/A.

Approval date of Registry and Registration No. of the study/trial: N/A.

Animal Studies: N/A.

Figure S1 | Flowchart of the included studies.

Click here for additional data file.

Table S1 | Search details

Click here for additional data file.

Table S2 | Risk of bias assessment for non‐randomized studies

Click here for additional data file.

Doc S1 | Methods. Risk of bias and sensitivity analysis.

Click here for additional data file.