The impact of dry eye disease on corneal nerve parameters: A systematic review and meta‐analysis

Dry eye disease (DED) is a growing global health problem with a significant impact on the quality of life of patients. While neurosensory abnormalities have been recognised as a contributor to DED pathophysiology, the potential role of in vivo corneal confocal microscopy in detecting nerve loss or damage remains unclear. This systematic review with meta‐analysis (PROSPERO registered CRD42022381861) investigated whether DED has an impact on sub‐basal corneal nerve parameters.


INTRODUC TION
Dry eye disease (DED) is a growing unmet problem affecting up to one in 10 of the global population. 1 It costs millions in healthcare economic burden worldwide, 2 with substantial adverse impacts on affected individuals' quality of life. 3 Notably, the impact of severe DED has been shown to be similar to that of moderate to severe angina. 4,5 Efforts to seek the optimal DED treatment have risen over the years; however, the multifactorial pathophysiological mechanisms underlying DED contribute to the difficulties in managing this chronic condition. In addition to tear film instability, hyperosmolarity and inflammation, the Tear Film and Ocular Surface Society Dry Eye Workshop (TFOS DEWS) II has identified neurosensory abnormalities as contributors to the development or progression of DED. 6 However, assessment methods of neuronal damage or dysfunction for the diagnosis and management of DED remain limited in clinical settings.
In vivo confocal microscopy is capable of imaging corneal microstructures en face across different layers in a non-invasive and rapid manner. Notably, the cornea is known to be one of the most densely innervated regions of the body. 7 Small nerve fibres in this region are responsible for signalling pain, temperature and mechanical sensations. 8 The sub-basal nerve plexus, which is situated between the basal epithelial layer and Bowman's layer, is often the region of interest in terms of imaging as the nerves in this layer are organised in a relatively homogenous manner compared with other nerve plexi in the cornea. 7 While there have been several iterations of in vivo confocal microscopy over the past few decades, laser scanning confocal microscopy has been the most widely adopted by researchers and clinicians due to its high resolution and magnification. 9 Dry eye disease has been shown to be associated with neuronal damage or abnormalities, with neurobiological changes associated with neurogenic inflammation and disturbances to the activity of peripheral ocular sensory nerve fibres. 8 Given the impact of DED on these nerves, characterising morphological changes in the sub-basal nerve plexus in a quantitative manner may aid in dry eye diagnosis and potentially guide treatment decisions. This may also impact the identification of more sensitive endpoint measures in evaluating the neuroregenerative or neuroprotective capabilities of current and future therapies. Hence, there has been increasing interest in the investigations of whether structural loss of corneal nerves could be observed in DED. Given the rise in interest and use of the instrument in both clinical and research settings, there is need for deeper discussions of limitations and future directions for in vivo corneal confocal microscopy in addition to providing a timely synthesis of more recent literature. 10 Hence, a systematic review and quantitative analysis was conducted to explore the following question: Do patients with DED have corneal nerve parameter changes as observed with in vivo corneal confocal microscopy?

METHODS
This systematic review and meta-analysis was prospectively registered on PROSPERO (CRD42022381861) and conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement 11 and guidelines provided by Rudnicka and Owen. 12

Study eligibility criteria
The inclusion criteria were studies, including observational studies (cross-sectional or case-control studies), which compared patients diagnosed with DED of any severity or type (primarily aqueous deficient or evaporative) with individuals without DED (healthy controls) with a sample size n ≥ 10 participants in each group. Only studies published in the English language were included. Studies must also have explicitly mentioned that DED was diagnosed with a combination of symptoms and signs or made reference to published criteria, which involved these combined diagnostic methods. Only studies using laser scanning confocal microscopy (Heidelberg Engineering GmbH, heide lberg engin eering.com) with outcome measures including quantitative corneal nerve parameters of the sub-basal nerve plexus (including length, density, area, width, fractal dimension and tortuosity) produced by manual, semi-automated or automated procedures were included.
Exclusion criteria were review articles, conference abstracts as data are limited from these sources, studies that did not compare patients with DED with healthy controls and studies with sample size <10 participants in either group. Studies involving animals or investigating patients with only other ocular surface issues or conditions such as contact lens discomfort from either soft or rigid contact

Key points
• Neurosensory abnormalities have been recognised as a contributor to the pathophysiology of dry eye disease; however, the role of corneal nerve imaging in detecting nerve damage or loss remains unclear. • This systematic review and meta-analysis demonstrated a reduction in several corneal nerve parameters, particularly with the use of a semiautomated image analysis method, although there was substantial heterogeneity between studies. • Further investigation is required to examine the diagnostic and prognostic capabilities of corneal nerve imaging and to improve its applicability and practicality in real-world clinical settings. lenses, ocular surgeries or injuries, ocular infections, ocular tumours, other ocular inflammatory conditions such as uveitis or neurodegenerative conditions such as glaucoma requiring active treatment, corneal dystrophies or corneal oedema were also excluded from this review. Studies involving patients only affected by systemic comorbidities which may impact corneal nerve parameters were also excluded, including diabetes and Sjögren syndrome. Studies investigating corneal nerve changes qualitatively without quantitative parameters and those that only assessed other corneal layers, microneuromas or non-neuronal features of the sub-basal nerve plexus such as dendritic cells were excluded.

Literature search strategy
Comprehensive literature searches were conducted using three electronic databases: PubMed, Embase (Ovid) and Web of Science Core Collection. Databases were searched from inception to the date of the search (9 December 2022). Search strategies were developed with assistance from an information specialist with expertise in systematic reviews and are provided in Table S1. The reference lists of included studies were also examined.

Study selection process
Citation results from each database were first imported into EndNote 20 software (endno te.com), and duplicate entries between database search results were identified by the software and subsequently removed. The Covidence systematic review software (Veritas Health Innovation, covid ence.org) was used for study screening. Two members from the review team (JCBC and VT) worked independently to screen studies by referring to the title and abstract of each study. The same members then independently assessed screened studies for eligibility by assessing the full-text article of each study.

Data extraction
Two members of the review team (JCBC and VT) extracted data from relevant studies while adhering to standardised data extraction forms developed for this systematic review. For each study, the following information was extracted: Where studies divided DED groups according to type or severity (potentially multiple levels), data were pooled from these groups as findings for an overall DED group, if this was not already provided in the relevant study. The primary or corresponding authors of investigations that did not explicitly provide exact numerical findings of corneal nerve parameters in their article or supporting information were contacted to provide these data.

Outcomes
The primary outcome included quantitative sub-basal corneal nerve plexus parameters (including density, length, width, area, fractal dimension and tortuosity) as measured with manual, semi-automated or automated procedures.

Risk of bias assessment
Two members from the review team (JCBC and VT) worked independently to assess the risk of bias of the included studies. Any discrepancies at any stage of the review were resolved through discussion and consensus. Risk of bias assessment (Table S2) was adapted from tools including the QUADAS-2 13 and Newcastle-Ottawa scale. 14 To provide a more informative assessment of the risk of bias, each element was assessed in terms of the potential for bias (high, low or unclear), instead of one overall determination for the domain or overall study.

Qualitative synthesis
Prior to conducting a quantitative synthesis, included studies were evaluated in terms of their relevance, potential discordance between studies and appropriateness to be synthesised in subsequent meta-analyses.

Quantitative synthesis and statistical analysis
To facilitate subsequent quantitative synthesis of results in this review, terminology and measures with common definitions were classified under an overarching, harmonising term and results converted to a uniform unit of measurement where possible. This also included converting standard errors of the mean or median and interquartile range to mean and standard deviation. 15 To ensure consistency between study findings, terminologies used in studies with insufficient detail in their provided definitions to determine comparability with parameters in other studies were noted and not grouped under the harmonising term (Table 1). Meta-analyses were performed using the Review Manager (RevMan) version 5.4.1 (train ing.cochr ane.org/ onlin e-learn ing/core-softw are/revman) if a specific corneal nerve parameter had been investigated by a sufficient number of studies using similar methods (≥2). A randomeffects meta-analysis with inverse variance method was used for each corneal nerve parameter to assess mean differences (or standardised mean differences if results could not be converted to a uniform unit of measurement) between the dry eye and healthy control groups. This method provides more conservative estimates particularly with the potential presence of significant study heterogeneity. Meta-analyses were grouped by the corneal nerve parameter investigated. Subgroup analyses were also conducted, grouping studies according to the analytical procedure used for measuring corneal nerve parameters (manual, semi-automated or automated) and presented as subtotals in the forest plots because magnitudes of outcome measures may differ substantially across these methods. [16][17][18] Statistical heterogeneity was assessed using τ 2 , χ 2 and I 2 . In cases where meta-analysis was not possible, the individual study findings were investigated. Statistical significance was considered as p < 0.05.

Study selection
The electronic searches yielded 509 unique reports following duplicate removal. Full-text articles were obtained and assessed for eligibility for 69 studies following the screening stage by title or abstract. A PRISMA flow diagram of the study selection process is provided in Figure 1, which culminated in 22 included studies for this systematic review (Table S3). There were two unique reports identified from the reference lists of the included studies; however, none were eligible to be included in this systematic review ( Figure 1).
Only two studies specified the number of personnel involved in imaging the cornea using in vivo corneal confocal microscopy: one study involved one imager 36 and one study involved two imagers. 22 Twenty studies specified T A B L E 1 Harmonising terminology used for corneal nerve parameters with referencing of the included studies in the meta-analyses.

Risk of bias in studies
The risk of bias assessment is summarised in Table S6. To provide a more detailed assessment in relation to study characteristics and in vivo corneal confocal microscopy methodology, elements under each domain were judged individually. There were no studies that had low risk of bias across all elements assessed. The element with the most studies judged as having high risk of bias (12 of 22 studies) [19][20][21][23][24][25]30,[33][34][35]37,40 was in the domain of comparability between groups, specifically assessing whether there was controlling or matching, or adjusting for both age and sex between groups in statistical analysis. Four of these 12 studies reported matching of groups for age but not sex, 24,30,37,40 while seven of these 12 studies reported no statistical significance between groups in terms of age and/or sex. 19,21,25,33,35,37,40 The elements with the most studies judged as having an uncertain risk of bias

Primary outcomes and qualitative synthesis
All studies analysed quantitative corneal nerve parameters, albeit with a variety of terminology; however, one did not provide exact numerical findings for these outcomes and hence was excluded from subsequent metaanalyses. 24 One study did not specify the definition for the term 'number of sub-basal nerves' 20 while another study did not explicitly define corneal nerve maximum length 25 ; hence, these findings were not included in the subsequent meta-analyses. One study reported a dichotomous result for beading presence; hence, this was not included in the meta-analysis on the number of beadings. 39 Only one study investigated corneal nerve reflectivity, showing no significant difference between DED and healthy control groups. 28 For the study that used both NeuronJ and ACCMetrics to measure the same corneal nerve parameters from participants, 34 NeuronJ was chosen as the primary data included in the relevant meta-analyses as more corneal nerves are known to be detected through this method. The meta-analyses were then repeated with the NeuronJ results replaced with the ACCMetrics data to evaluate any potential changes in the outcome of the analyses. Table 1 summarises the harmonising terms and units, with a total of 14 being used for the purposes of this review to facilitate comparability of findings between studies included in subsequent meta-analyses. The most variability in regards to terminology pertains to the total length of corneal nerves measured per unit area, with a majority of studies referring to this parameter as density (10/18 studies), 21,22,25,27,28,30,[36][37][38][39] while the other studies referred to this parameter as length. 19,23,26,31,[33][34][35]40 It is also common for studies to classify nerves as either main nerve trunks or branches, with the specific terminology, definitions and units used for corneal nerve parameters for each study provided in Table S7. For this review, density refers to the count of nerves, while length refers to the distance of the nerve path in an image.

Meta-analyses of corneal nerve parameters
This section outlines the meta-analyses conducted on each of the corneal nerve parameters reported. Forest plots of analysis of corneal nerve parameters investigated by a total of five or more studies are presented in this main article; otherwise, the plots are included as Supporting information.

Total corneal nerve density
Three studies 19,36,39 (199 DED and 90 healthy eyes) investigated total corneal nerve density. All these studies used NeuronJ for image analysis and showed evidence of a significant reduction in total corneal nerve density in DED eyes with a total mean difference of −70.86 number/mm 2 (95% CI −134.15, −7.58; p = 0.03; I 2 = 95%; Figure S7). The substantial heterogeneity may have been due to the varying sample sizes (range of 30-139 DED eyes and 16-42 healthy eyes), differences in the corneal location imaged and number of corneal nerve images analysed.

Corneal nerve fibre width
Five studies 19,23,26,28,31 (including 421 DED and 151 healthy eyes) investigated corneal nerve fibre width. There was no significant difference between DED and healthy eyes in this parameter (standardised mean difference: −0.03; 95% CI −0.39, 0.33; p = 0.88; I 2 = 70%; Figure 6). The subgroup analysis of either the two studies 19,28 which used NeuronJ or the three studies 23,26,31 with ACCMetrics also showed no significant difference. Some of the overall heterogeneity may be due to methodology differences, with studies using NeuronJ calculating a mean of three 19 or five 28 measurements of long nerve fibre thickness, while ACCMetrics 23,26,31 automatically measured the average thickness of all corneal nerve fibres per unit area. F I G U R E 4 Forest plot of corneal nerve branch density measured in number/mm 2 comparing dry eye disease (DED) and healthy control eyes. Heterogeneity: Tau² = 289.28; Chi² = 110.01, df = 3 (P < 0.00001); I² = 97% Test for overall effect: Z = 3.29 (P = 0.001)

F I G U R E 5
Forest plot of corneal nerve branch point density measured in number/mm 2 comparing dry eye disease (DED) and healthy control eyes.

Corneal nerve parameters change in DED
This systematic review and meta-analyses showed evidence of corneal nerve morphological changes in DED. This was consistently demonstrated with NeuronJ analysis in certain corneal nerve parameters including reduction in total corneal nerve length, corneal main nerve trunk density and corneal nerve branch density. The usage of this semi-automated method is known to identify and trace more nerves with the guide of an experienced assessor compared with automated methods, particularly the widely used ACCMetrics. 34,42 However, emerging deep learning techniques such as CNS-Net used by a study included in this review 25 and programs investigated by other studies may improve the capability of automated procedures in nerve detection and quantification in a more time-efficient and reliable manner compared with semiautomated procedures. 43,44 Other features of the corneal nerve plexus require further work More complex features of the sub-basal nerve plexus are also commonly investigated by studies, including beading formation and corneal nerve tortuosity. Increased beading formation on sub-basal nerves is thought to be indicative of heightened metabolic processing, which may be a stressor on the health of the corneal nerves. 45 More tortuous corneal nerves are also putative markers of aberrant nerve growth or regeneration following an insult. While most studies that have shown increased corneal nerve tortuosity in DED eyes were based on a crude manual grading scale, 41 emerging automated procedures are also showing similar findings with DED. 29 Corneal nerve reflectivity, fibre width and fibre area are highly dependent on image quality, which may impact analysis reliability. Hence, evidence for the diagnostic or prognostic potential of these parameters is currently limited.

The need for standardisation of methodology, terminology and images analysed
While evidence for corneal nerve loss seems to be evident in eyes with DED, most studies in this systematic review have uncertain to high risk of bias with substantial heterogeneity between studies. The number of images analysed across studies were not standardised. While greater numbers of minimally or non-overlapping images have been demonstrated to be more likely to represent the 'true mean' , 46 this may also depend on the extent of the sub-basal nerve plexus imaged beforehand. The experience of the imaging personnel and tolerability of the patient to the procedure may impact on the total area imaged. While most studies have reported masking of image assessors from the condition of the patients, masking of the imaging personnel may be more difficult in clinical research settings. More standardised widefield imaging with precise localisation capabilities may be required to facilitate repeatable monitoring, 47 akin to the technologies employed in optical coherence tomography. Investigators should clearly define and describe the terminology used to enhance the comparability between studies. While the harmonising terms included in this review were not meant to be prescriptive, the various F I G U R E 7 Forest plot of corneal nerve tortuosity comparing dry eye disease (DED) and healthy control eyes. terminology used by different studies highlight the heterogeneity in corneal nerve parameters analysed.

Limitations of the review
While DED diagnosis is becoming more standardised, 48 subtle differences still exist across studies which may impact on generalisability of the findings. As both age and sex are recognised risk factors of DED, 2 these factors should be matched or accounted for in studies comparing differences between groups or cohorts. The current metaanalyses included studies that did not explicitly report this, which may introduce some bias, although most studies did report similar age and sex between groups. Corneal dendritic cells, thought to be increased in DED and indicative of the inflammatory status of the cornea, 49 were also not investigated in this systematic review as the aim was to focus on corneal nerve morphology.

Future directions
Emerging automated deep learning methods may be crucial to improve the clinical applicability of corneal nerve imaging. Future studies on how in vivo corneal confocal microscopy could guide DED management in real-world clinical settings may also be beneficial. Longitudinal studies which track the development and progression of DED along with corneal nerve imaging may provide insight into the potential intraindividual sub-basal corneal nerve changes. Structure-function concordance is of particular clinical interest, with sparse evidence demonstrating that corneal nerve loss in DED could reduce corneal sensitivity or increase symptoms of ocular surface discomfort, 50,51 although a mouse model of chronic DED has also shown hypersensitivity with loss of intraepithelial corneal nerve terminals. 52 While evidence for the diagnostic accuracy of in vivo corneal confocal microscopy for DED is limited, there have been several studies that investigated the potential for corneal nerve imaging in monitoring the improvement in nerve morphology in both human clinical studies and animal models of DED. Recent randomised controlled trials of interventions including oral vitamin B1 and mecobalamin, 53 homologous serum eyedrops 54 and oral omega-3 essential fatty acid supplements 55 have demonstrated improvements in corneal nerve parameters as assessed with laser scanning confocal microscopy. However, each study involved different follow-up times, varying image selection and analysis and different sample sizes. The mouse model also showed that instillation of rebamipide 2% ophthalmic solution, a mucin secretagogue, protected nerve density (measured in pixels/frame) against environmental dry eye stress, while artificial tears and hyaluronic acid 0.1% ophthalmic solution did not have such protective effects. 56 An earlier randomised controlled trial further investigated the prognostic capability of in vivo corneal confocal microscopy, demonstrating that DED participants with higher total corneal nerve length at baseline (≥ 16.84 mm/mm 2 ; classified as near-normal) experienced improvements in clinical symptoms and corneal fluorescein staining following 4 weeks of either artificial tear eyedrops or loteprednol etabonate 0.5% eyedrops, compared with those having low total corneal nerve length (<16.84 mm/mm 2 ). 40 The central cornea has been the primary area of interest in imaging studies; however, other areas of the cornea may provide further insight into potential corneal nerve changes. These include corneal nerves in the far periphery, and the inferior whorl is an inferocentral anatomical landmark where most corneal nerves traverse and converge towards. 7 However, it should be noted that the translatability of terminology and definitions of branches or main nerve trunks as used in conventional central corneal nerve imaging may be difficult given the more complex nerve distributions in these pericentral or peripheral locations.

CONCLUSION
While the current systematic review and meta-analyses indicate evidence of corneal nerve loss in DED eyes particularly with the semi-automated procedure NeuronJ, more research is required to investigate whether the clinical applicability and practicality of corneal nerve imaging could be improved. Some evidence shows that in vivo corneal confocal microscopy may be useful in monitoring DED treatment effectiveness; however, whether it could be used to diagnose DED accurately and predict the optimal treatment approach for a particular patient requires further investigation.

FU N D I N G I N FO R M AT I O N
None.

CO N F L I C T O F I N T E R E S T S TAT E M E N T JSW is on the Executive of the Tear Film and Ocular Surface (TFOS) Society.
O R C I D Jeremy Chung Bo Chiang https://orcid. org/0000-0002-6133-7411 Vincent Tran https://orcid.org/0000-0002-5659-3859 James S. Wolffsohn https://orcid. org/0000-0003-4673-8927