Corneal collagen cross-linking epithelium-on vs. epithelium-off: a systematic review and meta-analysis

Background The purpose of the study was to determine the advantages and disadvantages of epi-on corneal cross-linking (CXL) techniques compared with standard epi-off CXL. Methods We searched MEDLINE and EMBASE for randomized controlled trials (RCTs) and non-randomized studies of interventions (NRSIs) and we evaluated the selected papers according to the Cochrane risk of bias tool. We considered, as primary outcomes, average Kmax flattening, changes in uncorrected and corrected distance visual acuity (UDVA and CDVA); as secondary outcomes, we considered changes in pachymetry values and endothelial cell density (ECD). We also investigated adverse events related to the treatments and treatment failure. Meta-analysis was conducted with a fixed or random-effects model using weighted mean difference (MD) with 95% confidence interval (CI) as the effect size. Results A total of 15 studies were included and among these 15 trials, 9 were RCTs and 6 were NRSIs, but only 4 studies showed no high risk of bias and were included in this meta-analysis. Our analysis revealed significant postoperative differences in CDVA (MD = 0.07; 95% CI 0.04 to 0.10; P < 0.001), and no significative differences in UDVA, Kmax, central corneal thickness (CCT) and ECD (P > 0.05). Epi-on CXL protocol was found to be significantly less prompt to have risks of delay in epithelial healing (P = 0.035) and persistent stromal haze (P = 0.026). Conclusion Epi-on CXL is as effective as epi-off CXL. Except for a higher significant improvement in CDVA with current epi-on protocols, our meta-analysis demonstrates that epi-on and epi-off CXL have comparable effects on visual, topographic, pachymetric, and endothelial parameters. Epi-on CXL has clinical advantages in terms of comfort and avoidance of complications as it reduces the risk of developing delay in epithelial healing and persistent stromal haze. Supplementary Information The online version contains supplementary material available at 10.1186/s40662-021-00256-0.

the first clinical results were published by Wallensak et al. in 2003 [6]. The original CXL procedure is known as the "Dresden protocol" and is an epithelium-off procedure. In the standard technique, after anesthetizing the eye, the central 8 mm of the corneal epithelium is removed to expose the collagen-rich stroma and riboflavin solution (0.1% riboflavin-5-phosphate and 20% dextran T-500) is applied to the surface of the cornea both 30 min before irradiation and at 5 min intervals during a 30 min exposure to 370 nm UVA with a fluence of 3 mW/cm 2 and a total irradiation dose of 5.4 J/cm 2 [7]. The downside of epithelial removal is that it causes significant pain and discomfort in the early postoperative period; epi-off CXL carries a small risk of viral reactivation, haze, melting, infectious ulceration and the development of permanent stromal scars [8].
Considering these situations, several variations of the standard CXL procedure have been proposed since its introduction. Transepithelial or epi-on CXL is one such variation. Leaving the corneal epithelium intact should reduce pain and complications associated with epithelial debridement, such as infectious keratitis and could also lead to a shorter interruption of contact lens wear [8]. Considering transepithelial procedures, it should be considered that riboflavin is a large hydrophilic molecule that cannot penetrate an intact epithelium; moreover, the epithelium blocks about 20% of the UV illumination administered [9]. To improve riboflavin penetration into the stroma via the intact epithelium, several approaches have been used and investigated to encourage riboflavin penetration to the stroma [9].
There is, however, a paucity of studies that have been constructed to answer the clinical question of relative benefit for these procedures, and thus the evidence available for evaluation is limited. The main objectives of this systematic review and meta-analysis were to determine, in an evidence-based manner and using the currently available literature, the advantages and disadvantages of epi-on CXL techniques compared with traditional epi-off CXL and to discuss their indications.

Methods
The review followed methods recommended by Cochrane and the Centre for Reviews and Dissemination [10] and is reported according to PRISMA guidelines [11]. Furthermore, it has been registered in the PROS-PERO database (CRD42020156072), where the review methods were established prior to the beginning of the review analysis.

Method of literature search
Two reviewers (FD and JLA) independently performed a systematic literature search in the MEDLINE/PubMed database and the EMBASE database from January 2014 to July 2021. This range was chosen to ensure that only contemporary and comparable procedures for CXL were included. The following keywords were used: "Keratoconus", "Cross-Linking Reagents", "Iontophoresis", "Riboflavin", "Epithelium", "Ultraviolet Rays", "Epi-On", "Epi-Off ", "Epithelium-On", "Epithelium-Off ". These were searched in the title and the abstract as well as MESH/EMTREE terms. Search terms are shown in Additional file 2. Only articles published in English, Italian, or Spanish with an available abstract were included in the systematic literature search (languages in which the authors had good command). The abstracts of related titles were reviewed, and the full articles retrieved if their title or abstract appeared to meet the objectives of this review. Studies were included if they met the inclusion criteria. Reference lists of the included papers were screened too. Moreover, we manually analyzed the following trial registries: clinicaltrials.gov and Cochrane Controlled Register of Trials as well as "grey literature".

Study selection, inclusion and exclusion criteria
All publications were screened by two authors (FD and JLA) and any disagreement was discussed by the two authors and resolved by consensus. The reports were screened according to the following selection criteria: (a) studies that compared the role of epi-on CXL to epioff CXL for progressive keratoconus patients; (b) clinical trial studies; (c) both randomized controlled trials (RCTs) and non-randomized studies of interventions (NRSI), to generate evidence that will guide medical decisions [12]. All the studies that we reviewed were classified based on their scientific level of evidence according to the General Guidelines for Methodologies on Research and Evaluation of Traditional Medicine of the World Health Organization [13]. Only articles with a level Ib or IIa scientific evidence were selected. All the articles that were found were carefully reviewed to select those that reported original clinical data pre-and postoperatively. Data from previously reported cases included in different articles were omitted to avoid duplication of data. Articles on corneal collagen cross-linking combined with other treatments, such as topography-guided photorefractive keratectomy or intrastromal corneal ring segments were excluded. Experimental animal studies and ex vivo investigational reports were excluded from the review analysis. We also excluded all studies with a follow-up period of less than 12 months.

Quality assessment
Two reviewers (FD and JLA) separately evaluated the studies based on the methods recommended in the Risk of Bias in Non-randomized Studies of Intervention (ROBINS-I) [10,14] for cohort studies based on seven domains (confounders, selection of participants into the study, classification of interventions, deviations from the intended intervention, missing data, measurement of outcomes, and selection of the reported results). The "Cochrane risk of bias tool" [12] was used to assess the methodological quality of RCTs, by examining the following domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other biases. The review authors looked for sources of funding for the studies included in the review. Any disagreement was discussed by the reviewers and resolved.

Data extraction and clinical outcome
Two reviewers achieved consensus on which data to extract from the included studies (FD and JLA, followed the previous methods to achieve a consensus) and included the name of the first author, the year of publication, the trial location, the study design, the number of eyes, the type of CXL protocols used, the follow-up durations and outcome measures. In this review we considered, as the main outcome, average Kmax flattening, considering that the primary objective of CXL is to stabilize the underlying disease process, changes in uncorrected distance visual acuity (UDVA) and corrected distance visual acuity (CDVA), changes in HOAs and comatic aberrations, and depth of the demarcation line as an indirect treatment indicator of effectiveness. The secondary outcome parameters investigated in this study were changes in pachymetry values, either central corneal thickness (CCT) or corneal thickness at the thinnest point and endothelial cell density (ECD). We also investigated adverse events related to the treatments and treatment failure, such as keratoconus progression, loss of ≥ 2 lines of CDVA, delay in epithelial healing, persistent stromal haze, sterile infiltrates and infections.

Statistical analysis
If all the included clinical characteristics were similar between groups, we think that it would be reasonable to combine these studies through a meta-analysis. Otherwise, a descriptive analysis will be carried out. If we performed the meta-analysis with our extracted data, following the AMSTAR-2 (Assessment of Multiple Systematic Reviews) checklist [15], we will only combine those works with no high risk of bias. For continuous outcomes, the weighted mean difference (MD) will be calculated for obtaining the absolute changes. This calculation will be carried out through a random-effects models, unless there is significant heterogeneity or if we have less than three studies [16]. Heterogeneity across studies will be estimated by using Cochran's Q test and I 2 , considering its presence when I 2 > 50% and/or P-value > 0. 10.
If significant heterogeneity existed among trials, we would explore sources of heterogeneity, using subgroup analysis and meta-regression models, so long as we have enough studies [16]. We will perform sensitivity analysis by repeating the calculations excluding groups of studies (those with unclear risk of bias and which have used iontophoresis) each time and compare the results obtained. A group of studies was considered influential if it varied the overall coefficient by at least 10%. Finally, we will analyze the asymmetry of funnel plots with the Egger test when we obtain at least 10 studies [17].
We set type I error at 5% and for each relevant parameter, its associated confidence interval (CI) was calculated. The statistical software for this meta-analysis was R 3.3.3 through the meta for package (Meta-Analysis Package for R).

Characteristics of included studies
We identified a total of 15 studies where epi-on CXL outcomes of progressive keratoconus patients were compared to epi-off CXL outcomes. Among these 15 trials, 9 were RCTs [48-50, 54, 55, 57, 59-61] and 6 were NRSIs [47, 51-53, 56, 58]. The country, study design, population and sample size, comparison, CXL regimen, outcome measures and follow-up of each clinical study were meticulously reviewed and summarized in Table 1. There were 511 eyes included in the standard epi-off CXL group and 574 eyes included in the epi-on CXL group. Studies were conducted in Italy, Jordan, Egypt, France, Peru, Netherlands, USA, Turkey, Russia and Saudi Arabia and were reported between 2015 and 2019. The duration of the follow-up ranged from 12 to 36 months. The study by Goodefroji et al. [54] derived data from a previously published RCT [61] to further investigate the development of HOAs and their effects on visual acuity, and thus only data concerning HOAs were extracted and included in the outcome measures (Tables 2, 3 and 4).

Secondary outcomes Pachymetry
Three studies compared the CCT between the epi-on and epi-off CXL groups (Fig. 2) [48,57,60]. There were no significant differences in CCT changes between the two groups (MD = 3.52; 95% CI − 12.61 to 5.58; P = 0.448) and no heterogeneity between studies (Q = 0.6306, df = 2, P = 0.7296; I 2 = 0%) (Fig. 2d). The sensitivity analysis demonstrated that the result remained unchanged after the removal of the studies with an unclear risk of bias or performed with iontophoresis (MD = − 2.0 and − 4.76, respectively).

Endothelial Cell Density
Four studies compared the ECD between the epi-on and epi-off CXL-treated groups (Fig. 2) [48,57,60,61]. There were no significant differences in ECD changes between the two groups (MD = 1.80; 95% CI − 23.25 to 26.86; P = 0.888) (Fig. 2e). Sensitivity analysis demonstrated that the result remained unchanged after removal of the study by Lombardo et al. [48] (MD = 3.0; 95% CI − 22.86 to 28.86; P = 0.468), but the result did change after the removal of the studies with an unclear risk of bias (MD = − 48.82; 95% CI − 180.79 to 83.15; P = 0.820). Potential publication bias for both the primary and secondary outcomes was not assessed due to the limited number of studies available for review.

Quality of the studies
The risk of bias assessment of RCTs and NRSIs is summarized in Additional file 1. Performance bias existed in all RCTs because personnel blinding to the intervention is impossible, but the review authors judge that the outcome is not likely to be influenced by lack of blinding. There was unclear evidence of random sequence generation in two studies [50,57] and high risk of bias in two other studies [49,55]. The allocation concealment was unclear in two studies [49,50]. A high risk of bias due to incomplete data was present in one study [50]. Three studies had high risk of bias [54,55,59] and two had unclear risk of bias [49,60] in selective reporting. Overall, only two studies [48,61] were of high quality with a low risk of bias, whereas two [57,60] were of unclear risk of bias and five [49,50,54,55,59] were of low quality with a serious risk of bias. All the NRSIs were at serious risk of bias due to confounding factors [47, 51-53, 56, 58]; moreover, Eraslan et al. [56] had a serious risk of bias in the selection of participants, Rossi et al. [51] had a serious risk of bias due to missing data and Jouve et al. [52] had a medium risk of bias in measurements of outcomes.

Discussion
CXL is now widely used to prevent the progression of keratoconus by strengthening the biomechanics of the human corneal stroma. Since its development, several clinical studies have reported the effectiveness of CXL using different transepithelial protocols and have compared the efficacy of both treatments. This metaanalysis of RCTs and NRSIs aimed at investigating the outcomes comparing different epi-on CXL protocols with standard epi-on CXL, whereas the difference in the epi-on CXL protocols used by the various authors has led to the lack of definitive evidence as to which technique is preferable.
Fifteen studies with 1085 eyes were included in the qualitative synthesis and a total of 4 studies with 264 eyes were included in the quantitative synthesis. Because the key treatment objective is to stabilize the underlying disease process, corneal topography (Kmax) was considered one of the primary outcome measures. Based on our meta-analysis, Kmax decreased slightly more after the epi-off CXL procedure but did not reach a statistically significant difference compared with epion CXL. Although CXL treatment is not intended to improve visual acuity, the induced changes in corneal topography may result in such a consequent improvement. Based on our systematic review and meta-analysis, the impact of CXL on visual acuity is remarkably different. CDVA showed a significant improvement with epi-on CXL (MD = 0.07; 95% CI 0.04 to 0.10). Based on the MD, patients treated with epi-on CXL had a larger 0.07 logMAR CDVA improvement as compared with control subjects; nevertheless UDVA was not different between both techniques. We assumed that patients treated with CXL protocols that preserved epithelium, probably based on the fewer numbers of postoperative corneal haze, had a higher improvement of CDVA [61]. In our study, we also found that ECD and CCT had no significant changes at long-term follow-up.
A previous systematic review and meta-analysis in 2018 [62] investigated the effectiveness of conventional CXL and transepithelial CXL based on RCTs and revealed no significant differences from the pooled results for the UDVA, CDVA, K-steepest, or ECD. Similarly, Wen et al. [63] compared standard epithelium-off CXL and transepithelial CXL for treating keratoconus and showed no differences in UDVA, CDVA and Kmax outcomes at 1 year. However, neither of the two systematic reviews investigated the RD for adverse events related to the procedures as well as the treatment failure rate, and we are unaware of any other similar analysis in previous systematic reviews. Based on our analysis, a significant delay in epithelial healing and persistent stromal haze were found in epi-off CXL if compared to epi-on CXL ( Table 5).
The standard epi-off CXL procedure in all included studies followed the Dresden protocol (central corneal epithelium was removed, riboflavin drops were instilled for 30 min, and eyes were irradiated with UVA for 30 min at an irradiance of 3 mW/cm 2 ).
We found that the methodology of the epi-on CXL protocol varied among the included studies. To increase epithelial permeability, Al Fayez et al. [59] used benzalkonium chloride and tetracaine; Soeters et al. [61] and Rossi et al. [51,60] used Ricrolin TE solution (SOOFT Italia); Al Zubi et al. [49] used 0.1% riboflavin in 20% dextran for 30 min; Iqbal et al. [50], Henriquez et al. [53], Rush et al. [55] and Eraslan et al. [56] used benzalkonium chloride and hydroxypropyl methylcellulose. Bikbova et al. [57] performed riboflavin soaking using an iontophoresis device with 1.0 mA/cm 2 current density for 10 min to induce absorption. Lombardo et al. preceded iontophoresis by removal of the precorneal mucin layer with the intent to increase epithelial permeability to riboflavin [48]. An iontophoresis device was also applied by Vinciguerra et al. [47,58], Rossi et al. [51,60] and Jouve et al. [52]. However, the total energy density was equal among studies. The disagreement between the epi-on CXL protocols used in the different RCTs has led to a lack of definitive evidence which can be overcome by the scientific methodology applied in the current meta-analysis.
Three of the 15 studies included in the current review treated pediatric patients with progressive keratoconus. Nevertheless, all the 4 RCTs included in the quantitative meta-analysis treated adult patients, and thus the results were not influenced by the patients' age.
The natural course of keratoconus can be long-lasting, with years of apparent stable keratometry readings after a period of latent progression. Only RCTs trials with adequate follow-up of at least one year have been included in this systematic analysis. Nevertheless, all the analyzed studies followed patients for less than 5 years. Consequently, questions about the long-term stability and efficacy of epi-on CXL compared to epi-off CXL beyond that period must be answered and RCTs comparing the long-term outcomes between these techniques are warranted.
Our study has several strengths. To the best of our knowledge, this is the first meta-analysis investigating the adverse events and treatment failure of conventional and transepithelial CXL for patients with keratoconus. Second, we update our review if compared to the previous systematic reviews on the topic: by including 3 RCTs and 1 NRSI published in 2019, we report the most updated evidence. Third, our quantitative evaluation was based only on prospective well-designed RCTs studies Table 4 Sensitivity analysis (mean differences) for the primary and secondary outcomes in randomized clinical trials assessing corneal collagen cross-linking (epi-on vs. epi-off ) CCT central corneal thickness; CDVA corrected distance visual acuity; CI confidence interval; D diopter; ECD endothelial cell density; logMAR logarithm of the minimum angle of resolution; MD mean differences; UDVA uncorrected distance visual acuity All the results were obtained through fixed-effects models because we only had two studies to summarize. If the value of the summary effect is negative, epi-on corneal collagen cross-linking would have more mean differences.  This systematic review also has several limitations that should be considered. First, like any meta-analysis, there is variability between the methodologies used for epi-on CXL among studies. However, our sensitivity analysis did not significantly alter the results when these trials using transepithelial or iontophoresis techniques were excluded. Second, we acknowledge the small number of cases per trial however, the total number of cases included in the meta-analysis gives these analyses high power. Moreover, among the 9 RCTs identified, three were published in open access journals that may or may not have limited the veracity of the data but these were not included in the meta-analysis thereby minimizing their influence on the results presented.

Conclusion
In conclusion, our systematic review and meta-analysis of the literature documenting the comparative efficacy of epi-on vs. epi-off CXL in halting the progression of keratoconus found that both protocols are equally effective. According to the studies included in the quantitative meta-analysis, the rate of keratoconus progression is not significantly different with the two protocols. The published evidence indicates that, except for a larger significant improvement in CDVA with current epi-on CXL protocols, the analysis of well-designed comparative studies available demonstrates that epi-on and epioff CXL have comparable effects on visual, topographic, pachymetric, and endothelial parameters at 1-2 years after surgery in adult patients. Epi-on CXL protocols, by preserving the epithelium, offers a significant reduction in epithelial healing delay and stromal haze formation and a faster postoperative recovery. Current evidence leaves the field open for final, confirmatory, well-designed unbiased RCTs to confirm the long-term efficacy and safety outcomes as well as similarities and differences between the epi-on and epi-off CXL techniques.