Comparison of Standard Versus Accelerated Epithelium-Off and Transepithelial Corneal CrossLinking in Pediatric Keratoconus: An Up-to-date Meta-Analysis

Purpose: To compare the ecacy of four corneal cross-linking (CXL) protocols: standard epithelium-off (SCXL), accelerated epithelium-off (ACXL), transepithelial (TCXL), and accelerated transepithelial (A-TCXL) for pediatric keratoconus. Methods: A comprehensive literature search on the ecacy of SCXL, ACXL, TCXL and A-TCXL in treating keratoconus patients age 18-year or under was conducted using PubMed and EMBASE up to March 2020. Primary outcomes included uncorrected visual acuity (UCVA) and maximum keratometry (Kmax). Secondary outcomes included best-corrected visual acuity (BCVA), central corneal thickness (CCT), and mean refractive spherical equivalent (MRSE). Estimations were analyzed by weighted mean difference (WMD) and 95% condence interval (95% CI) for the outcomes during observation periods from 6 to 36 months. Further investigation is required to compare the ecacy of different CXL protocols for the management of pediatric keratoconus.


Introduction
Keratoconus is a corneal ectatic disorder characterized by asymmetric, non-in ammatory, and progressive conical steepening and thinning. 1The prevalence of keratoconus varies among populations with an estimate of 1/2000 worldwide. 2In pediatric population, the prevalence of pediatric keratoconus is reported to be higher from 1/375 to 1/2000. 3Keratoconus commonly presents in the second decade and progresses until the third or fourth decade of life; compared with adults, pediatric keratoconus is more severe with higher risk of deterioration and faster progression. 4,5The clinical characteristics of keratoconus include progressive loss of vision and increasing irregular astigmatism, resulting from a more conical shape in thinning and steepening cornea. 6Vision loss is often caused by myopia and irregular astigmatism, and in rare cases the rupture of Descement's membrane, acute corneal edema and scarring.In comparison with adults, keratoconus in children appears more centrally located ectatic cornea, and often progresses asymmetrically, leading to good binocular visual performance until both eyes are affected.These factors may contribute to a late seeking in medical care and more deterioated visual function in pediatric patients. 3Visual impairment in keratoconus severely affects educational, economical, and social development, which may decrease patients' quality of life.Thus early and prompt intervention to halt the progression and improve visual quality is very important.
Multiple factors at cellular, physiological, biomechanical, and genetic levels contribute to the progression of keratoconus, and main changes among them are alterations in collagen ber, including the gradual loss of bril orientation, and weaken intra-and inter-brillary collagen cross-links. 7,8Based on an interaction between ribo avin (as a photosensitizer) and ultraviolet A (UVA) radiation, CXL aims to mitigate the progression of the disease by strengthening rigidity of corneal stroma, and avoid the need for corneal transplantation. 9,10though previous clinical trials have studied the postoperative e cacy of CXL in pediatric keratoconus and several have compared two or three protocols, none of them provided comprehensive comparison of four protocols including standard epithelium-off (SCXL), accelerated epithelium-off (ACXL), accelerated transepithelial (A-TCXL), and transepithelial CXL (TCXL).The small sample sizes in single study cast doubt on the validity of their conclusions.Meta-analysis about the comparison of epithelium-off versus transepithelial CXL in adult or mixed-age (both adult and children) patients suggests SCXL and TCXL might provide a comparable effect on visual, refractive, pachymetric and endothelial outcomes postsurgery, 11 however in pediatric population such an analysis is still unavailable.Given the publications of new trials and comparative observations for CXL of different protocols in long-term follow-ups, [12][13][14][15][16] an update of summery is necessary.Hence, we conducted a meta-analysis to compare the e cacy of SCXL with ACXL, TCXL and A-TCXL in pediatric keratoconus.

Evidence acquisition
This meta-analysis was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA checklist guidelines). 17

Search strategy and study selection
A comprehensive literature search was conducted in several databases including PubMed and Embase from earliest available dates to March 2020, in all languages.The keywords "keratoconus", "pediatric", and "corneal collagen cross-linking" OR "corneal collagen cross-linking" OR "CXL" were searched.The related-articles function were also applied to broaden results from the search engine.Reference lists from the publications were also checked for relevant studies.Retrieved papers were screened by two authors independently and duplicated studies were removed.Using inclusion and exclusion criteria described below, the papers were then assessed for meta-analysis.The literature search and selection were shown as owchart (Fig. 1).

Inclusion and exclusion criteria
Articles were retained if they met the following inclusion criteria: (1) observational comparative studies; (2) focused on keratoconus in patients aged 18 or younger; (3) involved at least two types of CXL protocols including SCXL, ACXL and epi-on; (4) outcomes containing at least UCVA (transferred to the log minimum angle of resolution, LogMAR), BCVA, Kmax, and CCT; (5) a minimum follow-up of 12 months.Articles were excluded if any of the following conditions existed: (1) studies with inadequate information for calculating data on outcomes; (2) duplicated report; (3) study with a sample size smaller than 10 eyes in each arm.If multiple studies by the same research team derived from different population were available, all of them were deemed eligible and included in the meta-analysis.

Data extraction
Two authors (YJL, YL) extracted data independently with a standardized form.The following information was retrieved in all the included publications: rst author name, year of publication, country, sample size of patients and eyes, mean age, gender, study design, type of CXL protocol, Amsler-Krumeich stage, follow-up time, UCVA, BCVA, Kmax, and CCT at the observation point (at speci c time-points or at the last follow-up).Any discrepancies in data extraction or disagreements in the data were resolved by discussion and re-assessment with the senior author (DW).

Quality assessment
Risk of bias of each included study was evaluated for the level of evidence according to criteria provided by the Center for Evidence-Based Medicine in Oxford. 18The quality of the randomized controlled trials (RCTs) was evaluated using Cochrane risk of bias tool, 19 and the quality of the comparative nonrandomized studies (CNSs) was evaluated using the ROBINS-I assessment tool. 20

Statistical analysis
Meta-analysis was performed using Review Manager Version 5.3.5 (Cochrane Collaboration, London, UK).
Dichotomous and continuous variables were compared using the odds ratio (OR) and weighted mean difference (WMD), respectively.Pooling estimates and their 95% con dential intervals (95% CI) were calculated.The xed-effect model was applied, and heterogeneity was quanti ed using the I 2 value, which represents the percentage of the total variation among studies.Cochrane Q-test P value > 0.1 was considered as no signi cant heterogeneity, and the random-effects model was used to calculate pooling estimates and address within-or between-study variances.For a clear visualization, forest plots were produced.An I 2 value of 25%-50%, 50%-75%, > 75% was de ned as low, moderate, and high heterogeneity, respectively.Statistically signi cance was measured by a P value of less than 0.05 (P < 0.05).Analyses were strati ed by types of CXL protocol and follow-up time.

Study selection
The ow diagram of the study selection is illustrated in Fig. 1.Initially, a total of 1722 articles were retrieved from databases.After removing the duplicates, 1595 potential papers left, and the titles and abstracts were reviewed.Among them, 1568 articles were excluded because of irrelevant topics.Full texts of the remaining 27 papers were assessed in their entirety, and 19 articles were excluded as no comparative data or not suitable for analysis.Eventually, 8 articles that provided detailed quantitative data were included in this meta-analysis.Informed consent was obtained in all included studies.

Characteristics of the included studies and quality assessment
Details of characteristics of included 8 studies are summarized in Table 1.The studies were published between 2013 and 2020, totally examining 704 eyes from 466 pediatric patients.The follow-up time of each research was ranging from 6 months to 48 months.−25 Each study compared the UCVA, BCVA, Kmax and CCT of different CXL protocols.The surgical procedures of the included studies were either SCXL, ACXL, TCXL or A-TCXL.The detailed information about the CXL procedures was listed in Table 2.There were 8 included studies involving the application of SCXL, 5 studies involving ACXL, 2 studies employing the TCXL and 2 studies the A-TCXL.The quality assessment of the RCTs is shown in Table 3.All CNSs were judged by the ROBINS-I assessment tool (shown in Table 4).
L, low risk of bias; M, moderate risk of bias
In addition to visual acuity, Kmax and CCT are important to assessment of the treatment effect.The ACXL resulted in statistically smaller Kmax as compared to that of the SCXL group at 1 year postoperation (Fig. 4A.WMD = 0.70, 95% CI: 0.24 to 1.17, P = 0.003, I 2 = 75%).At 2 years after the surgery the Kmax in ACXL group was also signi cantly smaller than that in SCXL group with high heterogeneity (WMD = 0.70, 95% CI: 0.25 to 1.14, P = 0.002, I 2 = 83%).The overall results were similar, with WMD = 0.76, 95% CI: 0.50 to 1.02, P < 0.00001, I 2 = 70%, which suggest that the ACXL may result in a smaller Kmax as compared to the SCXL.The CCT post-operation were not signi cantly different between the two types of CXL at either short or long follow-up term ranging from 6 months to 36 months (Fig. 4B.WMD = -2.98,95% CI: -6.40 to 0.44, P = 0.09, I 2 = 13%).
With respect to the postoperative values of Kmax, there was no signi cant difference between the SCXL group and the A-TCXL group at 6-month follow-up (Fig. 7A.WMD = -0.55,95% CI: -1.33 to 0.23, P = 0.16, I 2 = 55%).At 12-month visit, however, patients treated with the SCXL were as likely to achieve Kmax reduction as those with the A-TCXL (WMD = -0.93,95% CI: -1.71 to -0.15, P = 0.02, I 2 = 0%).Similarly, there was a superiority of the Kmax reduction in the SCXL group over the A-TCXL group in the pool analysis (WMD = -1.12,95% CI: -1.58 to -0.66, P < 0.00001, I 2 = 56%).In terms of the postoperative CCT, there was no signi cant difference between the two groups at all the follow-up time-point (Fig. 7B).

Discussion
of the asymmertric progression, and inverse correlation between onset age and severity of the disease, keratoconus in pediatric population are often more advanced at the time of diagnosis.Corneal cross-linking is recommended to apply to halt progression of the disease in mild-to-moderate corneal extasia. 3Conventional epithelium-off CXL, usually referred to as 'Dresden' protocol, was applied in the majority of keratoconus.However, recently increasing attention is being paid to the faster alternate, the accelerated protocols. 15Transepithelial approach is another choice which allows intact epithelium with newer preparation of ribo avin, seems to be ideal for pediatric cases to avoid epithelial debridement. 27hortening the treatment time and lowing the risk of postoperative pain or infections are essential for children patients, however, these adaptations in protocol may also lead to decreased permeability of ribo avin passing through intact epithelium.Thus it was reported that the epi-off CXL is less effective in stopping the disease as compared to the epi-off, 28 which also showed in our meta-analysis.Similarly, a multi-center trial by Iqbal et al showed that standard epithelium-off was more effective in pediatric keratoconus, attaining great stability as compared to either accelerated or transepithelial CXL. 12 On the other hand, Eissa et al reported that at 12-month follow-up, postoperative mean LogMAR UCVA, BCVA and Kmax of accelerated CXL were statistically less than those of conventional CXL in pediatric keratoconus eyes. 14The con icting results might result from varied follow-up duration and small sample sizes.Our meta-analysis could provide pooling data from multiple studies updated to 2020 and offer potential important insights into the CXL strategy prior to carrying out a large-scale clinical trial.
The included studies compared the e cacy and stability of either two or three of the standard epitheliumoff, accelerated epithelium-off, trans-epithelial and accelerated epithelium-off CXL in pediatric keratoconus.Although two among the eight studies were RCTs, most of the researches were relatively high quality, according to the quality assessment.These studies contained the newest surgical outcomes and observations of total 704 eyes (number of eyes in each protocols shown in Table 1) for the pediatric keratoconus treatment.An earlier meta-analysis by McAnena et al has evaluated 13 papers including 490 eyes of 401 pediatric patients with keratoconus, which compared the pre-and postoperative CXL outcomes in standard epithelium-off and trans-epithelial protocol. 15However, they only analyzed the outcomes in either protocol but not between the two groups, likely due to the lack of enough published data at that time.The McAnena study found that standard protocol might be effective in halting progression of pediatric keratoconus, with signi cant improvement in UCVA and BCVA at 1 year and statistical reduction in Kmax at 2 years.In their results, no signi cant vision gain or change in Kmax was observed in the trans-epithelium group at 1 year visit.Our study excluded the old publications and included some recent ones which contained comparative data about different protocols.Besides, most of the included studies have multiple observing time-points and longer term of postoperative follow-up, ranging from 12 months to 48 months (Table 1), which allowed subgroup analysis according to different follow-up time.This meta-analysis compared post-operative outcomes between different CXL protocols in both short-and long-term.
Our results indicated that the long-term best spectacle-corrected visual outcomes (BCVA, SE at 24 months) were in favor of SCXL as compared with ACXL (Fig. 2B and Fig. 3).Furthermore, the postoperative UCVA and BCVA in SCXL were signi cantly superior to those in A-TCXL group (Fig. 6).
These results are similar to those in Kobashi and Tsubota study, which focused on adult population and showed that ACXL has less effect on improving corrected visual acuity than SCXL after 1-year followup.33 Kmax in SCXL was also signi cant lower than those in A-TCXL group (Fig. 7A).Soeters et al reported that in adult population, transepithelial CXL might result in a continued keratoconus progression after 1 year. 28Ng et al also showed that standard CXL resulted in a signi cantly greater reduction in Kmax and Kmean than its accelerated counterpart. 29Our results in pediatric patients are in accordance with these previous studies.Although there was no signi cant difference of UCVA between the SCXL and TCXL group, the BCVA in the SCXL group was signi cantly superior over that in the TCXL group (Fig. 5A  and B).However, the data was documented in only two studies, and sample size was small, which required further clinical trials.The results indicated that in terms of halting progression and improving visual acuity, the standard epithelium-off CXL might be more e cacious than the others, in accordance with the conclusions of McAnena.
Corneal thickness was reported to increase after standard epithelium-off CXL possibly due to scattering formation. 15However, comparing to the thinning from natural progression of keratoconus, some other studies reported stable CCT after CXL. 30,31Different measurement techniques such as the Orbscan II and Pentacam HR might be the cause for variable.Previously it was reported that for adult patients with keratoconus, transepithelial CXL provided a more protective in uence on corneal thickness than standard CXL. 32Early meta-analysis showed that accelerated CXL might lead to a less reduction in CCT than standard CXL, 33,34 but recent studies suggests no difference between the ACXL and SCXL when comparing the CCT. 35,36In pediatric population, our meta-analysis also observed no statistical difference in CCT among different CXL protocols, regardless of different follow-up time (Fig. 4B, Fig. 5D, Fig. 7B).Endothelial cell count was also an important factor which affects the recovery of pediatric patients.
Salman 2013 reported no signi cant changes occurred in the endothelial cell count in either transepithelial CXL or conventional treatment group of pediatric patients. 37Another trial also showed that counting of endothelial cell did not change signi cantly during follow-up in iontophoretic CXL. 24This meta-analysis, however, did not enroll the endothelial cell count due to little records in the included studies.So, more clinical trials should be conducted in the future about speci c perioperative outcomes, especially the endothelial cell count that could affect the postoperative vision quality.Besides, although being relatively rare, adverse events such as corneal edema, transient haze, permanent scar, stertile ulcer, and infectious keratitis should be recorded to provide more detailed observations of CXL complications in pediatrics.A recent study showed that three of 968 eyes developed infectious keratitis and seven sterile in ltrates after accelerated CXL over 4 years follow-up, but the studied population included both children and adults. 38Maharana et al 2018 reported microbiological test results of the microbial keratitis after accelerated CXL, showing mixed and simple infection in the cases. 39The mixed infections included coagulase-negative Staphylococcus (CoNS) with Aspergillus fumigatus, Staphylococcus aureus with Mucor spp., Staph.aureus with Acanthamoeba, and simple infection included Staph.Aureus, CoNS, and Alternaria spp. 39ere are some limitations in this meta-analysis that should be discussed.Firstly, the surgical procedures of CXL in each protocol were not exactly the same, as shown in Table 2.In the Epi-off groups, the process of epithelium removal was different in concentration of ethanol and soaking time, and the ribo avin instilment interval during UVA exposure was not uniform.In the Epi-on groups, the concentrations of ribo avin in soaking, the UVA power, and the process of UVA exposure were varied.Second, most of the included studies were not randomized comparative.This would increase the risk of potential selection and publication bias.Lastly, some outcomes had high heterogeneity, such as the Kmax in subgroup analysis between SCXL and ACXL, and the UCVA and BCVA between SCXL and A-TCXL, which were possibly caused by different baseline features or surgical techniques.
Due to limited number of included studies, conclusions from this meta-analysis should be interpreted cautiously.The stability and e cacy of different CXL protocols may differ in different follow-up time.

Table 1
Characteristics of all included studies in the meta-analysis.

Table 2
Surgical procedures of the included studies in the meta-analysis.
NG, not given; Pulsed mode, 1 second on, 1 second off.