Introduction

Keratoconus is the most common corneal ectatic disease. It is a chronic progressive eye condition in which the cornea deforms to a more conical shape causing visual impairment.1 For the age group 10–44 years, the prevalence of keratoconus is 57 per 100 000 in Caucasians, but over four-fold higher in people originating from the Indian subcontinent. Conventionally spectacles, contact lenses, and corneal transplantation are the mainstay of treatment. The expected lifetime cost of management of keratoconus is $25 168 per patient.2 The factor that most influences health-care cost is the risk of initial corneal transplantation.

In 2003, Wollensak et al3 published a seminal article on corneal collagen crosslinking (CXL) describing the use of CXL to arrest the progression of keratoconus. They described a case series of patients with progressing keratoconus who had undergone epithelial removal (ER) CXL with riboflavin and ultraviolet A (UVA).3 In the 10 years since this paper, many investigators have reproduced these findings.4, 5, 6, 7, 8 In brief, this CXL procedure leads to photo-oxidation leading to additional covalent bonds between and within collagen fibrils of the cornea, which increases corneal stiffness, stabilises the keratoconus and, in some cases, improves refractive and topographic features.9, 10, 11 In this regard, riboflavin penetration into the corneal stroma is essential as this molecule absorbs UVA to achieve crosslinking and shields the underlying endothelium from its harmful effects. CXL is a relatively safe technique; however, complications related to epithelium removal and bandage contact lens use (such as corneal infiltrates,12, 13, 14 corneal melting,15, 16 infection,17, 18, 19, 20, 21 and scar formation22) have been reported.

The CXL technique has evolved rapidly over the last decade. There are studies reporting the use of pharmacological agents to loosen the epithelium before instillation of riboflavin,23, 24, 25, 26, 27, 28, 29, 30 iontophoretic experiments to enhance the riboflavin permeability,31, 32 partial disruption of epithelium,28, 33, 34 and even CXL with intact epithelium.27, 29, 30 As epithelial debridement is reported to be an essential step in the CXL reaction involving UVA and hydrophilic riboflavin,3 we performed this systematic review to analyse the differences in the safety and efficacy profiles of ER and transepithelial (TE) CXL techniques in the management of progressive keratoconus.

Materials and methods

We conducted a systematic review of studies in which CXL was used to treat progressing keratoconus. We aimed at including randomised controlled trials (RCTs) comparing the techniques ER or TE CXL. In the absence of any RCTs with direct comparison between the techniques, we decided to include RCTs comparing either ER or TE techniques with no treatment, as well as case series in which a minimum of 20 eyes were treated with either ER or TE technique and at least 12-month follow-up. These parameters were chosen to ensure only high quality studies were included. We accepted peer-reviewed articles of human studies only and included articles in any language. Conventional as well as accelerated treatments were included. Articles published online ahead of print were also included. We excluded animal and ex vivo studies, as well as studies investigating non-keratoconus corneal ectatic pathologies such as pellucid marginal degeneration and post-refractive surgery ectasia. We also excluded studies in which CXL was performed in combination with other surgical procedures such as intra-corneal segment insertion, excimer laser procedures, or iontophoresis techniques.

We performed a MEDLINE search for articles published to 26 January 2014 without stipulating any conditions on date or language of publication. We used the following search strategy:

  1. 1

    ‘crosslinking’ OR ‘cross-linking’ OR ‘crosslinkage’ OR ‘cross-linkage’ OR ‘cxl’ (48 663 results)

  2. 2

    ‘cornea’ OR ‘corneal’ (84 214 results)

  3. 3

    ‘collagen’ AND ‘keratoconus’ (453 results)

We then combined 1 AND 2 OR 3, producing 773 studies.

We assessed the titles and the abstracts resulting from the searches. We considered full-text copies of all possibly relevant studies to see whether they met the inclusion criteria. We extracted the data using a form developed by us on an Excel 2010 spreadsheet (Microsoft, Redmond, WA, USA) outlining efficacy and safety parameters. One review author entered the data on the spreadsheets. Any disagreements for inclusion or exclusion of the studies were resolved by discussion among us. There were no exclusions based on the randomisation methods in RCTs as long as the trial design was suitable for the conditions and procedures being studied. Publications in a language other than English were translated using Google Translate (Mountain View, CA, USA). Authors forming research teams were grouped together in tables to identify redundant articles. Redundant articles, in which identical data are published in a different language, were treated such that only one article was tabulated.

We predicted studies to have varying follow-up in each arm and so decided to present the data at their latest follow-up visit and the data on change in logMAR CDVA, change in refractive cylinder, and change in Kmax at 1 year for better comparison between ER and TE groups.

Statistical analysis was performed with SPSS Statistics 17.0 (IBM Corporation, Armonk, NY, USA) and utilised medians rather than means to overcome methodological problems involving redundant studies. All visual acuity data were converted to logMAR if presented in Snellen or decimal formats.

Results

We identified 45 ER4, 5, 6, 7, 8, 9, 10, 11, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71 and 6 TE27, 28, 29, 30, 36, 72 studies satisfying our entry criteria (Tables 1 and 2; Figure 1). Of the included studies, only one was an RCT comparing ER and TE crosslinking.36 The study designs of all included studies are described in Tables 1 and 2. The TE studies were all published in English, but five of the ER studies56, 59, 73, 74, 75 were published in German. Three73, 74, 75 of these were excluded as they were redundant articles presenting data identical to an included English article. A further redundant article in English was excluded,76 as was a large study that failed to present the sample size at follow-up.77 The analysis includes a total of 1990 eyes in the ER group and 215 eyes in the TE group. Excluded studies3, 51, 73, 74, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94 are listed in Table 3, together with the reason for their exclusion. As there was only one RCT comparing ER vs TE, meta-analysis was not possible.

Table 1 Systematic review of efficacy of epithelial removal (ER) corneal collagen crosslinking
Full size table
Table 2 Systematic review of efficacy of transepithelial (TE) corneal collagen crosslinking
Full size table
Figure 1
figure 1

Study flow diagram.

Full size image
Table 3 Prominent corneal collagen crosslinking trials not included in systematic review and reason for exclusion
Full size table

Crosslinking efficacy

Tables 1 and 2 present the crosslinking efficacy data for ER and TE studies, respectively. Articles are listed by alphabetical order as per first author and those by the same author listed together to highlight the possibility of data redundancy. ER articles were published from 2008 onwards, while TE studies were published from 2010 onwards. ER studies in general included a larger number of eyes and had longer follow-up (range: 12–72 months) vs TE studies (range: 12–24 months). The included RCT36 reported only pachymetry out of the included study parameters.

Uncorrected distance visual acuity

At the last follow-up visit 21 ER studies4, 6, 8, 10, 11, 35, 37, 38, 43, 46, 47, 50, 51, 52, 60, 61, 64, 67, 68, 69, 70 out of 45 studies reported uncorrected distance visual acuity (UDVA) (Table 1). Seventeen4, 6, 8, 10, 11, 35, 37, 38, 46, 50, 51, 60, 61, 64, 67, 68, 70 of these 21 ER studies (81% of studies) showed a median improvement of −0.17 logMAR UDVA (range: −0.37 to −0.01 logMAR), 3 studies43, 47, 52 showed median worsening of 0.18 logMAR UDVA (range: 0.03–0.25 logMAR) and 1 study69 showed no change in UDVA.

Three TE studies28, 29, 72 out of six reported UDVA (Table 2). Only two studies28, 29 out of these three (66.7% of studies) showed a median improvement of −0.27 logMAR (range −0.23 to −0.30) and one study76 showed worsening by 0.08 logMAR.

Corrected distance visual acuity

Thirty-three ER studies4, 5, 6, 7, 8, 9, 10, 11, 35, 37, 38, 40, 42, 43, 44, 45, 46, 47, 50, 51, 52, 53, 56, 58, 59, 60, 61, 62, 63, 67, 68, 69, 70 out of 45 reported corrected distance visual acuity (CDVA) (Table 1). Twenty-eight4, 5, 6, 7, 9, 10, 11, 35, 37, 38, 40, 42, 43, 44, 46, 50, 53, 56, 58, 59, 60, 61, 62, 63, 67, 68, 69, 70 these 33 studies (85% of studies) showed a median improvement of −0.09 logMAR (range: −0.23 to −0.01 logMAR), 4 studies8, 45, 47, 51 showed median worsening of 0.05 logMAR (range: 0.01–0.05) and 1 study52 did not show any change in CDVA.

Similarly, five TE studies27, 28, 30, 75, 76 out of six reported CDVA (Table 2). Four TE studies27, 28, 29, 30 of these five (80% of the studies) showed a median improvement of −0.08 logMAR (range: −0.12 to −0.04 logMAR) and one study72 showed no change in CDVA.

Mean myopic spherical equivalent

Fourteen ER studies4, 6, 8, 10, 11, 35, 37, 38, 43, 47, 52, 61, 63, 67 of 45 reported spherical equivalent (Table 1). Thirteen studies4, 6, 8, 10, 11, 37, 38, 43, 47, 52, 61, 63, 67 of these fourteen (93% of studies) showed a reduction in mean myopic spherical equivalent (median 0.85D, range 0.39–2.15D) and one study35 showed a −0.61D worsening.

Four TE studies27, 28, 30, 72 out of six reported spherical equivalent (Table 2). Three27, 28, 30 of these four studies (75% of studies) showed a reduction in mean myopic spherical equivalent (median 0.36D, range 0.21 to +0.74D) and one study72 showed no change.

Refractive astigmatism

Twenty-one ER studies4, 6, 8, 9, 10, 11, 35, 37, 38, 43, 44, 47, 51, 52, 59, 60, 61, 62, 63, 67, 69 out of forty-five reported refractive astigmatism (Table 1). Thirteen4, 6, 8, 9, 10, 11, 35, 37, 38, 44, 59, 60, 63 of these twenty-one ER studies (62% of studies) showed a reduction in refractive astigmatism (median −0.55D, range −1.64 to −0.08D) and eight ER studies43, 47, 51, 52, 61, 62, 67, 69 showed increased astigmatism (median 0.56D, range: 0.07–1.30 D).

Only three TE studies27, 28, 72 reported data on refractive astigmatism (Table 2). One study each reported reduced (−1.15D) (33% of the studies),28 increased (0.04D),27 and no change72 in refractive astigmatism post-crosslinking.

Maximum keratometry

Twenty-nine ER studies4, 5, 6, 7, 8, 9, 10, 11, 35, 37, 38, 39, 40, 42, 44, 45, 46, 50, 51, 53, 56, 58, 59, 61, 62, 67, 68, 69, 70 out of forty-five showed reported data on maximum keratometry (Kmax) (Table 1). Twenty-seven studies5, 6, 7, 9, 10, 11, 35, 36, 37, 38, 39, 41, 43, 44, 45, 47, 50, 52, 54, 57, 60, 62, 63, 69, 70, 71, 72 of these twenty-nine (93% of the studies) showed reduction (median −1.01D, range −0.14 to −6.16D) and two59, 73 out of twenty-nine studies showed an increase in Kmax (median 0.75D, range 0.4–1.1D).

Five TE studies27, 28, 29, 30, 72 out of six reported data on Kmax (Table 2). Two27, 30, 72 of these five TE studies (40% of the studies) reported reduction in Kmax (median −0.87D, range −1.17 to −0.57D) and three TE studies28, 29 reported an increase in Kmax (median 1.33D, range 0.51–1.55D).

Pachymetry

In the included RCT,36 corneal thickness increased in both TE and ER at 12 months but more after TE crosslinking (Tables 1 and 2). Twenty-four ER studies4, 6, 8, 11, 35, 37, 38, 39, 42, 43, 46, 47, 48, 50, 51, 52, 56, 58, 61, 62, 63, 69, 70, 71 out of forty-five reported data on pachymetry (Table 1). Fifteen11, 37, 38, 42, 43, 46, 51, 52, 56, 58, 62, 63, 69, 70, 71 of these twenty-five ER studies (60% of the studies) reported reduction in pachymetry (median change: −11.33 μm, range −66.46 to −0.24 μm) and 9 ER studies4, 6, 8, 35, 39, 47, 48, 50, 61 showed an increase in pachymetry (median change: 4.63 μm, range 0.6–37 μm).

Four TE studies27, 28, 29, 72 out of six reported data on pachymetry (Table 2). Three27, 29, 72 of these (75% of the studies) showed reduction in pachymetry (median change: −8.06 μm, range −32 to −0.4 μm) and one study28 showed increase in pachymetry (9 μm).

Treatment failure

Only five ER studies9, 42, 44, 59, 69 out of six reported treatment failure (median percentage of eye: 8.6%; range: 8.1–33.3%) (Table 1), and where this was done, the definitions were not consistent. Only one29 (out of six) TE studies reported 0% treatment failure (Table 2).

Retreatment rates

Only two ER studies reported retreatment rates of 5.4%44 and 8.6%59 (Table 1), whereas no TE study reported any retreatment rates (Table 2).

Conversion to deep anterior lamellar keratoplasty

Only one ER study reported 6.25%69 of patients progressing to deep anterior lamellar keratoplasty (DALK) and no TE study reported any conversion to DALK (Tables 1 and 2).

Crosslinking safety

Tables 4 and 5 present the safety data for the ER and TE studies, respectively. Overall, the ER studies reported more adverse events than TE studies, although reporting was haphazard in almost all studies. The included RCT36 reported outcomes on stromal oedema and change in endothelial counts only.

Table 4 Systematic review of safety of epithelium removal (ER) corneal collagen crosslinking
Full size table
Table 5 Systematic review of safety of transepithelial (TE) corneal collagen crosslinking
Full size table

Failure to re-epithelise

Nine ER studies5, 8, 38, 50, 51, 52, 54, 56, 59 reported data on this and showed no reports of failure to re-epithelise (Table 4). By definition, TE studies did not show any problems in this category (Table 5).

Stromal oedema

The included RCT36 reported 1.68% stromal oedema with ER compared with 0% with TE crosslinking at 12 months.

Six ER studies4, 5, 51, 68, 69, 70 reported data on stromal oedema with the median percentage of 17.5% (range: 0–70%) after treatment (Table 4) whereas no TE study (except the included RCT36) reported on stromal oedema (Table 5).

Sterile infiltrates

Only six TE studies5, 35, 42, 59, 67, 70 reported data on sterile infiltrate with median percentage of eyes of 2.5% (range: 2–4%). Hoyer et al59 noted sterile infiltrates which resolved on treatment with topical steroids (no percentage of eyes mentioned) (Table 4). None of the TE studies reported sterile infiltrate (Table 5).

Golden striae

Golden striae were reported by two ER studies from the same group in 43.5%38 to 62.0%6 of eyes (Table 4). There were no eyes with this complication in the TE group (Table 5).

Stromal haze

Twelve4, 5, 6, 8, 38, 45, 49, 56, 66, 68, 69, 70 of forty-five ER studies reported data on stromal haze as a phenomenon that was responsive to topical steroid treatment (median percentage of eyes: 9.8%; range: 0–100%). One66 of these twelve ER studies reported haze in their own grading system and hence it was not possible to include their stromal haze data in the calculations (Table 4).

Four TE studies27, 28, 29, 30 reported data on stromal haze (median percentage of eyes: 0%; range: 0–100%) (Table 5).

Corneal scar formation

Only 5 TE studies5, 8, 62, 68, 69 out of 45 reported corneal scar formation (median percentage eyes: 0%; range: 0–6%) (Table 4).

Four TE studies27, 28, 29, 30 reported data 0% scar formation (Table 5).

Incidence of microbial keratitis

Seven ER studies5, 8, 42, 46, 52, 59, 68 out of forty-five reported data on microbial keratitis (median percentage of eyes: 0%; range: 0–3%). One46 of these seven ER studies did not report microbial keratitis data specifically in eyes with keratoconus and hence the keratitis data from this study were not considered for calculation (Table 4).

Four TE studies27, 28, 29, 30 out of six reported data 0% incidence of microbial keratitis (Table 5).

Loss of CDVA

Six ER studies9, 47, 56, 58, 62, 69 of forty-five reported data on loss of CDVA (median percentage of eyes: 12.4%; range: 0–27%) (Table 4).

Whereas only two TE studies27, 30 out of six reported data (0% eyes with loss of CDVA) (Table 5).

Changes in endothelial cell count

The included RCT36 did not show significant difference in endothelial cell counts after ER or TE crosslinking.

Thirteen ER studies4, 6, 35, 37, 38, 39, 43, 46, 50, 51, 56, 69, 70 of forty-five reported on endothelial cell counts. Two46, 50 of these fourteen ER studies reported no change in endothelial cell counts, whereas nine ER studies4, 6, 35, 37, 38, 39, 43, 51, 69 reported reduction in endothelial cell counts (median: −24 cells/mm2; range: −131 to −12 cells/mm2) and two56, 70 of these fourteen ER studies reported a small increase in endothelial cell counts (median: 29.5 cells/mm2; range: 4–55 cells/mm2) (Table 4).

Three TE studies28, 29, 30 out of six reported data on endothelial cell counts. One36 of these four TE studies reported no change, two28, 29 reported reduction in endothelial cell counts (median: −82 cells/mm2; range: −130 to −34 cells/mm2) and one30 reported an increase in cell counts (27 cells/mm2) (Table 5).

Comparison of mean change in logMAR CDVA, refractive cylinder, and Kmax at 1 year

Thirty-three ER studies4, 5, 6, 7, 9, 10, 11, 35, 37, 38, 39, 40, 42, 43, 44, 45, 47, 48, 50, 51, 52, 53, 54, 56, 59, 60, 61, 62, 63, 67, 68, 69, 70 out of forty-five and five TE studies27, 28, 29, 30, 72 out of six reported one or more of these parameters at 1 year (Table 6).

Table 6 Systematic review of efficacy of epithelial removal (ER) and transepithelial corneal collagen crosslinking: change in CDVA, cylinder and maximum keratomtery at 12 months
Full size table

LogMAR CDVA at 1 year

Thirty ER studies4, 5, 6, 7, 9, 10, 11, 35, 37, 38, 40, 42, 43, 44, 45, 47, 50, 51, 52, 53, 56, 59, 60, 61, 62, 63, 67, 68, 69, 70 out of these thirty-three reported this parameter. Twenty-seven4, 5, 6, 7, 9, 10, 11, 35, 37, 38, 40, 42, 43, 44, 50, 52, 53, 56, 59, 60, 61, 62, 63, 67, 68, 69, 70 out of these thirty ER studies (90% of the studies) showed a median improvement of −0.09 logMAR CDVA (range: −0.58 to −0.01 logMAR) and the remaining three ER studies45, 47, 51 showed a median worsening of 0.05 logMAR CDVA (range, 0.01–0.05 logMAR) (Table 6).

Four TE studies27, 28, 29, 30 out of five (80% of the studies) reported a median improvement of −0.07 logMAR CDVA (range:−0.12 to −0.04 logMAR) at 1 year (Table 6).

Refractive cylinder at 1 year

Seventeen ER studies4, 9, 10, 11, 35, 37, 44, 47, 59, 60 out of thirty-three reported this parameter at 1 year. Ten ER studies4, 9, 10, 11, 35, 37, 43, 44, 47, 51, 52, 59, 60, 61, 62, 67, 69 out of these 17 (59% of the studies) reported a median reduction of −0.65D refractive cylinder (range: −1.02 to −0.02D) whereas remaining seven studies43, 51, 52, 61, 62, 67, 69 reported a median worsening of refractive cylinder (median: 0.25D; range: 0.07–0.99D) (Table 6).

Only two TE studies out of five reported data on refractive cylinder: one28 reported improvement of −1.15D (50% of the studies) and other27 reported worsening by 0.04D at 1 year (Table 6).

Kmax at 1 year

Twenty-two ER studies5, 7, 9, 10, 11, 35, 37, 39, 40, 42, 44, 45, 50, 51, 53, 59, 61, 62, 67, 68, 69, 70 out of thirty-three at 1 year reported this parameter. Twenty5, 7, 9, 10, 11, 35, 37, 39, 40, 42, 44, 45, 50, 51, 59, 61, 62, 67, 68, 69 of these twenty-two ER studies (91% of the studies) showed a median reduction in Kmax by −0.82D (range: −6.26 to −0.16D) and only two studies53, 70 showed a median worsening of Kmax by 0.48D (range: 0.4–0.56D) (Table 6).

Four TE studies27, 28, 30, 72 of five reported data on this parameter at 1 year and three27, 30, 72 of these (75% of the studies) showed a median worsening of 0.60D Kmax (range: 0.51–1.33D) and one study28 showed a reduction of −0.57D in Kmax at 1 year.

Discussion

Our systematic review highlights that although there is a paucity of TE studies in comparison with existing ER studies and follow-up remains relatively short in TE trials, the majority of eyes have improved visual acuity and reduced myopic spherical equivalent after ER or TE CXL. Nevertheless, although TE CXL has fewer complications, it is less effective, particularly in stabilising or improving Kmax.

The main conclusions of this review are listed below:

  1. 1

    Majority of the studies in ER (17 out of 21 studies=81%) and TE (2 out of 3 studies=66.7%) groups at the latest follow-up showed improvement in logMAR UDVA (Tables 1 and 2).

  2. 2

    Majority of the studies in ER (28 out of 33 studies=85%) and TE (4 out of 5 studies=80%) groups showed improvement in logMAR CDVA (Tables 1 and 2). This was similar at 1-year follow-up in ER (27 out of 30 studies=90%) and TE (4 out of 5 studies=80%) (Table 6).

  3. 3

    Majority of the studies in ER (13 out of 14 studies=93%) and TE (3 out of 4 studies=75%) groups showed reduction in mean myopic spherical equivalent (Tables 1 and 2).

  4. 4

    Over half of the studies in ER group (13 out of 21 studies=62%) and a third of the studies in TE group (1 out of 3 studies=33%) showed reduction in refractive cylinder (Tables 1 and 2). This was similar at 1-year follow-up in ER (10 out of 17 studies=59%) and TE (1 out of 2 studies=50%)(Table 6).

  5. 5

    Majority of the studies in ER (27 out of 29 studies=93%) showed reduction in Kmax whereas with TE, majority (3 out of 5 studies=75%) showed worsening in Kmax (Tables 1 and 2). This was similar at 1-year follow-up in ER (20 out of 22 studies=91%) showing improvement in Kmax and TE (3 out of 4 studies=75%) studies showing worsening (Table 6).

  6. 6

    Equal proportion of studies in ER (15 out of 25 studies=60%) and TE (3 out of 5 studies=60%) showed reduced pachymetry following CXL (Tables 1 and 2).

  7. 7

    Treatment failure (although this was defined variably in many studies), retreatment rates, and conversion to DALK were reported to be up to 33, 8.6, and 6.25%, respectively, in studies of ER group only (Tables 1 and 2). This may be due to significantly less number of TE studies reported until January 2014.

  8. 8

    Stromal oedema, haze, scarring, and risk of microbial keratitis were only seen in ER studies. Endothelial cell counts were variable in both ER and TE groups (Tables 4 and 5).

Since the publication of the first seminal study 10 years ago,3 CXL has revolutionised the treatment of keratoconus. Although many established therapies, such as rigid gas-permeable contact lenses and corneal transplantation, are effective in improving patient vision, no known therapy other than CXL is successful in halting the progression of disease. The work of the Dresden group revolutionised the field by showing that CXL could not only do this, but in some cases also leads to an improvement in many anatomical and refractive indices in keratoconus. However, despite the value of CXL in halting the progression of keratoconus, several investigators have raised concerns about its significant vision-threatening complications. These include corneal infiltrates,12, 13, 14 melting,15, 16 infection,17, 18, 19, 20, 21 and scar formation,22 all of which may lead to a reduction in CDVA.

Encouraged by the efficacy of ER CXL, some investigators concluded that CXL would prove significantly safer if the epithelium could be left in situ.95 This raised the problem of how riboflavin, a hydrophilic molecule, could be transported across the hydrophobic corneal epithelium. Several methods have now been shown to be helpful in achieving this, including the use of benzalkonium chloride23, 24, 96 EDTA,25 gentamicin,26 iontophoresis,31, 32 as well as minimal trauma (through epithelial poke marks) to the epithelium.33

Our review sought to answer the question of whether the new TE form of CXL is as effective as the standard ER form, and whether it is truly safer. This review certainly shows TE crosslinking lacks many of the significant complications of ER CXL. Despite lower numbers of TE studies published to date, the efficacy of ER and TE techniques appears to be comparable for most parameters with majority of the studies showing improvement of UDVA, CDVA, myopic spherical equivalent, and refractive astigmatism (Tables 1 and 2). However, whereas 93.1% (27 of 29 studies) showed Kmax to be stable or better with ER CXL, this figure was only 40.0% (2 of 5 studies) for TE CXL. This is of concern as Kmax is arguably the most important parameter when considering keratoconus progression, and hence, treatment failure. The greater efficacy of ER than TE CXL may be related to the deeper demarcation line observed after treatment.29

The TE CXL studies considered here had different surgical methodologies, with altogether disparate treatment effects. Most TE studies were able to achieve results that are comparable to ER studies.27, 28, 29, 30, 36, 72 Filippello et al29 used EDTA and trometamol as epithelial permeation enhancers, as well as a silicone corneal ring to help create a pool of enhanced riboflavin solution 30 min before UVA irradiation. This resulted in improvement of UDVA and CDVA by −0.23 and −0.11, respectively, and mean Kmax reduction of 1.17D. Moreover, Stojanovic et al28 used riboflavin solution without dextran, together with BAK, gentamicin and proparacaine as well as a polyvinyl alcohol sponge to increase epithelial permeability and riboflavin uptake. This led to significant improvements in UDVA and CDVA, as well as reduced mean myopic MSE by 0.74D and reduced mean Kmax by 0.57D.

This review has important strengths and limitations. It is, to our knowledge, the first systematic review of the safety and efficacy of CXL for the treatment of progressing keratoconus, and as it included both ER and TE treatments, helps summarise the published evidence to date. Our analysis had clear inclusion and exclusion criteria to collect specific and relevant data, and included trials published in languages other than English to ensure no relevant data were omitted. The analysis, however, is limited by the quality of reporting of study outcomes, which was inconsistent in many cases as is evident from the many gaps in our efficacy and safety tables (Tables 1, 2, 4, and 5). Furthermore, our work is completely reliant on the publication of conducted studies, and is therefore subjected to publication bias. Our review also makes significant assumptions where it compares studies with unequal follow-up durations, particularly between ER and TE CXL. However to address this, we identified three important parameters (change in CDVA, refractive astigmatism, and Kmax) and compared 12-month data of ER and TE CXL studies where these data were available (Table 6). The paucity of TE studies included also has significant potential for type II (beta) error, which is to overlook significant treatment effect due to a small sample size. Moreover, it was hard to analyse the data between the two groups categorising it as paediatric and adult. As evident from Tables 1 and 2, there were few studies where paediatric patients were involved and few of these had a heterogeneous age group consisting of paediatric and adult population.

A systematic review-based upon meta-analysis using the Cochrane Collaboration’s trusted and well-established methods would provide the ideal way to compare the efficacy and safety of ER and TE CXL. Unfortunately, this was not possible due to the paucity of RCTs comparing the two treatment modalities head-to-head. Furthermore, although two RCTs exist that compare ER CXL with observation alone,8, 35 there are currently no RCTs comparing TE CXL with observation. As a result of this, as well as the different study sizes and follow-up intervals, we employed medians and ranges to give the best statistically sound method of comparing treatment effects.

Our systematic review has important implications for research. We have highlighted the paucity of high-quality TE studies in the literature, as well as their relatively short follow-up. Techniques for TE CXL clearly need further modification and standardisation comparable to ER studies. We have demonstrated the inconsistency between CXL trials in reporting of important measures of efficacy and safety, and we recommend that all future trials report findings in terms of the headings used to assess the efficacy and safety in this review to aid standardisation. The variations in the treatment protocols of TE studies are envisaged to complicate the safety and efficacy data further as many researchers have now started questioning the standard Dresden protocol for TE and ER CXL and are employing permutation and combinations of settings to attain equivalent outcomes (an example of this is the recent introduction of rapid crosslinking protocols97, 98).

In summary, our study has significant implications for current clinical practice. It has shown that although further research is required in the field of ER and TE CXL to assess the efficacy and safety. A multitude of studies already testify to the efficacy of ER CXL in halting the progression of keratoconus, and recommending it as the standard of care. Additionally, our work has systematically brought together safety data on the treatment, such that patients may be counselled about complication rates to make informed decisions about their care.