Protocol for a systematic review, meta-analysis, and trial sequential analysis of clinical outcomes following accelerated versus conventional corneal collagen cross-linking for corneal ectasia

We will conduct a detailed search of the following electronic databases: MEDLINE, Embase, Web of Science, Cochrane CENTRAL, Cochrane DSR, CINAHL, OpenGrey, metaRegister of Controlled Trials, LILACS, ClinicalTrials.​gov, WHO Clinical Trials Database, WangFangData, CQVIP, INSPEC, COMPENDEX, and CNKI, from inception through February 1, 2019. We will use medical subject heading (MeSH) terms and keywords related to corneal ectasia and the cross-linking approach (accelerated or conventional) as well as clinical outcomes. Search strategies will be designed by a multidisciplinary research team composed of clinicians, researchers, and academic librarians with expertise in conducting systematic reviews. The proposed search strategy for the MEDLINE database is provided in Table 1. Our electronic search will be supplemented by manually screening the references of eligible articles, reviewing the proceedings of relevant meetings, and contacting clinical experts in the field. We will conduct our search without any restrictions on publication type, language, or time.

Search results will be evaluated independently by two reviewers (SN, CS) against predefined eligibility criteria to identify relevant studies (Fig. 1). Results from all searched databases will be exported as .RIS, .XML, or .CIW files containing the complete reference, and EndNote X9 (Clarivate Analytics, Philadelphia, USA) software will be used for reference management. Reviewers will screen titles and abstracts against inclusion criteria, and full articles will be retrieved for all references that meet these criteria, or where there is any ambiguity. In cases of ambiguity, the complete report of the specific study will be screened independently by both reviewers in order to reach a judgement on inclusion. Disagreements between reviewers will be resolved by deliberation and consensus, and, if needed, including an impartial third reviewer (AK), or contacting the trial authors.

Our inclusion criteria will be:

Data from included studies will be collected independently by two reviewers (SN, CS) and confirmed for accuracy by a third reviewer (AK). Prior to data extraction, the complete reports of all studies meeting our inclusion criteria will be collated, and reviewers will develop and pilot data extraction forms. For studies not published in English, the complete article will be translated into English and a clinical expert fluent in the original language of the study will be consulted. Discrepancies in data extraction will be resolved collaboratively by discussion amongst the two primary reviewers (SN, CS), conferring with an independent third reviewer (AK), or contacting the original trial authors. Where data in included studies is incomplete or ambiguous, we will contact study authors for further information and clarification.

We will extract data from eligible studies using forms with fields for the following: study first author, year of publication, journal of publication, language, study design, included centres, included countries, number of patients, number of males, number of females, recruitment period, eligibility criteria, method of randomisation, indication for CXL, number of patients in accelerated and conventional groups, age of patients in accelerated and conventional groups, procedure for follow-up, number of patients with disease progression 12 months following CXL in accelerated and conventional groups, K_max before CXL in accelerated and conventional groups, K_max 12 months after CXL in accelerated and conventional groups, K_max at longest follow-up after CXL in accelerated and conventional groups, UDVA before CXL in accelerated and conventional groups, UDVA 12 months after CXL in accelerated and conventional groups, CDVA before CXL in accelerated and conventional groups, CDVA 12 months after CXL in accelerated and conventional groups, central corneal thickness before CXL in accelerated and conventional groups, central corneal thickness 12 months after CXL in accelerated and conventional groups, endothelial cell density before CXL in accelerated and conventional groups, endothelial cell density 12 months after CXL in accelerated and conventional groups, intraocular pressure before CXL in accelerated and conventional groups, intraocular pressure 12 months after CXL in accelerated and conventional groups, corneal power before CXL in accelerated and conventional groups, corneal power 12 months after CXL in accelerated and conventional groups, spherical equivalent before CXL in accelerated and conventional groups, spherical equivalent 12 months after CXL in accelerated and conventional groups, stromal demarcation line depth after CXL in accelerated and conventional groups, method for epithelium removal, method for transepithelial riboflavin application, postoperative pain following accelerated and conventional CXL, time to best CDVA following accelerated and conventional CXL, time to best UDVA following accelerated and conventional CXL, power of UVA light in accelerated group, UVA light exposure time in accelerated group, incidence of significant complications (e.g., corneal melt, persistent epithelial defects, scarring, and persistent stromal haze) after accelerated and conventional CXL, and protocol preference as reported by authors.

All included trials will be assessed using the Cochrane Collaboration’s Risk of Bias Assessment Tool by two independent authors (SN, CS) [20]. Studies will be assessed to determine the risk of selection, performance, detection, attrition, reporting, and other biases. Discrepancies in quality assessment will be resolved collegially by deliberation and consensus, and consultation with an impartial third reviewer (AK). If there is insufficient information available to make a judgement on risk within an individual domain, we will rank that domain as ‘unclear’ and the original trial authors will be contacted for further information. Studies with one or more domains assessed to be ‘high risk’ will be categorised as having an overall high risk of bias.

Our primary outcome will be the change in K_max (in dioptres, D) over 12 months. Additional outcomes shall be the following: incidence of disease progression (defined as an increase of the K_max by ≥ 1.0 dioptres (D) over 12 months) after treatment, as well as change in: mean UDVA (in logMAR) at 12 months following treatment, mean CDVA (in logMAR) at 12 months following treatment, mean endothelial cell density (in cells/mm²) at 12 months following treatment, mean K_max (D) at longest follow-up after treatment, mean stromal demarcation line depth (in μm) after treatment, mean central corneal thickness (in μm) at 12 months following treatment, mean spherical equivalent (D) at 12 months following treatment, mean intraocular pressure (in mmHg) at 12 months following treatment, mean corneal power (D) at 12 months following treatment, and the mean K_max (D) at 12 months after treatment. We will also examine the incidence of serious adverse events (e.g., corneal melt, persistent epithelial defects, scarring, and persistent stromal haze).

Analyses will be conducted by the intention-to-treat principle using pooled study-level data. For dichotomous outcomes, such as incidence of ectatic disease progression and incidence of serious adverse events, we will summarise our analyses by calculating the relative risk (RR) with corresponding 95% confidence interval (CI). For continuous outcomes, such as change in K_max, mean K_max, mean UDVA, mean CDVA, mean central corneal thickness, mean stromal demarcation line depth, mean endothelial cell density, mean intraocular pressure, mean corneal power, and mean spherical equivalent, at 12 months following treatment, we will compute the weighted mean difference (MD) with 95% CI. Moreover, we will determine pooled estimates of all incidences across studies for the accelerated group, and then separately for the conventional group. The DerSimonian and Laird random effects model will be used to conduct our meta-analysis, and weights will be calculated using the inverse variance method [21]. The threshold for type I error for statistical significance shall be α = 0.05. Between-study heterogeneity will be assessed using Cochran’s Q test and quantified by the I² statistic, with I² values in excess of 25%, 50%, and 75%, graded as low, moderate, and high heterogeneity, respectively [22]. Publication bias will be examined qualitatively by visual inspection of funnel plot symmetry and quantified by Begg and Mazumdar’s [23] and Egger’s tests [24].

We will undertake prespecified subgroup analyses to investigate whether covariates exist and to examine heterogeneity in our primary outcome. Analyses will be performed for subgroups stratified by patient age and sex, CXL technique (transepithelial versus epithelium-off), severity of disease at time of treatment, and study risk of bias (low versus high). In addition, we will conduct a sensitivity analysis by performing trial sequential analysis (TSA) [25, 26]. TSA adjusts for the risk of type I error and constructs monitoring boundaries to determine whether an intervention is beneficial, harmful, or futile in comparison to control. We will perform TSA with an overall power of 80% and type I error risk of 5%.

The quality of available evidence will be summarised using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach [27]. Evidence will be assessed for the domains of risk of bias, consistency, precision, reporting bias, directness, and other bias. The evidence for each outcome will be ranked as being of high, moderate, low, or very low quality in accordance with criteria set by the GRADE working group.

All statistical analyses will be conducted using Comprehensive Meta-analysis v.3.3.070 (Biostat, Englewood, New Jersey, USA), and TSA will be performed using Trial Sequential Analysis v0.9.5.10 Beta (Copenhagen Trial Unit, Copenhagen, Denmark).