Eligibility criteria

The eligibility criteria are prespecified according to the P refer to the patient or population. The I refer to the intervention/exposure. The C refer to the comparison/control. The O refer to the outcome. The S refer to the study design. (PICO(S)) criteria.

Types of participants

All patients who underwent CVC insertion will be included. We will not impose an age restriction. Studies will be included if they use antimicrobial lock solutions for CRBSI. The participants must be using a tunnelled or non-tunnelled CVC as venous access for ≥48 hours. Patients with artery catheter or pacing wires were excluded regardless of the cause and duration of catheter use.

Types of interventions

RCTs of antimicrobial lock solutions used to prevent CRBSI will be included, regardless of whether the antimicrobials were tested between themselves (head-to-head) or against placebo/control intervention such as heparin. For antimicrobials, antibiotics, citrate, taurolidine and alcohol will be included regardless of their concentration. The anticoagulants (eg, heparin, citrate or EDTA (minocycline-EDTA lock solution)) could be given in all antimicrobial lock solutions. The comparator can be placebo, heparin or a different catheter lock solution. Comparing interventions of the same lock solution with different concentrations is beyond the scope of this study. For example, if the treatment group was administered 1000 IU/mL heparin, whereas the control group was administered 5000 IU/mL heparin, this study will be excluded.

We included all RCTs that met the above-mentioned inclusion criteria and included a control group (indirect evidence) or at least two interventions (direct evidence).

Types of outcome

Included studies must report CRBSI as one of the end points (defined below).

Types of studies

Eligible studies are RCTs assessing the effect of lock solutions in the prevention of CRBSI.

Search strategy

We will search PubMed, Embase, the Cochrane Library and Web of Science from inception until January 2019. We will create the search strategy to combine relevant Medical Subject Headings and keywords with synonyms and names of generic and brand names of all catheter lock solutions. The search strategy will be developed by JT, who has >10 years of experience as an information specialist. We have provided a draft of a full electronic search strategy for PubMed in online supplementary file named ‘Search Strategy’. We will also track the references in the included articles and relevant systematic reviews/meta-analyses to identify additional relevant studies. Meanwhile, we will perform weekly research in the above-mentioned electronic databases to identify any newly published articles until the submission of the present study. Non-English studies and duplicate publications of original research will be excluded. We will also exclude ongoing RCTs without extractable data.

Study selection

Literature search records will be imported into EndNoteX8 literature management software (Thomson Reuters, New York, New York, USA). Two researchers (FD and JT) will independently review the title and abstract of the studies and exclude those that clearly did not meet the inclusion criteria. Then, the remaining studies will be identified by reviewing the full text according to the prespecified inclusion criteria. Any disagreements will be resolved by discussion or through a third reviewer (HL).

Data extraction

Data will be independently extracted by two authors (FD and JT) from each included RCT according to a predesigned Excel spreadsheet created in Microsoft Excel 2016 (Microsoft, Redmond, Washington, USA). The study data will include study and publication details, participants and intervention characteristics as well as outcome. If disagreement persists, a third reviewer (HL) will make the decision.

Risk of bias assessment

The methodological risk of bias of included RCTs will be assessed according to the Cochrane Collaboration Risk of Bias Tool24 including the method of random sequence generation (selection bias), allocation concealment, blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), incomplete outcome data (attrition bias), selective reporting (reporting bias) and other biases. These items will be evaluated as having high, low or unclear risk of bias. Any disagreements will be resolved through a discussion and reaching a consensus with the third reviewer (HL).

Data synthesis and statistical methods

We planned this NMA to assess which lock solution for CRBSI is most effective. We will construct the networks for the primary and secondary outcomes separately.

Bayesian NMA

Bayesian NMA will be performed in WinBUGS 1.4.3 (V.1.4, Medical Research Council, UK, and Imperial College of Science, Technology and Medicine, University of Cambridge, UK) by using the Markov chain Monte Carlo method, and the other analyses will be performed and presented through Stata V.15.0 using the mvmeta command. The following two different chains will be used:

• chain 1: treatment effect: (d(k)=0); SD=1; mu(i)=0;

• chain 2: treatment effect: (d(k)=−1); SD=4; mu(i)=−3, where d(k)=treatment effect of experimental intervention ‘k’ compared with reference and mu(i)=treatment effect of the experimental intervention compared with control in the trial ‘i’. Model fit will be determined based on the deviance information criteria (DIC) for each outcome measure. It is best to choose the lowest model of DIC (differences>3 are considered meaningful).25 Two Markov chains will run simultaneously with different arbitrarily chosen initial values. We will generate the simulations for each chain if the convergence will be found to be adequate. Then, these simulations will be discarded as ‘burn-in’. To ensure convergence, the model convergence will be assessed by trace plots.26

Methods for indirect and direct comparisons

We will conduct the separate pairwise meta-analyses using the random-effects model to compare all the interventions with available direct evidence. Dichotomous data will be determined by using OR with 95% CI. Continuous outcomes will be analysed using weighted mean differences or standardised mean differences if different measurement scales are used. To present indirect comparisons of the catheter lock solutions, a network diagram will be drawn. Trials will be excluded if they are not connected by any treatments. The common comparator will be heparin as it is the current standard of care.27 In this network plot, nodes represent different interventions and edges represent the head-to-head comparisons between interventions. The size of the node reflects the sample size of the intervention, and the thickness of the edge reflects the number of included trials.

Dealing with missing data

We will attempt to obtain missing data from authors of the included RCTs by mail. We will explore the potential impact of the missing data by performing a sensitivity analyses and record the number of patients ‘loss to follow-up’. We will also extract data from studies that report outcomes using the intention-to-treat analysis.

Treatment ranking

To estimate the percentage of efficacy of each catheter lock solution for being at each possible rank, the surface under the cumulative ranking curve (SUCRA) values, which are expressed as percentages will be used. The SUCRA value is 100% for the best treatment and 0% for the worst treatment. The higher the SUCRA score, the more efficacious the treatment is expected to be.28 29

Assessment of heterogeneity

Clinical and methodological heterogeneity will be assessed by carefully examining the study population characteristics, interventions and outcomes of the included trials and comparing fit of the fixed-effects model and random-effects model. The statistical heterogeneity within each pairwise comparison will be measured by using the I² index.30 Substantial heterogeneity will be considered where I² is >50%. If considerable heterogeneity is found, network meta-regression or subgroup analysis will be used to explore possible sources of heterogeneity.

Subgroup and sensitivity analyses

Subgroup analyses and network meta-regression were planned to perform according to the type of catheter, catheter insertion site, catheter days/lock time, age and study quality. We will perform sensitivity analyses of the outcomes to determine the sensitivity of results to changes in the original assumptions by excluding trials with a high risk of bias in seven items, and excluding trials where the criterion of CRBSI diagnosis does not meet the above-mentioned definition.

Assessment of inconsistency

Inconsistency means the presence of disagreement between the different sources of evidence.31 We will evaluate the inconsistency between direct and indirect evidence when a loop is formed between treatments. The node-split method will be used to explore where the inconsistency is found.32 We will attempt to identify inconsistencies jointly from all possible sources in the network using the design-by-treatment interaction model and I² index, when possible.33 34

Publication bias

When each direct treatment comparison and outcome contains a sufficient number of studies (at least 10 studies), we will assess the possibility of publication bias by visually examining the asymmetry of funnel plots and Egger’s regression test.24 We will use a comparison-adjusted funnel plot to explore the potential small-study effects in the network and use contour-enhanced funnel plots to examine whether the funnel plot asymmetry is caused by a publication bias.35 If the publication bias is found, we will fit a selection model to obtain relative effects ‘adjusted’ for the impact of publication bias.36

Quality of the evidence

We will propose the following Grade of Recommendations Assessment, Development and Evaluation (GRADE) four-step approach37 to assess the quality of treatment effect estimates from NMA: (1) present direct and indirect treatment estimates for each comparison of the evidence network; (2) rate the quality of each direct and indirect effect estimate; (3) present the NMA estimate for each comparison of the evidence network and (4) rate the quality of each NMA effect estimate. According to the GRADE group, we will classify the quality of evidence into the following four levels: high quality, moderate quality, low quality and very low quality.

Potential limitations of the planned work

Some potential limitations of this NMA are predictable. First, excluding non-English studies may cause a publication bias. Second, we will only include RCT publications to support our intention to include only high-quality evidence, which indicate that there is a possibility that the study population will not be representative of the overall population. Besides, we only included patients with CRBSI defined according to the Infectious Diseases Society of America recommendations, which will increase the risk of bias and heterogeneity. Finally, if the number of included studies is small, our ability to explore heterogeneity, conduct a meta regression analysis and even perform NMA is limited. Meanwhile, if the degree of clinical heterogeneity is substantial, the internal validity will be threatened.

Ethics and dissemination

Given that the present meta-analysis is performed based on previous published studies, no ethical approval and patient safety considerations are required. This study commenced on 18 January 2019, and its expected completion date is 1 December 2019. We will disseminate the results of our network meta-analysis through an international peer-reviewed journal.