Types of studies

We intended to include all randomized controlled trials (RCTs) that evaluated the effectiveness of VASTs or specialist inserters for their impact on clinical outcomes. We excluded cluster RCTs, where the cluster represented randomization at the ward or hospital level.

We intended to include controlled clinical trials if we did not find any RCTs. Controlled clinical trials refer to quasi‐randomized studies where, although the trial involves testing an intervention with a control, with concurrent enrolment and follow‐up of test and control‐treated groups, the method of allocation is not considered truly random.

Types of participants

We included hospitalized or community participants requiring vascular access. Age was not an excluding factor.

Types of interventions

Intravenous/vascular access teams or specialist inserters (as in VAST) providing insertion or maintenance (or both) of VADs.

Types of outcome measures

Electronic searches

We identified RCTs through literature searching with systematic and sensitive search strategies as outlined in Chapter 6 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We applied no restrictions to language or publication status.

We searched the following databases for relevant trials.

1. Cochrane Central Register of Controlled Trials (CENTRAL; 2018, Issue 1) (Appendix 1)

2. MEDLINE (Ovid SP, 1966 to 7 February 2018) (Appendix 2)

3. Embase (Ovid SP, 1988 to 7 February 2018) (Appendix 3)

4. CINAHL (Cumulative Index to Nursing and Allied Health Literature) (EBSCO, 1982 to 7 February 2018) (Appendix 4)

5. ISI Web of Science (1990 to 7 February 2018) (Appendix 5)

We developed a subject‐specific search strategy in MEDLINE and used that as the basis for the search strategies in the other databases listed. Where appropriate, we expanded the search strategy with search terms for identifying RCTs.

We scanned the following trials registries for ongoing and unpublished trials (7 February 2018).

1. World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) (www.who.int/trialsearch)

2. ClinicalTrials.gov (www.clinicaltrials.gov)

3. Australian and New Zealand Clinical Trials Register (www.anzctr.org.au)

4. Current Controlled Trials (www.controlled‐trials.com/mrct)

5. HKU Clinical Trials Registry (www.hkclinicaltrials.com)

6. Clinical Trials Registry ‐ India (ctri.nic.in/Clinicaltrials/login.php)

7. UK Clinical Trials Gateway (www.controlled‐trials.com/ukctr/)

We developed the search strategy in consultation with the Information Specialist.

Searching other resources

We scanned the reference lists and citations of included trials and any relevant systematic reviews identified for further references to additional trials. We contacted trial authors for additional information when necessary.

Selection of studies

We merged all the electronically retrieved studies from individual databases into Covidence. Following this we removed any duplicate references found employing the filter using icons imports; manage imports then check duplicates. Two review authors (PC, NH) independently assessed titles and abstracts of retrieved studies for relevance. We retrieved full versions of all potentially eligible studies, which the same two review authors independently checked for eligibility. Any discrepancies between review authors were resolved either through mutual discussion, or on two occasions by consulting other review authors (CR, MC) to arbitrate.

Data extraction and management

Two review authors (PC, NH) intended to extract data from each study using our data extraction sheet (Appendix 6). The data extraction sheet was developed in conjunction with the Cochrane Anaesthesia, Critical and Emergency Care Group, and we had planned to pilot test the first two identified studies. Both PC and NH intended to independently extract data and then perform cross‐checking for accuracy and agreement. We intended to include only studies reported in one publication. If we located studies that had been published in duplicate, we intended to maximally extract data from all relevant publications but not to duplicate data in analyses. If we believed any data were missing from the papers, we planned to contact study authors to retrieve this missing information. See Appendix 6 for more information regarding the data that we intended to extract. As we did not locate any published studies, we could not perform data extraction.

Assessment of risk of bias in included studies

Two review authors (PC, NH), intended to independently assess the risk of bias for each of the studies using the 'Risk of bias' assessment tool described in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). The tool addresses six specific domains: sequence generation, allocation and concealment, blinding, incomplete outcome data, selective outcome reporting, and other potential sources of bias (Higgins 2011). We planned to express judgements as 'low risk', 'high risk', or 'unclear risk' of bias. We intended to resolve any disagreements by discussion; if we could not reach consensus, a third review author (CR) would arbitrate. We intended to conduct sensitivity analyses to determine whether excluding studies at high risk of bias affected the results of a planned meta‐analysis. We intended to report the 'Risk of bias' table as part of the 'Characteristics of included studies' table and present a 'Risk of bias' summary figure detailing all of the judgements made for all studies included in the review. As we did not locate any published studies, we were unable to assess the risk of bias.

Measures of treatment effect

We intended to calculate dichotomous outcomes using risk ratio (RR) with 95% confidence intervals (CI). We intended to calculate continuous outcomes using the mean difference (MD) with 95% CI. If the results were expressed as rate data (e.g. incidence rates), we planned to calculate a rate ratio. We intended to extract data from time‐to‐event (i.e. dwell time) studies, if the estimates were presented as log‐rank or Cox proportional models. We would not analyse time‐to‐event data that were incorrectly presented as continuous data. As we did not locate any published studies, we could not measure treatment effect.

Unit of analysis issues

Ideally, a study would be designed with participant‐level randomization and analysis, and only one device per participant (adjustment for clustering not necessary in this case). However, we expected to find studies that reported on multiple devices per participant, randomized or analysed at device level, or both, and unadjusted for clustering. We expected to find the following differences.

1. Number of devices and observed per participant: one or more than one (e.g. after the removal of the initial device the consecutive device(s) was (were) also observed).

2. Randomization methods: device level, participant level, or ward (or similar) level.

3. Unit of analysis: device level, participant level, or ward (or similar) level.

4. Analysis methods: adjusted for clustering or not adjusted for clustering.

In such cases, we intended to attempt to obtain the following from the study authors: participant‐level data or results; data or results for one device per participant; or device‐level data and perform multilevel regression to calculate the adjusted effect. We intended to combine the adjusted results in the meta‐analysis with those of participant‐level trials (using the generic inverse method) and perform sensitivity analyses (Higgins 2011). If we were unsuccessful in obtaining the additional necessary data, then we planned to exclude the study from the meta‐analysis. We excluded cluster RCTs where the cluster represents randomization at the ward or hospital level. We did not locate any published studies to assess unit of analysis.

Dealing with missing data

Whenever possible, we intended to contact the original investigators to request missing data and methodological details. If we considered that data were missing at random, we intended to analyse the available information. If we considered that data were not missing at random, we intended to analyse the available information and assess the potential impact of the missing data on the findings of the review in the Discussion section. However, we did not locate any published studies.

Assessment of heterogeneity

We intended to consider clinical, methodological, and statistical heterogeneity. We planned to undertake an assessment of comparability of the studies prior to meta‐analysis. We planned to assess heterogeneity of selected studies visually and by using the Chi² test with significance level set at P value less than 0.10. This assesses whether observed differences in results are compatible with chance alone. In addition, we planned to investigate the degree of heterogeneity by calculating the I² statistic (an equation combining the Chi² statistic relative to its degree of freedom). This describes the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error (chance). If studies were sufficiently similar to consider pooling, we planned to use a fixed‐effect model for low‐to‐moderate levels of heterogeneity (I² = 0% to 50%). Where appropriate, in the absence of clinical heterogeneity and in the presence of statistical heterogeneity (I² greater than 50%), we planned to use a random‐effects model. However, studies where heterogeneity exceeded 75% were not going to be pooled (Higgins 2011). As we did not locate any published studies, we did not need to assess heterogeneity.

Assessment of reporting biases

We intended to use visual asymmetry on funnel plots to assess reporting biases if at least 10 studies were available for a meta‐analysis (Sterne 2011). We intended to report each outcome separately. We planned to undertake an observation of small‐study effects if required.

We did not locate any published studies to assess reporting biases.

Data synthesis

We intended to enter into Review Manager 5 all trials included in the systematic review and insert and analyse quantitative data (RevMan 2014). The decision to pool data in a meta‐analysis depended upon the availability of outcome data and assessment of between‐trial heterogeneity. If we identified evidence of substantial heterogeneity (i.e. greater than 50%), we planned to explore potential causes and use a random‐effects model, and a fixed‐effect model to explore any differences between these two estimates. As no studies were included in the review, and synthesis was inappropriate, we have presented a structured narrative review in the Results of the search and Discussion sections.

Subgroup analysis and investigation of heterogeneity

If sufficient data were available, we planned to perform the following subgroup analyses.

1. Adult VAST versus paediatric VAST.

2. Device type: peripheral versus central.

3. Central device type: peripherally inserted central catheter versus CVC versus tunnelled VAD versus totally implanted.

4. VAST model: insertion only versus insertion and follow‐up care services.

5. VAST team versus individual specialists.

As we did not locate any studies, we did not undertake any subgroup analysis and investigation of heterogeneity.

Sensitivity analysis

We planned to initially perform a sensitivity analysis by excluding studies at high risk of bias. We intended to only include studies that were assessed as having a low risk of bias for the estimates of treatment effect in all key domains, namely adequate generation of the randomization sequence, adequate allocation concealment, and blinding of outcome assessor. We also planned to perform sensitivity analysis on:

1. size of study (fewer than 100 participants);

2. missing data (worst‐case/best‐case scenario).