Improving difficult peripheral intravenous access requires thought, training and technology (DART3): a stepped-wedge, cluster randomised controlled trial protocol

Background Peripheral intravenous catheters (PIVCs) are the most used invasive medical device in healthcare. Yet around half of insertion attempts are unsuccessful leading to delayed medical treatments and patient discomfort of harm. Ultrasound-guided PIVC (USGPIVC) insertion is an evidence-based intervention shown to improve insertion success especially in patients with Difficult IntraVenous Access (BMC Health Serv Res 22:220, 2022), however the implementation in some healthcare settings remains suboptimal. This study aims to co-design interventions that optimise ultrasound guided PIVC insertion in patients with DIVA, implement and evaluate these initiatives and develop scale up activities. Methods A stepped-wedge cluster randomized controlled trial will be conducted in three hospitals (two adult, one paediatric) in Queensland, Australia. The intervention will be rolled out across 12 distinct clusters (four per hospital). Intervention development will be guided by Michie’s Behavior Change Wheel with the aim to increase local staff capability, opportunity, and motivation for appropriate, sustainable adoption of USGPIVC insertion. Eligible clusters include all wards or departments where > 10 PIVCs/week are typically inserted. All clusters will commence in the control (baseline) phase, then, one cluster per hospital will step up every two months, as feasible, to the implementation phase, where the intervention will be rolled out. Implementation strategies are tailored for each hospital by local investigators and advisory groups, through context assessments, staff surveys, and stakeholder interviews and informed by extensive consumer interviews and consultation. Outcome measures align with the RE-AIM framework including clinical-effectiveness outcomes (e.g., first-time PIVC insertion success for DIVA patients [primary outcome], number of insertion attempts); implementation outcomes (e.g., intervention fidelity, readiness assessment) and cost effectiveness outcomes. The Consolidated Framework for Implementation Research framework will be used to report the intervention as it was implemented; how people participated in and responded to the intervention; contextual influences and how the theory underpinning the intervention was realised and delivered at each site. A sustainability assessment will be undertaken at three- and six-months post intervention. Discussion Study findings will help define systematic solutions to implement DIVA identification and escalation tools aiming to address consumer dissatisfaction with current PIVC insertion practices. Such actionable knowledge is critical for implementation of scale-up activities. Trial registration Prospectively registered (Australian and New Zealand Clinical Trials Registry; ACTRN12621001497897). Supplementary Information The online version contains supplementary material available at 10.1186/s12913-023-09499-0.


Contributors to the statistical analysis plan
The undersigned have reviewed this plan and approve it as final. They find it to be consistent with the requirements of the protocol as it applies to their respective areas. They also find it to be compliant with ICH-E9 principles and, in particular, confirm that this analysis plan was developed in a completely blinded manner (i.e., without knowledge of the effect of the intervention(s) being assessed).

Study synopsis
DART 3 is a multi-centre stepped wedge cluster randomised trial designed to determine the extent to which first attempt peripheral intravenous catheter (PIVC) insertion success, and other associated outcomes, are improved among in Difficult Intravenous Access (DIVA), and non-DIVA patients, within hospital wards and departments using the DIVA Identification and Ultrasound Escalation Pathways, compared with hospital wards and departments offering usual care.

Study objectives
(1) To develop hospital-based interventions that support DIVA identification and escalation, adapted to the local health care context.
(2) To implement co-designed DIVA identification and escalation pathways, and evaluate their clinical effectiveness, cost effectiveness and implementation strategies.
(3) Develop recommendations for future strategic options for scale-up and optimised implementation of DIVA identification and escalation pathways in different settings.

Primary hypothesis
The primary objective is to determine the extent to which implementation of the DIVA Identification and Ultrasound Escalation Pathways impacts first attempt PIVC insertion success for DIVA patients.

H1
: Implementation of the DIVA Identification and Ultrasound Escalation Pathways will significantly increase first attempt PIVC insertion success for DIVA patients

Secondary hypotheses
The secondary objectives are to determine the extent to which implementation of the DIVA Identification and Ultrasound Escalation Pathways: impacts first attempt PIVC insertion success for all patients; is feasible; is sustainable; is cost-effective; impacts incidence of ultrasound-guided insertion among DIVA patients; impacts number of PIVC insertion attempts; impacts rates of successful PIVC placement and time-to-therapy; impacts post-insertion failure and complications; and impacts patient/carer and staff satisfaction with insertion procedure.

Exclusion criteria
Exclusion criteria are emergencies (e.g., Medical Emergency Team call) requiring intraosseous access. Exclusion areas are operating theatres, radiology, rehabilitation, or psychiatric units because they either have all expert inserters so have less insertion failure, or rarely insert PIVCs. Screening data will not be collected for excluded patients.

Primary outcome
The primary outcome is first attempt PIVC insertion success for DIVA patients. The primary outcome is defined as one needle puncture, by one inserter, to achieve PIVC insertion for DIVA patients.
Data for the primary outcome will be collected in 2-month increments during the baseline and implementation phases of the study (corresponding to each study step), and 3-month increments during the sustainability phase of the study. During each window of time (or step), data will be collected at randomly selected study wards on specific days throughout the time-period. All outcomes will be analysed collectively once data collection has been completed.

Secondary outcomes
Secondary outcome 1 is first attempt PIVC insertion success for all patients. Secondary outcome 1 is defined as one needle puncture, by one inserter, to achieve PIVC insertion for all patients.
Secondary outcome 2 is ultrasound adoption. Secondary outcome 2 is defined as the proportion of DIVA patients with ultrasound used at first, or any attempt.
Secondary outcome 3 is number of PIVC insertion attempts. Secondary outcome 3 is defined as numbers of skin punctures to attempt PIVC placement.
Secondary outcome 4 is successful procedural outcome defined as successful PIVC insertion 4 Secondary outcome 5 is time from PIVC referral to PIVC insertion (censored at 48 hours).
Secondary outcome 6 is use of alternate device.
Secondary outcome 7 is use of alternate route (e.g., oral).
Secondary outcome 8 is PIVC failure. Secondary outcome 11 is defined as a composite measure of local infection, primary bloodstream infection, occlusion, infiltration/extravasation, dislodgement (includes leaking), thrombosis and/or phlebitis.
Secondary outcome 9 is insertion/post-insertion complications. Secondary outcome 9 is defined as bruising, haematoma, nerve injury, arterial puncture, or skin injury as well as the individual components of PIVC failure (above).
Secondary outcome 10 is PIVC dwell time. Secondary outcome 10 is defined as time from PIVC insertion to PIVC removal (in hours).
Secondary outcome 11 is sustainability of first attempt insertion success after implementation of the DIVA Identification and Escalation Pathways. Sustainability is defined as first attempt insertion success at +3 or +6 months compared to full implementation (month 10).
Secondary outcome 12 is PIVC necessity. Secondary outcome 12 is defined as PIVC used for fluids or medications within 24 hours (excluding patients who require prophylactic PIVC in situ as part of their treatment e.g., status epilepticus). Secondary outcome 13 is cluster level routinely collected rates of primary BSI and S. Aureus BSI.
Secondary outcome 14 is patient/carer/parent satisfaction with insertion procedure. Secondary outcome 14 is defined according to a 0-10 numeric rating scale.
Secondary outcome 15 is patient/carer/parent pain with insertion procedure. Secondary outcome 15 is defined according to a 0-10 numeric rating scale.
Secondary outcome 16 is inserter (initial, successful, and/or abandoned inserter) satisfaction with insertion procedure. Secondary outcome 16 is defined according to a 0-10 numeric rating scale. As with the primary outcome, data on secondary outcomes will be collected in 2-month increments during the baseline and implementation phases of the study (corresponding to each study step), and 3-month increments during the sustainability phase of the study. During each window of time (or step), data will be collected at randomly selected study wards on specific days throughout the timeperiod. All outcomes will be analysed collectively once data collection has been completed.

Intervention
A stepped-wedge cluster randomised trial will be used to implement and evaluate the DIVA Identification and Ultrasound Escalation Pathways at 3 South East Queensland hospitals. Each hospital will utilise 4 wards/departments for a total of 12 clusters. Within each cluster, 20 patients undergoing PIVC insertion procedures will be recruited. Sustainability will be measured at 3 and 6 months post full implementation. In total, it is expected that 600 patients will be recruited into the control condition, and 1080 patients will be recruited into the intervention condition.
In the control steps (Figure 1.), clusters will continue usual care. This will vary by site, however, in general usual care will be landmark insertion by default; with no/minimally implemented pathway to assess level of difficulty; and ad hoc referral to ultrasound skilled practitioners, typically after failed landmark attempts. During implementation, Research Nurses (ReNs) will work with investigators, local educators, and clinical managers to deliver implementation strategies to embed the DIVA Identification and Ultrasound Escalation Pathways (Figure 2.). All staff will be trained in the use of the DIVA tool and Escalation Pathway via a mix of written, didactic, and online learning modalities. ReNs will act as 'change champions', promoting consistent use of the tool and pathway. Advanced inserters (capable of ultrasound guided PIVC insertion) will be identified in the Escalation Pathway and be available to the study wards to provide a point of escalation for DIVA patients. Inserters will be the local accredited workforce (not ReNs).

Randomisation and blinding
As displayed in Figure 1., all 12 hospital clusters (wards/departments) will start the trial in the control phase for two months, with baseline measures taken. Following this, one cluster per hospital will randomly (computer generated) step-up to implementation each month over an 8-month period, until all are fully implemented. Sustainability will be measured by proportion of first-time insertion success and proportion following the pathway, at 3 and 6 months post full implementation. At each time point, a random sample of 20 PIVC insertion procedures will be studied per cluster. This will be achieved by randomly selecting dates for data collection, rostering ReNs on those dates and collecting the first 20 PIVC insertions that occur sequentially while the ReN is available. To fully understand health system implementation, we will assign up to one in four ReN shifts to occur in the early mornings, evenings or at weekends.

Sample size
The study has been designed to have adequate statistical power to test the primary hypothesis. We expect that approximately 40% of recruited patients will be DIVAs (8 per cluster sample) and 60% non-DIVAs (12 per cluster sample). Using Hemming's method for cluster randomised controlled trials, including correction for intra class correlation (r=0.03 constant throughout study), a two-sample test of proportions with 4 steps, 3 clusters randomised at each step, and 20 patients/cluster (Hemming & Girling, 2018). For patients classified as DIVA, we assume the current percentage of first attempt insertion success of 50% (based on prior meta-analysis (van Loon, 2018) and consistent with our pilot (Jauncey-Cooke, 2018), and that there will be a 50% relative improvement to 75% of first insertion success. With 20 patients recruited per cluster we will have >90% power to detect a statistically significant difference in first attempt insertion success before and after exposure to the intervention for DIVA patients (alpha=0.05).
When considering the effect of the intervention on all recruited patients, we assume the current first insertion success percentage in non-DIVA patients is 70%, and that this will rise to 80% postintervention. This is equivalent to an overall change from 62% first attempt insertion success to 78% first attempt insertion success. With 240 patients at each of the 5 data collection periods (baseline plus 4 steps), we have >90% power to detect a statistically significant difference between pre-and post-implementation of pathways (alpha=0.05).
To measure sustainability of the primary endpoint we will continue to assess 240 patients at each sustainability time point (3 and 6 months). We will have data on 168 DIVA patients receiving the pathway at steps 3 and 4, and 192 DIVA patients at the 3-and 6-month follow-up periods combined. This will allow the identification of a drop in first attempt insertion success from 75% to <60% postintervention with 80% power (alpha=0.05) (van Loon, 2018; Jauncey-Cooke 2018). When considering all patients regardless of DIVA status we have 420 at steps 3 and 4 and 480 at follow-up, giving 80% power to identify an overall reduction in first time insertion success from 78% to <70% postintervention (alpha=0.05).

Statistical software
All statistical analyses will be performed using Stata v13 (StataCorp. Stata Statistical Software. College Station, TX: StataCorp LP).

Reporting conventions
P-values will be reported to 2 decimal places to p=0.01, then 3 decimal places to p=0.001. All smaller values will be reported as p<0.001.

Data cleaning approach
Data collection and quality management will be completed through use of the REDCap database.

Data definitions/derivations including, but not limited to, visit windows definitions
Data collection will be classified by study steps, which will be defined according to calendar month.

Confidence intervals and P-values
For each outcome variable, statistical significance will be assessed at the 0.05 level and 95% confidence intervals will be reported.

Interim analyses
No interim analyses will be carried out for this study.

Multiplicity adjustment
No formal adjustments for multiplicity are planned, however significant test results will be interpreted considering the multiple comparisons made. To express uncertainty about the treatment effect, 95% confidence intervals will be used.

Data sets to be analysed: Adherence and protocol deviations
The primary analyses will be 'intention-to-treat'.
Adherence to the intervention will be defined as the successful delivery of implementation strategies by ReNs, investigators, local educators, and clinical managers to embed the DIVA Identification and Ultrasound Escalation Pathways. Adherence to the intervention will be assessed by the extent to which application of the DIVA Escalation Pathway reflects the study protocol.
Protocol deviations will be defined as inclusion of ineligible patients, recruitment outside designated timeframe, monitoring is missed, or any other deviations from the study protocol.
All protocol deviations will be summarised by type, including a description of the deviation and the corrective action taken.

Subject disposition
Details on recruitment are outlined in the participant flowchart ( Figure 6).
Consent will not be sought due to the observational and quality improvement nature of the study, and routine collection of these data in the provision of patient care. As such, participants will not be eligible to withdraw from the intervention. Loss to follow-up is considered unlikely and not expected to significantly impact data collection.
Should loss to follow-up occur prior to completion of the DIVA Identification and Escalation Pathways a new patient will be recruited in replacement ensuring potential for complete data collection for H1-8. In the unlikely event that loss to follow-up should occur after completion of the Escalation Pathway but prior to device removal, data collection for H9-10 may be impacted.
Any events of loss to follow-up will be summarised by reason, alongside any resulting study amendments.
DART Step 3 Assessed for eligibility (n=) Received intervention (n=no of clusters, average cluster size, variance of cluster sizes) Did not receive intervention, give reasons (n=no of clusters, average cluster size, variance of cluster size) Step 4 Assessed for eligibility (

Patient characteristics and baseline comparisons
A descriptive comparison of the patients observed under each condition will be conducted on the baseline characteristics presented in the following table.

Baseline characteristic Categories
Age -years -

Compliance to study intervention(s)
Compliance to the study invention(s) includes: i) The extent to which the intervention is implemented as intended (implementation fidelity), over time and across different sites. ii) How staff understand and respond to the intervention, over time and across different sites. iii) Context over time and across different sites and factors (including managerial, economic, organisational, and work level) that affect implementation.
Variables which will be used to define compliance during implementation include: 3.8 Analysis of the primary outcome

Primary outcome
The secondary outcome of interest is first attempt PIVC insertion success among DIVA patients. Patients identified as either 'DIVA' or 'potential DIVA' in the DIVA identification and Escalation Pathways will be included in the subgroup analysis. The measurement to be used to summarise data is frequency [percentage]. To derive the outcome, calculations will be used to determine the cumulative incidence in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for the primary outcome will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of first attempt PIVC insertion success among DIVA patients will use mixed-effects logistic regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-totreat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate presented will be odds ratio with 95% confidence interval.
Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Multicollinearity will be explored through Stata's 'collin' command. Linearity between the log odds of the dependent variable and the independent variables will be tested through the Box-Tidwell test. If assumptions for the preferred analysis are not met nonparametric binary logistic regression model using the local likelihood logit estimation method will be used.
Missing data will be reported when evident. <5% missing data is expected for the primary outcome as additional observations will be recruited when an observation is found to have missing data related to the primary outcome.
3.9 Analysis of secondary outcomes

Secondary outcome 1
The secondary outcome of interest is first attempt PIVC insertion success among all patients. The measurement to be used is cumulative incidence [percentage]. To derive the outcome, calculations will be used to determine the cumulative incidence in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 1 will be collection of the last participant at sustainability timepoint 2. The individual will be the unit of analysis. Between group comparison of first attempt PIVC insertion success among all patients will use mixed-effects logistic regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'.
The trial hypotheses will be evaluated through a framework of superiority. The effect estimate presented will be odds ratio with 95% confidence interval.
Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Multicollinearity will be explored through Stata's 'collin' command. Linearity between the log odds of the dependent variable and the independent variables will be tested through the Box-Tidwell test. If assumptions for the preferred analysis are not met, nonparametric binary logistic regression model using the local likelihood logit estimation method will be used Missing data will be reported when evident. <10% is expected for secondary outcome 1. If missing data is ≥20% imputation will be used.

Secondary outcome 2
The outcome of interest is proportion of PIVC insertions having completed the DIVA Identification and Ultrasound Escalation Pathways. The measurement to be used is cumulative incidence [percentage].
To derive the outcome, calculations will be used to determine the cumulative incidence in the intervention condition. The endpoint for secondary outcome 2 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Proportion of PIVC insertions having completed the DIVA Identification and Ultrasound Escalation Pathways will be compared to the 80% feasibility threshold. The analysis will be 'per protocol'. The trial hypotheses will be evaluated through a framework of noninferiority.
Missing data will be reported when evident. <10% is expected for secondary outcome 2. If missing data is ≥20% imputation will be used.

Secondary outcome 3
The secondary outcome of interest is intervention sustainability at 3 and 6 months. The measurement to be used is cumulative incidence [percentage]. To derive the outcome, calculations will be used to determine the cumulative incidence in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 3 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Follow-up to sustainability comparison of first attempt PIVC insertion success among DIVA patients and all patients will use mixed-effects logistic regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of non-inferiority. The effect estimate presented will be odds ratio with 95% confidence interval.
Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Multicollinearity will be explored through Stata's 'collin' command. Linearity between the log odds of the dependent variable and the independent variables will be tested through the Box-Tidwell test. If assumptions for the preferred analysis are not met, nonparametric binary logistic regression model using the local likelihood logit estimation method will be used Missing data will be reported when evident. <10% is expected for secondary outcome 3. If missing data is ≥20% imputation will be used.

Secondary outcome 4
The secondary outcome of interest is proportion of DIVA patients with ultrasound used at first or any attempt. The measurement to be used is cumulative incidence [percentage]. To derive the outcome, calculations will be used to determine the cumulative incidence in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 4 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of ultrasound use at first, or any attempt among DIVA patients will use mixed-effects logistic regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-totreat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate presented will be odds ratio with 95% confidence interval.
Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Multicollinearity will be explored through Stata's 'collin' command. Linearity between the log odds of the dependent variable and the independent variables will be tested through the Box-Tidwell test. If assumptions for the preferred analysis are not met, nonparametric binary logistic regression model using the local likelihood logit estimation method will be used.
Missing data will be reported when evident. <10% is expected for secondary outcome 4. If missing data is ≥20% imputation will be used.

Secondary outcome 5
The secondary outcome of interest is number of PIVC insertion attempts. The measurement to be used is median count [IQR]. To derive the outcome, calculations will be used to determine the difference between control and intervention conditions. The endpoint for secondary outcome 5 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of number of PIVC insertion attempts will use mixed-effects linear regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase (control/implementation) and time. The primary analysis will be 'intention-totreat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate presented will be mean difference with 95% confidence interval.
Linear relationships will be explored through scatter plots and Q-Q-Plots. Normality will be explored through histograms and Shapiro-Wilk tests. Homoscedasticity will be explored through a scatterplot of residuals versus predicted values. Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Errors in variables will be explored through Delgado and Gonzalez Manteiga's 'dgmtest' test in Stata. Multicollinearity will be explored through Stata's 'collin' command. If assumptions for the preferred analysis are not met, median regression will be substituted.
Missing data will be reported when evident. <10% is expected for secondary outcome 5. If missing data is ≥20% imputation will be used.

Secondary outcome 6
The secondary outcome of interest is rate of successful PIVC placement. The measurement to be used is cumulative incidence [percentage]. To derive the outcome, calculations will be used to determine the cumulative incidence in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 6 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of successful PIVC placement will use mixed-effects logistic regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate presented will be odds ratio with 95% confidence interval. Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Multicollinearity will be explored through Stata's 'collin' command. Linearity between the log odds of the dependent variable and the independent variables will be tested through the Box-Tidwell test. If assumptions for the preferred analysis are not met, nonparametric binary logistic regression model using the local likelihood logit estimation method will be used.
Missing data will be reported when evident <10% is expected for secondary outcome 6. If missing data is ≥20% imputation will be used.

Secondary outcome 7
The secondary outcome of interest is time-to-therapy. The measurement to be used is median elapsed time (in hours) [IQR]. To derive the outcome, calculations will be used to determine the time to event in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 7 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of time-to-therapy will use a multilevel survival model with cluster included as a random effect to account for probable nonindependence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate will be hazard ratio with 95% confidence interval.
Proportional hazards assumptions will be tested through Schoenfeld residuals. Nonlinearity will be assessed through martingale residuals. Influential observations will be examined through deviance residuals (or symmetric transformation of the martingale residuals). If assumptions for the preferred analysis are not met, Kaplan-Meier estimates and log-rank tests will be used.
Missing data will be reported when evident. <10% is expected for secondary outcome 7. If missing data is ≥20% imputation will be used.

Secondary outcome 8
The secondary outcome of interest is rate of alternate device use. The measurement to be used is cumulative incidence [percentage]. To derive the outcome, calculations will be used to determine the cumulative incidence in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 8 will be collection of the last participant at sustainability timepoint 2. The individual will be the unit of analysis. Between group comparison of alternate device use will use mixed-effects logistic regression with cluster included as a random effect to account for probable nonindependence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate presented will be odds ratio with 95% confidence interval.
Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Multicollinearity will be explored through Stata's 'collin' command. Linearity between the log odds of the dependent variable and the independent variables will be tested through the Box-Tidwell test. If assumptions for the preferred analysis are not met, nonparametric binary logistic regression model using the local likelihood logit estimation method will be used.
Missing data will be reported when evident <10% is expected for secondary outcome 8. If missing data is ≥20% imputation will be used.

Secondary outcome 9
The secondary outcome of interest is rate of alternate route use. The measurement to be used is cumulative incidence [percentage]. To derive the outcome, calculations will be used to determine the cumulative incidence in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 9 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of alternate route use will use mixed-effects logistic regression with cluster included as a random effect to account for probable nonindependence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate presented will be odds ratio with 95% confidence interval.
Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Multicollinearity will be explored through Stata's 'collin' command. Linearity between the log odds of the dependent variable and the independent variables will be tested through the Box-Tidwell test. If assumptions for the preferred analysis are not met, nonparametric binary logistic regression model using the local likelihood logit estimation method will be used.
Missing data will be reported when evident <10% is expected for secondary outcome 9. If missing data is ≥20% imputation will be used. The secondary outcome of interest is rate of unnecessary PIVCs. The measurement to be used is cumulative incidence [percentage]. To derive the outcome, calculations will be used to determine the cumulative incidence in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 10 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of rates of unnecessary PIVC will use mixed-effects logistic regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate presented will be odds ratio with 95% confidence interval.
Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Multicollinearity will be explored through Stata's 'collin' command. Linearity between the log odds of the dependent variable and the independent variables will be tested through the Box-Tidwell test. If assumptions for the preferred analysis are not met, nonparametric binary logistic regression model using the local likelihood logit estimation method will be used.
Missing data will be reported when evident <10% is expected for secondary outcome 10. If missing data is ≥20% imputation will be used.

Secondary outcome 11
The secondary outcome of interest is post-insertion PIVC failure. The measurement to be used is elapsed time (in hours) [IQR]. To derive the outcome, calculations will be used to determine the time to event in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 11 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of time-to-failure will use a multilevel survival model with cluster included as a random effect to account for probable nonindependence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate will be hazard ratio with 95% confidence interval.
Proportional hazards assumptions will be tested through Schoenfeld residuals. Nonlinearity will be assessed through martingale residuals. Influential observations will be examined through deviance residuals (or symmetric transformation of the martingale residuals). If assumptions for the preferred analysis are not met, Kaplan-Meier estimates and log-rank tests will be used. Missing data will be reported when evident. <10% is expected for secondary outcome 11. If missing data is ≥20% imputation will be used.

Secondary outcome 12
The secondary outcome of interest is PIVC dwell time. The measurement to be used is elapsed time (in hours) [IQR]. To derive the outcome, calculations will be used to determine the time to event in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 12 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of dwell time will use a multilevel survival model with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate will be hazard ratio with 95% confidence interval.
Proportional hazards assumptions will be tested through Schoenfeld residuals. Nonlinearity will be assessed through martingale residuals. Influential observations will be examined through deviance residuals (or symmetric transformation of the martingale residuals). If assumptions for the preferred analysis are not met, Kaplan-Meier estimates and log-rank tests will be used.
Missing data will be reported when evident. <10% is expected for secondary outcome 12. If missing data is ≥20% imputation will be used.

Secondary outcome 13
The secondary outcome of interest is rate of insertion/post-insertion complications. The measurement to be used is cumulative incidence [percentage]. To derive the outcome, calculations will be used to determine the cumulative incidence in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 13 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of insertion/post-insertion complications will use mixed-effects logistic regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate presented will be odds ratio with 95% confidence interval. Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Multicollinearity will be explored through Stata's 'collin' command. Linearity between the log odds of the dependent variable and the independent variables will be tested through the Box-Tidwell test. If assumptions for the preferred analysis are not met, nonparametric binary logistic regression model using the local likelihood logit estimation method will be used.
Missing data will be reported when evident <10% is expected for secondary outcome 13. If missing data is ≥20% imputation will be used.

Secondary outcome 14
The secondary outcome of interest is patient/carer satisfaction with insertion procedure. The measurement to be used is a 0-10 rating on an 11-point Likert scale [IQR]. To derive the outcome, calculations will be used to determine the difference between control and intervention conditions. The endpoint for secondary outcome 14 will be collection of the last participant at sustainability timepoint 2.
3.9.14.1 Main analysis The individual will be the unit of analysis. Between group comparison of patient/carer satisfaction with insertion procedure will use mixed-effects linear regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'.
The trial hypotheses will be evaluated through a framework of superiority. The effect estimate will be mean (or median) difference with 95% confidence interval.
Linear relationships will be explored through scatter plots and Q-Q-Plots. Normality will be explored through histograms and Shapiro-Wilk tests. Homoscedasticity will be explored through a scatterplot of residuals versus predicted values. Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Errors in variables will be explored through Delgado and Gonzalez Manteiga's 'dgmtest' test in Stata. Multicollinearity will be explored through Stata's 'collin' command. If assumptions for the preferred analysis are not met, median regression will be substituted.
Missing data will be reported when evident. <10% is expected for secondary outcome 14. If missing data is ≥20% imputation will be used.

Secondary outcome 15
The secondary outcome of interest is patient/carer pain with insertion procedure. The measurement to be used is a 0-10 rating on an 11-point Likert scale [IQR]. To derive the outcome, calculations will be used to determine the difference between control and intervention conditions. The endpoint for secondary outcome 15 will be collection of the last participant at sustainability timepoint 2. The individual will be the unit of analysis. Between group comparison of patient/carer pain with insertion procedure will use mixed-effects linear regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate will mean (or median) difference with 95% confidence interval.
Linear relationships will be explored through scatter plots and Q-Q-Plots. Normality will be explored through histograms and Shapiro-Wilk tests. Homoscedasticity will be explored through a scatterplot of residuals versus predicted values. Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Errors in variables will be explored through Delgado and Gonzalez Manteiga's 'dgmtest' test in Stata. Multicollinearity will be explored through Stata's 'collin' command. If assumptions for the preferred analysis are not met, median regression will be substituted.
Missing data will be reported when evident. <10% is expected for secondary outcome 15. If missing data is ≥20% imputation will be used.

Secondary outcome 16
The secondary outcome of interest is staff satisfaction with insertion procedure. The measurement to be used is a 0-10 rating on an 11-point Likert scale [IQR]. To derive the outcome, calculations will be used to determine the difference between control and intervention conditions. The endpoint for secondary outcome 16 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of staff satisfaction with insertion procedure will use mixed-effects linear regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate will mean (or median) difference with 95% confidence interval.
Linear relationships will be explored through scatter plots and Q-Q-Plots. Normality will be explored through histograms and Shapiro-Wilk tests. Homoscedasticity will be explored through a scatterplot of residuals versus predicted values. Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Errors in variables will be explored through Delgado and Gonzalez Manteiga's 'dgmtest' test in Stata. Multicollinearity will be explored through Stata's 'collin' command. If assumptions for the preferred analysis are not met, median regression will be substituted.
Missing data will be reported when evident. <10% is expected for secondary outcome 16. If missing data is ≥20% imputation will be used.

Secondary outcome 17
The secondary outcome of interest is rate of primary BSI and S. Aureus BSI. The measurement to be used is cumulative incidence [percentage]. To derive the outcome, calculations will be used to determine the cumulative incidence in each of the control and intervention conditions, as well as the difference between control and intervention conditions. The endpoint for secondary outcome 17 will be collection of the last participant at sustainability timepoint 2.

Main analysis
The individual will be the unit of analysis. Between group comparison of primary BSI and S. Aureus BSI will use mixed-effects logistic regression with cluster included as a random effect to account for probable non-independence of observations within clusters. Fixed effects included in the model will be study phase and time. The primary analysis will be 'intention-to-treat'. The trial hypotheses will be evaluated through a framework of superiority. The effect estimate presented will be odds ratio with 95% confidence interval.
Independence will be explored through plotting residuals against predicted values and use of Breusch-Pagan tests. Multicollinearity will be explored through Stata's 'collin' command. Linearity between the log odds of the dependent variable and the independent variables will be tested through the Box-Tidwell test. If assumptions for the preferred analysis are not met, nonparametric binary logistic regression model using the local likelihood logit estimation method will be used.
Missing data will be reported when evident <10% is expected for secondary outcome 17. If missing data is ≥20% imputation will be used.    Step 3 Assessed for eligibility (n=) Received intervention (n=no of clusters, average cluster size, variance of cluster sizes) Did not receive intervention, give reasons (n=no of clusters, average cluster size, variance of cluster size) Step 4 Assessed for eligibility (