Primary outcomes

For pressure ulcer prevention, the primary outcome is pressure ulcer incidence reported as proportion of participants developing a new pressure ulcer of any stage, or time to pressure ulcer development.

For pressure ulcer treatment, the primary outcome is complete healing of existing pressure ulcers (i.e. 100% granulation and re‐epithelialisation) reported as the proportion of participants with healed pressure ulcers or time to pressure ulcer healing.

Secondary outcomes

We will include the following secondary outcomes in this overview of reviews. The network meta‐analyses will only analyse and report primary outcomes.

• Patient‐support surface‐associated comfort. The definition and measurement of this outcome may vary from one review to another; for example, the proportion of participants who report comfort, or comfort measured by a scale with continuous (categorical) numbers.

• All reported adverse events (measured using survey/questionnaire/data capture process or visual analogue scale). We will consider the assessment of any event in general defined as adverse by participants, health professionals, or both.

• Health‐related quality of life (HRQOL) (measured using a standardised generic questionnaire such as EQ‐5D (Herdman 2011), or pressure‐ulcer‐specific questionnaires such as the PURPOSE Pressure Ulcer Quality of Life (PU‐QOL) questionnaire (Gorecki 2013)). We will present evidence on overall scores of questionnaires used rather than reporting multiple domain scores from the same measure.

• Cost‐effectiveness. Data extracted are incremental mean cost per incremental gain in benefit (incremental cost‐effectiveness ratio (ICER)) and other measures of relative cost‐effectiveness (e.g. net monetary benefit, net health benefit).

We will record outcome data from any time points specified in eligible reviews. Eligible reviews are likely to consider outcome data at a single time point. However, if a review specified an outcome measure at multiple time points, we will consider outcome measures at three months as the primary interest of this overview (Schoonhoven 2007), regardless of the time points specified as being of primary interest by the review itself. If the review did not report three‐month outcome measures, we will consider those closest to three months in this overview. Where a review only reported a single time point or did not specify a time point for their outcome measurement, we will consider these data in this overview. For all outcomes, we will class:

• one week or less to eight weeks as short‐term follow‐up;

• more than eight weeks to 16 weeks as medium‐term follow‐up;

• more than 16 weeks as long‐term follow‐up.

Assessment of risk of bias in included reviews for overview

Two researchers who are not authors of the Cochrane Reviews included in the overview will independently assess risk of bias using the ROBIS tool (Whiting 2016). ROBIS assesses risk of bias in three phases: first, assessing relevance (optional); second, identifying concerns with the review process; and third, forming an overall judgement of the risk of bias. In the second phase, concerns with the review process can be identified for four specific domains.

• Study eligibility criteria: assessing whether eligibility criteria of included reviews were pre‐specified, clear, and appropriate to the review question.

• Identification and selection of studies: assessing whether any trials that would have met the inclusion criteria of a review were not included in the review.

• Data collection and study appraisal: assessing whether bias may have been introduced during the data collection or risk of bias assessment processes.

• Synthesis and findings: assessing whether, when the review authors decided to pool data from the included trials (either in a quantitative or qualitative synthesis), the review authors have used appropriate methods to do so (Whiting 2016).

Concerns can be graded 'low', 'high' or 'unclear'. We will note the rationale for decisions at each stage. As this overview will only include Cochrane Reviews and relevance will be considered as part of our screening and selection process, we will not assess relevance using the ROBIS tool (an optional first phase).

Any disagreements between two overview authors will be resolved by discussion and a third reviewer where necessary.

Assessment of risk of bias in included studies for network meta‐analyses

For RCTs in the network meta‐analyses, we will use the results of overall risk of bias of the included trials that have already been assessed by the review authors.

Where any RCTs in the network meta‐analyses have not had risk of bias assessment, two independent reviewers will undertake this using Cochrane's 'Risk of bias' tool (Higgins 2017). We note further details about risk of bias assessment in Appendix 3.

Measures of treatment effect

For dichotomous outcome data (e.g. pressure ulcer incidence), we will present the risk ratio (RR) with its 95% CI. For continuous outcome data, we will present the mean difference (MD) with 95% CIs for studies that use the same assessment scale. If studies reporting continuous data use different assessment scales, we will report the standardised mean difference (SMD) with 95% CIs.

For time‐to‐event data (e.g. time to pressure ulcer development), we will present the hazard ratio (HR) with its 95% CI. If included studies reporting time‐to‐event data do not report an HR then, when feasible, we will estimate this using other reported outcomes, such as numbers of events, through employing available statistical methods (Parmar 1998; Tierney 2007).

For network meta‐analyses, we will also present the relative ranking of each bed, overlay and mattress support surface based on the estimated probability of that surface being the most effective (in terms of ulcer prevention or healing). This value is a cumulative probability called the Surface Under the Cumulative RAnking (SUCRA) (Salanti 2011). A SUCRA value can range from 0% to 100% and the larger the SUCRA value, the better the ranking of a bed or mattress support surface for the outcome of interest (Chaimani 2013; Salanti 2011).

Methods for overview data presentation and synthesis

The aim of this overview is to present a detailed summary of evidence on beds and mattresses for pressure ulcer prevention and treatment. We will present all eligible comparisons grouped by intervention type. We will use tabular formats and narrative techniques to present evidence summaries alongside the GRADE assessment for each comparison; and if the GRADE assessment is not available then the review authors will undertake it. Where possible we will also present results of meta‐analysis (including results of available sub‐group analysis), along with details of effects model and measures of statistical heterogeneity (i.e. Chi² tests and relevant P values, and I² statistics). Where meta‐analysis has not been undertaken, we will report study‐level effects. We will also present results of subgroup and sensitivity analyses.

We will present the certainty of evidence for each eligible outcome and comparison from each included Cochrane Review in a 'Summary of findings' table. The table will be designed according to the 'Summary of findings' table template proposed in Yepes‐Nuñez 2019 for network meta‐analysis. The table will include participants, interventions, comparators, outcomes, settings, the number of studies, the number of total participants, effect sizes and 95% CIs, anticipated absolute effects and 95% CIs of each group and difference between groups, certainty of evidence, and interpretation of findings.

Methods for network meta‐analyses

We will conduct four separate network meta‐analyses where possible: one for the proportion of participants developing a new pressure ulcer, one for time to pressure ulcer development, one for the proportion of participants with healed pressure ulcers, and one for time to pressure ulcer healing.

Unit of analysis, missing data, homogeneity and transitivity assumptions, and reporting bias have impacts on the validity of network meta‐analysis. Prior to carrying out network meta‐analysis, we will deal with those issues using methods described in Appendix 2.

We will synthesise included RCT data using the published network commands and network graph packages of STATA for network meta‐analysis and graphically present results (Chaimani 2013; Chaimani 2015; Schwarzer 2015; White 2015). Where applicable, we will estimate the relative effectiveness of any two interventions as a function of each intervention relative to the reference intervention (foam surfaces).

Using STATA (networkplot), we will produce a network plot of the included beds and mattresses to understand the geometry of the evidence base and to inform the analysis. We will exclude studies with one eligible arm where the arm cannot be connected to the network in any way.

We will perform network meta‐analyses using multivariate meta‐regression models in STATA (network meta) to estimate the relative effects for network contrasts. This modelling approach addresses correlations between the effect sizes from multi‐arm studies. We will fit a consistency model that assumes an agreement between direct and indirect evidence and will assume that a bed or mattress support surface has the average effect size for a range of similar populations (random‐effects model).

Methods for network meta‐analyses' relative rankings

On the basis of relative effect estimates of each bed and mattress, we will calculate the SUCRA percentages to estimate the relative rankings of bed and mattress support surfaces in STATA (sucra) and also present rankograms. We will estimate the relative rankings for each network meta‐analysis that we will undertake. We will present a rankogram for each bed and mattress ‒ this is a plot of the probabilities of assuming each of the possible rankings (Chaimani 2013).

Assessing the certainty of evidence and 'Summary of findings' tables for network meta‐analyses

We will assess the certainty of evidence for the network meta‐analyses using the GRADE approach proposed by Salanti 2014 via the Confidence in the Results of Network Meta‐Analysis (CINeMA) tool (Nikolakopoulou 2020). We will assess the certainty of evidence in two ways—for each network contrast and separately—for the network as a whole (assessing the certainty of the relative ranking).

The GRADE assessment using CINeMA involves consideration of six domains: within‐study bias; across‐studies bias; indirectness; imprecision; heterogeneity; and incoherence. To make the within‐study bias judgement, CINeMA evaluates the contributions of the included studies to each network contrast, producing the percentage contribution matrix (Nikolakopoulou 2020). These contributions are then used to weight study‐level risk of bias results to estimate the network contrast‐level within‐study bias (Nikolakopoulou 2020). The certainty of evidence can be assessed as being high, moderate, low or very low. RCT evidence has the potential to be high certainty.

We will present a separate 'Summary of findings' table for each comparison evaluated in the network meta‐analyses. We will present the following primary outcomes in the 'Summary of findings' tables.

• Proportion of participants developing a new pressure ulcer

• Time to pressure ulcer development

• Proportion of participants with healed pressure ulcers

• Time to pressure ulcer healing.

Assessment of statistical heterogeneity and inconsistency

We will assess the presence of the common network statistical heterogeneity in network meta‐analyses using the I² measure and its 95% CIs and Tau² measure (see Appendix 2 for further details). We will also assess inconsistency at levels of local loops, and the whole network using methods in Appendix 2. Inconsistency refers to statistical disagreement between direct and indirect evidence and is a manifestation of non‐transitivity (Cipriani 2013).

Investigation of heterogeneity (including sub‐group analysis)

When important inconsistency or heterogeneity occur, we will follow steps proposed by Cipriani 2013 to investigate further. Where necessary, we will perform subgroup analyses for binary and categorical factors (or meta‐regression for continuous factors) by the following four study‐level characteristics.

• Risk of bias (binary: low or unclear risk of bias; and high risk of bias) (Schulz 1995)

• Settings (categorical: acute care and other hospital settings; long‐term care settings; operating theatre setting; and intensive care unit)

• Baseline skin status (categorical: participants at risk, other skin status or non‐reporting; non‐blanchable erythema; existing ulcers of stage 2 or more severe) (Shi 2018b)

• Follow‐up duration (continuous) (Schoonhoven 2007)