Arm‐specific outcomes/arm‐level data: raw outcome data (e.g. mean (SD) or risk) for each arm of the trial (see treatment contrast). Assumptions for NMA: in common with all meta‐analysis, the true treatment effect across trials is assumed to be described by a fixed‐effect or random‐effects model. Additionally, transitivity is assumed and, concurrently, exchangeability and consistency. Baseline risk: the absolute risk of the outcome in the 'control' group. This is affected by the presence of prognostic factors. Some authors have used the baseline risk as a proxy effect modifier, but in general the effect estimate (RR/OR/HR) is independent of the baseline risk; on the other hand, the absolute risk difference depends on baseline risk. Bayesian approach: the explicit quantitative use of external evidence in the design, monitoring, analysis, interpretation of a health‐care evaluation. In the Bayesian paradigm, prior beliefs about parameters in the models are specified and factored into the estimation. Posterior distributions of model parameters are then derived from the prior information and the observed data. In NMA, it is common to use non‐informative priors for effect estimates. Coherence/consistency: the direct effect estimate (e.g. mean difference or log odds ratio) is the same as the sum of the indirect effect estimates. Connected network: a group of linked interventions, such that every trial in the network has at least one intervention in common with at least one other trial. Sometimes individual comparisons are not connected to the rest of the network (disconnected network) and can sometimes be joined in by extending the network to include supplementary interventions. Contour‐enhanced funnel plot: contour‐enhanced funnel plots show areas of statistical significance, and they can help in distinguishing publication bias from other possible reasons for asymmetry. In a network of interventions, each study estimates the relative effect of different interventions, so asymmetry in the funnel plot cannot be judged. To account for this, an adaptation of the funnel plot can be used, in which the standard error is plotted against an adjusted effect size for each study: the adjusted effect size for a comparison is the study‐specific effect size minus the mean for the meta‐analysis for that comparison. Contrast/comparison/study‐level data: outcome data for the comparison (e.g. mean difference, odds ratio). Credible interval (CrI): the 95% credible interval is the range within which the mean value lies with posterior probability of 95%. Decision space/decision set: the interventions in the decision set are the focal treatments of interest to systematic review authors. Direct evidence/direct comparison/direct contrast: head‐to‐head comparison of two treatments, for example, A versus B (see indirect evidence). Edge: line representing a direct contrast on a network diagram. Effect modifier: effect modification occurs when the effect of A versus B (as the RR/OR/HR for binary outcomes) is significantly different in two or more subgroups, and this leads to heterogeneity, either within trials or between trials, or both. Factors that give rise to subgroup effects are called effect modifiers, and it is important to identify potential effect modifiers and allow for them in the analysis. The identification of significant effect modifiers may lead to stratification (separate analyses for each subgroup) or to a decision not to combine data from different trials in a meta‐analysis. In general, trials have different distributions of effect modifiers (e.g. proportion of people with and without diabetes), leading to inconsistency between trials in the treatment effect. This is often magnified when there is a network of different contrasts. Exchangeability: it is assumed that treatments in a NMA are exchangeable, so, if treatment B had been given to participants in the indirect A versus C trials and if A had been given in the B versus C indirect trials, then the true AB differences in these indirect studies would be identical to the true AB difference in direct A versus B trials, or at least from the same common distribution. Furthermore, if participants in other trials within the wider linked network (e.g. D versus E trials) were given A and B, the AB differences would also be the same or from the same distribution. This assumption breaks down when there are effect modifiers. Fixed‐effect: the true treatment effect is assumed to be constant across trials (fixed‐effect) ‐ see also random‐effects and transitivity. Global inconsistency: inconsistency across a network is described as global inconsistency. It can be evaluated statistically by fitting models that allow and do not allow for inconsistency. See also: Inconsistency/incoherence: Heterogeneity in a NMA: participants are not randomised to different trials. Therefore, there may be systematic differences in study characteristics or the distribution of participant characteristics across trials. If these characteristics influence the treatment effects (i.e. are effect modifiers), then there are systematic differences in treatment effects across trials, which is called between‐trial heterogeneity. There may also be within‐trial heterogeneity if there are subgroups of an effect modifier for which results are reported separately. In a NMA, the term, 'heterogeneity' applies to variation in effect modifiers within a single contrast (e.g. A versus B); the term, 'inconsistency' refers to the imbalance in effect modifiers between contrasts. Heterogeneity variance parameter (tau²): in a random‐effects model we assume there is heterogeneity for each pairwise comparison (e.g. A versus B) with variance (tau²AB), but in a NMA we often assume that there is a common heterogeneity amongst all the contrasts in the network; this common heterogeneity has a variance (tau²), which is called the 'heterogeneity variance parameter'. It can be compared with empirical distributions of heterogeneity values typically found in meta‐analyses (Salanti 2014; Turner 2012). Inconsistency/incoherence: this occurs when the effect estimate derived from an indirect contrast is not the same as the effect estimate derived from a direct contrast. For example, in a network of three interventions, there is inconsistency if dAB(direct) ǂ dAB(indirect), where dAB(indirect) = dAC(direct) ‐ dBC(direct); the effect estimates are given as mean differences or log(odds ratios/risk ratios/hazard ratios). Note that in order to investigate inconsistency there must be both indirect and direct evidence (loops in the network). See also global inconsistency. Inconsistency factor: this is the absolute difference between the direct and indirect estimates on the log scale (or the logarithm of the ratio of the two odds/hazard ratios) for one of the contrasts in a loop. A statistically low powered z‐test and a 90% or 95% confidence interval (CI) of the inconsistency is computed to determine whether this difference is significant. Indirect evidence/indirect comparison/indirect contrast: comparison of two treatments, for example, A versus B, obtained from combinations of other comparisons (e.g. trials comparing A versus C and trials comparing B with C) (see direct evidence). Indirect comparison meta‐analysis: meta‐analysis of a set of treatments that are linked via common comparator(s), but none are compared directly; evidence is combined in a single internally consistent model. Leverage: this is the effective number of parameters of the model, which is calculated differently for fixed‐effect and random‐effects models, with the latter having greater complexity. Likelihood (function): the likelihood function is a tool for inferring the underlying distribution of the observed data. To do this, we propose a model to represent the data ‐ often a parametric distribution is assumed (e.g. binomial) ‐ and unknown parameters of that distribution are determined, given the data, by maximising the likelihood (the larger the likelihood, the closer the model fit). Loop (of evidence): combination of direct and indirect evidence, such that the interventions in the network diagram can be linked to form a closed loop. Meta‐analysis: a statistical synthesis of the results from two or more separate studies. Methods involve calculating a weighted average of effect estimates from the separate studies. Mixed treatment comparison meta‐analysis: another name for network meta‐analysis. Model: a statistical model is a (simplified) mathematical representation of the system we wish to learn about, and which generates our observed data. The model will usually depend on some known factors, such as other variables measured alongside the data, and some unknown parameters that we wish to determine. Then having determined the unknown parameters, the model should be able to simulate data that are an approximation of the real data, allowing us to make inferences from the data. Multi‐arm trial: individual trial that compares more than two interventions. Network: trials must be linked in a network of interventions, such that every trial in the network has at least one intervention in common with at least one other trial. Network diagram: graphical representation of the interventions in the network. It consists of nodes representing the interventions and edges representing the contrasts. The amount of available information can be presented by 'weighting' the nodes and edges using different node sizes and line thicknesses according to the number of studies reporting that treatment or contrast respectively. Other types of weighting are discussed in Chaimani 2013b. Network meta‐analysis (NMA): NMA is the simultaneous combination of data from randomised comparisons of multiple competing treatments (A versus B, A versus C, A versus D, B versus D, and so on), to deliver an internally consistent set of estimates while respecting the randomisation in the evidence. The use of indirect estimates can provide information on contrasts for which no trials exist. It can also improve the precision of the direct estimate by reducing the width of the CIs compared with the direct evidence alone. Node: intervention represented on a network diagram, usually by a circle of weighted size. Node splitting: a method of assessing inconsistency. A 'leave‐one comparison‐out' approach, often called 'node splitting,' is applied, with each direct contrast being excluded from the network and then estimating the difference between this direct evidence and the indirect evidence from the network. Pairwise meta‐analysis: meta‐analysis of one or more trials of direct comparisons (e.g. A versus B) ‐ see direct evidence. Prognostic factors: population or study characteristics that affect the risk of the outcome. In a sufficiently large randomised trial that is free from bias, prognostic factors are distributed evenly between intervention groups and do not affect the effect estimate (RR/OR/HR for binary outcomes) unless they are effect modifiers, but they do affect the baseline risk and absolute risk difference. Random‐effects: trial‐specific treatment differences are assumed to be from a common distribution ‐ see also fixed‐effect and transitivity. Ranking: ordering of treatments according to their relative effectiveness. Rankogram: graph of probability versus rank order for a particular treatment. Rankograms are based on the uncertainty in the effect estimates. So if two treatments A and B are each compared with the reference intervention and the CIs for the effect overlap, then each treatment will have some probability of being the most effective. If there is no overlap and A is better than B, then the probability of A being the best will be 1. Sparse data: data with wide CIs because of few events as a consequence of small studies or short follow‐up periods. Study‐level data: see contrast. SUCRA: Surface Under the Cumulative RAnking. This is a measure of the probability that the given treatment is the best. Thus, a SUCRA would be 1 (or 100%) when a treatment was certain to be the best and 0 (0%) when a treatment was certain to be the worst. Supplementary set (of interventions): interventions added to the network to provide additional evidence on relative treatment effects of the decision set. This may be to connect an otherwise unconnected network of treatments, to increase the precision of the treatment effect estimates or to help address between‐trial heterogeneity. Transitivity: NMA requires a transitivity assumption, such that there is no imbalance in the distribution of effect modifiers across the different types of treatment contrasts (see also exchangeability). 'Unadjusted' meta‐analysis: meta‐analysis of all the treatment arms for a particular treatment (e.g. all A arms). This breaks the randomisation and should not be done. References include: Caldwell 2005; Caldwell 2014; Chaimani 2013a; Chaimani 2013b; Cipriani 2013; Dias 2013; Dias 2016; Grant 2013; Jansen 2013; Lu 2004; Salanti 2008; Salanti 2011; Salanti 2014; Soares 2014; Thorlund 2012; Tu 2012; White 2012. The Cochrane Central Register of Controlled Trials (CENTRAL) #1MeSH descriptor: [Bandages] explode all trees #2MeSH descriptor: [Alginates] explode all trees #3MeSH descriptor: [Hydrogels] explode all trees #4MeSH descriptor: [Honey] explode all trees #5MeSH descriptor: [Silver] explode all trees #6MeSH descriptor: [Silver Sulfadiazine] explode all trees #7MeSH descriptor: [Charcoal] explode all trees #8MeSH descriptor: [Silicones] explode all trees #9(dressing* or pad or pads or gauze or tulle or film or bead or foam* or non‐adherent or "non adherent" or hydrocolloid* or "sodium hyaluronate" or alginat* or hydrogel* or silver* or honey* or matrix or iodine* or "protease modulat*" or "capillary action" or charcoal or silicon* or polymer*):ti,ab,kw #10((odour or odor) near/3 absorb*):ti,ab,kw #11(primapore or curasorb or seasorb or sorbsan or advadraw or vacutex or tegaderm or opsite or allevyn or biatain or medihoney or activon tulle or granuflex or "nu derm" or aquacel or iodoflex or iodozyme or xeroform or carboflex or cutimed sorbact or promogran or acticoat or "urgosorb silver" or mepitel or urgotul):ti,ab,kw #12{or #1‐#11} #13MeSH descriptor: [Metronidazole] explode all trees #14metronidazole:ti,ab,kw #15MeSH descriptor: [Anti‐Bacterial Agents] explode all trees #16MeSH descriptor: [Administration, Topical] explode all trees #17{and #15‐#16} #18(topical near/2 (antibiotic* or antimicrobial* or antibacterial*)):ti,ab,kw #19MeSH descriptor: [Iodophors] explode all trees #20{and #16, #19} #21((topical near/2 iodin*) or ("cadexomer iodine")):ti,ab,kw #22MeSH descriptor: [Collagenases] explode all trees #23{and #16, #22} #24(topical near/2 collagen*):ti,ab,kw #25MeSH descriptor: [Phenytoin] explode all trees #26{and #16, #25} #27(topical near/2 phenytoin):ti,ab,kw #28MeSH descriptor: [Zinc Oxide] explode all trees #29{and #16, #28} #30(topical near/2 zinc):ti,ab,kw #31(iodosorb or actiformcool or aquaflo or flamazine or silvadene):ti,ab,kw #32MeSH descriptor: [Ointments] explode all trees #33(ointment* or lotion* or cream* or powder* or gel or gels):ti,ab,kw #34(topical next (agent* or preparation* or therap* or treatment*)):ti,ab,kw #35{or #13‐#14, #17‐#18, #20‐#21, #23‐#24, #26‐#27, #29‐#34} #36{or #12, #35} #37MeSH descriptor: [Pressure Ulcer] explode all trees #38(pressure next (ulcer* or sore* or injur*)):ti,ab,kw #39(decubitus next (ulcer* or sore*)):ti,ab,kw #40((bed next sore*) or bedsore*):ti,ab,kw #41{or #37‐#40} #42{and #36, #41} in Trials Ovid MEDLINE 1 exp Bandages/ 2 exp Alginates/ 3 exp Hydrogels/ 4 exp Honey/ 5 exp Silver/ 6 exp Silver Sulfadiazine/ 7 exp Charcoal/ 8 exp Silicones/ 9 (dressing* or pad or pads or gauze or tulle or film or bead or foam* or non‐adherent or "non adherent" or hydrocolloid* or "sodium hyaluronate" or alginat* or hydrogel* or silver* or honey* or matrix or iodine* or "protease modulat*" or "capillary action" or charcoal or silicon* or polymer*).tw. 10 ((odour or odor) adj3 absorb*).tw. 11 (primapore or curasorb or seasorb or sorbsan or advadraw or vacutex or tegaderm or opsite or allevyn or biatain or medihoney or activon tulle or granuflex or "nu derm" or aquacel or iodoflex or iodozyme or xeroform or carboflex or cutimed sorbact or promogran or acticoat or "urgosorb silver" or mepitel or urgotul).tw. 12 or/1‐11 13 exp Metronidazole/ 14 metronidazole.tw. 15 exp Administration, Topical/ 16 exp Anti‐Bacterial Agents/ 17 and/15‐16 18 (topical adj2 (antibiotic* or antimicrobial* or antibacterial*)).tw. 19 exp Iodophors/ 20 and/15,19 21 ((topical adj2 iodin*) or "cadexomer iodine").tw. 22 exp Collagenases/ 23 and/15,22 24 (topical adj2 collagen*).tw. 25 exp Phenytoin/ 26 and/15,25 27 (topical adj2 phenytoin).tw. 28 exp Zinc Oxide/ 29 and/15,28 30 (topical adj2 zinc).tw. 31 (iodosorb or actiformcool or aquaflo or flamazine or silvadene).tw. 32 exp Ointments/ 33 (ointment* or lotion* or cream* or powder* or gel or gels).tw. 34 (topical adj (agent* or preparation* or therap* or treatment*)).tw. 35 or/13‐14,17‐18,20‐21,23‐24,26‐27,29‐34 36 or/12,35 37 exp Pressure Ulcer/ 38 (pressure adj (ulcer* or sore* or injur*)).tw. 39 (decubitus adj (ulcer* or sore*)).tw. 40 (bedsore* or bed sore*).tw. 41 or/37‐40 42 and/36,41 43 randomized controlled trial.pt. 44 controlled clinical trial.pt. 45 randomi?ed.ab. 46 placebo.ab. 47 clinical trials as topic.sh. 48 randomly.ab. 49 trial.ti. 50 or/43‐49 51 exp animals/ not humans.sh. 52 50 not 51 53 and/42,52 Ovid Embase 1 exp "bandages and dressings"/ 2 exp honey/ 3 exp hydrogel/ 4 exp Calcium Alginate/ 5 (dressing* or pad or pads or gauze or tulle or film or bead or foam* or non‐adherent or "non adherent" or hydrocolloid* or "sodium hyaluronate" or alginat* or hydrogel* or silver* or honey* or matrix or iodine* or "protease modulat*" or "capillary action" or charcoal or silicon* or polymer*).ti,ab. 6 ((odour or odor) adj3 absorb*).ti,ab. 7 (primapore or curasorb or seasorb or sorbsan or advadraw or vacutex or tegaderm or opsite or allevyn or biatain or medihoney or activon tulle or granuflex or "nu derm" or aquacel or iodoflex or iodozyme or xeroform or carboflex or cutimed sorbact or promogran or acticoat or "urgosorb silver" or mepitel or urgotul).ti,ab. 8 or/1‐7 9 exp metronidazole/ 10 metronidazole.ti,ab. 11 topical drug administration/ 12 exp Antibiotic Agent/ 13 and/11‐12 14 (topical adj2 (antibiotic* or antimicrobial* or antibacterial*)).ti,ab. 15 exp cadexomer iodine/ 16 and/11,15 17 "cadexomer iodine".ti,ab. 18 exp silver/ or exp sulfadiazine silver/ 19 and/11,18 20 exp collagenase/ 21 and/11,20 22 (topical adj2 collagen*).ti,ab. 23 phenytoin/ 24 and/11,23 25 (topical adj2 phenytoin).ti,ab. 26 exp zinc oxide/ 27 and/11,26 28 (topical adj2 zinc).ti,ab. 29 (iodosorb or actiformcool or aquaflo or flamazine or silvadene).ti,ab. 30 exp ointment/ 31 (ointment* or lotion* or cream* or powder* or gel or gels).ti,ab. 32 (topical adj (agent* or preparation* or therap* or treatment*)).ti,ab. 33 or/9‐10,13‐14,16‐17,19,21‐22,24‐25,27‐32 34 or/8,33 35 exp decubitus/ 36 (pressure adj (ulcer* or sore* or injur*)).tw. 37 (decubitus adj (ulcer* or sore*)).tw. 38 (bedsore* or bed sore*).tw. 39 or/35‐38 40 and/34,39 41 Randomized controlled trials/ 42 Single‐Blind Method/ 43 Double‐Blind Method/ 44 Crossover Procedure/ 45 (random* or factorial* or crossover* or cross over* or cross‐over* or placebo* or assign* or allocat* or volunteer*).ti,ab. 46 (doubl* adj blind*).ti,ab. 47 (singl* adj blind*).ti,ab. 48 or/41‐47 49 exp animals/ or exp invertebrate/ or animal experiment/ or animal model/ or animal tissue/ or animal cell/ or nonhuman/ 50 human/ or human cell/ 51 and/49‐50 52 49 not 51 53 48 not 52 54 and/40,53 EBSCO CINAHL Plus S57 S43 AND S56 S56 S44 OR S45 OR S46 OR S47 OR S48 OR S49 OR S50 OR S51 OR S52 OR S53 OR S54 OR S55 S55 TI allocat* random* or AB allocat* random* S54 MH "Quantitative Studies" S53 TI placebo* or AB placebo* S52 MH "Placebos" S51 TI random* allocat* or AB random* allocat* S50 MH "Random Assignment" S49 TI randomi?ed control* trial* or AB randomi?ed control* trial* S48 AB ( singl* or doubl* or trebl* or tripl* ) and AB ( blind* or mask* ) S47 TI ( singl* or doubl* or trebl* or tripl* ) and TI ( blind* or mask* ) S46 TI clinic* N1 trial* or AB clinic* N1 trial* S45 PT Clinical trial S44 MH "Clinical Trials+" S43 S37 AND S42 S42 S38 OR S39 OR S40 OR S41 S41 TI decubitus or AB decubitus S40 TI ( bed sore* or bedsore* ) or AB ( bed sore* or bedsore* ) S39 TI ( pressure ulcer* or pressure sore* ) or AB ( pressure ulcer* or pressure sore* ) S38 (MH "Pressure Ulcer+") S37 S13 OR S36 S36 S14 OR S15 OR S18 OR S19 OR S21 OR S22 OR S24 OR S25 OR S27 OR S28 OR S30 OR S31 OR S33 OR S34 OR S35 S35 TI (topical N3 agent* or topical N3 preparation* or topical N3 therap* and topical N3 treatment*) OR AB (topical N3 agent* or topical N3 preparation* or topical N3 therap* and topical N3 treatment*) S34 TI (ointment* or lotion* or cream* or powder* or gel or gels) OR AB (ointment* or lotion* or cream* or powder* or gel or gels) S33 (MH "Ointments") S32 TI (iodosorb or actiformcool or aquaflo or flamazine or silvadene) OR AB (iodosorb or actiformcool or aquaflo or flamazine or silvadene) S31 TI (topical N2 zinc) OR AB (topical N2 zinc) S30 S16 AND S29 S29(MH "Zinc Oxide") S28TI (topical N2 phenytoin) OR AB (topical N2 phenytoin) S27S16 AND S26 S26(MH "Phenytoin+") S25 TI (topical N2 collagen*) OR AB (topical N2 collagen*) S24 S16 AND S23 S23 (MH "Collagen") S22 TI "cadexomer iodine" OR AB "cadexomer iodine" S21 S16 AND S20 S20 (MH "Iodophors+") S19 TI (topical N2 (antibiotic* or antimicrobial* or antibacterial*)) OR AB (topical N2 (antibiotic* or antimicrobial* or antibacterial*)) S18 S16 AND S17 S17 (MH "Antiinfective Agents+") S16 (MH "Administration, Topical+") S15 TI metronidazole OR AB metronidazole S14 (MH "Metronidazole") S13 S1 OR S2 OR S3 OR S4 OR S5 OR S6 OR S7 OR S8 OR S9 OR S10 OR S11 OR S12 S12 AB (primapore or curasorb or seasorb or sorbsan or advadraw or vacutex or tegaderm or opsite or allevyn or biatain or medihoney or activon tulle or granuflex or "nu derm" or aquacel or iodoflex or iodozyme or xeroform or carboflex or cutimed sorbact or promogran or acticoat or "urgosorb silver" or mepitel or urgotul) S11 TI (primapore or curasorb or seasorb or sorbsan or advadraw or vacutex or tegaderm or opsite or allevyn or biatain or medihoney or activon tulle or granuflex or "nu derm" or aquacel or iodoflex or iodozyme or xeroform or carboflex or cutimed sorbact or promogran or acticoat or "urgosorb silver" or mepitel or urgotul) S10 TI odor N3 absorb* or AB odor N3 absorb* S9 TI odour N3 absorb* or AB odour N3 absorb* S8 AB (dressing* or pad or pads or gauze or tulle or film or bead or foam* or non‐adherent or "non adherent" or hydrocolloid* or "sodium hyaluronate" or alginat* or hydrogel* or silver* or honey* or matrix or iodine* or "protease modulat*" or "capillary action" or charcoal or silicon* or polymer*) S7 TI (dressing* or pad or pads or gauze or tulle or film or bead or foam* or non‐adherent or "non adherent" or hydrocolloid* or "sodium hyaluronate" or alginat* or hydrogel* or silver* or honey* or matrix or iodine* or "protease modulat*" or "capillary action" or charcoal or silicon* or polymer*) S6 (MH "Silver") or (MH "Silver Sulfadiazine") S5 (MH "Honey") S4 (MH "Charcoal") S3 (MH "Silicones") S2 (MH "Alginates") S1 (MH "Bandages and Dressings+") We used the following specialist NMA STATA routines in addition to the standard STATA meta‐analysis routines of metan and mvmeta.
We also used STATA routines to visually display the data
Individual interventions were mapped onto the group categories as shown in Table 10. Based on an assumption that dressings within certain categories can be used interchangeably, we grouped the individual interventions into the following pre‐specified categories: basic wound dressings, advanced dressings and antimicrobial dressings (as described in the BNF 2016), and we kept the different types of specialist dressings (e.g. protease‐modulating matrix dressings) and the different topical agents as separate categories. Mapping from individual to group interventions
We excluded 15 of the 51 included studies from the group analysis because they compared interventions in the same group: 14 had advanced dressings in both arms (Aguilo Sanchez 2002; Bale 1997a; Banks 1994a; Banks 1994c; Belmin 2002; Brod 1990; Brown‐Etris 1996; Brown‐Etris 2008; Darkovich 1990; Motta 1999; Muller 2001; Seeley 1999; Sopata 2002; Thomas 1997a); and one had two antimicrobial dressings (Yapucu Güneş 2007). Three additional studies not joined into the individual network were also isolated from the group network (Imamura 1989; Payne 2004; Van De Looverbosch 2004). Two studies previously excluded from the individual network were included initially in the group network (Ashby 2012; Sipponen 2008), but only the Sipponen 2008 study was included in the network. We excluded the Ashby 2012 study and nine others from the network because they were joined to an ineligible intervention that did not link two or more interventions in the network (Gorse 1987; Hondé 1994; Nisi 2005; Nussbaum 1994; Price 2000; Ramos‐Torrecillas 2015; Rees 1999; Serena 2010; Thomas 2005). We also excluded the Sebern 1986 study from the group network, as for the individual network. The group NMA generated results for 45 mixed treatment contrasts. The network was dominated by the contrast advanced dressing versus basic dressing and the rest of the data were sparse. Figure 7 shows all NMA results, with the all‐domain risk of bias shown alongside the forest plot contrasts. As in the individual network, the evidence for the majority of contrasts was informed by studies at high risk of bias, and CIs were wide or very wide, such that we downgraded all evidence at least once for imprecision. There was also heterogeneity or inconsistency, or both, for some contrasts. Consequently, evidence was of low or very low certainty, with the exception of one contrast, for which we assessed the evidence to be of moderate certainty. As for the individual network, this moderate‐certainty evidence should be interpreted in the light of the very low‐certainty evidence for the network as a whole. We report the representative set of contrasts of each intervention versus basic dressings (Table 4 and Figure 7 first subgroup). Further details of the GRADE assessment can be found in Appendix 8 and Appendix 9. It is not clear whether protease‐modulating dressings increase the probability of healing compared with basic dressings (RR 1.49; 95% CI 0.91 to 2.46, moderate quality evidence). This corresponds to an absolute risk difference of 94 more people healed per 1000 (95% CI 17 fewer to 279 more). We downgraded the evidence once for imprecision (low risk of bias). For each of two contrasts (collagenase ointment and tripeptide copper gel) it is unclear whether the intervention increases the probability of healing compared with basic dressings (collagenase: RR 2.01, 95% CI 1.05 to 3.88 and tripeptide copper gel: RR 3.39, 95% CI 0.94 to 12.30); Figure 7). This was low certainty evidence, downgraded once for risk of bias and once for imprecision for collagenase ointment, and twice for imprecision for tripeptide copper gel (for which the direct evidence involved only six participants experiencing five events (complete healing)). It is unclear whether the remaining six interventions affect the probability of healing compared with basic dressings (advanced dressings, advanced and antimicrobial dressings; antimicrobial dressings, dextranomer, phenytoin and sugar plus egg white) because the evidence is of very low certainty (downgraded mainly for risk of bias (once) and imprecision (twice), although two contrasts (phenytoin and advanced dressings versus basic dressings) had inconsistency. Ranking of treatmentsThe rank probability data are shown in Figure 18 and Table 12. The rankograms have maximum probabilities more sharply defined in the group network compared with the individual network for the treatments advanced dressings, advanced‐antimicrobial dressings, collagenase ointment and dextranomer, but there is still overlap of the rankograms for different treatments (Figure 8). The mean rank was 2.0 for dextranomer and 2.3 for tripeptide copper gel, and two treatments had a mean rank of 10 (out of 10): advanced‐antimicrobial dressings and sugar plus egg white. However, no SUCRA value was 0 or 1, indicating uncertainty in the group network.  Group network ‐ rankograms Ranks of interventions ‐ group network
As with the individual network, the results must be interpreted in the light of the considerable uncertainty in the network and individual estimates, which can give misleading results. Numerically, dextranomer and tripeptide copper gel had the highest probabilities of being the best treatments (55% and 34%, respectively), but these high rankings are likely to be an artificial result. . Across all treatments there was very low certainty in the ranking of interventions (see quality assessment below). Comparison of results from standard meta‐analysis versus NMA findingsWe compared the NMA results with the direct comparison (pairwise) results for the proportion completely healed for the eight different comparisons informing the group network (Table 11). Three comparisons had two or more direct comparison studies (Analysis 2.1). The direct evidence shows some heterogeneity for the comparison of advanced dressing versus basic dressing (I² = 52%, P = 0.02 and some variation in the point estimates). Appendix 11 shows direct evidence results for the time‐to‐healing outcome for three comparisons in five studies. Analysis Comparison 2 Direct evidence group intervention, number with complete healing, Outcome 1 Intervention 1 vs intervention 2. Certainty/quality assessment of the networkOverall we downgraded the evidence certainty three times for the network as a whole, because of risk of bias (once), imprecision (once) and inconsistency and publication bias (once): the weighted average risk of bias across the network was high (Appendix 8). For inconsistency, the global Wald test was borderline significant at the 90% significance level (P value was 0.095) (see Appendix 9), however, there were relatively few contrasts with conflicting results for direct and indirect estimates. We downgraded the evidence once for imprecision: there is some overlap of the individual rankograms and no SUCRA value was zero or 1, suggesting uncertainty around treatment estimates and ranking in this network. A contour‐enhanced funnel plot (Figure 9) suggested there may be some asymmetry in the plot (which may be a consequence of publication bias). Overall, we have little confidence in the findings in this group network, either in terms of the effect estimates or in the ranking of interventions. Sensitivity analyses ‐ group networkWe did not pre‐specify sensitivity analyses for the group network, mainly because the group network itself is based on the assumption that a variety of dressings can be grouped as advanced dressings or basic dressings (as defined by the BNF 2016). We attempted to investigate this assumption by examining the network contrasts for the individual network that compared two advanced dressings, expecting the effect estimates to be close to 1 if the assumption was valid. Results can be seen in Figure 4. Most point estimates were fairly close to 1, but CIs were usually wide or very wide around the estimates. Without exception, the risk of bias for each contrast was either high or very high, and the CI crossed at least one GRADE default MID. Thus, there is no clear evidence either to support or refute the group assumption. A post‐hoc sensitivity analysis (Appendix 12) examined the original assumption of combining topical agents and dressings in the same NMA, by restricting the group NMA to dressings. There may have been less imprecision in the network as a whole, but results for the contrasts with basic dressings were similar to those in the full group network. 1. Was the allocation sequence randomly generated? (Part of 'Selection bias') Low risk of bias The investigators describe a random component in the sequence generation process such as: referring to a random number table; using a computer random‐number generator; coin tossing; shuffling cards or envelopes; throwing dice; drawing of lots. High risk of bias The investigators describe a non‐random component in the sequence generation process. Usually, the description would involve some systematic, non‐random approach, for example: sequence generated by odd or even date of birth; sequence generated by some rule based on date (or day) of admission; sequence generated by some rule based on hospital or clinic record number. Unclear Insufficient information about the sequence generation process provided to permit a judgement of low or high risk of bias. 2. Was the treatment allocation adequately concealed? (Part of 'Selection bias') Low risk of bias Participants and investigators enrolling participants could not foresee assignment because one of the following, or an equivalent method, was used to conceal allocation: central allocation (including telephone, web‐based and pharmacy‐controlled randomisation); sequentially‐numbered drug containers of identical appearance; sequentially‐numbered, opaque, sealed envelopes. High risk of bias Participants or investigators enrolling participants could possibly foresee assignments and thus introduce selection bias, such as allocation based on: using an open random allocation schedule (e.g. a list of random numbers); assignment envelopes were used without appropriate safeguards (e.g. if envelopes were unsealed or non opaque or not sequentially numbered); alternation or rotation; date of birth; case record number; any other explicitly unconcealed procedure. Unclear Insufficient information provided to permit a judgement of low or high risk of bias. This is usually the case if the method of concealment is not described or not described in sufficient detail to allow a definite judgement, for example if the use of assignment envelopes is described, but it remains unclear whether envelopes were sequentially numbered, opaque and sealed. 3. Blinding ‐ was knowledge of the allocated interventions adequately prevented during the study? (Performance bias for blinding of participants and caregivers; detection bias for outcome assessors) Low risk of bias Any one of the following.
High risk of bias Any one of the following.
Unclear Either of the following.
4. Were incomplete outcome data adequately addressed? (Attrition bias) Low risk of bias Any one of the following.
High risk of bias Any one of the following.
Unclear Either of the following.
5. Are reports of the study free of suggestion of selective outcome reporting? (Outcome reporting bias) Low risk of bias Either of the following.
High risk of bias Any one of the following.
Unclear Insufficient information provided to permit judgement of low or high risk of bias. It is likely that the majority of studies will fall into this category. 6. Other sources of potential bias Low risk of bias The study appears to be free of other sources of bias. High risk of bias There is at least one important additional risk of bias. For example, the study:
Unclear There may be a risk of bias, but there is either:
The 51 included studies compared 39 different interventions in 59 comparisons, summarised in Table 13. Interventions in the included studies
One study (Ashby 2012) had a range of advanced dressings as the comparator, so was ineligible for the individual intervention NMA (but was eligible for the group NMA). Thus there were 50 eligible studies for the individual NMA, comparing 37 different interventions in 58 comparisons with 2952 participants.There were nine interventions eligible for the individual network, which, on their own, did not meet the inclusion criteria of the review, but were compared with interventions that did meet the inclusion criteria ‐ we have called these 'ineligible interventions'. The network diagram for all interventions in the individual NMA is shown in Figure 19. Ten interventions (including one ineligible intervention) were isolated from the network and are shown in red in Figure 19. Two further ineligible interventions (radiant heat and skin substitute) linked two of the eligible interventions and so were included in the joined network; the other six ineligible interventions (shown in blue) did not link eligible interventions and their studies were therefore omitted from thejoined network. Two of these omitted studies were three‐arm trials each with two ineligible interventions (Nussbaum 1994; Ramos‐Torrecillas 2015). Therefore, ten studies could not be included in the joined network (Gorse 1987; Imamura 1989; Nisi 2005; Nussbaum 1994; Payne 2004; Ramos‐Torrecillas 2015; Rees 1999; Sipponen 2008; Van De Looverbosch 2004; Yapucu Güneş 2007). The results for the isolated studies in which both interventions were eligible are reported in Analysis 5.1. Another study (Sebern 1986) only partially reported their results and was not considered further. Analysis Comparison 5 Direct evidence ‐ non‐network comparisons, Outcome 1 Intervention 1 vs intervention 2. Thus 39 studies were included in the individual network (Aguilo Sanchez 2002; Alm 1989; Bale 1997a; Banks 1994b; Banks 1994a; Banks 1994c; Barrois 1992; Belmin 2002; Brod 1990; Brown‐Etris 1996; Brown‐Etris 1997; Brown‐Etris 2008; Burgos 2000b; Colwell 1993; Darkovich 1990; Graumlich 2003; Hollisaz 2004; Hondé 1994; Kaya 2005; Kraft 1993; Matzen 1999; Meaume 2003; Motta 1999; Muller 2001; Neill 1989a; Oleske 1986; Parish 1979; Payne 2009; Piatkowski 2012; Price 2000; Romanelli 2001; Seeley 1999; Serena 2010; Sopata 2002; Thomas 1997a; Thomas 1998; Thomas 2005; Xakellis 1992; Zeron 2007). The percentage contributions to the mixed treatment contrasts from each direct contrast are shown in Figure 20 for the individual network contrasts versus saline gauze. We calculated the risk of bias for each contrast in the NMA and results for each mixed treatment contrast are shown in the last column of the table in Figure 20 and represented on the forest plot (see Assessment of risk of bias in included studies). Contributions matrix ‐ interventions versus saline gauze (independent network) Key: 1 = saline gauze dressing; 2 = alginate dressing; 3 = sequential hydrocolloid alginate dressings; 4 = basic wound contact dressing; 5 = collagenase ointment; 6 = dextranomer; 7 = foam dressing; 8 = hydrocolloid dressing; 9 = hydrocolloid +/‐ alginate (hydrocolloid with/without alginate filler); 10 = hydrogel dressing; 11 = ineligible intervention: radiant heat; 12 = ineligible intervention: skin substitute; 13 = iodine‐containing dressing; 14 = phenytoin; 15 = protease‐modulating dressing; 16 = PVP + zinc oxide; 17 = silicone + foam dressing; 18 = soft polymer dressing; 19 = sugar + egg white; 20 = tripeptide copper gel; 21 = vapour‐permeable dressing. The contributions to the whole network from each direct contrast are given in the last row of the table in Figure 20. The overall risk of bias was high and the largest contributions to the network were from the direct contrasts: foam versus hydrocolloid; collagenase ointment versus sugar plus egg white; and protease‐modulating dressings versus alginate dressings (but all contributions were still 10% or less). The percentage contributions to the mixed treatment contrasts from each direct contrast are shown in Table 14 for the group network contrasts versus basic dressings. We calculated the risk of bias for each contrast in the NMA and results for each mixed treatment contrast are shown in the last column of Table 14. Contributions matrix ‐ group network
Contributions to the whole network are given in the last row. The overall risk of bias was high and the largest contributions to the network were for the direct contrasts: collagenase ointment versus advanced dressings (19.6%); antimicrobial dressings versus advanced dressings (14.9%) and dextranomer versus collagenase ointment (10.5%). 8.1.1. Inconsistency for each contrast (local inconsistency)Firstly, we examined inconsistency factors, comparing results from the direct evidence with those from the indirect evidence for each contrast informed by a loop. We reported results as the ratio of risk ratios (RoRR), with its 90% confidence interval (CI) for the seven loops (Table 15), assuming a common heterogeneity estimate within each loop. At the 90% significance level, there appeared to be inconsistency in the saline gauze‐hydrogel‐phenytoin loop (RoRR 3.90, 90% CI 1,19 to 12.77). The results also suggested some non‐significant potential inconsistency in all other loops (because the 90% CI crosses 2 or 0.5), except for the loop comprising foam, hydrocolloid and hydrogel. In particular, the loop comprising saline gauze‐foam‐hydrogel had potential inconsistency (RoRR 1.64, 90% CI 0.27 to 9.99); this loop also had a fairly high tau² (0.512), suggesting heterogeneity within that loop. We also examined inconsistency factors using the assumption of a common heterogeneity estimate across the network (Veroniki 2013) (Table 15): for the individual network, tau² (network) was 0.043, and the RoRR for the saline gauze‐hydrogel‐phenytoin loop was 3.75 (90% CI 1.14 to 12.29), that is, similar to the original assumption. For this analysis, the 90% CI crossed 0.5 or 2.0 for all loops except foam‐hydrocolloid‐hydrogel. Inconsistency factors ‐ individual network
–Secondly, a node‐splitting approach was taken. The results following node‐splitting for indirect and direct NMA estimates are shown in Table 16, together with the ratio of risk ratios (RoRR) (indirect/direct) with its 90% CI (the 90% significance level was chosen for this test because of its lack of power). The 'indirect' estimate is the result when the NMA is run in the absence of the direct evidence for that contrast. This is only meaningful if the two interventions in the contrast are joined indirectly through the rest of the network; therefore, we report node splitting results for only 12 (of 27) direct contrasts. We made the following observations: Inconsistency: node splitting ‐ individual network
However, all the CIs were wide and there was uncertainty around whether there was inconsistency or not. We also compared inconsistency versus consistency assumptions for each contrast, examining any differences between different designs (3‐arm and 2‐arm), and between inconsistency and consistency NMA results (Table 17). We reported results only for contrasts with two 'core' interventions; there were no differences between models for 'peripheral' contrasts. Only one contrast had more than one type of design (hydrogel versus saline gauze) and there appeared to be large non‐significant differences in the results for different designs. One contrast had non‐significant differences between the NMA results using inconsistency and consistency assumptions (phenytoin versus saline gauze). Inconsistency and consistency NMA results
8.1.2. Inconsistency in the network as a wholeWe conducted both consistency and inconsistency analyses. The latter showed that the six inconsistency parameters (IP) for the individual network were as follows: (1) IP = 0: design is the 3‐arm trial comparing saline gauze versus hydrogel versus phenytoin (2) IP = 0: design is 2‐arm trials comparing foam and hydrocolloid (3) IP = 0: design is 2‐arm trials comparing foam and hydrogel (4) IP = 0: design is 2‐arm trials comparing hydrocolloid and hydrogel (5) IP = 0: design is 2‐arm trials comparing hydrogel and protease‐modulating dressings (6) IP = 0: design is 2‐arm trials comparing hydrogel and iodine‐containing dressings The global Wald test for inconsistency gave: Chi²(6) = 3.59 and P value 0.7314. 8.2.1. Inconsistency for each contrast (local inconsistency)Inconsistency factors are reported as the RoRR with its 90% CI in Table 18. There are two loops, basic dressing ‐ advanced dressing ‐ phenytoin; and basic dressing ‐ advanced dressing ‐ protease‐modulating dressing. At the 90% significance level, there did not appear to be inconsistency in these loops, but the results suggested some non‐significant potential inconsistency in both loops (because the 90% CI crossed 2): for the loop containing phenytoin, the RoRR was 3.04 (0.71 to 13.06) and for the loop containing protease‐modulating dressing the RoRR was 1.36 (0.39 to 4.73). Inconsistency factors ‐ group network
The results following node‐splitting are shown in Table 19 for five (of 11) direct contrasts. The results suggested that there may be inconsistency at the 90% confidence level for two contrasts: phenytoin versus basic dressing (RoRR 12.51, 90% CI 1.87 to 83.55; P = 0.029) and phenytoin versus advanced dressing (RoRR 0.08, 90% CI 0.01 to 0.53; P = 0.029). However, the CIs were wide. Inconsistency: node splitting ‐ group network
We also compared inconsistency versus consistency assumptions for each contrast, examining any differences between different designs (3‐arm and 2‐arm), and between inconsistency and consistency NMA results (Table 20). Results were reported only for contrasts with two 'core' interventions; there were no differences between models for 'peripheral' contrasts. Only one contrast had more than one type of design (advanced dressings versus basic dressings) and there appeared to be large non‐significant differences in the results for different designs. One contrast had non‐significant differences between the NMA results using inconsistency and consistency assumptions (phenytoin versus basic dressings). Inconsistency and consistency NMA results ‐ group network
8.2.2. Inconsistency in the network as a wholeThere were two inconsistency parameters (IP) for the group network: IP = 0: design is the 3‐arm trial comparing basic dressing, advanced dressing and phenytoin The global Wald test for inconsistency gave: Chi²(2) = 4.66 and P‐value 0.0975. This is borderline significant at the 90% confidence level. Data for each intervention were shown as the probability that each intervention is the best, second best, third best treatment, etc. (see Figure 5, and Table 21). There was substantial overlap of the individual rankograms, illustrated in Figure 21, which intentionally shows the confusion, together with some indication that dextranomer and tripeptide copper gel may be the best treatments and that the worst treatments may be the sequential hydrocolloid‐alginate dressings and sugar plus egg white. Across all treatments there was considerable uncertainty in the ranking of interventions and no intervention had more than 50% probability of being the best treatment. This, together with the mean rank being no higher than 3.6 and no lower than 18.6 (out of 21), and no SUCRA value being 0 or 1, reinforces our view of the considerable uncertainty around treatment estimates in this network. Rankograms combined ‐ individual network
Ranks of interventions ‐ individual network
The duration of follow‐up ranged from 3 to 26 weeks, but the distribution was insufficient to allow modelling of time dependence in the network. Seven studies (Alm 1989; Brod 1990; Graumlich 2003; Muller 2001; Payne 2009; Thomas 2005; Xakellis 1992) reported time‐to‐event data. We calculated the hazard ratio using the method and spreadsheet from Tierney 2007; one study (Xakellis 1992) reported the hazard ratio directly, adjusted for exudate level. The time‐to‐healing data are shown in Analysis 3.1 and summary statistics for the time‐to‐healing and the proportion healed are compared in Table 22 for the studies that report both healing outcomes. Analysis Comparison 3 Direct evidence: individual interventions, time‐to‐healing data, Outcome 1 Time‐to‐healing (survival analysis). Direct evidence: comparison of time‐to‐event outcomes and dichotomous data
In the individual network, two studies in 95 participants suggested that the time to healing may have been quicker for hydrocolloid versus saline gauze (HR 1.75, 95% CI 1.00 to 3.05); there was no heterogeneity (unlike the risk ratio). One study in 24 participants suggested healing may have been quicker for collagenase ointment compared with hydrocolloid (HR 2.58, 95% CI 1.00 to 2.65). In the other studies, the CI showed much uncertainty. There was some suggestion of a time dependent effect because there were qualitative and quantitative differences between the HR and the RR: for shorter studies (4‐6 weeks), the HR gave a smaller effect than the RR, but for the medium and longer term studies the HR gave a larger effect than the RR, suggesting that wounds that heal do so relatively quickly. For the group network, time‐to‐healing data are shown in Analysis 4.1. Three studies in 131 participants suggested that the time to healing may have been quicker for advanced dressing versus basic dressing (HR 1.57, 95% CI 0.97 to 2.55); there was no heterogeneity. This compared with the same three studies analysed as the proportion healed: RR (random‐effects) 1.48 (95% CI 0.79 to 2.77); I² = 66%, P = 0.05. Analysis Comparison 4 Direct evidence: group interventions, time‐to‐healing data, Outcome 1 Time‐to‐healing (survival analysis). Page 2 |
Which dressing should the nurse use to protect and absorb moisture when providing care to a patient with a pressure ulcer?
zusammenhängende Posts
Urheberrechte © © 2024 frojeostern Inc.
