Dressings and topical agents for treating venous leg ulcers

Summary of findings for the main comparison. NMA evidence: proportion with complete healing

NMA evidence for base‐case network: proportion with complete healing
Patient or population: people with venous leg ulcers Intervention: dressing or topical agent Comparator: alternative dressing or topical agent Settings: hospital, community or care home, or combinations
Contrasts	Relative effect (95% CI)	*Anticipated absolute effects (95% CI) ‐ from median of control groups in direct evidence**		Certainty of the evidence (GRADE)	Comments
		Median CGR	With intervention
Sucralfate versus nonadherent	RR 6.80 (2.24 to 20.7)	242 per 1000	1000 per 1000 (542 to 1000)	⊕⊕⊝⊝ Low^a,b	Base‐case: RR 17.2 (95% CI 1.52 to 193). Large differences between base‐case and extended base‐case. The calculated absolute effect for the intervention is more than 1000 per 1000 for the point estimate and its upper confidence limit; and so the corresponding values for the absolute risk difference are also approximated by 1000 per 1000.
		1000 more people healed per 1000 (300 to 1000 more)
Sucralfate versus foam	RR 5.94 (1.96 to 18.0)	376 per 1000	1000 per 1000 (737 to 1000)	⊕⊕⊝⊝ Low^a,b	Base‐case: RR 14.8 (95% CI 1.30 to 169) Large differences between base‐case and extended base‐case. The calculated absolute effect for the intervention is more than 1000 per 1000 for the point estimate and its upper confidence limit; and so the corresponding values for the absolute risk difference are also approximated by 1000 per 1000.
		1000 more people healed per 1000 (361 to 1000 more)
Sucralfate versus hydrocolloid	RR 6.51 (2.17 to 19.6)	433 per 1000	1000 per 1000 (940 to 1000)	⊕⊕⊝⊝ Low^a,b	Base‐case: RR 16.24 (95% CI 1.43 to 185) Large differences between base‐case and extended base‐case The calculated absolute effect for the intervention is more than 1000 per 1000 for the point estimate and its upper confidence limit; and so the corresponding values for the absolute risk difference are also approximated by 1000 per 1000.
		1000 more people healed per 1000 (507 to 1000 more)
Silver versus nonadherent	RR 2.43 (1.58 to 3.74)	242 per 1000	588 per 1000 (382 to 905)	⊕⊕⊕⊝ Moderate^a
		346 more people healed per 1000 (140 to 663 more)
Silver versus foam	RR 2.12 (1.46 to 3.07)	376 per 1000	797 per 1000 (549 to 1000)	⊕⊕⊝⊝ Low^c	Direct evidence: Analysis 1.24
		421 more people healed per 1000 (173 to 786 more)
Silver versus hydrocolloid	RR 2.32 (1.58 to 3.41)	433 per 1000	1000 per 1000 (684 to 1000)	⊕⊕⊝⊝ Low^a,d
		567 more people healed per 1000 (251 to 1000 more)
Sucralfate versus silver	RR 2.80 (0.88 to 8.97)	81 per 1000	225 per 1000 (71 to 722)	⊕⊝⊝⊝ Very low^a,e	Base‐case: RR 6.99 (95% CI 0.60 to 82.0) Large differences between base‐case and extended base‐case
		145 more people healed per 1,000 (10 fewer to 642 more)
Foam versus hydrocolloid	RR 1.10 (0.93 to 1.28)	433 per 1000	476 per 1000 (402 to 554)	⊕⊝⊝⊝ Very low^f,g,h	Direct evidence: Analysis 1.18
		43 more people healed per 1000 (from 31 fewer to 121 more)
Foam versus nonadherent dressing	RR 1.15 (0.91 to 1.44)	242 per 1000	278 per 1000 (220 to 348)	⊕⊕⊝⊝ Low^a,h
		36 more people healed per 1000 (from 22 fewer to 106 more)
Hydrocolloid versus nonadherent dressing	RR 1.04 (0.85 to 1.29)	242 per 1000	251 per 1000 (206 to 312)	⊕⊝⊝⊝ Very low^a,h,i	Direct evidence: Analysis 1.6
		9 more people healed per 1000 (from 36 fewer to 70 more)
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparator group and the relative effect of the intervention (and its 95% CI). CGR: control group risk; CI: confidence interval; NMA: network meta‐analysis; RR: risk ratio
GRADE Working Group grades of evidence High certainty (quality): we are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty (quality): we are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty (quality): our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty (quality): we have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
^a. NMA risk of bias from contributions matrix and direct evidence risk of bias (downgrade once) ^b. Imprecision ‐ direct evidence involving sucralfate: 1 study 43/50 events (sucralfate); 5 events (hydrogel) (downgrade once) ^c. Heterogeneity in point estimates for direct evidence; significant inconsistency in node splitting and in inconsistency factor (loop) (downgrade twice) ^d. Significant inconsistency in node splitting and in inconsistency factor (loop) (downgrade once) ^e. Imprecision ‐ CI crosses one MID (1.25) and direct evidence involving sucralfate: 43/50 events (sucralfate) and 5 events (hydrogel) (downgrade twice) ^f. NMA risk of bias from contributions matrix and direct evidence risk of bias (downgrade twice) ^g. Slight heterogeneity in point estimates for direct evidence; significant inconsistency in node splitting and inconsistency factor (downgrade once) ^h. Imprecision ‐ CI crosses one MID (1.25) (downgrade once) ^i. High heterogeneity in direct evidence (downgrade twice)

Background

available in

Description of the condition

Venous leg ulcers are common and recurring complex wounds that heal by secondary intention (that is by the growth of new tissue rather than by primary closure). Problems with the leg veins (such as damage to the valves, or blockages) reduce the efficient return of blood to the heart and increase the pressure in the veins (Ghauri 2010), which may result in venous leg ulcers. The precise chain of events that links high venous pressures (chronic venous hypertension) with skin breakdown and a chronic wound is not fully understood (Coleridge Smith 1988; Valencia 2001).

Venous leg ulcers commonly occur on the gaiter region of the lower leg (from just below the ankle up to mid‐calf ). A venous leg ulcer is defined as any break in the skin that has either been present for longer than six weeks or occurs in a person with a history of venous leg ulceration. Differential diagnosis of the type of leg ulcer (i.e. the underlying cause) is made by taking a clinical history, physical examination, laboratory tests and haemodynamic assessment (RCN 2013; SIGN 2010). True venous ulcers are moist, shallow and irregularly shaped and lie wholly or partly within the gaiter area of the leg. Leg ulcers can be associated with venous disease in combination with vascular disease, which impairs arterial blood supply; in these instances they are said to have a 'mixed' aetiology (to have more than one cause). Open skin ulceration due solely to limb ischaemia from vascular disease is less common.

Accurate, current estimates of leg ulcer prevalence are hard to identify because most surveys do not differentiate between causes of leg ulceration, or do so per limb but not per person (Moffatt 2004; Srinivasaiah 2007; Vowden 2009b). Estimates of the prevalence of open leg ulceration (any cause) range from 4 to 48 cases per 10,000 (Graham 2003; Johnson 1995; Walker 2002), with the point prevalence of venous leg ulceration in Australian and European studies being between 10 per 10,000 and 30 per 10,000 (Nelzen 2008). A recent estimate suggests that venous ulceration has a point prevalence of 2.9 cases per 10,000 in the United Kingdom (UK), whilst mixed arterial/venous leg ulceration has a point prevalence of 1.1 per 10,000 (Hall 2014).

Venous disease is a chronic condition which can be characterised by periods of ulceration (i.e. an open wound) followed by healing and then recurrence. An early cross‐sectional survey reported that half of current or recent ulcers had been open for up to nine months and that 35% of people with leg ulcers had experienced four or more episodes (Callam 1987b). This picture was supported by a subsequent cross‐sectional study (Nelzen 1994). More recent analysis of almost 1200 people with venous leg ulcers documented a 24‐week healing rate of 76% and a recurrence at one year of 17% (Gohel 2005).

Venous ulcers are painful, can be malodorous and prone to infection, and may severely affect people's mobility and quality of life. The presence of leg ulceration has been associated with pain, restriction of work and leisure activities, impaired mobility, sleep disturbance, reduced psychological well‐being and social isolation (Herber 2007; Maddox 2012; Persoon 2004). In severe cases, ulceration can lead to limb amputation, although this may be more common in people with comorbid arterial insufficiency (Dumville 2009; Nelzen 1997; Valencia 2001). Recent research suggests that people with complex wounds, including those with venous leg ulcers, commonly see complete wound healing as the most important outcome to them (Cullum 2016; Madden 2014).

The financial cost of treating an unhealed leg ulcer in the UK has most recently been estimated at around GBP 1700 per year (price year 2012) (Ashby 2014). An earlier evaluation estimated the average cost of treating a venous leg ulcer in the UK (based on costs for material for dressing changes) as between EUR 814 and EUR 1994 and, in Sweden as lying between EUR 1332 and EUR 2585 (price year 2002), with higher costs associated with larger and more chronic wounds (Ragnarson 2005). In Bradford, UK, GBP 1.69 million was spent on dressings and compression bandages, and GBP 3.08 million on nursing time (estimates derived from resource use data for all wound types) during the financial year 2006 to 2007 (Vowden 2009a). Data from a German study, which estimated total costs including those classified as indirect or intangible costs, estimated mean annual costs of leg ulcers as EUR 9060 per patient (price year 2006). This figure is higher than other estimates because it includes non‐health service costs to the patient and to society (Augustin 2012). These data are all derived from high‐income countries and thus may not be a true reflection of costs elsewhere, which may be higher or lower.

Description of the intervention

The review includes all dressings and topical agents applied directly onto or into wounds and left in situ. This contrasts with products used to irrigate, wash or cleanse wounds and that are only in contact with wounds for a short period. First‐line treatment for venous leg ulcers is compression therapy in the form of bandages, stockings or mechanical devices (Nelson 2014; O'Meara 2012). This application of external pressure around the lower leg assists venous return and reduces venous reflux (Woo 2013). We therefore anticipated that wound dressings would commonly be used in combination with compression therapy.

Dressings are widely used in wound care with the aim of protecting the wound and promoting healing by influencing the local wound environment (Bradley 1999), typically by physical means, such as thermal insulation, absorption of exudate and physical protection. Dressings may also have pharmacological, immunological or metabolic actions. Topical agents include hydrogel gels, ointments and creams that are placed in contact with the wound and left in situ.

Dressings

The classification of dressings usually depends on the key material used in their construction, and whether additional substances are added to the dressing. Several attributes of an ideal wound dressing have been described (BNF 2016), including the ability of the dressing to:

absorb and contain exudate without leakage or strike‐through, in order to maintain a wound that is moist but not macerated;
achieve freedom from particulate contaminants or toxic chemicals left in the wound;
provide thermal insulation, in order to maintain the optimum temperature for healing;
allow permeability to water, but not bacteria;
optimise the pH of the wound;
minimise wound infection and avoid excessive slough;
avoid wound trauma on dressing removal;
accommodate the need for frequent dressing changes;
provide pain relief; and
be comfortable.

There is a wide range of types of dressings available which may be used for treating wounds including venous leg ulcers; some of these and their properties are described below (BNF 2016). Impregnated dressings may have a range of bases, such as foams or alginates.

Absorbent dressings are applied directly to the wound and may be used as secondary absorbent layers in the management of heavily exuding wounds. Examples include Primapore (Smith & Nephew); this can be lifted off at dressing removal, or removed by irrigation. Bonding to a secondary viscose pad increases absorbency. Examples include: Curasorb (Covidien), SeaSorb (Coloplast) and Sorbsan (Unomedical).

Capillary‐action dressings consist of an absorbent core of hydrophilic fibres held between two low‐adherent contact layers. Examples include: Advadraw (Advancis) and Vacutex (Protex).

Permeable film and membrane dressings are permeable to water vapour and oxygen, but not to water or micro‐organisms. Examples include Tegaderm (3M) transparent film and OpSite (Smith & Nephew).

Foam dressings contain hydrophilic polyurethane foam and are designed to absorb wound exudate and maintain a moist wound surface. There are a variety of versions and some include additional absorbent materials, such as viscose and acrylate fibres, or particles of superabsorbent polyacrylate, which are silicone‐coated for non‐traumatic removal. Examples include: Allevyn (Smith & Nephew), Biatain (Coloplast) and Tegaderm (3M) foam adhesive and non‐adhesive dressings.

Honey‐impregnated dressings contain medical‐grade honey that is purported to have antimicrobial and anti‐inflammatory properties and can be used for acute or chronic wounds. Examples include: Medihoney (Medihoney) and Activon Tulle (Advancis).

Hydrocolloid dressings are usually composed of an absorbent hydrocolloid matrix on a vapour‐permeable film or foam backing. Examples include: Granuflex (ConvaTec) and NU DERM (Systagenix). Fibrous alternatives that resemble alginates and are not occlusive have also been developed: Aquacel (ConvaTec).

Iodine‐impregnated dressings release free iodine, which is thought to act as a wound antiseptic when exposed to wound exudate. Examples include Iodoflex (Smith & Nephew) and Iodozyme (Insense).

Low‐adherence dressings and wound contact materials usually consist of cotton pads that are placed directly in contact with the wound. They can be non‐medicated (e.g. paraffin gauze dressing, saline gauze dressing) or medicated (e.g. containing povidone iodine or chlorhexidine). Examples include paraffin gauze dressing, BP 1993 and Xeroform (Covidien) dressing ‐ a nonadherent petrolatum blend with 3% bismuth tribromophenate on fine mesh gauze.

Odour‐absorbent dressings contain charcoal and are used to absorb wound odour. Often this type of wound dressing is used in conjunction with a secondary dressing to improve absorbency. An example is CarboFLEX (ConvaTec).

Other antimicrobial dressings are composed of a gauze or low‐adherent dressing impregnated with an ointment thought to have antimicrobial properties (e.g. chlorhexidine gauze dressing (Smith & Nephew)). Alternatively, a dressing such as Cutimed Sorbact (BSN Medical) uses a hydrophobic layer to bind micro‐organisms to the dressing surface, allowing them to be removed from the wound when the dressing is changed.

Protease‐modulating matrix dressings alter the activity of proteolytic enzymes in chronic wounds. Examples include: Promogran (Systagenix).

Silver‐impregnated dressings are used to treat infected wounds, as silver ions are thought to have antimicrobial properties. Silver versions of most dressing types are available, including silver impregnated dressings (e.g. silver hydrocolloid etc). Examples include: Acticoat (Smith & Nephew) and Urgosorb Silver (Urgo).

Soft polymer dressings are composed of a soft silicone polymer held in a nonadherent layer; these are moderately absorbent. Examples include: Mepitel (Mölnlycke) and Urgotul (Urgo).

Topical agents

The following types of topical agents are considered as interventions in this review.

Cadexomer‐iodine paste consists of a water‐soluble, modified starch polymer containing iodine. It releases free iodine when exposed to wound exudate. The free iodine acts as an antiseptic on the wound surface, and the cadexomer absorbs wound exudate and encourages de‐sloughing. Examples include: Iodosorb (Smith & Nephew) ointment and powder.

Collagenase‐containing ointment is an enzymatic debriding ointment. Collagenase is thought to digest collagen in necrotic tissue and to contribute to granulation and epithelialisation (the final stage of wound healing).

Hydrogels consist of a starch polymer and up to 96% water. They can absorb wound exudate or rehydrate a wound depending on the wound moisture levels. Hydrogels are often considered to be dressings, but are also topical in nature. They are supplied in either flat sheets, an amorphous hydrogel or as beads. Examples include: ActiformCool (Activa) and Aquaflo (Covidien).

Topical phenytoin is thought to promote wound healing by a number of mechanisms, including stimulation of fibroblast proliferation, facilitation of collagen deposition and antibacterial activity.

Silver sulfadiazine cream is a topical antimicrobial cream that is used to treat and prevent infection in wounds by damaging bacterial cell membranes. Examples include Flamazine (Smith & Nephew) and Silvadene (Pfizer).

We did not consider studies evaluating any products containing growth factors, platelet‐rich plasma or other platelet‐derived products and colony‐stimulating factors.

How the intervention might work

Animal experiments conducted over 40 years ago suggested that acute wounds heal more quickly when their surfaces are kept moist rather than left to dry and scab (Winter 1962; Winter 1963a; Winter 1963b). A moist environment is thought to provide optimal conditions for the cells involved in the healing process with faster revascularisation (Dyson 1992), and development of granulation tissue (Svensjö 2000), as well as allowing autolytic debridement (removal of dead tissue by natural processes), which is thought to be an important part of the healing pathway (Cardinal 2009).

The desire to maintain a moist wound environment is a key driver for the use of wound dressings and related topical agents. Whilst a moist environment at the wound site has been shown to aid the rate of epithelialisation in superficial wounds, excess moisture at the wound site can cause maceration (breakdown) of the surrounding skin (Cutting 2002), and it has also been suggested that dressings that permit fluid to accumulate might predispose wounds to infection (Hutchinson 1991). Wound treatments vary in their level of absorbency, so that a very wet wound can be treated with an absorbent dressing (such as a foam dressing) to draw excess moisture away and avoid skin damage, whilst a drier wound can be treated with a more occlusive dressing or a hydrogel to maintain a moist environment.

Some dressings are now also formulated with an 'active' ingredient (e.g. silver, honey or protease modulators).

Why it is important to do this review

Venous leg ulcers are a relatively common type of complex wound that have a negative impact on people’s lives and incur high costs for health services and society. Leg ulcers are painful, sometimes malodorous, prone to infection, and may severely affect people's mobility and quality of life, and in severe cases, there is a risk of limb amputation. There are a number of treatments for venous leg ulcers, but many ulcers prove hard to heal, although healing is a key outcome for patients.

We conducted an open consultation with consumers to ask them which treatments for treating venous leg ulcers they would like to see considered. Respondents self‐selected through their response to a short questionnaire posted on the Cochrane Wounds website and Facebook page. Although some identified compression as the main consideration, others mentioned specific types of dressings. These included many of the dressing types listed in Description of the intervention, including charcoal‐containing (odour‐absorbing) dressings, dressings designed to reduce formation and presence of biofilms (bacteria which grow on a surface to form a film of cells) and dressings with antimicrobial properties and debriding actions. Also specifically identified as being of interest was Unna's boot; a specialised dressing which consists of gauze wraps impregnated with zinc oxide and calamine, sometimes in combination with other agents.

The diversity of dressings and related materials available to health professionals for treating venous leg ulcers makes evidence‐based decision‐making difficult when determining the optimum treatment regimen for a particular patient (NICE 2016a). With increasingly sophisticated technology being applied to wound care, practitioners need to know the relative effectiveness and cost‐effectiveness of these sometimes expensive dressings. Even where cost is not an issue, the most effective treatment may not be available (e.g. in some developing countries) or may be difficult or to use, so that information on the second and third best treatments is important too (Salanti 2011).

There are a number of existing or ongoing evidence syntheses on venous leg ulcer treatments, including Cochrane reviews of different types of dressings or topical treatments (Briggs 2012; O'Meara 2013; O'Meara 2014; O'Meara 2015; Ribeiro 2013; Ribeiro 2014; Westby 2016). There are also wider reviews of particular types of treatment for all wound types which include data on venous leg ulcers for treatments such as honey, silver, aloe Vera, and phenytoin (Dat 2012; Jull 2015; Shaw 2007; Vermeulen 2007). Other reviews on non‐healing or chronic ulcers have also included a substantial number of relevant trials (Greer 2013; AHRQ 2013), and there are also older general reviews (e.g. Bouza 2005; O'Donnell 2006).

Guidance drawing on reviews available at the time has also been published (Robson 2006; SIGN 2010). The SIGN 2010 guideline recommended that low‐adherent dressings be used routinely but that alternative dressings (hydrocolloids, alginates or hydrogels) may be considered to assist with pain, exudate and slough respectively. Earlier guidance (Robson 2006), recommended that maintaining a moist wound environment be prioritised in dressing choice. Most recently the UK National Institute for Health and Care Excellence (NICE) issued advice on the use of advanced and antimicrobial dressings for chronic wounds including venous leg ulcers (NICE 2016b). This updated the SIGN 2010 guidance to include the findings of the most recent systematic reviews.

However, despite the existence of high‐quality recent systematic reviews, there is insufficient evidence to support the use of any particular type of advanced or antimicrobial dressing or treatment as the direct evidence is of low certainty and no network meta‐analysis (NMA) has previously been undertaken in this area. Decision‐makers currently have to consider the findings of a plethora of pairwise randomised controlled trials (RCTs) simultaneously and to make qualitative judgements across these in the face of uncertainty, when considering the evidence on dressing use.

NMA is the simultaneous comparison of linked, multiple, competing treatments in a single statistical regression model (Caldwell 2005; Lu 2004; Salanti 2008). NMA utilises evidence from both 'direct' (head‐to‐head or 'pairwise') comparisons (e.g. trials directly comparing treatments A and B) and 'indirect' comparisons (e.g. the combination of trials comparing A with C and trials comparing B with C). If both direct and indirect estimates are available, they can be meta‐analysed, preserving within‐trial randomisation (Grant 2013; Thorlund 2012; Tu 2012).

Where there are relevant common comparators, NMA produces a set of effect estimates for each treatment linked into the network, relative to every other, whether or not they have been compared in head‐to‐head trials: thus, NMA is a method of obtaining estimates for comparisons for which there is no (direct) trial evidence. Even when direct evidence is available there may not be much of it, so pooling it with data from indirect comparisons generally gives more robust evidence and reduces uncertainty in the estimates of effect (Higgins 1996; Thorlund 2012). It is also possible to calculate the probability of one treatment being the best for a specific outcome, reflecting the precision surrounding the estimates (Caldwell 2014; Salanti 2011).

A glossary of NMA terms is given in Appendix 1.

This review comprised a network meta‐analysis (NMA) for the outcome of venous leg ulcer healing, for alternative dressings and topical agents for the treatment of venous leg ulcers. We drew on methods previously used in related work (Soares 2014; Westby 2017). The NMA was expected to enable us to determine which (if any) dressing or topical agent is the most effective for healing venous leg ulcers, taking into account direct and indirect evidence simultaneously. We also presented uncertainty around treatment estimates, and explored assumptions being made in the analysis.

Objectives

available in

Methods

available in

Criteria for considering studies for this review

Types of studies

We included published and unpublished randomised controlled trials (RCTs), irrespective of language of report. We only included cross‐over trials that reported outcome data at the end of the first treatment period and prior to cross‐over. We excluded studies using quasi‐random methods of allocation (such as alternation). We highlighted trials in which three or more interventions were randomised and included all relevant arms.

Types of participants

We included trials recruiting adults (aged at least 18 years) described as having venous leg ulcers, managed in any setting. We accepted study authors' definitions of venous leg ulcers. Where wounds were described only as "leg ulcers" without information as to aetiology, we assumed that they were venous in origin. Trials in which a minority of leg ulcers are described as having a mixed or arterial pathology were included provided that these were fewer than 25% of participants. Trials including other types of mixed wound populations were not included. We included participants at any stage of their treatment process ‐ for example, participants with or without ulcers described as being hard to heal or clinically infected.

Types of interventions

The interventions evaluated are all those that can be directly applied as dressings or topical agents to open venous leg ulcers. We presented results for these interventions and included them in summary tables. In the context of a network of competing treatments, there are no 'comparators'. We used the term 'comparison' to mean two interventions compared in a single study and the term 'contrast' to mean two interventions compared across all studies with that comparison. A contrast may be represented by a single study, a simple direct meta‐analysis or by the NMA.

We considered trials for which at least one of the interventions was (1) any dressing, including impregnated dressings or saline‐moistened dressings or combination dressings*, or (2) any topical agent applied directly to an open venous leg ulcer and left in situ. The treatment of interest had to be the only systematic difference between treatment groups. We did not take into account secondary dressings. We also considered 'no dressing' as a valid intervention, where the wound is left open/covered only by compression bandaging.

* 'combination dressings' means two or more dressings applied sequentially over time (e.g. hydrocolloid for four weeks followed by alginate for four weeks), or a product containing two or more types of dressing material (e.g. a multilayer product comprising silicone polymer and hydrocolloid).

Some of the interventions we considered are as follows; we used the categories listed below as the basis for grouping the treatments used in individual studies:

basic wound contact dressings (includes low‐adherence (including paraffin gauze) or absorbent dressings (of any absorbency));
saline‐moistened gauze (all degrees of moistness);
hydrogel dressing (includes hydrogel sheet or hydrogel application (amorphous) or sodium hyaluronate);
vapour‐permeable films and membranes (includes adhesive film (semi‐permeable) or adhesive film with absorbent pad);
soft polymer dressings (with/without absorbent pad or cellulose);
hydrocolloid dressing (with/without adhesive border or matrix hydrocolloid);
fibrous (spun) hydrocolloid;
foam dressings (all absorbencies);
alginate dressings;
capillary action dressings;
alginate dressing with charcoal;
other charcoal‐containing dressing;
honey sheet dressing or topical honey;
cadexomer Iodine ointments;
iodine‐containing dressings;
soft polymer dressing (with silver);
hydrocolloid (with silver);
foam dressings (with silver);
alginate dressings (with silver);
silver sulfadiazine (SSD) cream;
protease‐modulating matrix (PMM) dressings;
collagenase‐containing ointment;
topical phenytoin;
topical zinc oxide;
no dressing (wound left exposed); and
other treatments considered by the review team (with additional clinical advice where required) to be dressings or topical agents applied directly to the wound and left in situ.

The following interventions were excluded from evaluation: treatments in which dressings were attached to external devices such as negative pressure wound therapies, skin grafts, growth factor treatments, platelet gels and larval therapy. We also excluded interventions which, although topical, are not delivered as a physical presence (liquid or solid) on the wound surface such as oxygen, ultrasound, laser or radiant heat therapies. These treatments were considered to be outside the scope of a review focused on dressings and topical treatments used in place of dressings. Where studies compared an eligible with an ineligible intervention we included them if they usefully linked the network of studies evaluating two eligible treatments. Data from these linking studies were fully extracted and they were assessed for risk of bias. Studies which evaluated only one eligible intervention and did not perform this linking function were treated as excluded studies and are clearly identified in the list of excluded studies (Characteristics of excluded studies). Where studies used a placebo comparator for an eligible intervention, we included them and treated the placebo as being the vehicle used to deliver it; for example as an emollient cream, an inactive powder or a hydrogel. For example, a comparison of a cream containing an antibiotic with a placebo would be treated as a comparison of topical antibiotic with an emollient cream.

We grouped together dressings in the same class, for example, all hydrocolloid dressings were grouped together regardless of whether they were adhesive or non‐adhesive (BNF 2016). This grouping was regardless of a particular brand's stated absorbency, size, concentration of active component or degree of moistness. Thus, where studies only compared two dressings from the same class (for example, two alginates or two foam dressings), we excluded them from the review as they contributed no information about the effectiveness of the class. We considered an impregnated dressing to be in a different class from a non‐impregnated dressing. Judgements about whether particular dressings belonged to the same class were made on the basis of British National Formulary (BNF) classifications (BNF 2016), and clinical expert advice where there was remaining uncertainty. Evidence from comparisons between dressings of the same class can be found in the individual Cochrane reviews of particular types of dressings. Trials of this type are also identified as such in the list of excluded studies.

We anticipated that the great majority of participants would be treated with concurrent compression therapy and noted the type of compression therapy used. We also included any RCT in which other concurrent therapies were given (e.g. antibiotics, debridement), provided that these treatments were delivered in a standardised way across the trial arms of the individual trial (such that the treatment of interest is the only systematic difference). We did not treat separately comparisons with and without concurrent therapies, that is, we considered intervention 1 + concurrent therapy versus intervention 2 + concurrent therapy to be the same as intervention 1 versus intervention 2.

We assumed that the interventions are exchangeable, that is, participants in the network could, in principle, be randomised to any of the treatments being compared. For example, that a person with a venous leg ulcer could be equally likely to be randomised to a silver dressing, a polyurethane foam dressing, honey or saline gauze. Depending on the wound requirements for the dressing (e.g. highly absorbent), this may not always be a good assumption for individual wounds, but may be reasonable across the population in the trials.

Types of outcome measures

We reported outcome measures at the last time point available (assumed to be at the end of follow‐up if not specified) and the time point specified in the methods as being of primary interest (if this was different from latest time point available). Initially, we noted when studies reported results at other time points, or whether they included Kaplan‐Meier plots, or both.

Primary outcomes

The primary outcome for this review is complete wound healing.

We regarded the following as providing the most relevant measures of outcome for the analyses:

the proportion of wounds healed (frequency of complete healing: arm‐level data);
time to complete healing (survival data: study‐level data reported as a hazard ratio (HR) with standard error (SE)).

We accepted the authors' definitions of what constitutes a healed wound.

Secondary outcomes

We did not consider any secondary outcomes here, however they are considered in other relevant reviews (Briggs 2012; O'Meara 2013; O'Meara 2014; O'Meara 2015; Westby 2016) and ongoing reviews (Ribeiro 2013; Ribeiro 2014).

Search methods for identification of studies

Electronic searches

We searched the following electronic databases to identify reports of relevant randomised clinical trials:

Cochrane Wounds Specialised Register (searched 29 March 2017);
Cochrane Central Register of Controlled Trials (CENTRAL; 2017, Issue 2) (searched 29 March 2017);
Ovid MEDLINE (1946 to 29 March 2017);
Ovid MEDLINE (In‐Process & Other Non‐Indexed Citations, to 29 March 2017);
Ovid Embase (1974 to 29 March 2017);
EBSCO CINAHL Plus (1937 to 29 March 2017).

The search strategies for the Cochrane Wounds Specialised Register, CENTRAL, Ovid MEDLINE, Ovid Embase and EBSCO CINAHL Plus can be found in Appendix 2. We combined the Ovid MEDLINE search with the Cochrane Highly Sensitive Search Strategy for identifying randomised trials in MEDLINE: sensitivity‐ and precision‐maximising version (2008 revision) (Lefebvre 2011). We combined the Embase search with the Ovid Embase filter developed by the UK Cochrane Centre (Lefebvre 2011). We combined the CINAHL searches with the trial filters developed by the Scottish Intercollegiate Guidelines Network (SIGN 2018). There were no restrictions with respect to language, date of publication or study setting.

An updated search was conducted on 16 March 2018; these results have been added to Studies awaiting classification and Ongoing studies, and will be incorporated into the review at the next update.

Searching other resources

We tried to identify other potentially eligible trials or ancillary publications by searching the reference lists of retrieved included studies as well as relevant systematic reviews, meta‐analyses, guidelines and health technology assessment reports. We used any additional unpublished data for included studies obtained by previous reviews, contacting review authors where appropriate, and undertook cross‐checking to ensure that all relevant studies with evaluable outcome data were included.

Data collection and analysis

Data collection and analysis were carried out according to methods stated in the published protocol (Norman 2017), which were based on the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a).

Selection of studies

Two review authors independently assessed the titles and abstracts of the citations retrieved by the searches for relevance. After this initial assessment, we obtained full‐text copies of all studies considered to be potentially relevant. Two review authors independently checked the full papers for eligibility; disagreements were resolved by discussion and, where required, the input of a third review author. Where required and possible, we attempted to contact study authors where the eligibility of a study was unclear. We recorded all reasons for exclusion of studies for which we had obtained full‐text copies. We completed a PRISMA flowchart to summarise this process (Liberati 2009).

Where studies were reported in multiple publications/reports we sought to obtain all publications. Whilst the study was included only once in the review, we extracted data from all reports to ensure maximal relevant data were obtained.

Data extraction and management

We extracted the following information from each included study:

interventions being compared, including any ineligible interventions randomised to additional trial groups;
duration of the intervention;
details of any co‐interventions;
unit of randomisation (e.g. participant or ulcer);
number of ulcers per person;
unit of analysis (including any selection methods for people with multiple ulcers);
number of participants in each arm;
hazard ratio (HR) and its 95% confidence interval (CI) (or any data that will allow its calculation (Parmar 1998; Tierney 2007)) for comparisons between arms);
number of participants who healed in each arm, both at the latest time point and (if different) at another time specified as of primary interest in the study's methods section;
all other follow‐up times reported;
if a Kaplan Meier plot is displayed;
missing data rates per arm, and reasons for 'missingness', including the number of people dying.

Data on potential effect modifiers

We are not aware of any population‐specific effect modifiers for this research question: there is no existing evidence to suggest that one type of dressing works better than another for certain subgroups, such as different baseline ulcer characteristics (e.g. size and duration of ulcer), although it may be the case that some dressings are evaluated only in particular groups (e.g. those classed as having 'hard‐to‐heal' ulcers).

However, we extracted from each included study data that may act as effect modifiers (in this context):

type of funding (e.g. industry, academic, government); this was grouped into not‐for‐profit and other where reported;
risk of bias; this was classed as low or unclear, high or very high.

We did not give more weight to any individual domains of the 'Risk of bias' assessment.

Other data

We also extracted the following baseline and study data, reporting separately for each intervention arm if possible:

care setting;
age of participants;
duration of leg ulcer(s);
size of venous leg ulcer(s) (area/volume);
wound status (e.g. sloughy, necrotic, infected, 'hard‐to‐heal').

Assessment of risk of bias in included studies

We assessed risk of bias for each included study, and calculated separately the overall risk of bias for each direct pairwise meta‐analysis for the complete healing data. Two review authors independently assessed included studies using the Cochrane tool for assessing risk of bias (Higgins 2011b); a third review author was consulted where consensus could not be reached. The Cochrane risk of bias tool addresses six specific domains: sequence generation, allocation concealment, blinding of outcome assessors, incomplete outcome data, selective outcome reporting and other issues (Appendix 3). We then summarised data for the key biases reflected by these domains: selection bias, detection bias, attrition bias, reporting bias and other bias. We also noted the comparability of participant characteristics at baseline across the two groups, including whether an adjusted analysis was conducted. We used these data to help inform decisions on the risk of selection bias. For the category of "other bias" we paid particular attention to unit of analysis errors since they are highly prevalent in wounds research. We recorded all problems of unit of analysis, for example, where participants with multiple wounds were randomised and each of their wounds contributed outcome data.

We interpreted the overall risk of bias for each contrast of the network meta‐analysis, drawing on both indirect and direct data (see the section on Quality Assessment of Evidence (GRADE 2013), below).

Overall risk of bias and linking to GRADE assessment

In order to link these Cochrane risk of bias ratings to the GRADE assessment for study limitations (downgrading 0, 1 or 2 times), we used a two‐stage process. Firstly, we obtained an all‐domain (overall) risk of bias classification for each study and then we used this to produce an overall risk of bias for each contrast.

All‐domain risk of bias for each study

We summarised data for each of the key domains of selection bias, detection bias, attrition bias, reporting bias and other bias, assigning one of four ratings: low, unclear, high and very high. For example, selection bias was informed by sequence generation, allocation concealment and comparability of baseline characteristics.

In an adaption of the GRADE approach (Guyatt 2011), we produced an all‐domain risk of bias, with four ratings defined as:

'very high' ‐ two or more key domains with a high risk of bias or a single domain with very high levels of uncertainty (e.g. very high degree of differential missing data);
'high' ‐ high risk of bias for any one domain or 'almost high' risk of bias across more than one domain;
'low' ‐ low risk of bias for each of the key domains;
'unclear' ‐ insufficient information for at least one key domain (with the other domains being at low risk of bias).

We included this all‐domain risk of bias in the summary 'Risk of bias' figure, by adding additional columns to the 'Risk of bias' figure for each study. For the purposes of the GRADE assessment, we then grouped together studies with low and unclear all‐domain risks of bias.

Overall risk of bias for a direct comparison (the comparison of two intervention in one or more trials)

Where a single study contributed to a comparison, the overall risk of bias was that of the all‐domain risk of bias assigned to that study. Where more than one study contributed to a comparison, we assigned an overall comparison risk of bias by calculating a weighted average based on the inverse variance‐derived weights from the meta‐analysis, and using this in conjunction with the overall risk of bias (where numerical values were assigned to the all‐domain ratings for each study: low/unclear (1), high (2) and very high (3)). We aligned comparison 'Risk of bias' assessment with the GRADE categories of no limitations (not downgraded for risk of bias), serious limitations (downgraded once), and very serious limitations (downgraded twice) (Guyatt 2011; Salanti 2014). We presented the overall risk of bias associated with each direct estimate in a network diagram using colours to represent different ratings.

Overall risk of bias in the network

Each direct contrast in the network contributed differently to the estimation of each NMA summary effect (each NMA comparison). The contribution of each piece of indirect evidence to a mixed treatment contrast depends on its point estimate, precision and relative location within the network, and on that of any direct evidence or other indirect evidence (Chaimani 2013; Salanti 2014). A recently published tool, Krahn 2013, allows the contribution of each direct estimate to be determined for each contrast in the network informed by mixed evidence (direct and indirect), or when multiple loops of indirect evidence inform the same link. We used the CINeMA web tool (CINeMA 2017) to calculate the percentage contribution of each direct contrast to each network estimate. The overall risk of bias for each NMA comparison estimate is a composite measure of the risks of bias for all the direct contrasts contributing to that NMA comparison and was determined by calculating a weighted average risk of bias using the percentage contributions and the all‐domain risks of bias for all the direct contrasts. We acknowledge that this approach returns approximate weights.

Measures of treatment effect

Relative treatment effects

We were not able to calculate the hazard ratio (HR) for the majority of studies, and therefore presented the risk ratio (RR) (95% CI) for the proportion of people healed. In order to conduct these analyses (see Data synthesis), we used outcome data reported in individual studies, as raw data at the latest time point, unless otherwise stated. If there had been sufficient data, we had planned to calculate the HR with 95% CI and to model time‐to‐event data.

Unit of analysis issues

We expected the main unit of analysis issues to occur when participants had more than one wound per person. We treated the participant as the unit of analysis when the number of wounds assessed appeared equal to the number of participants (e.g. one wound per person). This included studies in which participants were randomised to treatments and there was more than one wound per person, but results were reported for one selected wound; we considered whether there was risk of bias in the selection process.

Where studies randomised at the participant level, we used the allocated treatment on multiple wounds per participant, and measured and analysed outcomes at the wound level, (e.g. wound healing), there were unit of analysis issues if the data were not correctly analysed. In these cases, we assessed whether it was possible and appropriate to approximate the correct analyses in accordance with Chapter 16 of the Cochrane Handbook for Systematic Reviews of Interventions, using information adapted from Higgins 2011c. Where this was not possible, we made a decision about inclusion of data in the analysis, and recorded these studies as being at high risk of bias if the number of participants and the mean number of wounds per person were judged to warrant this.

If cluster‐randomised trials had been identified, we would have decided the analytical approach based on the type and volume of cluster data. We accounted for the correlation between the effect sizes from multi‐arm studies in the analysis.

Dealing with missing data

It is common to have data missing from trial reports. Excluding participants post‐randomisation, or ignoring those participants who withdraw from the trial or are lost to follow‐up, compromises the randomisation and potentially introduces bias into the trial. Where there were missing data for the primary outcome of proportion of ulcers healed, we assumed participants did not have the outcome (i.e. they will be considered in the denominator but not the numerator). We considered examining this assumption in a sensitivity analysis but decided this was not necessary given the small numbers of trials with differences in attrition between treatment groups.

Assessment of heterogeneity

Assessment of clinical and methodological heterogeneity within treatment comparisons

We assessed the presence of clinical heterogeneity within each pairwise comparison (i.e. the degree to which studies vary in terms of participant, intervention and outcome characteristics) by comparing data extracted for included studies. We focused on key variables that are potential effect modifiers, such as whether studies were at high risk of bias in key domains and the source of funding for the study. We also considered the generalisability of our findings with reference to participant characteristics such as ulcer size and duration.

Assessment of transitivity across treatment comparisons

'Transitivity' refers to the situation in which an intervention effect measured using an indirect comparison is valid and equivalent to the intervention effect measured using a direct comparison. Thus, where there are differences in effect modifiers across comparisons, the transitivity assumption may not be met and there will be inconsistency in the network (Grant 2013; Jansen 2013). We did not identify any potential effect modifiers from the literature, and therefore had to assume that there is transitivity with respect to known effect modifiers across the pairwise comparisons. There are also limited underlying theoretical reasons to consider effect modification for these treatments ‐ however, in preparing the network we explored the effect of differences in risk of bias as possible effect modifiers across the network. We investigated inconsistency in the network (see Data synthesis).

We had also planned to investigate the effect of funding source as a potential effect modifier. However although many studies reported funding by a manufacturer of one of the assessed interventions, a substantial number of studies did not report the funding source. Only a minority of trials clearly reported a third sector or public funding source; a much smaller number reported non‐industry funding or a mixture of industry and non‐industry sources. In view of this imbalance and the high level of uncertainty around trials which did not report funding sources we did not attempt this analysis.

Assessment of reporting biases

We assessed the presence of reporting bias using a contour‐enhanced funnel plot, (Peters 2008; Salanti 2014).

Data synthesis

General methods

We performed pairwise meta‐analyses in a frequentist framework using the statistical software STATA 13 (STATA 2011; Salanti 2014). Experience (Westby 2017) suggested that there were likely to be insufficient data for us to model the impact of follow‐up duration on estimates of effect. We therefore conducted analyses based on binary data, analysed using risk ratios (RRs). We had planned to extract or calculate HRs where possible using established methods (Parmar 1998; Tierney 2007), and would have considered modelling the hazard function (Dias 2014; Soares 2014) using WINBUGS (WinBUGS 2016). However, there were insufficient HR data.

We used STATA 13 (STATA 2011) to calculate the contributions matrix for the network and used the results of this together with the evaluation of risk of bias (see Assessment of risk of bias in included studies) to inform a GRADE evaluation for the entire network (Salanti 2014). We summarised the findings according to GRADE principles (GRADE 2013; Schünemann 2011a; Schünemann 2011b). Where there were zero events in any trial arm, we followed the general approach taken by STATA and added 0.5 to the numerator and 1 to the denominator for each arm in the trial.

Methods for standard meta‐analysis

We performed pairwise meta‐analyses in a frequentist framework using Review Manager 5 (RevMan 2014) or STATA 13 (STATA 2011) as appropriate, using inverse variance weighting and a random‐effects model, and only analysing trials reporting that pairwise comparison. We also presented the data for these direct comparisons from the network in forest plots (Schünemann 2011a); for reasons of space we did not present all possible comparisons. While we report treatment effects for all data (see appendices), we focus on discussing selected comparisons chosen for their clinical relevance.

Methods for network meta‐analysis

We used STATA 13 to produce a network diagram based on all included studies in order to inform the analysis plan (Chaimani 2013). We excluded from the analysis two‐arm studies in which one of the interventions could be described as 'standard care' or 'mixed care'. These are treatment arms where the 'intervention' involves the choice of more than one treatment: they are unlikely to be consistently applied. We had anticipated that such interventions might have been acceptable for a grouped sensitivity analysis (see section on Sensitivity analysis), but experience (Westby 2017) led us to conclude that this was unlikely to be informative; such studies are therefore summarised in Appendix 4, but not considered further. We also excluded from the main analysis studies that had one intervention of direct interest (e.g. hydrocolloid) compared with one ineligible intervention (e.g. ultrasound), unless we found, after examining the network diagram, that the ineligible intervention linked two or more interventions of direct interest; such interventions were included in a sensitivity analysis looking at an expanded base‐case.

We performed multivariable network meta‐analysis using STATA 13. We used the 'mvmeta' command and adopted a random‐effects approach and a consistency model. We used per‐arm data (see Data extraction and management) throughout. The STATA routine took into account correlations between the effect sizes from multi‐arm studies. The NMA results were reported for all 'mixed treatment contrasts', which means the meta‐analysis involved both direct evidence and indirect evidence from across the whole network. The output was reported as pooled RRs, with their 95% CIs. If there were sufficient data we had also planned to perform an analysis of time‐to‐event data using the log HR with its standard error (SE).

We carried out analyses for network comparisons (where indirect evidence alone, or both direct and indirect evidence contributes) in a frequentist framework as above. Where required, we accounted for correlations induced by multi‐arm studies. We also presented the data in forest plots.

We obtained a treatment hierarchy using the surface under the cumulative ranking curve (SUCRA) and mean ranks (Salanti 2011) for each treatment. Both these measures are based on an assessment of the probability of each treatment being best, second best, etc. in terms of being the most likely to heal venous leg ulcers (when compared with all other evaluated treatments). We used the STATA methods described by Chaimani 2013.

We had planned to present two different networks: one for individual treatments and a sensitivity analysis in which interventions were grouped in broader clinically relevant categories. In practice, there were many different dressings and a wide range of topical agents too, and we decided, post‐hoc, to restrict the main analysis to treatments that were considered most important and widely used. Selection of treatments for analysis was decided by two review authors working independently, with guidance from a clinical review author who had not seen the data.This set of interventions was termed the 'base‐case network'.

Interventions which were considered in the base‐case were: alginate, cadexomer iodine, film, foam, gentian violet, hyaluronic acid, hyaluronic‐acid with povidone iodine, hydrocolloid, hydrofibre, hydrogel, ibuprofen‐releasing foam, nonadherent, octenidine, paste bandage, saline gauze, phenytoin, povidone iodine, protease‐modulating matrix (PMM), PMM silver, silver sulfadiazine (SSD), sucralfate, silver and zinc oxide. Only one of these ‐ phenytoin ‐ could not subsequently be joined into the network. Sensitivity analyses explored the impact of extending the number of treatments included or further restricting it (see Sensitivity analysis).

Comparisons of two eligible interventions not joined into the network remained in the review and we reported the direct evidence.These included comparisons between a specified intervention such as cadexomer iodine, silver or honey and "standard care" as well as comparisons between two individual interventions where one or both were only partly relevant to the network or could not be joined to the network.

There was a very large number of contrasts in the NMA and we decided to focus our reporting of the analysis firstly on results for the network as a whole, and then in the 'Summary of findings' table to report the treatment effect data for some specific treatment comparisons. This was done in order to maximise the clinical utility of the NMA and the accessibility of the review. We decided, post‐hoc to focus on the two treatments with the highest probabilities for being one of the best treatments and to examine in detail the results of their comparisons with three of the most common and widely used treatments (foam, hydrocolloid and nonadherent dressings). The results for all contrasts are also shown in forest plots.

Subgroup analysis and investigation of heterogeneity

Assessment of statistical heterogeneity

We assessed the presence of heterogeneity within each pairwise comparison using the I² statistic that measures the percentage of variability that cannot be attributed to random error (Higgins 2003). We also took into account the overlap of confidence intervals and the variability in the point estimates. We regarded effect estimates where an I² was less than 50% as having low levels of heterogeneity, given the potential for wide confidence intervals in pairwise comparisons within a network, which we had anticipated may be sparse.

Assessment of statistical inconsistency

We assessed inconsistency in two main ways: determining local inconsistencies (around particular contrasts in the network) and assessing inconsistency for the network as a whole. These tests are often underpowered so we carried out the assessment using the 90% significance level.

Local approaches to evaluating inconsistency

To evaluate the presence of inconsistency locally we used two main approaches. Firstly, we considered a loop‐specific approach. This method evaluates the consistency assumption in each closed loop of the network separately as the difference between direct and indirect estimates for a specific comparison in the loop (inconsistency factor, IF). Then, the magnitude of the inconsistency factors and their 90% CIs can be used to make inferences about the presence of inconsistency in each loop. We assumed a common heterogeneity estimate within each loop.

Secondly, we considered a 'node splitting' approach (Dias 2010; Salanti 2014). This method was applied, singly, to each direct contrast (called a 'node' by Dias 2010). A STATA routine was used to calculate an indirect estimate using the rest of the network, by running the NMA after excluding the direct evidence for that contrast. The indirect estimates were then compared with the respective direct estimates.

For both approaches a ratio of risk ratios (RoRR) with its 90% CI was calculated for each contrast. If the CI excluded 1, there is statistically significant inconsistency. We also considered whether the CI included 2 or more (or 0.5 or less). This would mean that the direct estimate could be twice as large (or half as big) as the indirect estimate, which is an indication of potential inconsistency (Chaimani 2013).

Where we detected serious inconsistency, either in the direct evidence or between the direct and indirect evidence for a contrast, we downgraded the evidence for that contrast.

Global approaches to evaluating inconsistency

We evaluated consistency in the entire network simultaneously, by extending the analysis to include an inconsistency model that omits consistency equations (Dias 2013). This used a design‐by‐treatment interaction model, which allows for different trial designs (Higgins 2012; White 2012). This approach produced a set of inconsistency parameters. After fitting the inconsistency model we tested the null hypothesis of consistency by globally testing the set of inconsistency parameters using a global Wald test. This test may lack power and we considered a significance level of P < 0.1. Inconsistency in the entire network was considered a reason for downgrading the certainty of the evidence which the network, as a whole, represented.

Investigation of heterogeneity and inconsistency

Where sufficient studies were available, we planned to perform network meta‐regression (data permitting) or subgroup analyses using funding source and risk of bias as possible sources of inconsistency or heterogeneity, or both. In the event we were able to perform an analysis using risk of bias as a possible source of heterogeneity.

Sensitivity analysis

We re‐analysed the network with studies removed if they were considered to be at high risk of bias for any one or more of selection, attrition or detection bias (Appendix 3).

We considered a sensitivity analysis to assess the possible impact of missing outcome data on the network estimates, via assessment of risk of attrition bias (as defined in Appendix 3), testing the assumption of imputation of no event for missing data.

Where one or more studies were clearly outliers (i.e. in terms of direction or size of relative treatment effect, or both, or as flagged in inconsistency testing), we had planned to conduct a sensitivity analysis where the study was removed from the network, as long as the network was still analysable; in the event we did not need to do this.

We had planned to conduct a sensitivity analysis, in which dressings interventions were grouped in broader categories, with clinical guidance, but this was not conducted. Instead, we conducted two post‐hoc sensitivity analyses for the base‐case network: one restricted the dataset to a narrower set of clinically appropriate interventions; the other included additional treatments outside the base‐case, which reinforced the network with more links. The reduced network excluded the following interventions which were included in the base‐case: gentian violet, hyaluronic‐acid with povidone iodine, ibuprofen‐releasing foam, octenidine, phenytoin and sucralfate. The expanded base‐case added nine trials and the following supplementary interventions to the base‐case decision set: blood product (non‐eligible intervention); emollient cream; and growth factor (non‐eligible intervention). We conducted this sensitivity analysis to investigate the impact of strengthening the network through indirect evidence provided by comparisons of key decision set interventions such as saline gauze and hydrogel with these supplementary interventions.

Quality assessment of evidence (GRADE) generated from the network meta‐analysis (NMA)

We summarised the findings according to GRADE principles (Schünemann 2011a; Schünemann 2011b). The quality and certainty of the data included in any synthesis model are key to determining the validity of the results and of inferences made. We explored the application of GRADE methodology to NMA, focusing on the approach of Salanti 2014. We assessed evidence quality in two main ways, for each contrast and separately, for the network as a whole, in order to assess the quality of the ranking order. We assessed individual GRADE factors as follows.

Risk of bias: contributions for each particular contrast were considered, and used to assess the overall risk of bias for that contrast. We assessed overall risk of bias per contrast and also for the network as a whole (see Assessment of risk of bias in included studies).
Indirectness: this was assessed as without limitations because we did not identify any effect modifiers.
Inconsistency: at the level of the contrast, we considered both heterogeneity in the direct evidence for that comparison and inconsistency related to different routes of analysis for the comparison (e.g. direct versus indirect evidence). We noted that inconsistency can only be assessed where there is both direct and indirect evidence. GRADE inconsistency was assessed as a serious limitation if there was heterogeneity in the direct estimate or inconsistency in the network with respect to that comparison. Very serious limitations were attributed to the comparison if there was severe heterogeneity or severe inconsistency or limitations with both heterogeneity and inconsistency. At the level of the network, we considered the global Wald test for inconsistency (see Data synthesis; Assessment of heterogeneity). Tests of this nature are typically underpowered, so a P value less than 0.1 was considered significant. Additionally, if several contrasts showed direct and indirect results that would have led to different clinical decisions, we considered inconsistency to be present.
Imprecision: at the level of the contrast, we assessed imprecision for each pairwise comparison using the GRADE default minimally important difference (MID) values of 1.25 and 0.75 for the RR. For contrasts that were not part of the 'core' of the network, we also took into account the number of events informing the direct evidence and considered it in relation to the optimal information size. At the level of the network, we assessed the overlap of the rankograms and the magnitude of the SUCRA estimates.
Publication bias: was assessed for each pairwise comparison using standard GRADE (where there were 10 or more studies); we used contour‐enhanced funnel plots where appropriate to examine publication bias in the network as a whole.

'Summary of Findings' tables

We presented the main results of the review in a 'Summary of findings' table, reporting the results for a representative set of contrasts, with one row for each contrast. We focused on interventions which the SUCRA suggested were likely to be high ranked and the comparisons between these and commonly‐used types of intervention. This table presents key information concerning the certainty of the evidence, the magnitude of the effects for the contrasts examined, and the sum of the available data (Schünemann 2011a). The 'Summary of findings' table also includes an overall grading of the evidence using the GRADE approach.

For calculating absolute risk differences for the probability of healing we used a 'control group risk', calculated as the median of the risks for the comparator across all direct evidence studies with these comparators.

Results

Description of studies

Results of the search

Electronic searches identified 1836 records after deduplication. Of these, we excluded 1024 after initial screening of title and abstract. Full‐text screening of 812 records led to the identification of 127 relevant reports of 78 studies (see Figure 1).

Figure 1

Study flow diagram.

We included studies that compared two eligible interventions (see criteria for inclusion ‐ interventions (Types of interventions). We also included studies that assessed only one eligible intervention, but which provided linking for the network of eligible studies. Therefore there were three types of included study:

studies which compared two eligible interventions and which were included in the NMA;
studies which compared two eligible interventions but which could not be joined into the NMA;
studies which compared an eligible intervention with one or more ineligible interventions but which strengthened the network by linking other two or more eligible interventions.

A total of 78 studies with 7014 randomised participants was included in one or more of these categories.

An updated search in March 2018 retrieved 100 additional records. Of these 23 required consideration in detail. Two records were added as additional publications to studies already identified as excluded studies in the review. Ten studies (11 records) could be clearly excluded (see excluded studies) and nine studies were added to Studies awaiting classification. One study was added to ongoing studies (see Characteristics of ongoing studies).

Included studies

There were 47 studies that we joined into the network with two relevant interventions as outlined in Data synthesis: (Armstrong 1997; Backhouse 1987; Banerjee 1997; Blair 1988a; Blair 1988b; Bowszyc 1995; Brandrup 1990; Callam 1992; Casoni 2002; Charles 2002; Dimakakos 2009; Fogh 2012; Gottrup 2008; Hanft 2006; Hansson 1998; Harding 2001; Humbert 2013; Ivins 2006; Jørgensen 2005; Kelechi 2012; Kucharzewski 2013; Lanzara 2008; Leaper 1991; Meaume 2012; Meredith 1988; Moffatt 1992a; Moffatt 1992b; Nelson 2007; Norkus 2005; Ohlsson 1994; Ormiston 1985; Petkov 1997; Romanelli 2015a; Rubin 1990; Schulze 2001; Scurr 1994; Senet 2014; Smith 1992; Smith 1994; Sopata 2016; Stacey 2000; Taddeucci 2004; Thomas 1997; Tumino 2008; Vanscheidt 2012; Vin 2002; Zuccarelli 1992).

A further 13 studies (Alvarez 2012; Beckert 2006; Bishop 1992; Caprio 1992; De Araujo 2016; Dereure 2012a; Greguric 1994; Luiza 2015; Kalis 1993; Moss 1987; Romero‐Cerecero 2012; Solovastru 2015; Tarvainen 1988), all assessed comparisons between two eligible treatments, which could be linked to the network but where one or both interventions was considered to be only partly relevant and therefore only the direct evidence was considered, or the trial was included only in a sensitivity analysis. These interventions included dextranomer, A. Pinchinsensis extract, ozonated oil, shale oil, papain, magnesium sulphate and cellulose. Summaries of these comparisons are provided in Appendix 4.

There were a number of studies that evaluated relevant interventions but which could we could not connect into the network. This included the following studies that compared a particular treatment with 'standard care' (which was either not specified or included a range of different dressings or topical treatments): Arnold 1994; Brown 2014; Jull 2008; Harcup 1986; Lindsay 1986; Michaels 2009; Steele 1986. Other studies not joined into the network were Hokkam 2011, which compared two interventions which did not otherwise link to the network: phenytoin with no treatment and Salim 1992, which compared sulphadryl powders to inactive powder. Summaries of these comparisons are also provided in Appendix 4.

We included nine studies that had only one relevant intervention in an expanded base‐case to strengthen the network (Arenbergerova 2013; Biland 1985; Rasmussen 1991; Robson 1995; Robson 2001; Robson 2004; Senet 2003; Senet 2011; Stacey 1997). These were all two‐arm trials with one relevant intervention from the base‐case or partly relevant interventions such as emollient cream or an ineligible intervention.

Summary details of all trials in the review are shown in Table 1; a summary of the status of individual studies within the review and the networks is shown in Table 2, which clearly denotes which trials are included in the base‐case and the sensitivity analyses and which are included only in the review and not in the network.

Table 1. Summary characteristics of individual studies

Study characteristic	Details of studies
Publication	Abstract or poster only: Caprio 1992; Casoni 2002; Hanft 2006; Ivins 2006; Kalis 1993; Lanzara 2008; Petkov 1997; Zuccarelli 1992. All other studies had a full publication
Multiple interventions	Three arms: Bishop 1992; De Araujo 2016; Hansson 1998; Robson 1995. All other studies had two arms
Unit of randomisation	Ulcer: Caprio 1992 Leg: Stacey 1997 Unclear: Hanft 2006; Kalis 1993; Leaper 1991 All other studies used participants as the unit of randomisation
Funding	Industry: Armstrong 1997; Backhouse 1987; Beckert 2006; Bishop 1992; Bowszyc 1995; Charles 2002; Dereure 2012a; Fogh 2012; Gottrup 2008; Hansson 1998;Humbert 2013; Jørgensen 2005; Kelechi 2012; Lanzara 2008; Leaper 1991; Meaume 2012; Moffatt 1992a; Moffatt 1992b; Moss 1987; Nelson 2007; Norkus 2005; Robson 1995; Scurr 1994; Senet 2003; Senet 2014; Smith 1992; Smith 1994; Stacey 2000; Vin 2002; Zuccarelli 1992. Others did not report funding source or reported no funding or a non‐industry source
Follow‐up time	4 weeks: Bishop 1992; Ivins 2006; Jørgensen 2005; Schulze 2001; 30 days: Solovastru 2015; 6 weeks: Armstrong 1997; Biland 1985;Fogh 2012; Gottrup 2008; Leaper 1991; Meredith 1988; Ohlsson 1994; Robson 1995; Scurr 1994; Senet 2014; Smith 1994;Steele 1986; 8 weeks: Brandrup 1990; Caprio 1992; Meaume 2012; Taddeucci 2004; Tarvainen 1988; 60 days: De Araujo 2016; Dereure 2012a; Humbert 2013; 9 weeks: Dimakakos 2009; Kalis 1993 10 weeks: Arnold 1994;12 weeks: Backhouse 1987; Blair 1988a; Blair 1988b; Callam 1992; Charles 2002; Hanft 2006; Hansson 1998; Harding 2001;Lanzara 2008; Luiza 2015; Moffatt 1992a; Moffatt 1992b; Ormiston 1985; Rasmussen 1991; Robson 2001; Romanelli 2015a; Salim 1992; Senet 2003; Senet 2011; Vanscheidt 2012; Vin 2002; Zuccarelli 1992; 3 months: Casoni 2002; 90 days: Tumino 2008; 13 weeks:Arenbergerova 2013; Thomas 1997; 4 months: Smith 1992; 20 weeks: Beckert 2006; Kelechi 2012; Robson 2004; 24 weeks: Nelson 2007;26 weeks: Moss 1987; Petkov 1997; 9 months: Stacey 1997; Stacey 2000; 10 months: Romero‐Cerecero 2012; 12 months; Norkus 2005; Rubin 1990; Unclear/till healing: Greguric 1994; Sopata 2016 (max 40 weeks) Kucharzewski 2013 (max 16 weeks)
Included < 25% non venous leg ulcers	Included non‐venous leg ulcers: Armstrong 1997; Biland 1985; Brandrup 1990; Norkus 2005; Ohlsson 1994; Rasmussen 1991 Unclear: Backhouse 1987; Humbert 2013; Ivins 2006; Jørgensen 2005; Leaper 1991; Luiza 2015; Romero‐Cerecero 2012; Senet 2011; Tarvainen 1988; Zuccarelli 1992. All others enrolled only participants with VLU

VLU: venous leg ulcers

Table 2. Studies: status in network/review

Study	Interventions	No eligible interventions	Expanded base‐case	Base‐case	Sensitivity analysis	Risk of bias
Alvarez 2012^c	Cellulose Nonadherent	2	X	X	X	High
Arenbergerova 2013^b	Hydrofibre Blood product	1	√	X	X	Very high
Armstrong 1997^a	Alginate Hydrofibre	2	√	√	√	High
Arnold 1994^c	Hydrocolloid Iodine OR nonadherent	2	X	X	X	Very high
Backhouse 1987^a	Nonadherent Hydrocolloid	2	√	√	√	Low/Unclear
Banerjee 1997^a	Nonadherent Film	2	√	√	√	Very high
Beckert 2006^c	Shale oil Hydrogel	2	X	X	X	Low/unclear
Biland 1985^b	Blood product Emollient cream	1	√	X	X	High
Bishop 1992^b	Tripeptide copper Emollient cream SSD	3	√	X	X	Low/unclear
Blair 1988a^a	Nonadherent Hydrocolloid	2	√	√	√	High
Blair 1988b^a	Nonadherent SSD	2	√	√	√	High
Bowszyc 1995^a	Foam Hydrocolloid	2	√	√	√	High
Brandrup 1990^a	Hydrocolloid Zinc oxide	2	√	√	√	High
Brown 2014^c	Silica gel Alternative traditional dressings	2	X	X	X	Very high
Callam 1992^a	Nonadherent Foam		√	√	√	Very high
Caprio 1992^c	Hydrocolloid Collagen	2	X	X	X	Low/unclear
Casoni 2002^a	Nonadherent Hyaluronic+Povidone	2	√	√	X	Low/unclear
Charles 2002^a	Foam Hydrocolloid	2	√	√	√	Low/unclear
De Araujo 2016^b	Blood product Hydrogel Papain	3 (2 in network)	√	X	X	Low/unclear
Dereure 2012a^b	Hyaluronic acid Emollient cream	2	√	X	X	Low/unclear
Dimakakos 2009^a	Foam Silver	2	√	√	√	Low/unclear
Fogh 2012^a	Foam Silver	2	√	√	√	High
Gottrup 2008^a	Foam Silver	2	√	√	√	High
Greguric 1994^c	Magnesium sulphate Hydrocolloid	2	X	X	X	High
Hanft 2006^a	PMM silver Hydrocolloid	2	√	√	√	High
Hansson 1998^a	Nonadherent Cadexomer iodine Hydrocolloid	3	√	√	√	High
Harcup 1986^c	Standard care Cadexomer iodine	2	X	X	X	Low/unclear
Harding 2001^a	Alginate Hydrofibre	2	√	√	√	Very high
Hokkam 2011^c	Phenytoin No treatment	2	X	X	X	Low/unclear
Humbert 2013^a	Hyaluronic acid Saline gauze	2	√	√	√	High
Ivins 2006^a	Foam Silver	2	√	√	√	Low/unclear
Jørgensen 2005^a	Foam Silver	2	√	√	√	Low/unclear
Jull 2008^c	Honey Standard care	2	X	X	X	Very high
Kalis 1993^c	Hydrocolloid Dextranomer	2	X	X	X	Very high
Kelechi 2012^a	Nonadherent Hydrofibre	2	√	√	√	High
Kucharzewski 2013^a	Hydrocolloid Silver	2	√	√	√	High
Lanzara 2008^a	PMM silver Foam	2	√	√	√	High
Leaper 1991^a	Nonadherent Hydrocolloid	2	√	√	√	High
Lindsay 1986^c	Standard care Cadexomer iodine	2	X	X	X	Low/unclear
Luiza 2015^c	Papain Hydrogel	2	X	X	X	High
Meaume 2012^a	PMM Foam	2	√	√	√	High
Meredith 1988^a	Nonadherent Hydrocolloid	2	√	√	√	Low/unclear
Michaels 2009^c	Silver non‐silver	2	X	X	X	High
Moffatt 1992a^a	Nonadherent Hydrocolloid	2	√	√	√	Low/unclear
Moffatt 1992b^a	Alginate Nonadeherent	2	√	√	√	Low/unclear
Moss 1987^c	Cadexomer iodine Dextranomer	2	X	X	X	Very high
Nelson 2007^a	Nonadherent Hydrocolloid	2	√	√	√	Very high
Norkus 2005^a	Foam Hydrocolloid	2	√	√	√	Very high
Ohlsson 1994^a	Hydrocolloid Saline gauze	2	√	√	√	Low/unclear
Ormiston 1985^b	Cadexomer iodine gentian violet	2	√	√	X	High
Petkov 1997^a	PMM Alginate	2	√	√	√	Low/unclear
Rasmussen 1991^b	Growth factor Hydrocolloid	1	X	X	√	High
Robson 1995^b	Growth factor Nonadherent	1	X	X	√	Low/unclear
Robson 2001^b	Growth factor Nonadherent	1	X	X	√	High
Robson 2004^b	Growth factor Nonadherent	1	X	X	√	Low/unclear
Romanelli 2015a^a	PMM Alginate	2	√	√	√	Low/unclear
Romero‐Cerecero 2012^c	A. Pichinchensis Alginate	2	X	X	X	High
Rubin 1990^a	Foam paste bandage	2	√	√	√	High
Salim 1992^c	Sulphadryl Inactive powder	2	X	X	X	High
Schulze 2001^a	Foam alginate	2	√	√	√	Very high
Scurr 1994^a	Hydrocolloid Alginate	2	√	√	√	Low/unclear
Senet 2003^b	Blood product Hydrocolloid	1	X	X	√	Low/unclear
Senet 2011^b	Growth factor Hydrogel	1	X	X	√	High
Senet 2014^a	silver foam	2	√	√	√	High
Smith 1992^a	Hydrocolloid Povidone iodine	2	√	√	√	Very high
Smith 1994^a	Hydrocolloid alginate	2	√	√	√	Very high
Solovastru 2015^c	Ozonated oil Emollient cream	2	X	X	X	Low/unclear
Sopata 2016^a	hydrocolloid foam	2	√	√	√	Very high
Stacey 1997^a	Paste bandage alginate	2	√	√	√	High
Stacey 2000^b	Blood product Saline gauze	1	X	X	√	High
Steele 1986^c	Standard care Cadexomer iodine	2	X	X	X	Low/unclear
Taddeucci 2004^a	Nonadherent Hydrogel	2	√	√	√	Very high
Tarvainen 1988^c	Cadexomer iodine Dextranomer	2	X	X	X	Very high
Thomas 1997^a	Foam Hydrocolloid	2	√	√	√	Very high
Tumino 2008^a	Sucralfate Hydrogel	2	√	√	√	High
Vanscheidt 2012^a	Octenidine Foam	2	√	√	X	High
Vin 2002^a	PMM nonadherent	2	√	√	√	Low/unclear
Zuccarelli 1992^a	Foam Hydrocolloid	2	√	√	√	Low/unclear

Abbreviations: PMM: protease modulating matrix; SSD: silver sulphadiazine

^aStudy in original base‐case

^bStudy only included in sensitivity analysis

^cStudy included in review but not in network

Interventions

Included studies evaluated a wide range of dressings and topical treatments. A total of 20 different types of dressings were evaluated; this included dressings which were impregnated with agents such as ibuprofen, silver, povidone iodine or zinc oxide. Sixteen different topical treatments were included. Although the majority of trials compared two dressings or two topical treatments (and most of these compared two dressings), some compared a dressing with a topical treatment (e.g. a hydrocolloid dressing compared with silver sulfadiazine (SSD)). A minority of trials compared arms which included more than one treatment option and these included both dressings and topical treatments.

The number and types of Interventions are fully detailed in the effects of interventions section (Effects of interventions) and in supplementary tables (Table 2; Table 3; Table 4), which also show the status of each trial in the review and network analyses.

Table 3. Direct comparisons for individual interventions compared with NMA results

Contrast/comparison	Number of studies (participants)	Studies	RR (95% CI) direct evidence. Random effects (inverse variance) Heterogeneity statistics	NMA results (extended base‐case; consistency assumption): RR (95% CI)
Comparisons with nonadherent: RR > 1 indicates greater proportion healing with specified alternative treatment
Alginate	1 (113)	Moffatt 1992b	1.08 (0.86 to 1.36)	1.21 (0.92 to 1.60)
Cadexomer iodine	1 (105)	Hansson 1998	1.00 (0.39 to 2.56)	1.16 (0.50 to 2.69)
Film	1 (71)	Banerjee 1997	1.34 (0.61 to 2.92)	1.34 (0.61 to 2.95)
Foam	1 (124)	Callam 1992	1.35 (0.89 to 2.05)	1.15 (0.91 to 1.44)
Hyaluronic acid plus povidone iodine	1 (55)	Casoni 2002	1.93 (0.95 to 3.92)	1.93 (0.94 to 3.96)
Hydrocolloid	7 (662)	Backhouse 1987; Blair 1988a; Hansson 1998; Leaper 1991; Meredith 1988; Moffatt 1992a; Nelson 2007	1.26 (0.92 to 1.72) I² = 69%; P = 0.004	1.04 (0.85 to 1.29)
Hydrofibre	1 (82)	Kelechi 2012	1.47 (0.88 to 2.46)	1.39 (0.93 to 2.08)
Hydrogel	1 (24)	Taddeucci 2004	2.00 (0.21 to 19.23)	0.79 (0.39 to 1.62)
PMM	1 (74)	Vin 2002	1.42 (0.80 to 2.51)	1.31 (0.93 to 1.84)
SSD	1 (60)	Blair 1988b	0.79 (0.57 to 1.10)	0.81 (0.57 to 1.15)
Growth factor^a	3 (460)	Robson 1995; Robson 2001; Robson 2004	0.96 (0.81 to 1.14) I² = 0%; P = 0.65	0.95 (0.72 to 1.25)
Comparisons with alginate: RR > 1 indicates greater proportion healing with specified alternative treatment
Foam	1 (113)	Schulze 2001	0.55 (0.10 to 2.86)	0.94 (0.72 to 1.23)
Hydrocolloid	2 (80)	Scurr 1994; Smith 1994	0.72 (0.15 to 3.42) I² = 52%; P = 0.15	0.86 (0.68 to 1.11)
Hydrofibre	2 (175)	Armstrong 1997; Harding 2001	1.47 (0.48 to 4.47) I² = 54%; P = 0.14	1.15 (0.77 to 1.72)
Paste bandage	1 (133)	Stacey 1997	1.22 (0.91 to 1.63)	1.39 (1.01 to 1.90)
PMM	2 (140)	Petkov 1997; Romanelli 2015a	1.10 (0.84 to 1.46) I² = 0%; P = 0.87	1.08 (0.83 to 1.40)
Comparisons with cadexomer iodine: RR > 1 indicates greater proportion healing with specified alternative treatment
Hydrocolloid	1 (104)	Hansson 19980	0.73 (0.26 to 2.08)	0.90 (0.39 to 2.10)
Gentian violet	1 (60)	Ormiston 1985	0.58 (0.27 1.28)	0.58 (0.26 to 1.29)
Comparisons with foam: RR > 1 indicates greater proportion healing with specified alternative treatment
Hydrocolloid	6 (458)	Bowszyc 1995; Charles 2002; Norkus 2005; Sopata 2016; Thomas 1997; Zuccarelli 1992	0.92 (0.77 to 1.08) I² = 0%; P = 0.84	0.91 (0.78 to 1.07)
Ibuprofen	2 (242)	Fogh 2012; Gottrup 2008	0.88 (0.48 to 1.61) I² = 0%; P = 0.79	0.88 (0.48 to 1.62)
Octenidine	1 (126)	Vanscheidt 2012	1.03 (0.56 to 1.90)	1.03 (0.55 to 1.92)
Paste bandage	1 (36)	Rubin 1990	2.30 (1.29 to 4.10)	1.47 (0.99 to 2.17)
PMM	1 (187)	Meaume 2012	0.87 (0.30 to 2.48)	1.14 (0.82 to 1.60)
PMM silver	1 (30)	Lanzara 2008	1.57 (0.84 to 2.92)	1.15 (0.78 to 1.71)
Silver	4 (397)	Dimakakos 2009;Ivins 2006; Jørgensen 2005; Senet 2014	1.65 (1.08 to 2.52) I² = 0%; P = 0.77	2.12 (1.46 to 3.07)
Comparisons with hyaluronic acid: RR > 1 indicates greater proportion healing with specified alternative treatment
Saline gauze	1 (88)	Humbert 2013	0.52 (0.23 to 1.17)	0.57 (95% CI 0.28 to 1.14)
Emollient cream	1 (101)	Dereure 2012a	1.31 (0.31 to 5.55)	1.75 (0.87 to 3.52)
Comparisons with hydrocolloid: RR > 1 indicates greater proportion healing with specified alternative treatment
PMM silver	1 (49)	Hanft 2006	1.07 (0.69 to 1.67)	1.27 (0.87 to 1.85)
Povidone iodine	1 (200)	Smith 1992	0.92 (0.69 to 1.23)	0.92 (0.68 to 1.26)
Saline gauze	1 (28)	Ohlsson 1994	0.29 (0.07 to 1.14)	0.34 (95% CI 0.15 to 0.8)
Silver	1 (58)	Kucharzewski 2013	4.39 (2.23 to 8.65) Note 100% events in silver arm	2.32 (1.58 to 3.41)
Zinc oxide	1 (43)	Brandrup 1990	0.95 (0.27 to 3.33)	0.95 (0.27 to 3.35)
Blood product^a	1 (13)	Senet 2003	0.86 (0.07 to 10.96)	0.38 (95% CI 0.17 to 0.88)
Growth factor^a	1 (29)	Rasmussen 1991	1.83 (0.22 to 15.51)	0.91 (0.71 to 1.17)
Comparisons withhydrogel: RR > 1 indicates greater proportion healing with specified alternative treatment
Sucralfate	1 (100)	Tumino 2008	8.60 (3.72 to 19.90)	8.60 (3.68 to 20.07)
Blood product^a	1 (44)	De Araujo 2016	0.47 (0.14 to 1.58)	0.51 (CI 0.21 to 1.23)
Growth factor^a	1 (59)	Senet 2011	1.38 (0.64 to 3.01)	1.20 (0.61 to 2.35)
Comparisons with blood product : RR > 1 indicates greater proportion healing with specified alternative treatment
Saline gauze	1 (67)	Stacey 2000	0.93 (0.74 to 1.16)	0.89 (0.68 to 1.17)
Emollient cream	1 (147)	Biland 1985	0.76 [0.55, 1.06]	0.79 (0.56 to 1.11)
Comparisons with emollient cream: RR > 1 indicates greater proportion healing with specified alternative treatment
SSD	1 (57)	Bishop 1992	6.21 (0.80 to 48.38)	2.56 (1.01 to 6.53)

Abbreviations: PMM: protease modulating matrix; RR: relative risk; SSD: silver sulphadiazine

^aNon‐eligible linking intervention

See: Summary of findings for the main comparison NMA evidence: proportion with complete healing

Table 4. Interventions in the included studies

Intervention	Number of included studies	Included studies	Number of participants in included studies
A. Pichinchensis	1	Romero‐Cerecero 2012	34
Alginate	10	Armstrong 1997; Harding 2001; Moffatt 1992b; Petkov 1997; Romanelli 2015a; Romero‐Cerecero 2012; Schulze 2001; Scurr 1994; Smith 1994; Stacey 1997	735
Blood product^a	5	Arenbergerova 2013; Biland 1985; De Araujo 2016; Senet 2003; Stacey 2000	431
Cadexomer iodine	7	Hansson 1998; Harcup 1986; Lindsay 1986; Moss 1987; Ormiston 1985; Steele 1986; Tarvainen 1988	433
Cellulose	1	Alvarez 2012	48
Collagen	2	Caprio 1992; Robson 1995	132
Dextranomer	3	Kalis 1993; Moss 1987; Tarvainen 1988	171
Emollient cream	3	Biland 1985; Bishop 1992; Dereure 2012a	384
Film	1	Banerjee 1997	56
Foam	18	Bowszyc 1995; Callam 1992; Charles 2002; Dimakakos 2009; Fogh 2012; Gottrup 2008; Ivins 2006; Jørgensen 2005; Lanzara 2008; Meaume 2012; Norkus 2005; Rubin 1990; Schulze 2001; Senet 2014; Sopata 2016; Thomas 1997; Vanscheidt 2012; Zuccarelli 1992	1672
Gentian violet	1	Ormiston 1985	60
Growth factor^a	5	Rasmussen 1991; Robson 1995; Robson 2001; Robson 2004; Senet 2011	560
Honey	1	Jull 2008	368
Hyaluronic acid	2	Dereure 2012a; Humbert 2013	189
Hyaluronic acid + povidone iodine	1	Casoni 2002	65
Hydrocolloid	25	Backhouse 1987; Blair 1988a; Bowszyc 1995;Brandrup 1990; Caprio 1992; Charles 2002; Greguric 1994; Hanft 2006; Hansson 1998; Kalis 1993; Kucharzewski 2013; Leaper 1991; Meredith 1988; Moffatt 1992a; Nelson 2007; Norkus 2005; Ohlsson 1994; Rasmussen 1991; Scurr 1994; Senet 2003; Smith 1992; Smith 1994; Sopata 2016; Thomas 1997; Zuccarelli 1992	2044
Hydrofibre	4	Arenbergerova 2013; Armstrong 1997; Harding 2001; Kelechi 2012	329
Hydrogel	6	Beckert 2006; De Araujo 2016; Luiza 2015; Senet 2011; Taddeucci 2004; Tumino 2008	393
Ibuprofen	2	Fogh 2012; Gottrup 2008	222
Magnesium sulphate	1	Greguric 1994	110
Nonadherent	20	Alvarez 2012; Arnold 1994; Backhouse 1987; Banerjee 1997; Blair 1988a; Blair 1988b; Callam 1992; Casoni 2002; Hansson 1998; Kelechi 2012; Leaper 1991; Moffatt 1992a; Moffatt 1992b; Meredith 1988; Nelson 2007; Robson 1995; Robson 2001; Robson 2004; Taddeucci 2004; Vin 2002	1725
Non silver	1	Michaels 2009	208
No treatment	1	Hokkam 2011	104
Octenidine	1	Vanscheidt 2012	106
Ozonated oil	1	Solovastru 2015	29
Papain	2	De Araujo 2016; Luiza 2015	70
Paste bandage	2	Rubin 1990; Stacey 1997	149
Phenytoin	1	Hokkam 2011	104
Povidone iodine	1	Smith 1992;	200
PMM	4	Meaume 2012; Petkov 1997; Romanelli 2015a; Vin 2002	400
PMM‐silver	2	Hanft 2006; Lanzara 2008;	79
Saline gauze	3	Humbert 2013; Ohlsson 1994; Stacey 2000	202
Shale oil	1	Beckert 2006	119
Silica gel fibre	1	Brown 2014	120
Silver	6	Dimakakos 2009; Ivins 2006; Jørgensen 2005; Kucharzewski 2013; Michaels 2009; Senet 2014;	663
SSD	2	Bishop 1992; Blair 1988b	146
Standard care/mixed treatments	6	Arnold 1994; Brown 2014; Harcup 1986; Jull 2008; Lindsay 1986; Steele 1986	715
Sucralfate	1	Tumino 2008	100
Suphadryl	1	Salim 1992	137
Tripeptide copper	1	Bishop 1992	86
Zinc oxide	2	Brandrup 1990; Solovastru 2015	72

Abbreviations: PMM: protease modulating matrix; SSD: silver sulphadiazine

^aIneligible intervention included in expanded base‐case to improve network connectivity

Characteristics of participants in included studies

See Characteristics of included studies for full details

Most studies included only people with venous leg ulcers; six studies also included some participants with mixed aetiology or arterial ulcers (although we excluded those with more than 25% of such participants); in 10 studies it was not clear whether a minority of people with non‐venous ulcers were included. The mean or median age range reported for participants ranged between 46 and 81 years. Almost all studies enrolled a majority of women; there were no single sex studies. The mean sizes of ulcers at baseline varied by up to a factor of 10 but were typically between 5 cm² and 10 cm². The mean duration of ulceration at enrolment ranged between one month and 75 months. Many studies excluded participants with either any type of infection or with a specified severity of infection (typically requiring systemic antibiotics); only one study specified that the participant must have an infected ulcer at baseline (Dimakakos 2009). Reporting of other types of ulcer characteristics such as level of slough or exudate was limited. All studies reported some use of compression although the methods and the specificity of the reporting of this varied.

Characteristics of studies

Where funding was reported, it was often industry funding by a manufacturer of one of the assessed interventions (30 studies). However, a substantial number of studies reported no funding or did not report the funding source. A minority of trials reported a third sector or public funding source. Most studies used participants as the unit of both randomisation and analysis, only two reported data at the ulcer or leg level Caprio 1992; Stacey 1997), while a small number appeared to randomise at the level of the person but analyse at the level of the ulcer; in each case these were dealt with in the ''Risk of bias' assessment. Follow‐up ranged between four weeks and 12 months but most trials had follow‐up of three months or less.

For more details on study characteristics see Table 1.

Excluded studies

A large number of records were rapidly excluded after reading the full‐text. A list of these studies is available on request from the authors (see Figure 1). Some studies were excluded after more detailed consideration. These studies are listed with reasons for their exclusion in Characteristics of excluded studies. An additional ten studies were excluded from records retrieved by an update search in March 2018.

Two studies are awaiting classification (Belcaro 2011; Polignano 2010) from the original search. A further nine studies are awaiting classification following an update search in March 2018 (Alvarez 2017; Cavalcanti 2017; Colenci 2016; Cullen 2017; Glukhov 2017; Moreno‐Eutimio 2017; Oliveira 2017; Robinson 1988; Somani 2017). One ongoing study was identified in the update search (Jull 2018).

Risk of bias in included studies

Allocation

Risk of selection bias is assessed based on generation of randomisation sequence and allocation concealment. Many studies were at unclear risk of bias for one or both of these, most commonly for allocation concealment. High risk of bias for randomisation was documented for only one study where errors were noted to have compromised the process. However only a minority (20 studies) were considered to have a low risk of bias. The remainder did not report the processes used clearly enough for us to determine the risk of bias. The number of studies considered to be at low risk for allocation concealment was even lower, with only 12 considered to be clearly at low risk of bias.

Blinding

Many studies were at high or unclear risk of performance bias. Although only a minority (18 studies) were clearly at high risk, many more had an unclear risk. Only 10 studies were considered to be at low risk. For detection bias, we observed a similar pattern although more studies clearly had outcomes determined by blinded observers; 20 were considered to be at low risk of detection bias.

Incomplete outcome data

Twenty‐six studies were considered to be at high risk of attrition bias. However, a larger number had a low risk of bias and only ten were considered to be at unclear risk.

Selective reporting

Only four studies were at high risk of selective reporting bias; a further 16 had an unclear risk in this domain; the remainder were considered to be at low risk of bias.

Other potential sources of bias

Thirteen studies were considered to be at high risk from other forms of bias, mostly due to issues with the analysis. A further 27 had an unclear risk of bias, again primarily related to the reporting of the analysis.

All‐domain risk of bias

All‐domain (overall) risk of bias was assessed for each study. In total 51 studies were considered to have a high or very high all‐domain risk of bias (Figure 2) and 27 studies were considered to be at unclear or low overall risk of bias (these were grouped together for analysis purposes). No study was at low overall risk of bias since all studies had an unclear rating for one or more domains.

Figure 2

'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.

Effects of interventions

Interventions and comparisons: base‐case network and sensitivity analyses

The base‐case network comprised 47 studies assessing 22 interventions: 12 eligible dressings (foam, hydrocolloid, hydrofibre, alginate, ibuprofen‐releasing foam, nonadherent, paste bandage, protease‐modulating (PMM), PMM‐silver, silver‐containing, film, saline gauze); and 10 topical agents (hydrogel, cadexomer iodine, gentian violet, hyaluronic acid, hyaluronic‐acid with povidone iodine, octenidine, povidone iodine, silver sulfadiazine (SSD), sucralfate and zinc oxide). One study was a three‐arm trial (Hansson 1998; hydrocolloid, nonadherent and cadexomer iodine). The total number of comparisons was 49, encompassing a total of 4026 participants, who experienced a total of 1479 events (complete healing).

The sensitivity analysis using an extended base case contained 59 studies assessing 25 interventions in 5156 participants with 1925 events; added interventions were blood product, emollient cream and growth factor. This explored the impact of strengthening the network with more links by including trials which contained an eligible intervention compared to one of three ineligible interventions.

An additional sensitivity analysis looked at a narrower set of 17 interventions assessed in 41 studies that included 3435 participants with 1331 events; removed interventions were ibuprofen‐releasing foam, gentian violet, hyaluronic‐acid with povidone iodine, octenidine and sucralfate. This explored the impact of restricting the network to a narrower set of interventions which excluded interventions that are not widely used in clinical practice.

In the base‐case network, there were 31 different direct contrasts and 12 triangular loops; the extended base‐case sensitivity analysis had 40 direct contrasts, 15 triangular loops and six quadratic loops; and the narrower network had 26 direct contrasts and 12 triangular loops.

We carried out network meta‐analysis for the base‐case and the two sensitivity analyses (Appendix 5). The extended base‐case sensitivity analysis identified instability in the base‐case results for contrasts of some treatments and in the rank order of treatments. Additionally, in the extended base‐case, the point estimates and confidence intervals (CIs) for contrasts with sucralfate were often considerably reduced compared with the base‐case; and the direction of effect was reversed for most contrasts with hydrogel. This instability for some treatments is likely to occur because, in the base‐case, the direct evidence (from single small studies) had an important contribution. As a consequence, we placed more reliance on the extended base‐case sensitivity analysis and therefore report the results for this sensitivity analysis in the rest of the results section. Full details and results for the base‐case and both sensitivity analyses are given in Appendix 5.

The network diagram for the extended base‐case is shown in Figure 3. We weighted node (circle) size by the number of studies reporting each intervention and weighted the thickness of the edge lines according to the inverse variance of the treatment effect estimates for the direct evidence contrast (Chaimani 2013).

Figure 3

Network diagram ‐ extended network, by risk of bias (3 categories)

Key: green = low/unclear; yellow = high; red = very high overall risk of bias for the contrast. The number of studies for each contrast is given in Table 3.

Most treatments in the extended base‐case were part of at least one loop ('core interventions') and eight interventions were 'hanging' treatments (film, gentian violet, hyaluronic acid plus povidone iodine, ibuprofen dressing, octenidine, povidone iodine, sucralfate and zinc oxide).

Risk of bias for the extended base‐case network

We report risk of bias in three ways (see Methods: Assessment of risk of bias in included studies):

for each study, as the all‐domain risk of bias ‐ taking into account selection bias, detection bias, attrition bias, reporting bias and other bias;
for each direct comparison of two interventions, as an overall risk of bias ‐ taking into account the all‐domain risk of bias for the studies (1 above) and the weighting in the meta‐analysis for that comparison;
for each contrast in the network (any pair of interventions in the network) as the overall risk of bias ‐ taking into account the risk of bias for each direct comparison (2 above) and their percentage contributions to the network estimate. We also calculated the overall risk of bias in the network as a whole.

All‐domain risk of bias for each study is shown in Figure 2. For the extended base‐case network , we judged no included studies to be at low risk of bias and 21 at unclear risk of bias (Backhouse 1987; Bishop 1992; Casoni 2002; Charles 2002; De Araujo 2016; Dereure 2012a; Dimakakos 2009; Ivins 2006; Jørgensen 2005; Meredith 1988; Moffatt 1992a; Moffatt 1992b; Ohlsson 1994; Petkov 1997; Robson 1995; Robson 2004; Romanelli 2015a; Scurr 1994; Senet 2003; Vin 2002; Zuccarelli 1992). Twelve were at very high risk of bias (Arenbergerova 2013; Banerjee 1997; Callam 1992; Harding 2001; Nelson 2007; Norkus 2005; Schulze 2001; Smith 1992; Smith 1994; Sopata 2016; Taddeucci 2004; Thomas 1997), and the rest we assessed to be at high risk of bias. We grouped the low and unclear categories together.

We have indicated the overall risk of bias for each direct comparison in the network diagram in Figure 3, using colour for three risk of bias ratings: low/unclear (green), high (yellow), very high (red). There is a substantial amount of direct evidence at high or very high risk of bias. For selected contrasts in the network, we calculated the overall risk of bias as described in Appendix 6.

Network meta‐analysis results

We examined the results in two ways: as risk ratios (RRs) with their 95% CIs for each intervention compared with every other intervention in the network (NMA effect estimates); and for the network as a whole, giving the rank order for the interventions in the network and the probability that a particular intervention is the best, second best, etc treatment.

There are 300 mixed treatment contrasts in the extended network, so we report results for the rank order first, and then, for the NMA effect estimates, we focus on contrasts involving the top two treatments and three common and widely used treatments. In Appendix 5, we report results for all contrasts in the extended network, and give the full rank orders for the base‐case and the two sensitivity analyses.

Extended base‐case network

The NMA generated results for 300 mixed treatment contrasts (i.e. all possible pairwise combinations of the interventions). There were 40 direct contrasts, of which 32 were informed by only one study and the average number of events per mixed treatment contrast was around six (1925/300). The data were sparse and there was uncertainty around the estimates.

As a consequence of the sparseness in the network, only 55 of 300 contrasts had precise estimates. The majority of CIs were wide or very wide, crossing at least one default minimally important difference (MID); i.e. the value of 0.75 or 1.25 was included in the CI (see Sensitivity analysis, GRADE assessment). Fifty‐four contrasts with precise estimates had the whole of the CI above the default MID (i.e. the whole confidence interval lay above 1.25), but 21 of these involved treatments for which the direct evidence comprised one study and had small numbers of events in at least one arm ('fragility'): this applied to contrasts with sucralfate. Overall, 89% of the contrasts were considered to have imprecise results: the exceptions (ignoring contrasts with ineligible interventions) were silver versus each of the following: nonadherent, alginate, foam, hydrocolloid, hydrogel, povidone iodine, saline gauze, SSD; hydrocolloid versus foam; and saline gauze versus alginate,foam, hydrofibre, hyaluronic acid/povidone iodine, paste bandage, PMM and PMM silver.

Ranking of treatments

The NMA produced a large number of estimates. An alternative way of presenting and interpreting data from the whole NMA was to summarise using rankograms: data for each intervention were shown as the probability that each intervention is the best, second best, third best treatment, etc. These probabilities are based on uncertainty, reflecting the effectiveness from the network contrasts and the precision around the estimates. The closer the probability of a rank to 100% (or 0%) and the narrower the distribution across different ranks, the greater the confidence in the ranking. Results are given in Figure 4, Figure 5 and Appendix 5 and summarised here, but must be interpreted in the light of the uncertainty and sparseness in the network.

Figure 4

Rankograms for the extended base‐case showing the probability that each intervention is the best, second best, third best treatment, etc.

Figure 5

Rankograms for all treatments in the extended base‐case network showing the probability that each intervention is the best, second best, third best treatment, etc.

Numerically, sucralfate had by far the highest probability of being the best treatment (93%), and saline gauze was most likely to be the worst treatment (33%). However, the sucralfate ranking is likely to be artificially high: sucralfate is connected to the core of the network via hydrogel and the direct evidence for sucralfate versus hydrogel involves one study with 43 (of 50) healing events for sucralfate and five healing events for hydrogel. The NMA results for all comparisons with sucralfate have very wide CIs and large point estimates. Consequently, sucralfate (versus other interventions) has a high probability of having a very large effect estimate (at the upper confidence limit), in turn leading to an artificially high probability of being the best treatment. Silver also had a high probability of being among the most effective treatments (50% at rank 2). Surface under the cumulative ranking curve (SUCRA) values were generally between 0.3 and 0.8, but one treatment had a SUCRA value of 1 or 0 (sucralfate was 1), with another two treatments having values of 0.9 or 0.1 (silver 0.9 and saline gauze 0.1).

The rankograms for many treatments are broad and uninformative (Figure 4, Figure 5). Of the eligible interventions in the extended network, only five had a maximum probability above 20%. The mean ranks for these treatments were: sucralfate 1.1, silver 2.7, hyaluronic acid plus povidone iodine 5.3, paste bandage 5.4 and saline gauze 23.0.

Certainty/quality assessment of the evidence across the whole network

Further details of information used for GRADE assessment can be found in Appendix 5, Appendix 6 and Appendix 7.

The risk of bias across the extended base‐case network was estimated to be high (Appendix 6). There appeared to be little inconsistency in the network (Appendix 7) and there were relatively few contrasts with conflicting results for direct and indirect or NMA estimates, so across the network we did not downgrade for inconsistency. We downgraded the evidence once for imprecision: in addition to the sparseness (and probably as a consequence of it), there is some overlap of the individual rankograms (see Appendix 5). A contour‐enhanced funnel plot is shown in Figure 6. There does not appear to be a small‐studies effect. Overall, we classed the evidence for the whole network as being of low certainty (downgraded once for risk of bias and once for imprecision).

Figure 6

Contour‐enhanced funnel plot for the extended base‐case network showing comparison‐specific pooled effect sizes

1=non‐adherent, 2=alginate, 3=blood product, 4=cadexomer iodine, 5=emollient cream, 6=film, 7=foam, 8=gentian violet, 9=growth factor,
10=hyaluronic acid + povidone iodine, 11=hyaluronic acid, 12=hydrocolloid, 13=hydrofibre, 14=hydrogel, 15=ibuprofen, 16=octenidine,
17=paste bandage, 18=PMM, 19=PMM silver, 20=povidone iodine, 21=saline gauze, 22=silver, 23=SSD, 24=sucralfate, 25=zinc oxide

Results and quality assessment for selected individual comparisons

Here we focus on the treatment effect data for some specific treatment combinations to provide further insights into the results of the NMA. We considered comparisons of sucralfate, silver, foam, hydrocolloid and nonadherent dressings. These represent the two with the highest probabilities for ranks 1 to 3 in Figure 4. (sucralfate and silver) and three common and widely used treatments (foam, hydrocolloid and nonadherent dressings). These widely used treatments were selected by authors who did not have knowledge of the precise results of the network. The results for the extended base‐case are shown in Table 5. We calculated absolute risk differences using the median risk for the comparator, which was obtained from the risks for that comparator in all direct evidence studies. For all four comparators, the risk varied widely across studies. We report GRADE assessment of selected contrasts in summary of findings Table for the main comparison. Most of the evidence for these individual contrasts was of low or very low certainty.

Table 5. Comparison of NMA results for base‐case and two sensitivity analyses

NMA contrast	Base‐case RR (95% CI)	Narrow sensitivity analysis RR (95% CI)	Extended sensitivity analysis RR (95% CI)
Sucralfate versus hydrogel	8.60 (3.66 to 20.2)	‐‐‐	8.60 (3.68 to 20.1)
Sucralfate versus silver	6.99 (0.60 to 82.0)	‐‐‐	2.80 (0.88 to 8.97)
Sucralfate versus foam	14.83 (1.30 to 169)	‐‐‐	5.94 (1.96 to 18.0)
Sucralfate versus hydrocolloid	16.24 (1.43 to 185)	‐‐‐	6.51 (2.17 to 19.6)
Sucralfate versus nonadherent	17.15 (1.52 to 193)	‐‐‐	6.80 (2.24 to 20.7)
Hydrogel versus silver	0.81 (0.08 to 8.19)	0.81 (0.08 to 8.20)	0.33 (0.15 to 0.72)
Hydrogel versus foam	1.73 (0.18 to 16.9)	1.72 (0.18 to 16.9)	0.69 (0.34 to 1.41)
Hydrogel versus hydrocolloid	1.89 (0.19 to 18.4)	1.88 (0.19 to 18.4)	0.76 (0.38 to 1.53)
Hydrogel versus nonadherent	1.99 (0.21 to 19.3)	2.00 (0.21 to 19.4)	0.79 (0.39 to 1.62)
Silver versus foam	2.12 (1.46 to 3.09)	2.12 (1.45 to 3.10)	2.12 (1.46 to 3.07)
Silver versus hydrocolloid	2.32 (1.58 to 3.43)	2.32 (1.57 to 3.44)	2.32 (1.58 to 3.41)
Silver versus nonadherent	2.45 (1.58 to 3.82)	2.47 (1.58 to 3.86)	2.43 (1.58 to 3.74)
Foam versus hydrocolloid	1.10 (0.93 to 1.28)	1.09 (0.93 to 1.29)	1.10 (0.94 to 1.28)
Foam versus nonadherent	1.16 (0.91 to 1.47)	1.16 (0.91 to 1.49)	1.15 (0.91 to 1.44)
Hydrocolloid versus nonadherent	1.06 (0.84 to 1.32)	1.06 (0.85 to 1.33)	1.04 (0.85 to 1.29)

Abbreviations: CI: confidence interval; RR: relative risk

For the contrast of the two interventions with the highest mean ranks ‐ sucralfate and silver dressing ‐ it is unclear whether there is a difference in the probability of venous leg ulcer healing (RR 2.80, 95% CI 0.88 to 8.97; very low‐certainty evidence, downgraded once for risk of bias and twice for imprecision).

Silver dressings may increase the probability of venous leg ulcer healing, compared with nonadherent dressings (RR 2.43, 95% CI 1.58 to 3.74; moderate‐certainty evidence, downgraded for risk of bias). This corresponds to an absolute risk difference of 346 more people healed per 1000 (95% CI 140 to 663 more), for a nonadherent median probability of healing of 242 per 1000. Although this contrast was assessed as itself representing moderate‐certainty evidence, it sits in the context of a network which was, overall, judged to represent low‐certainty evidence, and should therefore be considered with appropriate caution. We also note that many of the trials which contributed to the contrast (and the direct comparison) were at an unclear risk of bias. Therefore, although there is no clear high risk of bias, there is also a lack of clarity about the true risk of bias.

For each of six contrasts the low certainty of the evidence means it is unclear whether the intervention increases the probability of healing; for two more the certainty of the evidence was very low:

sucralfate versus foam dressing (RR 5.94, 95% CI 1.96 to 18.0);
sucralfate versus hydrocolloid dressing (RR 6.51; 95% CI 2.17 to 19.6);
sucralfate versus nonadherent dressing (RR 6.80, 95% CI 2.24 to 20.7);
silver dressing versus foam dressing (RR 2.12; 95% CI 1.46 to 3.07);
silver dressing versus hydrocolloid dressing (RR 2.32; 95% CI 1.58 to 3.41);
foam dressing versus nonadherent dressing (RR 1.15; 95% CI 0.91 to 1.44);
foam dressing versus hydrocolloid dressing (RR 1.10; 95% CI 0.93 to 1.28);
hydrocolloid dressing versus nonadherent dressing (RR 1.04; 95% CI 0.85 to 1.29).

In each of these six contrasts, the evidence was graded as low certainty; downgraded either once for imprecision and once for risk of bias (sucralfate versus foam; sucralfate versus hydrocolloid; sucralfate versus nonadherent dressing; foam versus nonadherent) or twice for inconsistency (silver versus foam); or once for risk of bias and once for inconsistency (silver versus hydrocolloid).

It is unclear whether there is a difference in the probability of healing for the remaining two contrasts because the evidence is of very low certainty (downgraded for risk of bias (twice) and imprecision (once) or for risk of bias, imprecision and inconsistency): foam versus hydrocolloid; and hydrocolloid versus nonadherent dressing. The contrasts with sucralfate were informed by one study with 100 participants in the direct evidence, with 43/50 events for sucralfate and five events for hydrogel; we therefore downgraded further for imprecision to allow for the fragility this invoked.

Comparison of results from the NMA with the direct evidence

Of the eight contrasts with more than one study, five had an I² of 0%; the remaining three were downgraded for inconsistency; one was downgraded twice for inconsistency. Details are given in Table 3.

Discussion

available in

Summary of main results

We conducted a network meta‐analysis (NMA) of dressings and topical agents for healing venous leg ulcers. The network included 59 studies with 5156 participants. The systematic review that underpins the NMA includes 78 RCTs involving a total of 7014 participants, comparing different dressings or topical agents or combinations of treatments for the healing of venous leg ulcers. This included a range of treatments from the most widely‐used categories of dressings to experimental treatments assessed by a single research study.

We treated each topical agent as a separate intervention, but grouped dressings by class as described in the BNF 2016 (e.g. alginates, hydrocolloids). There were many interventions, often involving small single studies with atypical or experimental treatments. In order to simplify and rationalise the NMA, we produced a list of important and more widely‐used treatments with clinical direction and input from review authors who had not seen the results. This led to the 'base‐case' NMA, which we extended following sensitivity analysis, adding three linking 'ineligible' interventions to obtain greater robustness.

Alongside the analysis, we have applied a new method of GRADE assessment (Salanti 2014), which allows us to view the results in the light of the certainty of their findings. Using this approach, we found the evidence for the network as a whole was of low certainty (downgraded for risk of bias and imprecision). The network presents results derived from 59 studies of 25 interventions evaluating 40 direct comparisons: we highlight the results from contrasts involving the two treatments with the highest mean ranks (sucralfate and silver): the majority of the evidence for individual contrasts was of low or very low certainty, and was mainly downgraded for risk of bias and imprecision; there was a limited degree of inconsistency for some contrasts (see Quality of the evidence).

In summary:

overall findings reflect the uncertainty of the component evidence and the sparseness of the network. For the network as a whole, the evidence was of low certainty. With so many interventions that appeared to have similar efficacies, there was considerable uncertainty in the middle ranks, but numerically two treatments had more than 50% probability of being the best (sucralfate and silver dressings); ‐ see also Quality of the evidence.
for the head‐to‐head comparison of these two treatments with the highest mean ranks, it is very uncertain whether there is a difference between sucralfate and silver dressing in the probability of venous leg ulcer healing (very low‐certainty evidence);
silver dressings may increase the probability of venous leg ulcer healing, compared with nonadherent dressings: RR 2.43, 95% CI 1.58 to 3.74 (moderate‐certainty evidence in the context of a low‐certainty network);
in the other contrasts between these treatments with the highest probability of being best and the most widely‐used dressing classes, it was unclear whether the intervention increased the probability of healing; in each case this was low‐ or very low‐certainty evidence characterised by wide confidence intervals;
one of the sensitivity analyses highlighted some instability in key aspects of the network; this instability is likely to be due to sparseness. As a consequence, we reported the results of the extended sensitivity analysis.

Overall completeness and applicability of evidence

The studies included in the review do not represent all the studies which have been conducted on relevant interventions; substantial numbers of studies were excluded because they did not report the outcome of complete wound healing. However, this was an issue across treatments and did not appear to impact disproportionately on any particular treatment or comparison. As discussed below we believe that this approach was the appropriate one for the purpose of the review.

The populations represented in the included studies appear representative of the people who present with venous leg ulcers in clinical practice in terms of age, gender and ulcer characteristics at enrolment. However, although many studies specified characteristics such as ulcer dimensions, duration and infection in inclusion criteria or reported these in participant details, they were much less likely to specify or describe wound characteristics such as levels of slough, exudate or necrosis.

We identified a wide range of eligible interventions, and included both dressings and topical treatments; specific dressing types such as impregnated dressings and modern 'advanced' dressings were well represented. We conducted sensitivity analyses in order to assess the stability of the network and the impact of decreasing the number of included interventions to a smaller clinically‐defined set, or expanding it in order to increase the amount of evidence available for key interventions which were particularly poorly linked into the network.

The review included a substantial number of studies not included in the network; in particular, studies which compared specific interventions such as honey with standard care or choices of multiple treatments. The inclusion of these studies in the review means that they are easily identifiable for researchers who may wish to conduct alternative analyses using this type of data. We believe that the choices we have made concerning data to include in the network meta‐analysis are likely to maximise its relevance to clinical decision‐making, but acknowledge that this is balanced against the availability of only direct evidence for some comparisons.

Quality of the evidence

A high proportion of the included studies were considered to be at high risk of bias for one or more domains and a substantial number were at very high risk of overall bias. The principal reasons for a study to be considered at high risk of bias were lack of blinding of one or more groups of participants, professionals and outcome assessors, and attrition bias. However, many studies which were not considered to be at high risk of bias had unclear risks of bias for several or even all domains. Therefore, even when a contrast has not been downgraded due to high risk of bias in the contributions matrix, this does not mean that we are confident that there is a low risk of bias pertaining to the contrast, but merely that there is no known high or very high risk of bias.

Many comparisons (the majority) were informed by a single trial, and most trials were small and underpowered. Only a few comparisons ‐ between some of the most widely‐used dressing types ‐ were represented by multiple trials and substantial numbers of participants and events. This is reflected in the wide confidence intervals and therefore the imprecision of most contrasts in the NMA. Some contrasts were also judged to be affected by inconsistency. These factors, together with high risks of bias, meant that many key contrasts were judged to be low or very low certainty while the network as a whole was judged to represent low‐certainty evidence.

The inclusion criteria and the nature of the evidence included meant that we did not downgrade for indirectness and we also found no evidence of publication bias.

Potential biases in the review process

Although all the included studies were reported in English, we ordered a number of full‐texts in languages other than English; including Polish, German, Portuguese, Dutch, Norwegian, Chinese, Italian and Spanish. These were ultimately excluded as they did not meet the inclusion criteria, but would clearly have been included if they had proved eligible.

We searched a number of databases and checked the references of reviews and included studies; time constraints meant that the planned searches of trials registers were not conducted. We found no evidence of publication bias, and our focus on the single outcome of healing means that trials identified from registers were unlikely to have data which would have led to their inclusion in the network. We found a relatively small number of unobtainable records; close examination of the records for these led us to conclude that the studies they represented were unlikely to have been included in the review.

This NMA and review focused on the outcome of complete healing. The impact of including only studies reporting healing in this way was considerable; lack of these data was the single most common substantive reason for excluding a study. Complete healing is the outcome which is most important to people living with venous leg ulcers and therefore we believe that the decision to focus the network on this outcome was the right one. Other reviews include studies that focused on other outcomes considered important to people with lived experience of the condition; this review stands alongside those syntheses and does not seek to replicate them.

There is potential for bias in our choice of base‐case and sensitivity analysis and also our choice of studies with only one eligible intervention for the expanded base‐case sensitivity analysis. We made a post‐hoc decision to focus on a base‐case of interventions which were likely to be used in clinical practice. Clearly post‐hoc decisions of this nature could be a source of bias in their impact on which interventions were included. However, no interventions were excluded from the review on the basis of this decision: comparisons including interventions judged to be partly relevant are included in the review and the direct evidence is available to the reader. The decisions on which interventions should be included in the base‐case and the narrow sensitivity analysis were made on clinical grounds rather than on the basis of known results; they were made independently by two authors, one of whom had no access to the extracted data at that point, and who were in almost complete agreement when the decisions were compared; where there was a disagreement a more inclusive approach was adopted. The effect of the approach adopted was to remove some of the noise in what was a sparsely‐populated network and to increase our ability to examine the relative effectiveness of treatments relevant to clinical practice.

Our updated search in March 2018 identified nine studies, which may be eligible for inclusion but which have not yet been incorporated into the review. None of these was large in absolute terms but the results of these studies may nevertheless have some impact on our sparse network.

Agreements and disagreements with other studies or reviews

We have been unable to identify any other NMAs examining dressings and topical agents for healing venous leg ulcers. The high level of uncertainty around contrasts between most dressings reflects that in the most recent NICE guidance (NICE 2016a) and the most recent report by the AHRQ (AHRQ 2013); these reflect in part the findings of a number of Cochrane reviews of individual types of dressing (see Why it is important to do this review). The 2010 guidance by SIGN (SIGN 2010) recommended the use of nonadherent dressings with possible alternatives being hydrocolloids, alginates or hydrogels. The results of the NMA do not conflict with this advice, suggesting broadly comparable efficacy for complete healing in these dressing categories.

The finding that silver dressings may increase the number of ulcers healed does not take account of the largest trial available for silver (Michaels 2009). This is because both arms of this trial contained more than one treatment class (specifically, silver‐containing dressings and silver sulfadiazine (SSD)), and hence could not be integrated into the NMA. Michaels 2009 found no difference in overall healing between the silver and non‐silver arms of this study (an RR of 1.00, 95% CI 0.95 to 1.06 at one year's follow‐up), but we note that the 'silver' arm included 39% of participants receiving SSD, which may have substantially changed the effect in this study. Nevertheless, the data from this trial should be borne in mind when considering the results and when planning any further research on these treatments.

Figure 1

Study flow diagram.

Figure 2

'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.

Figure 3

Network diagram ‐ extended network, by risk of bias (3 categories)

Key: green = low/unclear; yellow = high; red = very high overall risk of bias for the contrast. The number of studies for each contrast is given in Table 3.

Figure 4

Rankograms for the extended base‐case showing the probability that each intervention is the best, second best, third best treatment, etc.

Figure 5

Rankograms for all treatments in the extended base‐case network showing the probability that each intervention is the best, second best, third best treatment, etc.

Figure 6

Contour‐enhanced funnel plot for the extended base‐case network showing comparison‐specific pooled effect sizes

Figure 7

Network diagram for the base‐case, coded by risk of bias (3 categories)

Key: green = low/unclear; yellow = high; red = very high overall risk of bias for the contrast. The number of studies for each contrast is given in Table 3

Figure 8

Rankograms for the base‐case network showing the probability that each intervention is the best, second best, third best treatment, etc.

Figure 9

Rankograms for all treatments in base‐case network showing the probability that each intervention is the best, second best, third best treatment, etc.

Figure 10

Rankograms for the narrow sensitivity analysis showing the probability that each intervention is the best, second best, third best treatment, etc.

Figure 11

Comparisons of alginate with other treatments in the extended base‐case network with risk of bias

Figure 12

Comparisons of cadexomer iodine with other treatments in the extended base‐case network with risk of bias

Figure 13

Comparisons of foam with other treatments in the extended base‐case network with risk of bias

Figure 14

Comparisons of hydrocolloid with other treatments in the extended base‐case network with risk of bias

Figure 15

Comparisons of hydrofibre with other treatments in the extended base‐case network with risk of bias

Figure 16

Comparisons of hydrogel with other treatments in the extended base‐case network with risk of bias

Figure 17

Comparisons of paste bandage with other treatments in the extended base‐case network with risk of bias

Figure 18

Comparisons of PMM and PMM silver with other treatments in the extended base‐case network with risk of bias

Figure 19

Comparisons of saline gauze with other treatments in the extended base‐case network with risk of bias

Figure 20

Comparisons of silver and SSD with other treatments in the extended base‐case network with risk of bias

Figure 21

Comparisons of nonadherent with other treatments in the extended base‐case network with risk of bias

Figure 22

Risk of bias contributions to each NMA comparison in the extended base‐case network (vertical scale 0% to 100%)
red = very high risk of bias, yellow = high risk of bias, green = low risk of bias

Analysis 1.1

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 1 Alginate vs nonadherent.

Analysis 1.2

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 2 Cadexomer iodine vs nonadherent.

Analysis 1.3

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 3 Film vs nonadherent.

Analysis 1.4

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 4 Foam vs nonadherent.

Analysis 1.5

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 5 Hyaluronic plus povidone vs nonadherent.

Analysis 1.6

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 6 Hydrocolloid vs non‐adherent.

Analysis 1.7

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 7 Hydrofibre vs nonadherent.

Analysis 1.8

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 8 Hydrogel vs nonadherent.

Analysis 1.9

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 9 PMM vs nonadherent.

Analysis 1.10

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 10 SSD vs nonadherent.

Analysis 1.11

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 11 Foam vs alginate.

Analysis 1.12

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 12 Hydrocolloid vs alginate.

Analysis 1.13

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 13 Hydrofibre vs alginate.

Analysis 1.14

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 14 Paste bandage vs alginate.

Analysis 1.15

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 15 PMM vs alginate.

Analysis 1.16

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 16 Gentian violet vs cadexomer iodine.

Analysis 1.17

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 17 Hydrocolloid vs cadexomer iodine.

Analysis 1.18

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 18 Hydrocolloid vs foam.

Analysis 1.19

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 19 Ibuprofen foam vs foam.

Analysis 1.20

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 20 Octenidine vs foam.

Analysis 1.21

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 21 Paste bandage vs foam.

Analysis 1.22

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 22 PMM vs foam.

Analysis 1.23

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 23 PMM silver vs foam.

Analysis 1.24

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 24 Silver vs foam.

Analysis 1.25

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 25 Saline gauze vs hyaluronic acid.

Analysis 1.26

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 26 PMM silver vs hydrocolloid.

Analysis 1.27

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 27 Povidone iodine vs hydrocolloid.

Analysis 1.28

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 28 Saline gauze vs hydrocolloid.

Analysis 1.29

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 29 Silver vs hydrocolloid.

Analysis 1.30

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 30 Zinc oxide vs hydrocolloid.

Analysis 1.31

Comparison 1 Direct evidence ‐ included in base‐case network, Outcome 31 Sucralfate vs hydrogel.

Analysis 2.1

Comparison 2 Direct evidence ‐ not in base case network, in expanded base case, Outcome 1 Blood product vs emollient.

Analysis 2.2

Comparison 2 Direct evidence ‐ not in base case network, in expanded base case, Outcome 2 Blood product vs hydrocolloid.

Analysis 2.3

Comparison 2 Direct evidence ‐ not in base case network, in expanded base case, Outcome 3 Blood product vs hydrogel.

Analysis 2.4

Comparison 2 Direct evidence ‐ not in base case network, in expanded base case, Outcome 4 Blood product vs saline gauze.

Analysis 2.5

Comparison 2 Direct evidence ‐ not in base case network, in expanded base case, Outcome 5 Hyaluronic vs emollient cream.

Analysis 2.6

Comparison 2 Direct evidence ‐ not in base case network, in expanded base case, Outcome 6 Growth factor vs hydrocolloid.

Analysis 2.7

Comparison 2 Direct evidence ‐ not in base case network, in expanded base case, Outcome 7 Growth factor vs hydrogel.

Analysis 2.8

Comparison 2 Direct evidence ‐ not in base case network, in expanded base case, Outcome 8 Growth factor vs nonadherent.

Analysis 2.9

Comparison 2 Direct evidence ‐ not in base case network, in expanded base case, Outcome 9 SSD vs emollient.

Analysis 3.1

Comparison 3 Direct evidence ‐ not in network, Outcome 1 A. Pichinchensis vs alginate.

Analysis 3.2

Comparison 3 Direct evidence ‐ not in network, Outcome 2 Non‐adherent vs cellulose.

Analysis 3.3

Comparison 3 Direct evidence ‐ not in network, Outcome 3 Phenytoin vs no treatment.

Analysis 3.4

Comparison 3 Direct evidence ‐ not in network, Outcome 4 Cadexomer iodine vs standard treatment.

Analysis 3.5

Comparison 3 Direct evidence ‐ not in network, Outcome 5 Honey vs standard treatment.

Analysis 3.6

Comparison 3 Direct evidence ‐ not in network, Outcome 6 Papain vs hydrogel.

Analysis 3.7

Comparison 3 Direct evidence ‐ not in network, Outcome 7 Shale oil vs hydrogel.

Analysis 3.8

Comparison 3 Direct evidence ‐ not in network, Outcome 8 Tripeptide copper vs hydrogel.

Analysis 3.9

Comparison 3 Direct evidence ‐ not in network, Outcome 9 Hydrocolloid vs collagen.

Analysis 3.10

Comparison 3 Direct evidence ‐ not in network, Outcome 10 Hydrocolloid vs dextranomer.

Analysis 3.11

Comparison 3 Direct evidence ‐ not in network, Outcome 11 Hydrocolloid vs magnesium sulphate.

Analysis 3.12

Comparison 3 Direct evidence ‐ not in network, Outcome 12 Hydrocolloid vs nonadherent or iodine.

Analysis 3.13

Comparison 3 Direct evidence ‐ not in network, Outcome 13 Ozonated oil vs zinc oxide.

Analysis 3.14

Comparison 3 Direct evidence ‐ not in network, Outcome 14 Cadexomer iodine vs dextranomer.

Analysis 3.15

Comparison 3 Direct evidence ‐ not in network, Outcome 15 Silica gel fibre vs standard care.

Analysis 3.16

Comparison 3 Direct evidence ‐ not in network, Outcome 16 Silver vs non‐silver.

Analysis 3.17

Comparison 3 Direct evidence ‐ not in network, Outcome 17 Sulphadryl vs inactive powder.

Analysis 3.18

Comparison 3 Direct evidence ‐ not in network, Outcome 18 Tripeptide copper vs emollient cream.

Analysis 3.19

Comparison 3 Direct evidence ‐ not in network, Outcome 19 Tripeptide copper vs SSD.

Summary of findings for the main comparison. NMA evidence: proportion with complete healing

NMA evidence for base‐case network: proportion with complete healing
Patient or population: people with venous leg ulcers Intervention: dressing or topical agent Comparator: alternative dressing or topical agent Settings: hospital, community or care home, or combinations
Contrasts	Relative effect (95% CI)	*Anticipated absolute effects (95% CI) ‐ from median of control groups in direct evidence**		Certainty of the evidence (GRADE)	Comments
		Median CGR	With intervention
Sucralfate versus nonadherent	RR 6.80 (2.24 to 20.7)	242 per 1000	1000 per 1000 (542 to 1000)	⊕⊕⊝⊝ Low^a,b	Base‐case: RR 17.2 (95% CI 1.52 to 193). Large differences between base‐case and extended base‐case. The calculated absolute effect for the intervention is more than 1000 per 1000 for the point estimate and its upper confidence limit; and so the corresponding values for the absolute risk difference are also approximated by 1000 per 1000.
		1000 more people healed per 1000 (300 to 1000 more)
Sucralfate versus foam	RR 5.94 (1.96 to 18.0)	376 per 1000	1000 per 1000 (737 to 1000)	⊕⊕⊝⊝ Low^a,b	Base‐case: RR 14.8 (95% CI 1.30 to 169) Large differences between base‐case and extended base‐case. The calculated absolute effect for the intervention is more than 1000 per 1000 for the point estimate and its upper confidence limit; and so the corresponding values for the absolute risk difference are also approximated by 1000 per 1000.
		1000 more people healed per 1000 (361 to 1000 more)
Sucralfate versus hydrocolloid	RR 6.51 (2.17 to 19.6)	433 per 1000	1000 per 1000 (940 to 1000)	⊕⊕⊝⊝ Low^a,b	Base‐case: RR 16.24 (95% CI 1.43 to 185) Large differences between base‐case and extended base‐case The calculated absolute effect for the intervention is more than 1000 per 1000 for the point estimate and its upper confidence limit; and so the corresponding values for the absolute risk difference are also approximated by 1000 per 1000.
		1000 more people healed per 1000 (507 to 1000 more)
Silver versus nonadherent	RR 2.43 (1.58 to 3.74)	242 per 1000	588 per 1000 (382 to 905)	⊕⊕⊕⊝ Moderate^a
		346 more people healed per 1000 (140 to 663 more)
Silver versus foam	RR 2.12 (1.46 to 3.07)	376 per 1000	797 per 1000 (549 to 1000)	⊕⊕⊝⊝ Low^c	Direct evidence: Analysis 1.24
		421 more people healed per 1000 (173 to 786 more)
Silver versus hydrocolloid	RR 2.32 (1.58 to 3.41)	433 per 1000	1000 per 1000 (684 to 1000)	⊕⊕⊝⊝ Low^a,d
		567 more people healed per 1000 (251 to 1000 more)
Sucralfate versus silver	RR 2.80 (0.88 to 8.97)	81 per 1000	225 per 1000 (71 to 722)	⊕⊝⊝⊝ Very low^a,e	Base‐case: RR 6.99 (95% CI 0.60 to 82.0) Large differences between base‐case and extended base‐case
		145 more people healed per 1,000 (10 fewer to 642 more)
Foam versus hydrocolloid	RR 1.10 (0.93 to 1.28)	433 per 1000	476 per 1000 (402 to 554)	⊕⊝⊝⊝ Very low^f,g,h	Direct evidence: Analysis 1.18
		43 more people healed per 1000 (from 31 fewer to 121 more)
Foam versus nonadherent dressing	RR 1.15 (0.91 to 1.44)	242 per 1000	278 per 1000 (220 to 348)	⊕⊕⊝⊝ Low^a,h
		36 more people healed per 1000 (from 22 fewer to 106 more)
Hydrocolloid versus nonadherent dressing	RR 1.04 (0.85 to 1.29)	242 per 1000	251 per 1000 (206 to 312)	⊕⊝⊝⊝ Very low^a,h,i	Direct evidence: Analysis 1.6
		9 more people healed per 1000 (from 36 fewer to 70 more)
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparator group and the relative effect of the intervention (and its 95% CI). CGR: control group risk; CI: confidence interval; NMA: network meta‐analysis; RR: risk ratio
GRADE Working Group grades of evidence High certainty (quality): we are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty (quality): we are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty (quality): our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty (quality): we have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
^a. NMA risk of bias from contributions matrix and direct evidence risk of bias (downgrade once) ^b. Imprecision ‐ direct evidence involving sucralfate: 1 study 43/50 events (sucralfate); 5 events (hydrogel) (downgrade once) ^c. Heterogeneity in point estimates for direct evidence; significant inconsistency in node splitting and in inconsistency factor (loop) (downgrade twice) ^d. Significant inconsistency in node splitting and in inconsistency factor (loop) (downgrade once) ^e. Imprecision ‐ CI crosses one MID (1.25) and direct evidence involving sucralfate: 43/50 events (sucralfate) and 5 events (hydrogel) (downgrade twice) ^f. NMA risk of bias from contributions matrix and direct evidence risk of bias (downgrade twice) ^g. Slight heterogeneity in point estimates for direct evidence; significant inconsistency in node splitting and inconsistency factor (downgrade once) ^h. Imprecision ‐ CI crosses one MID (1.25) (downgrade once) ^i. High heterogeneity in direct evidence (downgrade twice)

Summary of findings for the main comparison. NMA evidence: proportion with complete healing

Table 1. Summary characteristics of individual studies

Study characteristic	Details of studies
Publication	Abstract or poster only: Caprio 1992; Casoni 2002; Hanft 2006; Ivins 2006; Kalis 1993; Lanzara 2008; Petkov 1997; Zuccarelli 1992. All other studies had a full publication
Multiple interventions	Three arms: Bishop 1992; De Araujo 2016; Hansson 1998; Robson 1995. All other studies had two arms
Unit of randomisation	Ulcer: Caprio 1992 Leg: Stacey 1997 Unclear: Hanft 2006; Kalis 1993; Leaper 1991 All other studies used participants as the unit of randomisation
Funding	Industry: Armstrong 1997; Backhouse 1987; Beckert 2006; Bishop 1992; Bowszyc 1995; Charles 2002; Dereure 2012a; Fogh 2012; Gottrup 2008; Hansson 1998;Humbert 2013; Jørgensen 2005; Kelechi 2012; Lanzara 2008; Leaper 1991; Meaume 2012; Moffatt 1992a; Moffatt 1992b; Moss 1987; Nelson 2007; Norkus 2005; Robson 1995; Scurr 1994; Senet 2003; Senet 2014; Smith 1992; Smith 1994; Stacey 2000; Vin 2002; Zuccarelli 1992. Others did not report funding source or reported no funding or a non‐industry source
Follow‐up time	4 weeks: Bishop 1992; Ivins 2006; Jørgensen 2005; Schulze 2001; 30 days: Solovastru 2015; 6 weeks: Armstrong 1997; Biland 1985;Fogh 2012; Gottrup 2008; Leaper 1991; Meredith 1988; Ohlsson 1994; Robson 1995; Scurr 1994; Senet 2014; Smith 1994;Steele 1986; 8 weeks: Brandrup 1990; Caprio 1992; Meaume 2012; Taddeucci 2004; Tarvainen 1988; 60 days: De Araujo 2016; Dereure 2012a; Humbert 2013; 9 weeks: Dimakakos 2009; Kalis 1993 10 weeks: Arnold 1994;12 weeks: Backhouse 1987; Blair 1988a; Blair 1988b; Callam 1992; Charles 2002; Hanft 2006; Hansson 1998; Harding 2001;Lanzara 2008; Luiza 2015; Moffatt 1992a; Moffatt 1992b; Ormiston 1985; Rasmussen 1991; Robson 2001; Romanelli 2015a; Salim 1992; Senet 2003; Senet 2011; Vanscheidt 2012; Vin 2002; Zuccarelli 1992; 3 months: Casoni 2002; 90 days: Tumino 2008; 13 weeks:Arenbergerova 2013; Thomas 1997; 4 months: Smith 1992; 20 weeks: Beckert 2006; Kelechi 2012; Robson 2004; 24 weeks: Nelson 2007;26 weeks: Moss 1987; Petkov 1997; 9 months: Stacey 1997; Stacey 2000; 10 months: Romero‐Cerecero 2012; 12 months; Norkus 2005; Rubin 1990; Unclear/till healing: Greguric 1994; Sopata 2016 (max 40 weeks) Kucharzewski 2013 (max 16 weeks)
Included < 25% non venous leg ulcers	Included non‐venous leg ulcers: Armstrong 1997; Biland 1985; Brandrup 1990; Norkus 2005; Ohlsson 1994; Rasmussen 1991 Unclear: Backhouse 1987; Humbert 2013; Ivins 2006; Jørgensen 2005; Leaper 1991; Luiza 2015; Romero‐Cerecero 2012; Senet 2011; Tarvainen 1988; Zuccarelli 1992. All others enrolled only participants with VLU
VLU: venous leg ulcers

Table 1. Summary characteristics of individual studies

Table 2. Studies: status in network/review

Study	Interventions	No eligible interventions	Expanded base‐case	Base‐case	Sensitivity analysis	Risk of bias
Alvarez 2012^c	Cellulose Nonadherent	2	X	X	X	High
Arenbergerova 2013^b	Hydrofibre Blood product	1	√	X	X	Very high
Armstrong 1997^a	Alginate Hydrofibre	2	√	√	√	High
Arnold 1994^c	Hydrocolloid Iodine OR nonadherent	2	X	X	X	Very high
Backhouse 1987^a	Nonadherent Hydrocolloid	2	√	√	√	Low/Unclear
Banerjee 1997^a	Nonadherent Film	2	√	√	√	Very high
Beckert 2006^c	Shale oil Hydrogel	2	X	X	X	Low/unclear
Biland 1985^b	Blood product Emollient cream	1	√	X	X	High
Bishop 1992^b	Tripeptide copper Emollient cream SSD	3	√	X	X	Low/unclear
Blair 1988a^a	Nonadherent Hydrocolloid	2	√	√	√	High
Blair 1988b^a	Nonadherent SSD	2	√	√	√	High
Bowszyc 1995^a	Foam Hydrocolloid	2	√	√	√	High
Brandrup 1990^a	Hydrocolloid Zinc oxide	2	√	√	√	High
Brown 2014^c	Silica gel Alternative traditional dressings	2	X	X	X	Very high
Callam 1992^a	Nonadherent Foam		√	√	√	Very high
Caprio 1992^c	Hydrocolloid Collagen	2	X	X	X	Low/unclear
Casoni 2002^a	Nonadherent Hyaluronic+Povidone	2	√	√	X	Low/unclear
Charles 2002^a	Foam Hydrocolloid	2	√	√	√	Low/unclear
De Araujo 2016^b	Blood product Hydrogel Papain	3 (2 in network)	√	X	X	Low/unclear
Dereure 2012a^b	Hyaluronic acid Emollient cream	2	√	X	X	Low/unclear
Dimakakos 2009^a	Foam Silver	2	√	√	√	Low/unclear
Fogh 2012^a	Foam Silver	2	√	√	√	High
Gottrup 2008^a	Foam Silver	2	√	√	√	High
Greguric 1994^c	Magnesium sulphate Hydrocolloid	2	X	X	X	High
Hanft 2006^a	PMM silver Hydrocolloid	2	√	√	√	High
Hansson 1998^a	Nonadherent Cadexomer iodine Hydrocolloid	3	√	√	√	High
Harcup 1986^c	Standard care Cadexomer iodine	2	X	X	X	Low/unclear
Harding 2001^a	Alginate Hydrofibre	2	√	√	√	Very high
Hokkam 2011^c	Phenytoin No treatment	2	X	X	X	Low/unclear
Humbert 2013^a	Hyaluronic acid Saline gauze	2	√	√	√	High
Ivins 2006^a	Foam Silver	2	√	√	√	Low/unclear
Jørgensen 2005^a	Foam Silver	2	√	√	√	Low/unclear
Jull 2008^c	Honey Standard care	2	X	X	X	Very high
Kalis 1993^c	Hydrocolloid Dextranomer	2	X	X	X	Very high
Kelechi 2012^a	Nonadherent Hydrofibre	2	√	√	√	High
Kucharzewski 2013^a	Hydrocolloid Silver	2	√	√	√	High
Lanzara 2008^a	PMM silver Foam	2	√	√	√	High
Leaper 1991^a	Nonadherent Hydrocolloid	2	√	√	√	High
Lindsay 1986^c	Standard care Cadexomer iodine	2	X	X	X	Low/unclear
Luiza 2015^c	Papain Hydrogel	2	X	X	X	High
Meaume 2012^a	PMM Foam	2	√	√	√	High
Meredith 1988^a	Nonadherent Hydrocolloid	2	√	√	√	Low/unclear
Michaels 2009^c	Silver non‐silver	2	X	X	X	High
Moffatt 1992a^a	Nonadherent Hydrocolloid	2	√	√	√	Low/unclear
Moffatt 1992b^a	Alginate Nonadeherent	2	√	√	√	Low/unclear
Moss 1987^c	Cadexomer iodine Dextranomer	2	X	X	X	Very high
Nelson 2007^a	Nonadherent Hydrocolloid	2	√	√	√	Very high
Norkus 2005^a	Foam Hydrocolloid	2	√	√	√	Very high
Ohlsson 1994^a	Hydrocolloid Saline gauze	2	√	√	√	Low/unclear
Ormiston 1985^b	Cadexomer iodine gentian violet	2	√	√	X	High
Petkov 1997^a	PMM Alginate	2	√	√	√	Low/unclear
Rasmussen 1991^b	Growth factor Hydrocolloid	1	X	X	√	High
Robson 1995^b	Growth factor Nonadherent	1	X	X	√	Low/unclear
Robson 2001^b	Growth factor Nonadherent	1	X	X	√	High
Robson 2004^b	Growth factor Nonadherent	1	X	X	√	Low/unclear
Romanelli 2015a^a	PMM Alginate	2	√	√	√	Low/unclear
Romero‐Cerecero 2012^c	A. Pichinchensis Alginate	2	X	X	X	High
Rubin 1990^a	Foam paste bandage	2	√	√	√	High
Salim 1992^c	Sulphadryl Inactive powder	2	X	X	X	High
Schulze 2001^a	Foam alginate	2	√	√	√	Very high
Scurr 1994^a	Hydrocolloid Alginate	2	√	√	√	Low/unclear
Senet 2003^b	Blood product Hydrocolloid	1	X	X	√	Low/unclear
Senet 2011^b	Growth factor Hydrogel	1	X	X	√	High
Senet 2014^a	silver foam	2	√	√	√	High
Smith 1992^a	Hydrocolloid Povidone iodine	2	√	√	√	Very high
Smith 1994^a	Hydrocolloid alginate	2	√	√	√	Very high
Solovastru 2015^c	Ozonated oil Emollient cream	2	X	X	X	Low/unclear
Sopata 2016^a	hydrocolloid foam	2	√	√	√	Very high
Stacey 1997^a	Paste bandage alginate	2	√	√	√	High
Stacey 2000^b	Blood product Saline gauze	1	X	X	√	High
Steele 1986^c	Standard care Cadexomer iodine	2	X	X	X	Low/unclear
Taddeucci 2004^a	Nonadherent Hydrogel	2	√	√	√	Very high
Tarvainen 1988^c	Cadexomer iodine Dextranomer	2	X	X	X	Very high
Thomas 1997^a	Foam Hydrocolloid	2	√	√	√	Very high
Tumino 2008^a	Sucralfate Hydrogel	2	√	√	√	High
Vanscheidt 2012^a	Octenidine Foam	2	√	√	X	High
Vin 2002^a	PMM nonadherent	2	√	√	√	Low/unclear
Zuccarelli 1992^a	Foam Hydrocolloid	2	√	√	√	Low/unclear
Abbreviations: PMM: protease modulating matrix; SSD: silver sulphadiazine ^aStudy in original base‐case ^bStudy only included in sensitivity analysis ^cStudy included in review but not in network

Table 2. Studies: status in network/review

Table 3. Direct comparisons for individual interventions compared with NMA results

Contrast/comparison	Number of studies (participants)	Studies	RR (95% CI) direct evidence. Random effects (inverse variance) Heterogeneity statistics	NMA results (extended base‐case; consistency assumption): RR (95% CI)
Comparisons with nonadherent: RR > 1 indicates greater proportion healing with specified alternative treatment
Alginate	1 (113)	Moffatt 1992b	1.08 (0.86 to 1.36)	1.21 (0.92 to 1.60)
Cadexomer iodine	1 (105)	Hansson 1998	1.00 (0.39 to 2.56)	1.16 (0.50 to 2.69)
Film	1 (71)	Banerjee 1997	1.34 (0.61 to 2.92)	1.34 (0.61 to 2.95)
Foam	1 (124)	Callam 1992	1.35 (0.89 to 2.05)	1.15 (0.91 to 1.44)
Hyaluronic acid plus povidone iodine	1 (55)	Casoni 2002	1.93 (0.95 to 3.92)	1.93 (0.94 to 3.96)
Hydrocolloid	7 (662)	Backhouse 1987; Blair 1988a; Hansson 1998; Leaper 1991; Meredith 1988; Moffatt 1992a; Nelson 2007	1.26 (0.92 to 1.72) I² = 69%; P = 0.004	1.04 (0.85 to 1.29)
Hydrofibre	1 (82)	Kelechi 2012	1.47 (0.88 to 2.46)	1.39 (0.93 to 2.08)
Hydrogel	1 (24)	Taddeucci 2004	2.00 (0.21 to 19.23)	0.79 (0.39 to 1.62)
PMM	1 (74)	Vin 2002	1.42 (0.80 to 2.51)	1.31 (0.93 to 1.84)
SSD	1 (60)	Blair 1988b	0.79 (0.57 to 1.10)	0.81 (0.57 to 1.15)
Growth factor^a	3 (460)	Robson 1995; Robson 2001; Robson 2004	0.96 (0.81 to 1.14) I² = 0%; P = 0.65	0.95 (0.72 to 1.25)
Comparisons with alginate: RR > 1 indicates greater proportion healing with specified alternative treatment
Foam	1 (113)	Schulze 2001	0.55 (0.10 to 2.86)	0.94 (0.72 to 1.23)
Hydrocolloid	2 (80)	Scurr 1994; Smith 1994	0.72 (0.15 to 3.42) I² = 52%; P = 0.15	0.86 (0.68 to 1.11)
Hydrofibre	2 (175)	Armstrong 1997; Harding 2001	1.47 (0.48 to 4.47) I² = 54%; P = 0.14	1.15 (0.77 to 1.72)
Paste bandage	1 (133)	Stacey 1997	1.22 (0.91 to 1.63)	1.39 (1.01 to 1.90)
PMM	2 (140)	Petkov 1997; Romanelli 2015a	1.10 (0.84 to 1.46) I² = 0%; P = 0.87	1.08 (0.83 to 1.40)
Comparisons with cadexomer iodine: RR > 1 indicates greater proportion healing with specified alternative treatment
Hydrocolloid	1 (104)	Hansson 19980	0.73 (0.26 to 2.08)	0.90 (0.39 to 2.10)
Gentian violet	1 (60)	Ormiston 1985	0.58 (0.27 1.28)	0.58 (0.26 to 1.29)
Comparisons with foam: RR > 1 indicates greater proportion healing with specified alternative treatment
Hydrocolloid	6 (458)	Bowszyc 1995; Charles 2002; Norkus 2005; Sopata 2016; Thomas 1997; Zuccarelli 1992	0.92 (0.77 to 1.08) I² = 0%; P = 0.84	0.91 (0.78 to 1.07)
Ibuprofen	2 (242)	Fogh 2012; Gottrup 2008	0.88 (0.48 to 1.61) I² = 0%; P = 0.79	0.88 (0.48 to 1.62)
Octenidine	1 (126)	Vanscheidt 2012	1.03 (0.56 to 1.90)	1.03 (0.55 to 1.92)
Paste bandage	1 (36)	Rubin 1990	2.30 (1.29 to 4.10)	1.47 (0.99 to 2.17)
PMM	1 (187)	Meaume 2012	0.87 (0.30 to 2.48)	1.14 (0.82 to 1.60)
PMM silver	1 (30)	Lanzara 2008	1.57 (0.84 to 2.92)	1.15 (0.78 to 1.71)
Silver	4 (397)	Dimakakos 2009;Ivins 2006; Jørgensen 2005; Senet 2014	1.65 (1.08 to 2.52) I² = 0%; P = 0.77	2.12 (1.46 to 3.07)
Comparisons with hyaluronic acid: RR > 1 indicates greater proportion healing with specified alternative treatment
Saline gauze	1 (88)	Humbert 2013	0.52 (0.23 to 1.17)	0.57 (95% CI 0.28 to 1.14)
Emollient cream	1 (101)	Dereure 2012a	1.31 (0.31 to 5.55)	1.75 (0.87 to 3.52)
Comparisons with hydrocolloid: RR > 1 indicates greater proportion healing with specified alternative treatment
PMM silver	1 (49)	Hanft 2006	1.07 (0.69 to 1.67)	1.27 (0.87 to 1.85)
Povidone iodine	1 (200)	Smith 1992	0.92 (0.69 to 1.23)	0.92 (0.68 to 1.26)
Saline gauze	1 (28)	Ohlsson 1994	0.29 (0.07 to 1.14)	0.34 (95% CI 0.15 to 0.8)
Silver	1 (58)	Kucharzewski 2013	4.39 (2.23 to 8.65) Note 100% events in silver arm	2.32 (1.58 to 3.41)
Zinc oxide	1 (43)	Brandrup 1990	0.95 (0.27 to 3.33)	0.95 (0.27 to 3.35)
Blood product^a	1 (13)	Senet 2003	0.86 (0.07 to 10.96)	0.38 (95% CI 0.17 to 0.88)
Growth factor^a	1 (29)	Rasmussen 1991	1.83 (0.22 to 15.51)	0.91 (0.71 to 1.17)
Comparisons withhydrogel: RR > 1 indicates greater proportion healing with specified alternative treatment
Sucralfate	1 (100)	Tumino 2008	8.60 (3.72 to 19.90)	8.60 (3.68 to 20.07)
Blood product^a	1 (44)	De Araujo 2016	0.47 (0.14 to 1.58)	0.51 (CI 0.21 to 1.23)
Growth factor^a	1 (59)	Senet 2011	1.38 (0.64 to 3.01)	1.20 (0.61 to 2.35)
Comparisons with blood product : RR > 1 indicates greater proportion healing with specified alternative treatment
Saline gauze	1 (67)	Stacey 2000	0.93 (0.74 to 1.16)	0.89 (0.68 to 1.17)
Emollient cream	1 (147)	Biland 1985	0.76 [0.55, 1.06]	0.79 (0.56 to 1.11)
Comparisons with emollient cream: RR > 1 indicates greater proportion healing with specified alternative treatment
SSD	1 (57)	Bishop 1992	6.21 (0.80 to 48.38)	2.56 (1.01 to 6.53)
Abbreviations: PMM: protease modulating matrix; RR: relative risk; SSD: silver sulphadiazine ^aNon‐eligible linking intervention

Table 3. Direct comparisons for individual interventions compared with NMA results

Table 4. Interventions in the included studies

Intervention	Number of included studies	Included studies	Number of participants in included studies
A. Pichinchensis	1	Romero‐Cerecero 2012	34
Alginate	10	Armstrong 1997; Harding 2001; Moffatt 1992b; Petkov 1997; Romanelli 2015a; Romero‐Cerecero 2012; Schulze 2001; Scurr 1994; Smith 1994; Stacey 1997	735
Blood product^a	5	Arenbergerova 2013; Biland 1985; De Araujo 2016; Senet 2003; Stacey 2000	431
Cadexomer iodine	7	Hansson 1998; Harcup 1986; Lindsay 1986; Moss 1987; Ormiston 1985; Steele 1986; Tarvainen 1988	433
Cellulose	1	Alvarez 2012	48
Collagen	2	Caprio 1992; Robson 1995	132
Dextranomer	3	Kalis 1993; Moss 1987; Tarvainen 1988	171
Emollient cream	3	Biland 1985; Bishop 1992; Dereure 2012a	384
Film	1	Banerjee 1997	56
Foam	18	Bowszyc 1995; Callam 1992; Charles 2002; Dimakakos 2009; Fogh 2012; Gottrup 2008; Ivins 2006; Jørgensen 2005; Lanzara 2008; Meaume 2012; Norkus 2005; Rubin 1990; Schulze 2001; Senet 2014; Sopata 2016; Thomas 1997; Vanscheidt 2012; Zuccarelli 1992	1672
Gentian violet	1	Ormiston 1985	60
Growth factor^a	5	Rasmussen 1991; Robson 1995; Robson 2001; Robson 2004; Senet 2011	560
Honey	1	Jull 2008	368
Hyaluronic acid	2	Dereure 2012a; Humbert 2013	189
Hyaluronic acid + povidone iodine	1	Casoni 2002	65
Hydrocolloid	25	Backhouse 1987; Blair 1988a; Bowszyc 1995;Brandrup 1990; Caprio 1992; Charles 2002; Greguric 1994; Hanft 2006; Hansson 1998; Kalis 1993; Kucharzewski 2013; Leaper 1991; Meredith 1988; Moffatt 1992a; Nelson 2007; Norkus 2005; Ohlsson 1994; Rasmussen 1991; Scurr 1994; Senet 2003; Smith 1992; Smith 1994; Sopata 2016; Thomas 1997; Zuccarelli 1992	2044
Hydrofibre	4	Arenbergerova 2013; Armstrong 1997; Harding 2001; Kelechi 2012	329
Hydrogel	6	Beckert 2006; De Araujo 2016; Luiza 2015; Senet 2011; Taddeucci 2004; Tumino 2008	393
Ibuprofen	2	Fogh 2012; Gottrup 2008	222
Magnesium sulphate	1	Greguric 1994	110
Nonadherent	20	Alvarez 2012; Arnold 1994; Backhouse 1987; Banerjee 1997; Blair 1988a; Blair 1988b; Callam 1992; Casoni 2002; Hansson 1998; Kelechi 2012; Leaper 1991; Moffatt 1992a; Moffatt 1992b; Meredith 1988; Nelson 2007; Robson 1995; Robson 2001; Robson 2004; Taddeucci 2004; Vin 2002	1725
Non silver	1	Michaels 2009	208
No treatment	1	Hokkam 2011	104
Octenidine	1	Vanscheidt 2012	106
Ozonated oil	1	Solovastru 2015	29
Papain	2	De Araujo 2016; Luiza 2015	70
Paste bandage	2	Rubin 1990; Stacey 1997	149
Phenytoin	1	Hokkam 2011	104
Povidone iodine	1	Smith 1992;	200
PMM	4	Meaume 2012; Petkov 1997; Romanelli 2015a; Vin 2002	400
PMM‐silver	2	Hanft 2006; Lanzara 2008;	79
Saline gauze	3	Humbert 2013; Ohlsson 1994; Stacey 2000	202
Shale oil	1	Beckert 2006	119
Silica gel fibre	1	Brown 2014	120
Silver	6	Dimakakos 2009; Ivins 2006; Jørgensen 2005; Kucharzewski 2013; Michaels 2009; Senet 2014;	663
SSD	2	Bishop 1992; Blair 1988b	146
Standard care/mixed treatments	6	Arnold 1994; Brown 2014; Harcup 1986; Jull 2008; Lindsay 1986; Steele 1986	715
Sucralfate	1	Tumino 2008	100
Suphadryl	1	Salim 1992	137
Tripeptide copper	1	Bishop 1992	86
Zinc oxide	2	Brandrup 1990; Solovastru 2015	72
Abbreviations: PMM: protease modulating matrix; SSD: silver sulphadiazine ^aIneligible intervention included in expanded base‐case to improve network connectivity

Table 4. Interventions in the included studies

Table 5. Comparison of NMA results for base‐case and two sensitivity analyses

NMA contrast	Base‐case RR (95% CI)	Narrow sensitivity analysis RR (95% CI)	Extended sensitivity analysis RR (95% CI)
Sucralfate versus hydrogel	8.60 (3.66 to 20.2)	‐‐‐	8.60 (3.68 to 20.1)
Sucralfate versus silver	6.99 (0.60 to 82.0)	‐‐‐	2.80 (0.88 to 8.97)
Sucralfate versus foam	14.83 (1.30 to 169)	‐‐‐	5.94 (1.96 to 18.0)
Sucralfate versus hydrocolloid	16.24 (1.43 to 185)	‐‐‐	6.51 (2.17 to 19.6)
Sucralfate versus nonadherent	17.15 (1.52 to 193)	‐‐‐	6.80 (2.24 to 20.7)
Hydrogel versus silver	0.81 (0.08 to 8.19)	0.81 (0.08 to 8.20)	0.33 (0.15 to 0.72)
Hydrogel versus foam	1.73 (0.18 to 16.9)	1.72 (0.18 to 16.9)	0.69 (0.34 to 1.41)
Hydrogel versus hydrocolloid	1.89 (0.19 to 18.4)	1.88 (0.19 to 18.4)	0.76 (0.38 to 1.53)
Hydrogel versus nonadherent	1.99 (0.21 to 19.3)	2.00 (0.21 to 19.4)	0.79 (0.39 to 1.62)
Silver versus foam	2.12 (1.46 to 3.09)	2.12 (1.45 to 3.10)	2.12 (1.46 to 3.07)
Silver versus hydrocolloid	2.32 (1.58 to 3.43)	2.32 (1.57 to 3.44)	2.32 (1.58 to 3.41)
Silver versus nonadherent	2.45 (1.58 to 3.82)	2.47 (1.58 to 3.86)	2.43 (1.58 to 3.74)
Foam versus hydrocolloid	1.10 (0.93 to 1.28)	1.09 (0.93 to 1.29)	1.10 (0.94 to 1.28)
Foam versus nonadherent	1.16 (0.91 to 1.47)	1.16 (0.91 to 1.49)	1.15 (0.91 to 1.44)
Hydrocolloid versus nonadherent	1.06 (0.84 to 1.32)	1.06 (0.85 to 1.33)	1.04 (0.85 to 1.29)
Abbreviations: CI: confidence interval; RR: relative risk

Table 5. Comparison of NMA results for base‐case and two sensitivity analyses

Table 6. Ranks of treatments ‐ base‐case and two sensitivity analyses (ordered by mean rank)

Treatment	Base‐case (rank of 22) Mean rank (SUCRA) and maximum probability and its corresponding rank	Narrow sensitivity analysis (rank of 17) Mean rank (SUCRA) and maximum probability and its corresponding rank	Extended base‐case (rank of 25)# Mean rank (SUCRA) and maximum probability and its corresponding rank
Sucralfate	1.5 (1.0)_______91% (rank 1)	‐‐‐	1.1 (1.0)_______93% (rank 1)
Silver	3.2 (0.9)_______38% (rank 3)	1.9 (0.9)_______40% (rank 2)	2.7 (0.9)_______50% (rank 2)
Hyaluronic acid + povidone iodine	5.8 (0.8) _______32% (rank 21)	‐‐‐	5.3 (0.8)_______21% (rank 3)
Paste bandage	5.8 (0.8)_______19% (rank 5)	4.0 (0.8)_______26% (rank 3)	5.4 (0.8)_______22% (rank 4)
Hydrofibre	8.3 (0.7)_______14% (rank 7)	5.9 (0.7)_______17% (rank 5)	8.1 (0.7)_______16% (rank 6)
Hydrogel	8.9 (0.6)_______39% (rank 2)	6.4 (0.7)_______39% (rank 1)	16.9 (0.3)_______15% (rank 20)
PMM	9.4 (0.6)_______15% (rank 9)	7.0 (0.6)_______19% (rank 6)	9.0 (0.7)_______16% (rank 8)
PMM silver	9.5 (0.6)_______12% (rank 8)	6.8 (0.6)_______15% (rank 5)	8.9 (0.7)_______13% (rank 7)
Film	10.1 (0.6)_______9% (rank 5)	7.5 (0.6)_______10% (rank 3)	10.2 (0.6)_______9% (rank 5)
Alginate	10.9 (0.5)_______17% (rank 10)	8.1 (0.6)_______20% (rank 7)	10.5 (0.6)_______16% (rank 10)
Octenidine	11.4 (0.5)_______7% (rank 7)	‐‐‐	11.4 (0.6)_______9% (rank 6)
Foam	12.0 (0.5)_______18% (rank 11)	9.0 (0.5)_______20% (rank 9)	11.5 (0.6)_______17% (rank 11)
Cadexomer iodine	12.1 (0.5)_______9% (rank 19)	9.0 (0.5)_______9% (rank 4)	11.8 (0.5)_______8% (rank 19)
Zinc oxide	13.3 (0.4)_______13% (rank 20)	10.5 (0.4)_______14% (rank 15)	14.4 (0.4)_______8% (rank 25)
Ibuprofen‐releasing foam	14.1 (0.4)_______12% (rank 18)	‐‐‐	14.3 (0.4)_______9% (rank 18)
Hydrocolloid	14.3 (0.4)_______21% (rank 15)	11.0 (0.4)_______25% (rank 11)	14.0 (0.5)_______18% (rank 14)
Nonadherent	15.2 (0.3)_______18% (rank 15)	11.8 (0.3)_______25% (rank 13)	15.3 (0.4)_______15% (rank 15)
Povidone iodine	15.2 (0.3)_______14% (rank 17)	11.8 (0.3)_______17% (rank 13)	15.5 (0.4)_______12% (rank 17)
Hyaluronic acid	15.7 (0.3)_______18% (rank 4)	12 (0.3)_______38% (rank 16)	17.0 (0.3)_______19% (rank 22)
Gentian violet	17.4 (0.2)_______19% (rank 21)	‐‐‐	18.4 (0.3)_______16% (rank 25)
SSD	18.1 (0.2)_______23% (rank 19)	14 (0.2)_______28% (rank 15)	18.8 (0.3)_______16% (rank 19)
Saline gauze	21.0 (0)_______69% (rank 22)	16.3 (0)_______77% (rank 17)	23.0 (0.1)_______33% (rank 24)
Abbreviations:PMM: protease modulating matrix; SSD: silver sulphadiazine; SUCRA surface under the cumulative ranking curve # ranks for extra treatments not reported

Table 6. Ranks of treatments ‐ base‐case and two sensitivity analyses (ordered by mean rank)

Table 7. Contributions matrix

Mixed treatment comparisons ⇨ Direct comparisons (risk of bias) ⇩	Silver vs HC	HC vs NA	Foam vs NA	HC vs Foam	Silver vs foam	Sucralfate vs HC	Silver vs NA	Sucralfate vs NA	Sucralfate vs silver	Sucralfate vs foam
Hyaluronic + povidone iodine vs nonadherent (low)
Hydrocolloid vs nonadherent (high)	3.0	80.6	32.7	6.2	1.2	17.9	28.9	2.7	11.1	10.9
Hyaluronic acid vs saline gauze (high)
Hydrofibre vs nonadherent (high)	0.2	0.6	1.0	0.3	0.1	0.1	0.7	0.4	0.2	0.2
Hydrogel vs nonadherent (very high)						3.4		5.2	2.6	3.4
PMM vs nonadherent (low)	0.3	0.8	2.1	0.6	0.1	0.3	1.3	0.1	0.5	0.8
PMM silver vs hydrocolloid (high)	1.0	0.1	1.2	2.2	0.4	0.1	0.4		0.3	0.6
Hydrocolloid vs povidone iodine (very high)
Hydrocolloid vs saline gauze (low)		0.2	0.1			2.7	0.1	1.9	1.8	1.9
Silver vs hydrocolloid (high)	32.5	0.1	0.9	2.2	12.9	0.1	15.9		7.2	0.5
Hydrocolloid vs zinc oxide (high)
SSD vs nonadherent (high)		0.1				0.5		0.9	0.4	0.5
Cadexomer iodine vs nonadherent (high)		0.8	0.3			0.2	0.3		0.1	0.1
Sucralfate vs hydrogel (high)						25.1		32.4	18.7	21.4
Film vs nonadherent (very high)
Foam vs nonadherent (very high)	2.5	5.4	21.4	5.3	1.0	1.8	9.1	0.3	3.3	5.8
Growth factor vs nonadherent (low)		0.5	0.2			16.2	0.2	22.7	12.2	13.8
Hydrocolloid vs alginate (high)	0.3	0.5		0.5	0.1	0.3	0.1
Hydrofibre vs alginate (very high)	0.2	0.6	1.0	0.3	0.1	0.4	0.7	0.1	0.5	0.6
Paste bandage vs alginate (high)	0.4	0.4	1.3	0.7	0.2	0.2	0.8		0.5	0.6
PMM vs alginate (low)	0.1	0.4	0.7	0.1		0.1	0.5	0.1	0.2	0.2
Foam vs alginate (very high)	0.1	0.1	0.4	0.3	0.1	0.1	0.2		0.1	0.2
Hydrocolloid vs blood product (low)		0.1				1.2		0.8	0.9	0.9
Hydrofibre vs blood product (very high)						0.4		0.5	0.3	0.4
Hydrogel vs blood product (low)		0.2	0.1			4.8	0.1	4.2	3.4	3.7
Blood product vs saline gauze (high)		0.2	0.1			2.7	0.1	1.9	1.8	1.9
Blood product vs emollient cream (high)		0.1				0.6		1.0	0.5	0.5
Hydrocolloid vs cadexomer iodine (high)		0.8	0.3			0.2	0.3		0.1	0.1
Gentian violet vs cadexomer iodine (high)
Hyaluronic acid vs emollient cream (low)
SSD vs emollient cream (low)		0.1				0.5		0.9	0.4	0.5
Hydrocolloid vs foam (very high)	26.6	6.0	31.2	75.5	11.1	2.1	13.3	0.3	7.0	13.1
Ibuprofen foam vs foam (high)
Octenidine vs foam (high)
Paste bandage vs foam (high)	0.4	0.4	1.3	0.7	0.2	0.2	0.8		0.5	0.6
PMM vs foam (high)	0.3	0.4	1.4	0.5	0.1	0.1	0.8		0.4	0.5
PMM silver vs foam (high)	1.0	0.1	1.2	2.2	0.4	0.1	0.4		0.3	0.6
Silver vs foam (low)	31.0	0.1	0.9	2.2	71.9	0.1	24.7		11.6	0.5
Hydrocolloid vs growth factor (high)		0.4	0.1			0.7	0.1	0.3	0.5	0.5
Hydrogel vs hydrocolloid (high)		0.1	0.1			16.9	0.1	23.0	12.7	14.3
RISK OF BIAS FOR Mixed Treatment Comparison	High	High	High	Very high	Low	High	High	High	High	High
Abbreviations:HC: hydrocolloid; NA: nonadherent; PMM: protease modulating matrix; SSD: silver sulphadiazine

Table 7. Contributions matrix

Table 8. Inconsistency factors ‐ base‐case and extended base‐case

Loop	RoRR and 90%CI	P value	Loop heterogeneity tau² (loop)
Foam‐hydrocolloid‐silver	2.44 (90%CI 1.23 to 4.84)	0.033	0
Nonadherent‐alginate‐foam	2.28 (90%CI 0.54 to 9.67)	0.349	0
Alginate‐foam‐PMM	2.26 (90%CI 0.43 to 11.94)	0.419	0
Nonadherent‐cadexomer‐hydrocolloid	1.81 (90%CI 0.25 to 13.24)	0.625	0.104
Nonadherent‐alginate‐hydrocolloid	1.66 (90%CI 0.35 to 7.74)	0.59	0.103
Foam‐hydrocolloid‐PMM silver	1.60 (90%CI 0.83 to 3.08)	0.24	0
Alginate‐foam‐hydrocolloid	1.40 (90%CI 0.26 to 7.39)	0.74	0
Nonadherent‐alginate‐PMM	1.26 (90%CI 0.72 to 2.21)	0.503	0
Nonadherent‐foam‐PMM	1.25 (90%CI 0.43 to 3.62)	0.73	0
Nonadherent‐alginate‐hydrofibre	1.18 (90%CI 0.61 to 2.26)	0.684	0
Nonadherent‐foam‐hydrocolloid	1.06 (90%CI 0.55 to 2.06)	0.878	0.042
Alginate‐foam‐paste bandage	1.03 (90%CI 0.23 to 4.58)	0.974	0
Extended base‐case only
Nonadherent‐growth factor‐hydrogel	3.00 (90%CI 0.40 to 22.43)	0.370	0
Blood product‐hydrocolloid‐saline gauze	2.78 (90%CI 0.24 to 31.92)	0.491	0
Nonadherent‐growth factor‐hydrocolloid	2.23 (90%CI 0.21 to 23.65)	0.577	0.078
Quadratic loops
Alginate‐blood product‐hydrocolloid‐hydrofibre	7.34 (90%CI 0.12 to 460.27)	0.428	0.487
Nonadherent‐blood product‐hydrofibre‐hydrogel	4.7 (90%CI 0.15 to 148.15)	0.461	0
Nonadherent‐blood product‐hydrocolloid‐hydrofibre	4.09 (90%CI 0.03 to 493.5)	0.629	0.096
Blood product‐emollient cream‐hyaluronic acid‐saline gauze	3.68 (90%CI 0.89 to 15.16)	0.131	0
Blood product‐growth factor‐hydrocolloid‐hydrogel	1.38 (90%CI 0.07 to 28.86)	0.862	0
Nonadherent‐blood product‐hydrocolloid‐hydrogel	1.15 (90%CI 0.02 to 82.36)	0.957	0.096
Abbreviations:CI: confidence interval; PMM: protease modulating matrix; RoRR: ratio of relative risks

Table 8. Inconsistency factors ‐ base‐case and extended base‐case

Table 9. Node splitting

Comparison	Direct RR (95% CI)	Indirect RR (95% CI)	RoRR (90% CI)
Alginate vs nonadherent	1.08 (95% CI 0.86 to 1.36)	1.52 (95% CI 1.07 to 2.15)	0.71 (90% CI 0.50 to 1.02)
Foam vs nonadherent	1.35 (95% CI 0.87 to 2.08)	1.10 (95% CI 0.83 to 1.47)	1.22 (90% CI 0.79 to 1.89)
Hydrocolloid vs nonadherent	0.94 (95% CI 0.72 to 1.23)	2.01 (95% CI 0.56 to 7.23)	0.47 (90% CI 0.16 to 1.39)
Hydrofibre vs nonadherent	1.47 (95% CI 0.84 to 2.56)	1.35 (95% CI 0.71 to 2.56)	1.09 (90% CI 0.53 to 2.23)
Hydrogel vs nonadherent	2.00 (95% CI 0.21 to 19.1)	0.76 (95% CI 0.36 to 1.64)	2.62 (90% CI 0.35 to 19.5)
PMM vs nonadherent	1.46 (95% CI 0.80 to 2.67)	1.29 (95% CI 0.85 to 1.96)	1.13 (90% CI 0.61 to 2.11)
Foam vs alginate	0.55 (95% CI 0.10 to 2.87)	0.95 (95% CI 0.72 to 1.27)	0.57 (90% CI 0.14 to 2.36)
Hydrocolloid vs alginate	0.70 (95% CI 0.24 to 2.06)	0.87 (95% CI 0.67 to 1.14)	0.81 (90% CI 0.31 to 2.05)
Hydrofibre vs alginate	1.18 (95% CI 0.66 to 2.10)	1.09 (95% CI 0.57 to 2.10)	1.08 (90% CI 0.51 to 2.29)
Paste bandage vs alginate	1.22 (95% CI 0.91 to 1.63)	2.41 (95% CI 1.28 to 4.53)	0.51 (90% CI 0.28 to 0.91)
PMM vs alginate	1.08 (95% CI 0.76 to 1.53)	1.07 (95% CI 0.57 to 1.98)	1.01 (90% CI 0.55 to 1.85)
Hydrocolloid vs foam	0.92 (95% CI 0.76 to 1.12)	0.90 (95% CI 0.64 to 1.28)	1.02 (90% CI 0.73 to 1.42)
Paste bandage vs foam	2.30 (95% CI 1.29 to 4.09)	1.17 (95% CI 0.79 to 1.72)	1.97 (90% CI 1.10 to 3.55)
PMM vs foam	0.87 (95% CI 0.30 to 2.51)	1.19 (95% CI 0.81 to 1.74)	0.73 (90% CI 0.28 to 1.90)
PMM silver vs foam	1.57 (95% CI 0.83 to 2.96)	0.96 (95% CI 0.59 to 1.57)	1.64 (90% CI 0.83 to 3.21)
Silver vs foam	1.65 (95% CI 1.08 to 2.51)	4.12 (95% CI 2.06 to 8.22)	0.40 (90% CI 0.20 to 0.79)
PMM silver vs hydrocolloid	1.07 (95% CI 0.68 to 1.7)	1.75 (95% CI 0.91 to 3.37)	0.61 (90% CI 0.31 to 1.2)
Silver vs hydrocolloid	4.39 (95% CI 2.23 to 8.62)	1.76 (95% CI 1.12 to 2.75)	2.50 (90% CI 1.26 to 4.95)
Emollient cream vs blood product	0.76 (95% CI 0.54 to 1.09)	2.78 (95% CI 0.53 to 14.62)	0.28 (90% CI 0.07 to 1.15)
Hydrocolloid vs blood product	1.17 (95% CI 0.09 to 14.81)	2.39 (95% CI 0.91 to 6.32)	0.49 (90% CI 0.05 to 4.84)
Hydrofibre vs blood product	0.33 (95% CI 0.01 to 7.82)	3.65 (95% CI 1.31 to 10.19)	0.09 (90% CI 0.01 to 1.5)
Hydrogel vs blood product	2.13 (95% CI 0.63 to 7.25)	1.33 (95% CI 0.33 to 5.42)	1.6 (90% CI 0.33 to 7.69)
Saline gauze vs blood product	0.93 (95% CI 0.71 to 1.21)	0.34 (95% CI 0.1 to 1.18)	2.71 (90% CI 0.93 to 7.85)
Hyaluronic acid vs emollient cream	0.76 (95% CI 0.18 to 3.24)	2.78 (95% CI 1.13 to 6.8)	0.28 (90% CI 0.07 to 1.15)
Hydrocolloid vs growth factor	0.55 (95% CI 0.06 to 4.6)	1.11 (95% CI 0.84 to 1.46)	0.49 (90% CI 0.08 to 3.01)
Hydrogel vs growth factor	0.72 (95% CI 0.33 to 1.6)	1.55 (95% CI 0.39 to 6.11)	0.47 (90% CI 0.12 to 1.78)
Hyaluronic acid vs saline gauze	2.32 (95% CI 1.06 to 5.07)	0.64 (95% CI 0.14 to 2.89)	3.63 (90% CI 0.87 to 15.21)
Hydrocolloid vs saline gauze	3.5 (95% CI 0.87 to 14.06)	1.91 (95% CI 0.57 to 6.39)	1.84 (90% CI 0.39 to 8.7)
Abbreviations:CI: confidence interval; PMM: protease modulating matrix; RoRR: ratio of relative risks; RR: relative risk

Table 9. Node splitting

Comparison 1. Direct evidence ‐ included in base‐case network

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Alginate vs nonadherent Show forest plot	1	60	Risk Ratio (IV, Random, 95% CI)	1.08 [0.86, 1.36]

2 Cadexomer iodine vs nonadherent Show forest plot	1	105	Risk Ratio (IV, Random, 95% CI)	1.0 [0.39, 2.56]

3 Film vs nonadherent Show forest plot	1	71	Risk Ratio (IV, Random, 95% CI)	1.34 [0.61, 2.92]

4 Foam vs nonadherent Show forest plot	1	132	Risk Ratio (IV, Random, 95% CI)	1.35 [0.89, 2.05]

5 Hyaluronic plus povidone vs nonadherent Show forest plot	1	65	Risk Ratio (IV, Random, 95% CI)	1.93 [0.95, 3.92]

6 Hydrocolloid vs non‐adherent Show forest plot	7	662	Risk Ratio (IV, Random, 95% CI)	1.26 [0.92, 1.72]

7 Hydrofibre vs nonadherent Show forest plot	1	82	Risk Ratio (M‐H, Random, 95% CI)	1.47 [0.88, 2.46]

8 Hydrogel vs nonadherent Show forest plot	1	24	Risk Ratio (IV, Random, 95% CI)	2.0 [0.21, 19.23]

9 PMM vs nonadherent Show forest plot	1	74	Risk Ratio (IV, Random, 95% CI)	1.42 [0.80, 2.51]

10 SSD vs nonadherent Show forest plot	1	60	Risk Ratio (IV, Random, 95% CI)	0.79 [0.57, 1.10]

11 Foam vs alginate Show forest plot	1	113	Risk Ratio (IV, Random, 95% CI)	0.55 [0.10, 2.86]

12 Hydrocolloid vs alginate Show forest plot	2	80	Risk Ratio (IV, Random, 95% CI)	0.72 [0.15, 3.42]

13 Hydrofibre vs alginate Show forest plot	2	175	Risk Ratio (IV, Random, 95% CI)	1.47 [0.48, 4.47]

14 Paste bandage vs alginate Show forest plot	1	133	Risk Ratio (IV, Fixed, 95% CI)	1.22 [0.91, 1.63]

15 PMM vs alginate Show forest plot	2	140	Risk Ratio (IV, Random, 95% CI)	1.10 [0.84, 1.46]

16 Gentian violet vs cadexomer iodine Show forest plot	1	60	Risk Ratio (IV, Random, 95% CI)	0.58 [0.27, 1.28]

17 Hydrocolloid vs cadexomer iodine Show forest plot	1	104	Risk Ratio (IV, Random, 95% CI)	0.73 [0.26, 2.08]

18 Hydrocolloid vs foam Show forest plot	6	458	Risk Ratio (IV, Random, 95% CI)	0.92 [0.77, 1.08]

19 Ibuprofen foam vs foam Show forest plot	2	242	Risk Ratio (IV, Random, 95% CI)	0.88 [0.48, 1.61]

20 Octenidine vs foam Show forest plot	1	126	Risk Ratio (IV, Random, 95% CI)	1.03 [0.56, 1.90]

21 Paste bandage vs foam Show forest plot	1	36	Risk Ratio (IV, Random, 95% CI)	2.30 [1.29, 4.10]

22 PMM vs foam Show forest plot	1	187	Risk Ratio (IV, Random, 95% CI)	0.87 [0.30, 2.48]

23 PMM silver vs foam Show forest plot	1	30	Risk Ratio (IV, Random, 95% CI)	1.57 [0.84, 2.92]

24 Silver vs foam Show forest plot	4	397	Risk Ratio (M‐H, Random, 95% CI)	1.65 [1.08, 2.52]

25 Saline gauze vs hyaluronic acid Show forest plot	1	88	Risk Ratio (IV, Random, 95% CI)	0.52 [0.23, 1.17]

26 PMM silver vs hydrocolloid Show forest plot	1	49	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.69, 1.67]

27 Povidone iodine vs hydrocolloid Show forest plot	1	200	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.69, 1.23]

28 Saline gauze vs hydrocolloid Show forest plot	1	28	Risk Ratio (IV, Fixed, 95% CI)	0.29 [0.07, 1.14]

29 Silver vs hydrocolloid Show forest plot	1	58	Risk Ratio (M‐H, Random, 95% CI)	4.39 [2.23, 8.65]

30 Zinc oxide vs hydrocolloid Show forest plot	1	43	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.27, 3.33]

31 Sucralfate vs hydrogel Show forest plot	1	100	Risk Ratio (IV, Random, 95% CI)	8.60 [3.72, 19.90]

Comparison 1. Direct evidence ‐ included in base‐case network

Comparison 2. Direct evidence ‐ not in base case network, in expanded base case

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Blood product vs emollient Show forest plot	1	197	Risk Ratio (M‐H, Random, 95% CI)	1.31 [0.94, 1.82]

2 Blood product vs hydrocolloid Show forest plot	1	13	Risk Ratio (M‐H, Random, 95% CI)	0.86 [0.07, 10.96]

3 Blood product vs hydrogel Show forest plot	1	44	Risk Ratio (M‐H, Random, 95% CI)	0.47 [0.14, 1.58]

4 Blood product vs saline gauze Show forest plot	1	86	Risk Ratio (M‐H, Random, 95% CI)	1.08 [0.86, 1.35]

5 Hyaluronic vs emollient cream Show forest plot	1	101	Risk Ratio (M‐H, Random, 95% CI)	0.77 [0.18, 3.25]

6 Growth factor vs hydrocolloid Show forest plot	1	29	Risk Ratio (M‐H, Random, 95% CI)	1.83 [0.22, 15.51]

7 Growth factor vs hydrogel Show forest plot	1	59	Risk Ratio (M‐H, Random, 95% CI)	1.38 [0.64, 3.01]

8 Growth factor vs nonadherent Show forest plot	3	460	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.81, 1.14]

9 SSD vs emollient Show forest plot	1	57	Risk Ratio (M‐H, Random, 95% CI)	6.21 [0.80, 48.38]

Comparison 2. Direct evidence ‐ not in base case network, in expanded base case

Comparison 3. Direct evidence ‐ not in network

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 A. Pichinchensis vs alginate Show forest plot	1	34	Risk Ratio (M‐H, Random, 95% CI)	1.67 [1.03, 2.70]

2 Non‐adherent vs cellulose Show forest plot	1	48	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.38, 2.22]

3 Phenytoin vs no treatment Show forest plot	1	104	Risk Ratio (M‐H, Random, 95% CI)	1.25 [0.90, 1.74]

4 Cadexomer iodine vs standard treatment Show forest plot	3	157	Risk Ratio (M‐H, Random, 95% CI)	5.16 [1.56, 17.10]

5 Honey vs standard treatment Show forest plot	1	368	Risk Ratio (M‐H, Random, 95% CI)	1.12 [0.92, 1.36]

6 Papain vs hydrogel Show forest plot	2	70	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.25, 3.49]

7 Shale oil vs hydrogel Show forest plot	1	119	Risk Ratio (M‐H, Random, 95% CI)	1.49 [0.82, 2.68]

8 Tripeptide copper vs hydrogel Show forest plot	1	57	Risk Ratio (M‐H, Random, 95% CI)	0.07 [0.00, 1.26]

9 Hydrocolloid vs collagen Show forest plot	1	96	Risk Ratio (M‐H, Random, 95% CI)	0.77 [0.50, 1.18]

10 Hydrocolloid vs dextranomer Show forest plot	1	108	Risk Ratio (M‐H, Random, 95% CI)	0.77 [0.37, 1.60]

11 Hydrocolloid vs magnesium sulphate Show forest plot	1	110	Risk Ratio (M‐H, Random, 95% CI)	7.0 [0.37, 132.40]

12 Hydrocolloid vs nonadherent or iodine Show forest plot	1	70	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.42, 1.48]

13 Ozonated oil vs zinc oxide Show forest plot	1	29	Risk Ratio (M‐H, Random, 95% CI)	10.31 [0.62, 170.96]

14 Cadexomer iodine vs dextranomer Show forest plot	2	63	Risk Ratio (M‐H, Random, 95% CI)	1.33 [0.64, 2.75]

15 Silica gel fibre vs standard care Show forest plot	1	120	Risk Ratio (M‐H, Random, 95% CI)	0.63 [0.31, 1.26]

16 Silver vs non‐silver Show forest plot	1	213	Risk Ratio (M‐H, Random, 95% CI)	1.05 [0.94, 1.16]

17 Sulphadryl vs inactive powder Show forest plot	1	168	Risk Ratio (M‐H, Random, 95% CI)	1.31 [1.10, 1.56]

18 Tripeptide copper vs emollient cream Show forest plot	1	58	Risk Ratio (M‐H, Random, 95% CI)	0.33 [0.01, 7.86]

19 Tripeptide copper vs SSD Show forest plot	1	57	Risk Ratio (M‐H, Random, 95% CI)	0.07 [0.00, 1.26]

Comparison 3. Direct evidence ‐ not in network