Bacteria and bioburden and healing in complex wounds: A prognostic systematic review

The wound microbiome may play an important role in the wound healing process. We conducted the first systematic prognosis review investigating whether aspects of the wound microbiome are independent prognostic factors for the healing of complex wounds. We searched Medline, Embase, CINAHL and the Cochrane Library to February 2019. We included longitudinal studies which assessed the independent association of aspects of wound microbiome with healing of complex wounds while controlling for confounding factors. Two reviewers extracted data and assessed risk of bias and certainty of evidence using the GRADE approach. We synthesised studies narratively due to the clinical and methodological heterogeneity of included studies and sparse data. We identified 28 cohorts from 21 studies with a total of 38,604 participants, including people with diabetes and foot ulcers, open surgical wounds, venous leg ulcers and pressure ulcers. Risk of bias varied from low (2 cohorts) to high (17 cohorts); the great majority of participants were in cohorts at high risk of bias. Most evidence related to the association of baseline clinical wound infection with healing. Clinical infection at baseline may be associated with less likelihood of wound healing in foot ulcers in diabetes (HR from cohort with moderate risk of bias 0.53, 95% CI 0.33 to 0.83) or slower healing in open surgical wounds (HR 0.65, 95% CI 0.51 to 0.83); evidence in other wounds is more limited. Most other associations assessed showed no clear relationship with wound healing; evidence was limited and often sparse; and we documented gaps in the evidence. There is low certainty evidence that a diagnosis of wound infection may be prognostic of poorer healing in foot ulcers in diabetes, and some moderate certainty evidence for this in open surgical wounds. Low certainty evidence means that more research could change these findings.


| INTRODUCTION
Complex wounds heal by secondary intention-that is by the formation of new tissue rather than by the approximation of the wound edges (primary intention). 1 Complex wounds include pressure, leg, and foot ulcers, and open surgical wounds that, despite different aetiologies, share a common risk of infection and slow healing. These wounds are all usually managed with dressings, although other treatments vary (eg, venous leg ulcers are treated with compression and pressure ulcers with repositioning and specialist support surfaces). 2 In 2011, the point prevalence of complex wounds in the United Kingdom was estimated at 1.47 (1.38-1.56) per 1000 of the population, this figure being inferred from a large multiservice cross-sectional study of a single city (Leeds, population c. 750,000). 1 The most frequently documented wounds were pressure ulcers and leg ulcers.
Complex wounds take a substantial-but variable-time to heal, impose a considerable burden on people living with them and also have a societal and cost impact in lost activity and medical expenditure. 3 People with complex wounds report that complete wound healing is the most important outcome to them. 1 There are many potential risk factors or predictors for whether a complex wound will heal and how long this might take. Candidate factors can broadly be grouped as being at: the population level, the individual level, the whole wound level and the cellular level. Prognostic factors at each of these levels are likely to be inter-related in often complex and bidirectional relationships between predictive factors. 4 For example, known factors associated with the future healing of venous leg ulcers include wound size and wound duration at study entry 5 which may also predict pressure ulcer healing. 6 Such factors do not, however, explain much of the observed variation in healing trajectory, and may themselves have a bidirectional relationship with other factors such as the wound microbiome. 7 Elucidating independent relationships between prognostic factors and wound healing requires carefully designed studies which collect a range of clinical, biochemical and microbiological data.
All open wounds contain bacteria: initial colonisation is usually by commensal species from the skin with the potential to acquire or activate virulence factors. 8 Subsequent colonisation of wounds by pathogenic species is also a risk in open wounds. 9 Bacteria are often believed to be present in wounds in the form of biofilms. 10 Biofilms are generally defined as microbial cells surrounded by a polymer matrix of microbial and/or host origin, that adhere to surfaces or to themselves. Growing in this way can make bacteria more tolerant to bactericidal agents. 11,12 There are many possible associations between elements of the wound microbiota and wound progress including how long wounds take to heal completely. Suggested prognostic factors include the overall number of microbes (microbial load or bioburden) 13 ; the specific types of bacteria (including drug-resistant characteristics) in the wound, 14 and/or inter-bacterial interactions 15 and characteristics of resulting biofilms. 16 Most wounds heal despite the presence of microbes 17 so it has been suggested that the balance between bacterial activity and the responsiveness of the host immune system is likely to be of key importance. 18 This develops an earlier concept of critical colonisation, a precursor to clinical infection 19 ; this state is   often considered, by rather circular reasoning, to be indicated by delayed healing. Diagnosis of infection by contrast usually requires the   additional presence of one or more clinical signs such as local pain, heat, redness, swelling and secretion of pus.
Whilst there has been research on the association of the wound microbiome and healing, empirical evidence in this area has not been systematically reviewed. The Australian Wound Management Association states that 'the true extent of bacterial impairment of wound healing is unknown'. 20 Whilst prognostic association of the microbiome and healing has not been systematically reviewed, there are several treatments available which are suggested to promote healing based on purported antimicrobial activity, for example, silver dressings. 21 Use of these treatments in wounds without clinical infection is predicated on there being a relationship between reducing bacterial load (and perhaps preventing infection) and the wound healing trajectory. Evidence for the effectiveness of these treatments is also unclear. 22 Finding a predictive relationship between aspects of the wound microbiome and wound healing may not necessarily indicate a causal relationship but may, nevertheless, allow better identification of wounds which are likely to show poor healing. Scoping work indicated that the relevant literature was likely to be sparse and diverse, and this supported a decision to adopt broad objectives and inclusion criteria for this review.

| Objective
To determine whether aspects of the wound microbiome are independently prognostic for wound healing in people with the following types of complex wounds: pressure ulcers, venous leg ulcers, foot ulcers in people with diabetes, or surgical wounds healing by secondary intention.

| MATERIALS AND METHODS
The full protocol for this systematic review was registered on PROS-PERO and has the registration number CRD42019136141. 23

| Inclusion criteria
We included prospective or retrospective longitudinal studies, including those based on registry data (data collected through an organised registry system on patients with a common condition), which assessed whether a relevant potential bacterial prognostic factor was associated with healing in complex wounds in humans. Case-control studies would have been included in the absence of other studies for a particular prognostic factor association with the outcome.
We only included studies which aimed to measure the independent association with healing of one or more prognostic factors while controlling for other factors by use of a multivariable analysis or other appropriate method such as the use of propensity matching in the study design. Studies reporting only univariate analyses were noted but not included in the review.
In some cohorts, study authors undertook a multivariable analysis, and applied prespecified criteria for the inclusion of variables in their multivariable analyses. This could mean that the potential bacterial prognostic factor was not included in the final multivariable analysis because its association with healing did not reach criteria for significance-a potentially important finding. We therefore included all studies which carried out multivariable analyses where at least one potential bacterial prognostic factor was considered, including where the variables of interest to our review did not meet the criteria for multivariable analysis. Where this was the case, we noted that variables did not meet criteria for inclusion in the multivariable analysis and, if possible, reported the univariate associations. We considered the appropriateness of the criteria used to determine which variables were included in the multivariable analyses as part of our risk of bias assessment. Similarly, where there were primary outcome data for an association, we noted studies reporting only our secondary outcome of change in wound size, but did not analyse these.
Complex wounds were defined as: foot ulceration in people with diabetes (Wagner grade I or above); venous or mixed aetiology leg ulcers; pressure ulcers (stage II or above); surgical wounds healing by secondary intention (open surgical wounds). We did not further limit eligibility and included studies in which at least 75% of participants had a relevant wound. We excluded studies of burn wounds, which often heal by secondary intention but are outside the scope of this review because of the specific immunological issues associated with burn injuries.
We considered the following types of potential bacterial prognos-

| Outcomes
Primary outcomes were time to wound healing (survival analysis) and proportion of healed wounds at any specified timepoint. An association ratio measure less than 1 indicated a risk of poorer healing (fewer wounds healed or longer time to healing) when the potential prognostic variable was present.
The secondary outcome was the surrogate outcome of change or rate of change in wound size; we planned to analyse these secondary outcome data only if no primary outcome data were available for an association. As our primary outcome was reported for all associations analyses of secondary data are reported in Supporting Information Tables.

| Search
We searched the following databases in February 2019: Ovid Medline (from 1946), Ovid Embase (from 1946), CINAHL (from 1982) and Cochrane Central Register of Controlled Trials. We also checked the references of included studies and identified systematic reviews.
Details of the search strategy are provided in the Appendix.

| Selection of studies, data extraction and risk of bias and quality assessment
Data were extracted by one review author and checked by a second using a standardised data extraction sheet (piloted on a small number of studies). We extracted data on the following: country and setting, study design, eligibility criteria, participant baseline characteristics, treatment regimen, sampling method(s), prognostic factor with measurement and assessment methods, outcome data and measurement, follow-up duration, analysis details including adjustment (potential confounding) factors, association statistics with measures of variance, losses to follow-up. We extracted adjusted and, where appropriate, unadjusted measures of association and planned to convert effect sizes as necessary. Confounding factors were also considered as part of the risk of bias assessment.
Risks of bias were independently assessed by two review authors using the Quality in Prognosis Studies (QUIPS) tool. 24 Any disagreements were resolved through discussion and consensus. The QUIPS tool considers the following six domains: representative population, missing data, prognostic factor measurement, outcome measurement, confounding factors, analysis and reporting. We assigned overall risks of bias as low, moderate or high based on predetermined criteria across all assessed domains. Associations were considered to have a low risk overall risk of bias where all domains were low risk or where one was moderate but the rest were low; where two or more domains were at moderate risk of bias and none at high risk the overall risk of bias was considered to be moderate; where one or more domains was at high risk of bias the overall risk of bias was considered to be high.
For each study we summarised the results of the risk of bias assessment for each domain, and overall, in a table (see Supporting Information Tables); a full assessment which includes the comments for each study for each domain is available on request from the authors.
We used a modified GRADE framework to assess the quality of evidence for each prognostic factor-outcome combination. 25 GRADE assessments were carried out by one review author and checked by a second; disagreements were resolved through discussion and consensus. GRADE takes into account the results of the risk of bias assessment of contributing studies, but also whether there is imprecision, inconsistency, indirectness or publication bias in the evidence for each association. Evidence may be graded as high, moderate, low or very low certainty based on the assessment. In some cases we adopted a flexible approach where areas of concern across multiple domains were aggregated to produce a single downgrade; this approach is supported by the GRADE working group. 26  with that from studies with incomplete reporting. Where we needed to select an effect estimate for the Summary of Findings Table we gave priority to cohorts with low or moderate risk of bias over those with high risk of bias, while acknowledging the results of higher risk cohorts.

| Data synthesis
We conducted a narrative synthesis supported by structured tables, with evidence from studies grouped, as planned in the protocol, by the prognostic factor reported, the baseline infection status and the wound aetiology, because of the clinical relevance of these factors and the likely different treatment regimens in practice We had planned to implement a random-effects meta-analytic approach where appropriate 23

| Characteristics of cohorts
The characteristics of the included cohorts are summarised in  Table S1.
Of the 28 cohorts, 15 had a prospective cohort study design or were RCTs analysed as cohort studies, and the remainder were retrospective cohorts, or drawn from retrospectively analysed registry data. Of the 38,604 participants who contributed data to this review 90% (34,675) were from six cohorts using registry data; analyses of these data were reported in three overlapping studies. 4,6,35 Most other studies were small (total participants 3929; median 136; range 64-1340).

| Quality of the evidence
We assessed the overall quality of the evidence using the GRADE approach for prognostic factor evidence. 25 Risk of bias was one element considered in the quality assessment and this varied; nine cohorts were considered to produce findings at moderate risk of bias but two cohorts produced findings judged at low risk, 40,42 and 17 cohorts (from 10 reports) produced findings judged at high risk of bias. 4,6,13,27,28,33,35,36,39,43 The great majority of participants were in the majority of cohorts judged to produce findings at high risk of bias.
A summary of the QUIPS assessment is shown in Table S2. The domain which was most commonly judged to be at high risk of bias was analysis and reporting, whilst that most rarely judged to be at high risk of bias was prognostic factor measurement; high risk of bias was equally common across the other four domains. Inclusion of appropriate confounding factors is one of these domains and cohorts reported adjusting for a range of factors at the level of the participant and the wound. Age and gender were the most common person-level factors, while wound duration and measures of wound size were the most common wound level factors. Full details are given in Tables S3-S5. As previously noted, evidence was also assessed for inconsistency, imprecision, indirectness and publication bias. Evidence for most of the prognostic factors evaluated was downgraded at least once for risk of bias; other downgrading decisions varied but almost all evidence was assessed as either low or very low certainty, with many associations downgraded at least once for imprecision. Publication bias was considered likely in evidence for the association between infection and wound healing.  Table S3. As outlined, we prioritised data from fully reported adjusted analyses, where these were available, in forming our GRADE assessments, while taking into account the agreement with evidence from other cohorts and their data.      95% CI 0.98 to 2.40). 29 This was low certainty evidence downgraded twice for imprecision.

The most commonly assessed species was Staphylococcus aureus
where the association with healing was assessed in six cohorts (801 participants), followed by strains of Streptococcus (five cohorts, 731 participants) 17,29,[39][40][41] ; and Pseudomonas (five cohorts, 711 participants). 27,29,[39][40][41] Adjusted effect estimates were available for S. aureus for two cohorts (393 participants) 17,29 ; for Streptococcus these two cohorts also reported adjusted effects, 17,29 while for Pseudomonas only a single cohort (299 participants) did so. 29 In each case adjusted  We identified 21 studies reporting on 28 cohorts; most were prospective studies but a minority including six cohorts of registry data were retrospective. GRADE assessments judged almost all the evidence to be low or very low certainty, largely due to high all-domain risks of bias and imprecision.
There was some moderate certainty evidence that a diagnosis of

| Mixed wound infection status studies
Cohorts that recruited people with wounds that could be either and is probably an additional limitation for this association across wound types. 44 Another limitation was the fact that several studies did not report how they defined wound infection in their baseline assessment (see Tables S2 and S3). The evidence for wound infection being prognostic for wound healing had very low certainty in pressure ulcers and venous leg ulcers.
In determining the certainty of the evidence we considered carefully whether observed association sizes should prompt uprating of the evidence for moderate or large effect size, but the studies which might have given rise to this decision were at high risk of bias in more than one domain and the risk of bias might have led to inflated association statistics sizes. Evidence from these mixed infection status cohorts for other potential prognostic factors was very limited.

| Wounds infected at baseline
Where wounds are infected at baseline, studies evaluated several potential microbiological prognostic factors in foot ulcers in diabetes including the presence of multiple individual genera or species of bacteria but, in most cases, the there is no clear association with healing. This is low certainty evidence because of imprecision.
There is no evidence for any other type of wound where there is baseline infection.

| Wounds without baseline infection
In wounds without an infection at baseline, studies also evaluated several potential microbiome-related prognostic factors in foot ulcers in diabetes, including bacterial load and diversity as well as individual genera or species but in no case did any study find a clear association with healing. There is very limited and uncertain evidence in venous leg ulcers without baseline infection. There were no studies assessing potential prognostic factors in any other type of wound where there is no baseline infection.

| Relationship of the evidence base to clinical practice
There are many potential microbiological prognostic factors for wound healing in complex wounds. The studies we identified collectively assessed a large number of these. In almost all cases there was no clear association, or too much uncertainty to determine if there was an association, between the potential factor and wound healing.
The exception is evidence that clinical infection is associated with poorer healing for foot ulcers in diabetes (low certainty) and open surgical wounds (moderate certainty) but even this may be affected by publication bias. Even though this is low certainty evidence, considered across all wound types, the taking of steps to prevent wound infection remains clinically important for a range of reasons. Treating infections clearly also remains important.
Whilst wounds that are infected may be at increased risk of poor healing, this association requires further exploration and findings of this work might guide future development of treatment regimes. It is important to note that an association between a prognostic variable and an outcome does not imply that there is a causal relationship. 46 Clinical wound infection may be reflective of a sub-optimal wound status at baseline rather than a cause of it. Although the included studies adjusted for some factors in their analyses it is highly likely that underlying issues at the level of the wound or person played a role in determining both susceptibility to infection and healing trajectory in the wounds evaluated. We know that there is limited evidence for the effect of antimicrobial treatments on healing in the types of complex wounds included here. 22,47 We did not identify any studies that looked at the relationship between microbiological responses to treatment and wound healing, although antibacterial treatments were used and adjusted for in analyses in studies in infected wounds.
It has been proposed that other features of wound microbiomes (bacterial load, diversity and presence of particular types of bacteria or resistant strains of bacteria) may be predictive of or responsible for poor healing of complex wounds. 19 The evidence we identified was relatively limited and what data were available on potential factors assessed showed no clear relationship with wound healing. In particular, we did not find evidence to support a prognostic relationship of nonclinical wound features such as the concept of "critical colonisation" with wound healing. Bacterial load and bacterial diversity were assessed only in wounds without baseline infection and, in each case, there was limited evidence which did not show clear associations between load. 13,17 We identified only one study which specifically looked at colonisation, 39 and one which looked at presence of biofilm. 34 In both cases there were issues with the way in which these potential prognostic factors were assessed. There were no data which addresses whether "critical colonisation" may be associated with poorer healing, and the data on bacterial load were imprecise, being drawn from small cohorts.
This summary of the evidence shows that we do not know with any certainty whether the load or variety of bacteria, or the presence of particular groups of bacteria in complex wounds is independently associated with the healing of those wounds. Low numbers of participants contributed to substantial imprecision around most measures of association, and in many cases the data came from studies at high risk of bias. Our findings are in line with guidance which suggests restricting use of many antimicrobial dressings to wounds with clinical infection. 2 There is a particularly clear gap in the evidence for possible microbiological prognostic factors for the healing of pressure ulcers, especially given that the healing trajectory for these wounds can be slow and difficult to predict once mechanical factors are reflect underlying factors such as reduced immunological response or poor perfusion, which also lead to reduced healing. 48