Background

Chronic wounds affect about 1% of the population in Western industrialised countries, with much higher rates in inpatient settings, and pose a serious risk to patients’ health and quality of life [1,2,3,4]. Negative pressure wound therapy (NPWT), also called vacuum-assisted wound closure, was introduced into clinical practice in the early 1990s. With this technique, an open-cell foam dressing is placed into the wound cavity and a controlled subatmospheric pressure is applied to suck fluid from the wound, with the intention of improving wound healing [5]. In the past decades, the use of NPWT has increased considerably and it is currently applied across the world in both inpatient and outpatient settings for various surgical indications. Although multiple clinical benefits have been described, most clinical studies or evidence syntheses have failed to prove statistically significant or clinically relevant benefits versus standard wound therapy (SWT). For instance, in 2006, the German Institute for Quality and Efficiency in Health Care (Institut für Qualität und Wirtschaftlichkeit im Gesundheitswesen, IQWiG) conducted a health technology assessment (HTA) of NPWT studies [6] followed by a rapid report in 2007 [7] and found that “although there is some indication that NPWT may improve wound healing, the body of evidence available is insufficient to clearly prove an additional clinical benefit of NPWT. The large number of prematurely terminated and unpublished trials is the reason for concern” [8]. The IQWiG reports contained only a few small studies (all conducted in Western industrialised countries), all of poor methodological quality. In the meantime, considerably more evidence has accumulated on NPWT from randomised controlled trials (RCTs) conducted in multiple surgical indications and settings.

The aim of this systematic review of RCTs was therefore to assess the patient-relevant benefits and harms of NPWT versus SWT. Due to numerous and changing surgical indications and further developments in technology, our analysis considered all wounds healing by secondary intention.

Methods

Protocol and methodological approach

Our review formed part of a German-language HTA of the benefits and harms of NPWT in patients with wounds healing by secondary intention published by IQWiG in 2019. The full (German-language) protocol and report (Commission No. N17-01A) are available on the Institute’s website [9]. Both the preliminary protocol and the preliminary report underwent public commenting procedures. IQWiG’s responsibilities and methodological approach are described in its methods paper [10]. Only completed studies were used, so there was no need for ethical approval and patient consent. We adhered to the PRISMA statement [11] throughout this manuscript.

Eligibility criteria

We included both published and previously unpublished RCTs comparing NPWT for wounds healing by secondary intention with any kind of SWT and investigating at least one predefined patient-relevant outcome. In this context, the term “patient-relevant” refers to “how a patient feels, functions or survives” [12]. The detailed eligibility criteria are presented in Table 1

Table 1 Eligibility criteria for studies included
Full size table

Search strategy and study selection

This systematic review is based on two previous HTA reports of IQWiG [6, 7]. We conducted an update search for the period not covered by these reports (from 2006 onwards). We searched the following bibliographic databases: MEDLINE, Embase, the Cochrane Central Register of Controlled Trials, the Cochrane Database of Systematic Reviews, and the Health Technology Assessment Database. The peer-reviewed search strategy included a combination of subject headings and free text with terms such as “negative pressure wound therapy” and “vacuum-assisted closure” (see Additional file 1 for the full search strategy). In addition, we searched ClinicalTrials.gov and the International Clinical Trials Registry Platform Search Portal. The last search was run on July 24, 2018. The reference lists of relevant systematic reviews and HTA reports published between 2013 and 2018 were scrutinised to identify further studies. In order to obtain the most complete data set possible, we also asked NPWT manufacturers to supply unpublished studies and additional unpublished data from published studies (see Additional file 1 for the full list of manufacturers).

As a prerequisite for the use of unpublished data, IQWiG asked the manufacturers to sign an agreement requiring (1) the submission of a list of all sponsored published and unpublished studies investigating NPWT and (2) the submission of CONSORT-compliant documents (in general the complete clinical study reports, CSRs) on all relevant studies selected by IQWiG. This procedure was required to avoid bias through the selective provision of data. Furthermore, we contacted the investigators responsible for investigator-initiated trials (IITs) to obtain the current study status or even data from potentially completed studies identified in study registries. In addition, persons and parties who had submitted comments on the preliminary version of the IQWiG report in the written public hearing were asked to provide any additional relevant studies.

Two reviewers independently screened titles and abstracts of the retrieved citations to identify potentially eligible primary and secondary publications. The full texts of these articles were obtained and independently evaluated by the same reviewers. All documents retrieved from non-bibliographical sources were also screened for eligibility or relevant information on studies. Disagreements were resolved by consensus.

Literature searching and study selection were done in parallel for two HTA reports, one on NPWT in patients with wounds healing by primary intention and one on NPWT in patients with wounds healing by secondary intention. The results of the HTA report on wounds healing by primary intention will be reported separately.

Data extraction

The individual steps of the data extraction and risk-of-bias assessment procedures were always conducted by one person and checked by another; disagreements were resolved by consensus. Details of the studies were extracted using standardised tables.

We extracted information on:

  1. (1)

    Study characteristics, including the study design, length of follow-up, sample size, location, number of centres and period in which the study had been conducted.

  2. (2)

    Characteristics of the study participants, including inclusion and exclusion criteria, age, sex, wound characteristics at baseline, time since wound occurrence and dropout rate.

  3. (3)

    Characteristics of the test and control interventions, including treatment regimens and concomitant treatments.

  4. (4)

    Outcomes and type of outcome measures: outcomes as presented above; we did not limit the types of measures for a specific outcome, but rather analysed all measures used (e.g. wound healing (yes/no), time to wound healing).

  5. (5)

    Risk-of-bias items (see below).

    Information and data from publications were supplemented by publicly available results data from study registries and unpublished CSRs provided by manufacturers or IIT investigators.

Assessment of risk of bias in individual studies

Using the IQWiG methods, we assessed the risk of bias (high or low) on the study and outcome level [10]. Because of the large number of studies, we conducted a stepwise assessment: if the generation of a randomisation sequence and/or the allocation concealment were inadequate, we assigned a high risk of bias to the study. If we did not, the following items were assessed at study level across outcomes: blinding of patients and treating staff, reporting of all relevant outcomes independent of results and other aspects, such as differences in the length of follow-up.

A high risk of bias on the study level generally led to a high risk of bias on the outcome level. Otherwise, the following outcome-specific items were assessed: blinding of outcome assessors, appropriate application of the intention-to-treat principle, reporting of individual outcomes independent of results and other aspects.

Using the IQWiG methods, we graded the results of the (meta-)analysis into different categories: proof, indication and hint (or neither proof, nor indication nor hint) of a greater benefit of the test intervention. In short, proof of a greater benefit of the test intervention is inferred if a meta-analysis of at least 2 studies with a low risk of bias shows a statistically significant effect favouring the test intervention. An indication of a greater benefit is inferred if one single study with a low risk of bias shows a statistically significant effect favouring the test intervention or a meta-analysis of studies with a high risk of bias shows a statistically significant effect favouring the test intervention. A hint of a greater benefit is inferred if a single study with a high risk of bias shows a statistically significant effect favouring the intervention. No proof (or indication or hint) of a greater benefit or harm is inferred if there are no statistically significant differences between the test and control interventions, if relevant heterogeneity exists or if no suitable data are available.

If studies with both a low and a high risk of bias are available for a specific outcome, the studies with a low risk of bias are primarily used to derive proof, an indication or a hint of a greater benefit of NPWT.

In addition to IQWiG methods, we also applied the GRADE approach (Grading of Recommendations Assessment, Development and Evaluation) in order to describe the certainty of the evidence in a widely used framework.

Assessment of publication bias

Studies missing for analysis were defined as those that fulfilled the eligibility criteria listed above (except for the reporting of at least one patient-relevant outcome), had been completed at least 1 year before the last bibliographic and registry search and were not published. Studies that had been terminated prematurely or had an unclear study status (no update of study status in the 2 years before our literature search) were also counted as missing, as long as no contrary information was available (e.g. from author inquiries).

We assessed publication bias by comparing the estimated number of patients from missing studies with the number of all patients (from included and missing studies). If the proportion of missing data was < 10%, it was assumed that the impact of bias on the results introduced by the missing data was low and no action was taken. If the proportion of missing data was between 10 and 30%, it was assumed that the impact of bias on the results was high and all conclusions of proof, an indication or a hint of a greater benefit of NPWT were downgraded to an indication, a hint or no hint. If the proportion of missing data was > 30%, it was assumed that the impact of bias on the results was too high to be able to draw robust conclusions and no data analysis was performed.

Data analysis

If results for different time points were available, the most recent one was used for the analysis, if not stated otherwise. The mean values and standard deviations of continuous variables were derived from the median, minimum and maximum values or the first and third quartile using the method by Wan 2014 [13] or from standard errors or confidence intervals (CI). If no information was available, missing standard deviations were derived from the median values of the standard deviations of all control interventions.

Odds ratios (OR) were calculated to compare dichotomously measured outcomes, mean differences (MD) or Hedges’ g were calculated to compare continuously measured outcomes. In most cases, Hedges’ g was used to adjust for different wound types, different scales applied for outcome measurements or heterogeneity in the original scale. For all effect estimates, 95% CI were reported.

If feasible and meaningful, data were pooled by means of meta-analyses. An overall effect was calculated using the Knapp and Hartung method with the Paule-Mandel heterogeneity estimator [14]. If only 2 studies were available, a fixed-effect model with inverse variance [15] was used to combine the study results. We used the beta-binomial model [16] to calculate an overall effect estimate to account for studies with no events in both treatment arms (double-zero studies).

If relevant heterogeneity [15] was present (p < 0.05), no overall effect estimate was calculated and, if possible, a 95% prediction interval (PI) [17] was calculated instead.

The results of the meta-analysis were presented in a forest plot. If studies with a low risk of bias showed a statistically significant effect, they were presented separately within the same plot.

A p value of < 0.05 was considered statistically significant.

We also planned subgroup analyses for age, sex, type of indication/wound, type of health care setting, and type of NPWT device.

Results

Literature search

A total of 42 eligible studies were identified from 1195 references retrieved from bibliographic databases. Details of the study selection process from bibliographic databases are shown in Fig. 1. In addition, 20 potentially eligible studies were identified in further sources: 7 studies in the previous IQWIG reports, 5 in the reference lists of other relevant systematic reviews and one in a study registry. Two studies were identified in study registries and became eligible due to data provided by the investigator or manufacturer. Five additional and previously completely unpublished manufacturer-initiated studies finalised the initial study pool (details on data submission by the manufacturers are provided in the current IQWiG report [18]).

Fig. 1
figure 1

Flowchart of study selection (based on Moher et al. [11])

Full size image

The 62 eligible studies (all RCTs) included 14 studies (ActiVac [19], Dwivedi 2016 [20, 21], Eginton 2003 [22], Ford 2002 [23], ISAW [24,25,26] [27], Joseph 2000 [28], Keskin 2008 [29], Riaz 2010 [30], Sajid 2015 [31], Sun 2007 [32], Vaidhya 2015 [33] and Wanner 2003 [34]) that failed to provide any evaluable data on patient-relevant outcomes. As these studies fulfilled the eligibility criteria and for transparency reasons, they were formally included in the initial study pool but excluded from the analysis. Ultimately, 48 eligible studies with evaluable data on 4315 patients were analysed.

Potential publication bias

The literature search also identified 45 further studies without any published data on patient-relevant outcomes (of which 12 were planned or ongoing). These studies included 14 terminated, 9 completed studies and 10 studies with an unknown study status—30 out of these 33 studies should already have provided results, as the study had been completed at least 12 months before the search date of the present review. These 30 studies represented missing data of at least 1386 patients; for further details, see Table 2. Compared with the available evaluable data of 4315 patients, this results in a proportion of inaccessible data of 24% (1386/5701) of eligible patients. In consequence, we downgraded the certainty of our conclusions as described in the “Methods” section and refrained from doing subgroup analyses because the results would be hardly interpretable. Detailed documentation of all 45 studies without any published data on patient-relevant outcomes is given in Table 18 of the current IQWiG report [18].

Table 2 Unpublished studies considered for assessment of potential publication bias
Full size table

Characteristics of included studies

Table 3 presents the main characteristics of the 48 studies reporting evaluable results on patient-relevant outcomes. These studies included between 12 and 460 patients and were conducted worldwide between 1998 and 2016. The majority were 2-arm studies (n = 46); one study was 3-armed (Novinščak 2010 [112]) and one was 4-armed (TOPSKIN [122]). The study design was mostly monocentric (n = 35). The majority of studies were performed in an inpatient setting (n = 38). In 47 studies, patients were randomised and in one study (Moisidis 2004 [107]) wound halves were randomised. In 45 studies, one wound per patient and in 2 studies (Kakagia 2014 [100] and VAC 2001-06 [132]) at least one wound per patient was analysed.

Table 3 Characteristics of included studies
Full size table

The 48 studies included covered a wide range of different wounds of various causes: amputation wounds (n = 1, Liao 2012 [103]), pressure ulcers (n = 2, Ashby 2012 [70], VAC 2001-01 [125]), diabetic foot wounds (n = 6, Dalla Paola 2010 S-II [86], DiaFu [89], Karatepe 2011 [101], Nain 2011 [111], VAC 2001-07 [135], VAC 2001-08 [142]), diabetic ulcer wounds (n = 1, Novinščak 2010), foot wounds (n = 1, Chiang 2017 [83]), fasciotomy wounds due to compartment syndrome (n = 1, Kakagia 2014), necrotizing fasciitis wounds (n = 2, Huang 2006 [98], Xu 2015 [162]), open fractures (n = 7, Arti 2016 [68], Gupta 2013 [96], Jayakumar 2013 [99], Sibin 2017 [117], VAC 2001-06, Virani 2016 [154], WOLLF [158]), open abdominal wounds (n = 4, Bee 2008 [73], Correa 2016 [84], Rencüzoğulları 2015 [114], VAC 2002-10 [152]), pilonidal sinus wounds (n = 2, Banasiewicz 2013 [72], Biter 2014 [74]), open thorax wounds (n = 1, VAC 2002-09 [150]), traumatic wounds of various causes (n = 3, Llanos 2006 [104], Saaiq 2010 [115], Sinha 2013 [118]), leg ulcer wounds (n = 4, Leclercq 2016 [102], VAC 2001-02 [128], VAC 2001-03 [130], Vuerstaek 2006 [155]), burns (n = 2, of which one was in infants, TOPSKIN, Shen 2013 [116]), groyne wounds caused by infection (n = 1, Acosta 2013 [65]), and various other wounds due to diseases and/or traumatic or iatrogenic causes (n = 10, Braakenburg 2006 [76], CE/044/PIC [77], De Laat 2011 [87], Hu 2009 [97], Mody 2008 [105], Moisidis 2004, Mouës 2004 [108], Mohsin 2017 [106], Perez 2010 [113], SWHSI [119]). Comparators were mostly described as standard wound care or standard dressings. If specified, dressings were described as sterilised gauze or moist gauze. Very few sudies provided more detailed information such as alginate, hydrofiber, silver-dressing or polyurethanes.

Risk of bias in individual studies

The risk of bias at the study level was low in 7 (Ashby 2012, DiaFu, Llanos 2006, SWHSI, VAC 2001-07, Vuerstaek 2006 and WOLLF) out of 48 studies. As shown in Table 4, 41 studies were rated as having a high risk of bias at the study level due to an inadequate description of the randomisation procedure (n = 20) and/or the allocation concealment (n = 37), or selective reporting of outcomes (n = 4; TOPSKIN, VAC 2001-01, VAC 2001-02 and VAC 2001-08). Of the 7 studies rated as having a low risk of bias at the study level, only one also showed a low risk of bias for all outcomes (Llanos 2006). The studies by Ashby 2012 and SWHSI showed a low risk of bias, but not for all of the reported outcomes: pain, wound bleeding and infection (Ashby 2012) as well as adverse events, pain, duration of hospital stay and health-related quality of life (SWHSI) were rated as having a high risk of bias. In all other studies, the risk of bias at the outcome level was high. The detailed risk-of-bias assessments for all outcomes in all studies included are presented in Additional file 2.

Table 4 Risk of bias of included studies
Full size table

Effects of NPWT versus SWT

Wound closure was measured as wound healing (yes/no), time to wound healing (in days), wound healing and/or surgical wound closure (yes/no), time to wound healing and/or surgical wound closure (in days), time to wound healing after the intervention and surgical wound closure (< 6 weeks yes/no), and time to wound healing after the intervention and surgical wound closure (in days).

The results for wound healing (yes/no) were reported in 14 studies (Acosta 2013, Ashby 2012, Braakenburg 2006, CE/044/PIC, Dalla Paola 2010 S-II, Hu 2009, Leclercq 2016, Llanos 2006, Moisidis 2004, Novinščak 2010, VAC 2001-03, VAC 2001-07, VAC 2001-08 and Vuerstaek 2006). Two studies with a low risk of bias (Ashby 2012 and Llanos 2006) showed no statistically significant difference between the groups. The combined analysis of studies with a low and high risk of bias showed a statistically significant effect in favour of NPWT (OR 1.56, 95% CI 1.15 to 2.13, p = 0.008, see Fig. 2).

Fig. 2
figure 2

Forest plot of wound healing with overall effect estimation, NPWT vs. SWT. Abbreviations: CI confidence interval, n number of events, N number of patients, NPWT negative pressure wound therapy, OR odds ratio, SWT standard wound therapy

Full size image

Only 7 studies defined wound healing as 100% re-epithelialization (Acosta 2013, CE/044/PIC, Dalla Paola 2010 S-II, Moisidis 2004, VAC 2001-07, VAC 2001-08 and Vuerstaek 2006), while in the other 7, a definition was missing. We therefore conducted a sensitivity analysis with the definition of wound healing as a stratification factor. Studies with a proper definition showed a statistically significant effect in favour of NPWT (OR 1.53, 95% CI 1.04 to 2.23, p = 0.034). Studies without such a definition showed no statistically significant effect (OR 1.72, 95% CI 0.73 to 4.04, p = 0.163). However, the interaction test showed no statistically significant difference between the two effects (p = 0.743). There was an indication of a greater effect of NPWT on wound healing.

Six studies reported data on time to wound healing (in days) (Acosta 2013, Biter 2014, Karatepe 2011, Llanos 2006, Shen 2013 and Vuerstaek 2006). The study with a low risk of bias (Llanos 2006) (Hedges’ g − 1.33, 95% CI − 1.90 to − 0.77, see Fig. 3), as well as studies with a low and high risk of bias (Hedges’ g − 0.77, 95% CI − 1.19 to − 0.35, p = 0.005, see Fig. 3) showed a statistically significant difference in favour of NPWT. The results were classified as clinically relevant, since the upper limits of these 95% CIs were below the irrelevance threshold of − 0.2. A sensitivity analysis stratified by the definition of wound healing showed a statistically significant result for the 4 studies (Acosta 2013, Biter 2014, Shen 2013 and Vuerstaek 2006) with a proper definition (100% re-epithelialization; Hedges’ g − 0.69, 95% CI − 0.88 to − 0.49, p < 0.001) and the 2 studies with a missing definition (Hedges’ g − 0.95, 95% CI − 1.33 to − 0.58, p < 0.001). However, the interaction test showed no statistically significant difference between the two effects (p = 0.210). There was proof of a greater effect of NPWT on time to wound healing.

Fig. 3
figure 3

Forest plot of time to wound healing (in days) with overall effect estimation, NPWT vs. SWT. Abbreviations: CI confidence interval, n number of patients, NPWT negative pressure wound therapy, SD standard deviation, SWT standard wound therapy

Full size image

The meta-analysis of time to wound healing after the intervention and surgical wound closure (< 6 weeks yes/no) included 3 studies with a high risk of bias (Gupta 2013, Jayakumar 2013 and Sibin 2017). There was a statistically significant effect in favour of NPWT after 6 weeks (OR 16.07, 95% CI 3.19 to 80.97, p = 0.018, see Fig. 4).

Fig. 4
figure 4

Forest plot of time to wound healing after intervention and surgical wound closure (< 6 weeks) with overall effect estimation, NPWT vs. SWT. Abbreviations: CI confidence interval, n number of events, N number of patients, NPWT negative pressure wound therapy, OR odds ratio, SWT standard wound therapy

Full size image

The meta-analysis of time to wound healing after the intervention and surgical wound closure (in days) included 2 studies with a high risk of bias (Dalla Paola 2010 S-II and Hu 2009). There was a statistically significant difference in favour of NPWT (Hedges’ g − 1.14, 95% CI − 1.45 to − 0.84, p < 0.001, see Fig. 5). The effect was clinically relevant, since the upper limit of the 95% CI was below the irrelevance threshold of − 0.2. Overall, there was an indication of a greater effect of NPWT on time to wound healing after the intervention and surgical closure.

Fig. 5
figure 5

Forest plot of time to wound healing after the intervention and surgical wound closure (in days) with overall effect estimation, NPWT vs. SWT. Abbreviations: CI confidence interval, n number of patients, NPWT negative pressure wound therapy, SD standard deviation, SWT standard wound therapy

Full size image

Due to heterogeneity, no effect estimate was calculated for wound healing and/or surgical wound closure (yes/no) (21 studies; Acosta 2013, Arti 2016, Braakenburg 2006, Dalla Paola 2010 S-II, DiaFu, Gupta 2013, Hu 2009, Jayakumar 2013, Kakagia 2014, Mouës 2004, Nain 2011, Perez 2010, Saaiq 2010, Sibin 2017, SWHSI, VAC 2001-06, VAC 2001-07, VAC 2001-08, Virani 2016, WOLLF and Xu 2015) and time to wound healing and/or surgical wound closure (in days) (9 studies; Braakenburg 2006, Mouës 2004, Perez 2010, VAC 2001-07, VAC 2001-08, VAC 2002-09, VAC 2002-10, Virani 2016 and Vuerstaek 2006). In both analyses, the 95% PI included the zero effect (OR = 1 and Hedges’ g = 0); see Fig. 1 and 2 in Additional file 3. The heterogeneity for time to wound healing and/or surgical wound closure (in days) was mainly caused by Perez 2010. In a sensitivity analysis, the result of this study was shifted closer to zero, reducing heterogeneity. The mean value in the NPWT group was shifted towards the mean value in the SWT group until the test of heterogeneity was non-significant (p > 0.05). The resulting pooled effect was statistically significant (Hedges’ g − 0.39, 95 % CI − 0.66 to − 0.11, p = 0.013) but not clinically relevant, as the upper limit of the 95% CI was not below the irrelevance threshold of − 0.2. There was neither proof (nor indication nor hint) of a greater or smaller effect of NPWT on wound healing and/or surgical wound closure (yes/no) and time to wound healing and/or surgical wound closure (in days).

Overall, there was proof of a greater benefit of NPWT for wound closure. Due to the potential publication bias mentioned above, this conclusion was downgraded. There was thus an indication of a greater benefit of NPWT for wound closure.

Adverse events comprised additional measures required for direct wound closure (such as skin transplantation or sutures), re-interventions (such as regrafting or revision fixation), bleeding, infections, the overall rate of serious adverse events and study discontinuation due to adverse events.

Ten studies provided data on re-interventions (Chiang 2017, De Laat 2011, Liao 2012, Llanos 2006, Mohsin 2017, Moisidis 2004, Saaiq 2010, VAC 2001-07, VAC 2001-08 and WOLLF). One study with a low risk of bias (Llanos 2006) showed no statistically significant difference between groups. In the combined analysis of studies with a low and high risk of bias, NPWT significantly reduced the odds for re-interventions (OR 0.46, 95% CI 0.24 to 0.86, p = 0.021, see Fig. 3 in Additional file 3). One study (Perez 2010) measured the number of operations until wound closure and found a statistically significant difference in favour of SWT (MD 2.80, 95% CI 0.79 to 4.81, p = 0.008). Due to the small number of patients in this study, the result of the meta-analysis was not challenged. There was thus an indication of a greater effect of NPWT on re-intervention.

There were no statistically significant differences between groups with regard to additional measures required for direct wound closure (23 studies, see Fig. 4 in Additional file 3) and the overall rate of serious adverse events (12 studies, see Fig. 5 in Additional file 3). No effect estimate was calculated for infection (20 studies, see Fig. 6 in Additional file 3) and study discontinuation due to adverse events (7 studies, see Fig. 7 in Additional file 3) because of heterogeneity. In these 2 analyses, the 95% PI included the zero effect (OR = 1). For bleeding (5 studies, see Fig. 8 in Additional file 3), no events occurred in 3 out of 6 studies. The other 3 studies showed no statistically significant differences. No overall effect was therefore calculated. There was neither proof (nor indication nor hint) of a greater or smaller effect of NPWT on any of these outcomes. As the overall rate of serious adverse events was the primary analysis for adverse events, there was neither proff (nor indication nor hint) of greater benefit or harm of NPWT for adverse events.

Fig. 6
figure 6

Forest plot of hospital stay (in days) with overall effect estimation, NPWT vs. SWT. Abbreviations: CI confidence interval, n number of patients, NPWT negative pressure wound therapy, SD standard deviation, SWT standard wound therapy

Full size image
Fig. 7
figure 7

Forest plot of hospital stay (> 1 month) with overall effect estimation, NPWT vs. SWT. Abbreviations: CI confidence interval, n number of events, N number of patients, NPWT negative pressure wound therapy, OR odds ratio, SWT standard wound therapy

Full size image

Hospital stay and readmission was measured as hospital stay (in days) in 10 studies (Acosta 2013, Huang 2006, Liao 2012, Llanos 2006, Rencüzoğulları 2015, TOPSKIN, VAC 2001-06, VAC 2001-07, Vuerstaek 2006 and Xu 2015), hospital stay (> 1 month yes/no) in 4 studies (Gupta 2013, Jayakumar 2013, Saaiq 2010 and Sibin 2017), intensive care unit stay (in days) in 2 studies (Rencüzoğulları 2015 and TOPSKIN) and readmission (yes/no) in 4 studies (CE/044/PIC, De Laat 2011, SWHSI and VAC 2001-07). For all outcomes and studies (except one [Llanos 2006]), the risk of bias was high.

For hospital stay (in days), the study with a low risk of bias (MD − 3.50, 95% CI − 5.58 to − 1.42, see Fig. 6), as well as studies with a high and low risk of bias (MD − 4.78, 95% CI − 7.79 to − 1.76, p = 0.005, see Fig. 6) showed a statistically significant difference in favour of NPWT. There was a proof of a greater effect of NPWT on hospital stay (in days).

Hospital stay (> 1 month yes/no) also showed a statistically significant effect in favour of NPWT (OR 0.07, 95% CI 0.02 to 0.17, p = 0.003, see Fig. 7). There was an indication of a greater effect of NPWT on hospital stay (> 1 month).

Despite homogeneity between the studies (I2 = 0%, p = 0.441), no effect estimate was presented for intensive care unit stay (in days) because of the large range of the 95% CI (containing the zero effect MD = 0). There was no statistically significant effect for readmission (yes/no) (OR 1.01, 95% CI 0.44 to 2.31, p = 0.973, see Fig. 9 in Additional file 3). There was neither proof (nor indication nor hint) of a greater or smaller effect of NPWT on the 2 latter outcomes.

Overall, there was proof of a greater benefit of NPWT for hospital stay and readmission. Due to the high risk of publication bias, the evidence for this outcome was downgraded. There was thus an indication of a greater benefit of NPWT for hospital stay and readmission.

Data on mortality were available in 18 studies (Acosta 2013, Ashby 2012, Bee 2018, Braakenburg 2006, Correa 2016, DiaFu, Huang 2006, Mouës 2004, Rencüzogullari 2015, Saaiq 2010, VAC 2001-01, VAC 2001-07, VAC 2001-08, VAC 2002-09, VAC 2002-10, Vuerstaek 2006, WOLLF and Xu 2015). There was no statistically significant effect of NPWT compared to SWT (OR 1.20, 95 % CI 0.84 to 1.70, p = 0.290, see Fig. 10 in Additional file 3).

Data on amputation were available in 10 studies (Acosta 2013, Braakenburg 2006, Dalla Paola 2010 S-II, DiaFu, Hu 2009, Huang 2006, Mody 2008, SWHSI, VAC 2001-06 and WOLLF). There was no statistically significant effect of NPWT compared to SWT (OR 0.89, 95% CI 0.55 to 1.43, p = 0.588, see Fig. 11 in Additional file 3).

Pain was measured continuously in 6 studies (visual analogue scale [Banasiewicz 2013, Biter 2014, SWHSI and Vuerstaek 2006], visual analogue thermometer [TOPSKIN], numeric rating scale [DiaFu]), dichotomously (pain yes/no) in 3 studies (Ashby 2012, Mody 2008 and WOLLF), and as pain on application and removal of dressing (continuously [SWHSI], dichotomously [CE/044/PIC]). None of these outcomes showed a statistically significant difference between NPWT and SWT (see Figs. 12 and 13 and eTable 1 and eTable 2 in Additional file 3).

Two studies reported data on health-related quality of life using the Physical Composite Scale (PCS) and the Mental Health Composite Scale (MCS) derived from the SF-12 after 3 months (SWHSI) and 12 months (WOLLF), respectively. There was no statistically significant difference between NPWT and SWT for the MCS; heterogeneity was shown for the PCS (see Figs. 14 and 15 in Additional file 3).

Three studies provided data on physical function measured as time to resume work or school (Biter 2004), time to restoration of normal activity (Banasiewicz 2013) and the Disability Rating Index (DRI; [WOLLF]). A meta-analysis of the first 2 outcomes showed heterogeneity, while the DRI showed no statistically significant difference between NPWT and SWT (see Fig. 16 and eTable 3 in Additional file 3).

No study reported data on dependence on outside help or need for care. There was neither proof (nor indication nor hint) of a greater benefit or harm of NPWT for mortality, amputation, health-related quality of life, physical function and dependence on outside help or need for care.

Using the beta-binomial model to account for double-zero studies did not alter the results of the meta-analyses presented.

A summary of the results is presented in Table 5. An overview of the key findings according to GRADE methods can be found in Additional file 4.

Table 5 Summary of results
Full size table

Discussion

Summary of findings

This systematic review of NPWT versus SWT in patients with wounds healing by secondary intention showed some advantages of NPWT with regard to wound closure and hospital stay. No differences were shown for any other important outcomes such as infection or amputation (see Table 5).

Comparison with previous research

It is difficult to compare the present results with previous systematic reviews on NPWT due to their more or less restricted focus, in contrast to the rather wide question of the present review, which is wounds healing by secondary intention. Using only publications from 2013 to 2018, we identified 30 systematic reviews on NPWT for various wounds healing by secondary intention (see Additional file 5). Only 13 analysed wound closure, the key outcome for this type of intervention. Of these, 10 came to a positive conclusion [163,164,165,166,167,168,169,170,171,172], while 3 did not [173,174,175]. More importantly, only 14 [163,164,165,166, 168, 169, 171, 173, 175,176,177,178,179,180] of the 30 previous systematic reviews at least mentioned the risk of publication bias and none implemented any consequences for their conclusions in the event of this type of bias.

Data pooling

The question as to whether to pool clinical study data or not is fundamental in meta-analysis. In the assessment of NPWT, various medical aspects need to be considered before pooling the data into a common effect estimate. The type of wound investigated is the most obvious difference between the studies in our analysis. Further factors potentially affecting study results are the exact type of NPWT technique used (e.g. pressure applied), choice of control treatment, type of healthcare setting and study duration. However, against the background of the lack of a standardised nomenclature for wounds and the fact that if the underlying disease and the respective wound is prepared optimally, wounds are very similar, it seemed meaningful to pool the data with regard to the type of wound healing. Furthermore, innovation would become impossible if each new wound treatment had to be tested for each of the numerous different wound types. Since several meta-analyses showed no or only little heterogeneity, our data at least do not contradict this approach.

Published evidence tends to overestimate the benefits and underestimate the harms of medical interventions [181], and it is widely accepted that the results of all relevant studies must be fully available for an unbiased estimation of effects. The introduction of mandatory measures such as registration of studies and their results has increased data availability, but many studies (in the present case mostly IITs), especially on medical devices, still remain partly or fully unpublished. Further measures to ensure full disclosure are thus urgently required. As the present case shows that voluntary commitment does not work, further legal and regulatory action, in combination with sanctions, seems to be indispensable. For instance, IQWiG proposes that funders of clinical research and ethic committees exert stronger supervision over research projects by denying further funding if previous projects were not properly published [18].

Strengths and limitations

Strengths of the present research include the systematic literature search and the inclusion of several unpublished studies, which were in particular obtained from several NPWT manufacturers (and sometimes provided only after public pressure). This methodological rigour was essential in order to limit potential publication bias. Nevertheless, the present review included an insufficient proportion of the relevant clinical study data to draw highly robust conclusions, which is why our conclusions are cautious. In addition, we did not assess outcome reporting bias. We cannot exclude that the proportion of missing data would have been even higher if this type of bias had also been considered. It should also be noted that overall, the quality of the studies considered was low and the sample sizes were small. If larger, independently funded, multicentre trials had been performed, evaluation of NPWT would have been possible in an earlier, easier and more reliable way.

Implications for future research

The advantages of NPWT were modest, and the main conclusion with regard to wound closure was derived from only 14 of the 48 studies. Due to various definitions of wound closure and different time points of data collection (or even missing information on these items), no conclusions can be made with regard to the sustainability of wound closure. In addition, as stated, the size and quality of the studies were generally low. Therefore, it is certainly not unethical to conduct further (but high-quality) RCTs on this topic. We recommend that they systematically investigate and clearly define the key outcomes of wound healing, adverse events and health-related quality of life.

Conclusion

In summary, low-quality data indicate a greater benefit of NPWT versus SWT for the outcome of wound closure in patients with wounds healing by secondary intention. The length of hospital stay is also shortened. However, the data show no advantages or disadvantages of NPWT versus SWT for mortality, adverse events, amputation, pain or health-related quality of life. Although data on serious adverse events were not systematically collected in most of the primary studies, NPWT appears to be safe. Publication bias is an important problem in NPWT research, underlining that all clinical studies need to be fully reported regardless of funding source, premature study termination or study results.