Following characterisation of SWHSI patients and their treatment in workstream 1, the evidence gap regarding the cost-effectiveness of SWHSI treatments remained. Data collected in workstream 1 indicated that a variety of interventions are used when treating and managing SWHSIs. Although the majority of participants in the cohort study received dressings throughout follow-up, 29.3% (115/393) received a significantly more expensive technology: NPWT.
Negative-pressure wound therapy devices apply a carefully controlled negative pressure (or vacuum) to a dressing, creating an air-tight seal and removing tissue fluid from the treated wound area into a disposable canister.14 The canister is removed and replaced either when it becomes full or at least once a week. The device is generally used as part of the SWHSI treatment pathway, rather than to the point of healing, and can be administered and removed by both nurses and surgeons. It has been claimed that NPWT speeds up wound healing by removing excess fluid, increasing tissue perfusion and removing bacteria.26 However, although a recent Cochrane systematic review identified two RCTs investigating the relative effectiveness of NPWT on SWHSIs,14 it could not draw any conclusions, as the outcome data reported in both studies were very limited.
Given the current level of clinical use on SWHSIs, the cohort data collected in workstream 1 have been used here to evaluate the cost-effectiveness of NPWT for the healing of SWHSIs (the clinical outcome of wound treatments most valued by patients)27 and the value of a future RCT. From this work, a paper highlighting methodological challenges of analysing observational evidence in this context has been published (see Appendix 5). The current chapter presents the clinical implications of the research.
Objectives
- To estimate the cost-effectiveness of NPWT compared with wound dressings for the healing of SWHSIs, using cohort data from workstream 1.
- To assess whether or not investment in further research on treatments for SWHSIs is likely to be worthwhile.
Methods
In this workstream, we evaluated if, after adjustment, NPWT was clinically effective and cost-effective when compared with standard dressings. For clinical effectiveness, we used wound healing as the outcome; for cost-effectiveness, we quantified the impact of differences in time to healing on quality-adjusted life-years (QALYs). This approach assumed that no mortality differences existed and, thus, any improvements in health-related quality of life (HRQoL) were associated with accelerated time to healing only. We also noted that any additional HRQoL improvements that may have arisen from treatment (independent of healing status) were expected to be transient and not to have a significant impact on HRQoL (which is a multidimensional concept that includes domains related to physical, mental, emotional and social functioning). For this reason, such effects were not considered here.
We hypothesised that NPWT could be cost-effective by reducing healing time, improving HRQoL and reducing treatment costs (via reduced healing time). We envisaged that if the total costs of treatment were higher with NPWT than for comparator treatments, NPWT would only be cost-effective if the incremental costs to QALY gains fell below the standard NICE £20,000–30,000 per QALY threshold.
Specifically, the cost-effectiveness analysis used cost per QALY gained as the cost-effectiveness outcome. A 1 year time horizon was considered and a NHS perspective was adopted. Formally, we aimed to quantify the impact of NPWT in expected time to healing. The difference in QALYs between the interventions was derived from the expected time to healing and the EQ-5D index score while healed/unhealed, the latter estimated using the multilevel model (described in supplementary material in Appendix 5). Incremental disease costs were calculated in an analogous way to incremental QALYs, making use of the cost estimates derived in the multilevel two-part modelling approach (see Appendix 5). Treatment-related costs considered that patients received some form of treatment up to healing. The daily costs associated with the ‘non-NPWT’ group were based on average costs of dressings in participants who had received other treatments and were calculated by multiplying this cost by the difference in time to healing. For patients receiving NPWT, an average daily cost of NPWT was applied to the mean time on NPWT treatment in the cohort study. In the remaining time, patients were assumed to use dressings, with a daily cost equal to patients in the ‘non-NPWT’ group.
The cohort data were analysed for evidence regarding the various aspects of treatment and wound healing, including the impact of treatment (NPWT vs. dressings) on the time to healing and the impact of this SWHSI healing on disease-specific costs and HRQoL, as is presented in Figure 3. To estimate the effect of treatment on healing, we attempted to control the confounding (and selection bias) inherent when using observational data in this context. We adjusted for observed confounding (i.e. for characteristics observed within the cohort study sample that determine time to healing and may have differed between the groups). We also attempted to account for any potential confounding from unobserved sources.
FIGURE 3 Time to healing: descriptive Kaplan–Meier curve. Reproduced with permission from Saramago et al.28 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure. (Figure not reproduced here; see the original article.)
Health-related quality of life was assessed in the cohort study using the EQ-5D index score, obtained through a self-reported questionnaire, comprising five domains (mobility, self-care, usual activities, pain and discomfort, and anxiety and depression).22 Participants were also asked to complete a resource use questionnaire quarterly. This questionnaire was developed for this study to allow costing of NHS resources specifically attributable to the SWHSIs and included collection of number of contacts with general practitioners (GPs), nurses and doctors in NHS institutions or the patient’s home, as well as the number of nights spent as an inpatient between study follow-up time points. The costing used national sources and 2014 prices.29
Analyses required the application of econometric models to the cohort data. A Bayesian approach to inferences was utilised, computed using Markov chain Monte Carlo simulation, using WinBUGS version 1.4.3 (Medical Research Council Biostatistics Unit, Cambridge, UK).30
Identification of external evidence to supplement cohort data
To evaluate the existence of any external evidence to complement the cohort data, we also conducted the following range of systematic literature searches.
We conducted Cochrane systematic reviews of RCTs evaluating the treatment of SWHSIs, with any recognised form of NPWT.14,31
Using the general review methods and with the support of an information specialist and an expert systematic reviewer, we conducted systematic searches and identification of the following studies:
- any cohort studies evaluating the use of NPWT for SWHSIs
- any EQ-5D or related health utility data for people with SWHSIs
- any previous modelling work for the treatments of SWHSIs that might inform the planned work.
Details of the search strategies used are detailed in Appendix 6.
EuroQol-5 Dimensions index scores and disease-specific costs
A multilevel (or panel data) approach was used to model both EQ-5D index scores and quarterly costs that accounted not just for between-participant variation but also for within-participant variation across the quarterly observations. A time-dependent indicator of whether the participant had healed or not at each time point was used to capture the effect of healing. Changes from baseline EQ-5D index scores were modelled using a standard multilevel model based on the normal distribution (random intercept). For cost data, due to a high proportion of zero observations and the typical skewness associated with these data, a two-part model was used: the first part was a multilevel logistic regression (to explain the probability of observing zero costs over time) and the second was mixed gamma model with log-transformed costs (to evaluate the expected value of the non-zero observations). Both parts were multilevel models, to account for the repeated observations per participant.
We calculated the average daily cost of NPWT as £31.78 (according to centre-specific costings). For dressings, a unit cost of £2.39 per day was calculated (based on the mix of dressings used within the cohort data). The duration of NPWT treatment was evaluated from the cohort data (mean 37 days, SD 64.6 days) and was assumed to be independent of time to wound healing, as this treatment is rarely used up to the point of wound healing. NPWT patients were assumed to receive only dressings for the remainder of time unhealed.
Results indicated that healing was associated with a small (but statistically significant) increase in EQ-5D index score [mean 0.055, standard error (SE) 0.02]. Cost data, as presented in Table 11, indicated that healing significantly reduced quarterly wound-related costs (mean £865, SE £112).
| Model | Mean | 95% CrI |
|---|---|---|
| Expected EQ-5D (unhealed) | 0.091 | 0.036 to 0.146 |
| Expected EQ-5D (healed) | 0.091 + 0.055 = 0.146 | 0.098 to 0.193 |
| Difference in EQ-5D attributed to healing | 0.055 | 0.015 to 0.094 |
| Expected costs (£) (unhealed) | 1124.45 | 916.9 to 1359.1 |
| Expected costs (£) (healed) | 258.91 | 208.1 to 317.7 |
| Difference in costs (£) attributed to healing | 865 | 674.2 to 1126.5 |
Relative effectiveness (time to healing) of negative-pressure wound therapy and dressing-treated wounds
Based on the cohort data, we took two different approaches to modelling the time to healing and assessed robustness of our conclusions to the approach taken.
Modelling strategy A: ordinary least squares with imputation
Modelling strategy A used ordinary least squares (OLS) (or linear regression) to describe time to healing. OLS methods do not deal with censoring and 73 participants did not heal within the study follow-up period. To generate a complete data set on time to healing, the censored observations on time to wound healing were imputed, assuming their follow-up time plus an additional quantity, either 1 day (i.e. participants were conservatively assumed to have healed the day after they were censored) or expected healing time (data elicited from three experienced clinicians and pooled) (see Appendix 5). Note that by attributing an additional time to healing, elicited by clinical experts, to patients who have been censored in the study, we are also extrapolating beyond the study follow-up. Such extrapolation is required to allow achieving an unrestricted estimate of the effects of NPWT on mean time to healing, required for cost-effectiveness analysis.
Modelling strategy B: two-stage model
A second, more complex, modelling strategy (modelling strategy B) was also implemented, which explicitly considered the unhealed participants. Instead of OLS, it used a two-step modelling approach: a logistic regression estimated the probability of healing within the follow-up for each treatment, followed by linear regression, which estimated the expected time to healing for the population that healed within follow-up. This strategy reflected the possibility of there being a group of ‘long-term non-healers’ and allowed exploration of different determinants of the probability of healing and of time to wound healing, conditional on having healed in the adjustment for observables.
In terms of adjustment for observed confounding, and in the absence of evidence on determinants of healing in this patient group, a comprehensive variable selection process was undertaken using the cohort data sample. It aimed to identify predictors of treatment assignment with NPWT, predictors of healing and treatment effect modifiers. The selection criterion was based on goodness of fit (using the deviance information criteria at a widely used cut-off point of 5 deviance information criterion points).32 All the predictors identified were used as adjustment covariates in the regression model for time to healing.
To account for unobservable confounding, instrumental variable (IV) regression was used. This method uses a variable (the instrument) that is associated with treatment assignment but not with outcome (except for any indirect effect on treatment assignment) to adjust for unobservable confounding. A range of instruments focused on the decisions of health professionals to use NPWT, that is use of NPWT on previous patient treated, perceptions of NPWT as effective treatment, likelihood of NPWT use for patients, desire to use NPWT more frequently, perceptions of NPWT as affordable and value for money, and company representative input, were considered (see Appendix 5). The selection of instruments was based on frequentist tests aimed at assessing aspects of validity and relevance (see Appendix 5). Previous treatment used by the treating health professional was the best-performing instrument.
The IV model used two regressions (two stages). The first stage regressed treatment allocation by the instrument and the set of relevant predictors of outcome (logistic regression) to predict the probability of each individual participant being assigned to NPWT. The predictions were then used as a covariate in the second-stage regression(s), which, again, conditions on the set of relevant predictors of outcome. Inferences were obtained using a Bayesian framework, so that uncertainty over predictions of treatment allocation was fully considered. Any external evidence located (from previous research) was to have been used as prior evidence to update the data obtained from the cohort.
Cost-effectiveness of negative-pressure wound therapy
The cost-effectiveness analysis uses cost per QALY gained as the cost-effectiveness outcome. A 1-year time horizon was considered and a NHS perspective was adopted. Formally, we aimed to quantify the impact of NPWT in expected time to healing. The difference in QALYs between the interventions was derived from the expected time to healing and the EQ-5D index score while healed or unhealed, the latter estimated by the multilevel model (described in the supplementary material in Appendix 5). Incremental disease costs were calculated in an analogous way to incremental QALYs, making use of the cost estimates derived in the multilevel two-part modelling approach described above (see Appendix 5). Treatment-related costs considered that patients received some form of treatment up to healing. The daily costs associated with the ‘non-NPWT’ group were based on average costs of dressings in participants who had received other treatments and were calculated by multiplying this cost by the difference in time to healing. For patients receiving NPWT, an average daily cost of NPWT was applied to the mean time on NPWT treatment in the cohort study. In the remaining time, patients were assumed to use dressings, with a daily cost equal to that for patients in the ‘non-NPWT’ group.
Results
Identification of external evidence to supplement cohort data
We screened 1477 citations from the literature searches and found no existing health utility data collected from patients with SWHSIs that could be used. Neither did we find any relevant published models exploring either the clinical effectiveness or the cost-effectiveness of treatments for SWHSIs. In terms of relative effects for NPWT compared with alternative treatments on healing, we located four potentially relevant RCTs33–36 and no observational data.
Of the four trials, two explored the use of NPWT compared with standard dressings to treat foot wounds in patients with diabetes mellitus.33,34 One trial33 included 162 adult patients who had undergone a transmetatarsal amputation of the foot. It was not clear from the study whether or not the wounds had been left open following surgery. A second trial34 included 342 adult patients who had undergone debridement of a foot wound and were randomised to receive NPWT or dressings (predominantly hydrogels and alginates). Although it was not clear whether or not the debridement of these wounds was surgical, we were advised by local clinicians that it was likely to be. Both trials of NPWT in foot wounds reported time to healing data; however, the data appeared to include wounds that had been closed by surgery following treatment with NPWT or dressings. The two trials were considered to be at high risk of performance bias, as unblinded health professionals had made key decisions regarding the treatment of wounds, such as closure surgery. This issue has been noted previously and the validity of combining wounds closed by secondary intention and those closed by surgery has been questioned.37 The potential for high risk of bias in these studies, combined with the fact that the data from these trials related to only a subset of the heterogeneous group of patients of interest, meant that we felt that the relative healing data could not be reasonably combined with the adjusted estimates of the cohort data.
Two further trials were identified.35,36 The first study,35 with 20 participants, compared NPWT with an alginate dressing in the treatment of open, infected groin wounds. The second study,36 with 49 participants, compared NPWT with a silicone dressing in the treatment of excised pilonidal sinus. No useable healing data were reported in either trial. For this reason, these studies did not provide relevant evidence to complement the cohort data.
Effectiveness results when adjusting for observables
Modelling strategy A: ordinary least squares with imputation
Relevant associations found with predictors of treatment assignment with NPWT were wound area > 25 cm2, skin and subcutaneous tissue loss (a measure of tissue involvement) and inpatient (vs. outpatient) management. Relevant associations found with predictors of healing were inpatient (vs. outpatient) management and previous history of a SWHSI. The data also suggested that history of a SWHSI could be a relevant treatment effect modifier. Predictors of treatment assignment and predictors of healing were used to adjust all regressions within modelling strategy A.
In the model in which treatment effect modifiers were not considered, and censored observations were assumed to heal 1 day after follow-up, participants treated with NPWT were expected to take longer to heal than those who did not receive NPWT, on average 73 days longer [95% credible interval (CrI) 33.8 to 112.8 days longer] (Table 12). The addition of the interaction term between treatment and SWHSI history indicated that NPWT was still expected to increase time to healing compared with the use of dressings alone, but the magnitude of effect differed between the groups (181 days for participants with history of SWHSIs vs. 42 days for patients without history of a SWHSI). Results were similar when elicited expert opinion was used to impute censored healing times; however, the point estimates were even less favourable for NPWT (requiring, on average, 114 days additional days to wound healing compared with the use of standard dressings) and uncertainty was significantly increased.
| Model ID | Model | ||||
|---|---|---|---|---|---|
| 0 | 1 | 2 | 3 | 4 | |
| Instrument used, z | None | None | None | Previous treatment | Previous treatment with interaction |
| Equation | TTH | TTH | TTH | (Stage 2) TTH | (Stage 2) TTH |
| Adjustment covariates | None | Observables | Observables + interaction | Observables | Observables + interaction |
| Area (> 25 cm2), x1 | 18.0 (95% CrI –23.1 to 58.0) | 21.1 (95% CrI –20.3 to 62.1) | 23.8 (95% CrI 29.8 to 43.7) | 36.2 (95% CrI –23.7 to 56.3) | |
| Treatment location,a x2 | 45.3 (95% CrI 11.3 to 79.2) | 43.8 (95% CrI 9.9 to 76.8) | 48.9 (95% CrI 3.0 to 63.8) | 55.3 (95% CrI 10.8 to 70.8) | |
| Tissue involvement,b x3 | 31.9 (95% CrI –1.5 to 65.0) | 26.6 (95% CrI –5.3 to 58.7) | 33.8 (95% CrI –8.0 to 73.1) | 37.2 (95% CrI –2.0 to 50.6) | |
| History, w | 7.5 (95% CrI –28.4 to 42.9) | –29.3 (95% CrI –71.5 to 12.1) | 7.2 (95% CrI –29.5 to 44.6) | –10.3 (95% CrI –61.4 to 7.4) | |
| NPWT, t | 108.0 (95% CrI 73.6 to 142.2) | 73.2 (95% CrI 33.8 to 112.8) | 41.9 (95% CrI –1.1 to 84.1) | 56.9 (95% CrI –71.7 to 192.8) | 9.1 (95% CrI –120.2 to 55.9) |
| NPWT × history, w.t | 138.7 (95% CrI 56.9 to 221.8) | 69.4 (95% CrI –67.5 to 118.0) | |||
| Constant | 116.8 (95% CrI 97.9 to 135.2) | 81.4 (95% CrI 53.3 to 110.0) | 92.7 (95% CrI 64.5 to 120.8) | 81.1 (95% CrI 52.3 to 109.1) | 86.7 (95% CrI 56.7 to 97.0) |
| Observations | 393 | 372 | 372 | 372 | 372 |
Modelling strategy B: two-stage model
This modelling approach explicitly modelled the probability of healing within follow-up and, in addition to the predictors of treatment assignment mentioned in Modelling strategy A: ordinary least squares with imputation, it identified associations with body mass index (BMI) and wound location being the foot (vs. other locations). For those healing within follow-up, time to healing was associated with surgery type (colorectal vs. other surgeries), wound duration and diabetic foot wounds. No evidence of relevant interaction terms was found. As detailed in Table 13, results indicated that participants treated with NPWT were estimated to have lower probability of healing than those who did not receive NPWT [odds ratio (OR) of healing with NPWT was 0.59 (95% CrI 0.28 to 1.12)]. The expected time to healing of NPWT-treated participants who healed compared with participants treated with dressings alone was an additional 46 days (95% CrI 19.6 to 72.5 days).
| Model ID | Model | |||
|---|---|---|---|---|
| b0 | b1 | |||
| Equation | (Stage 2A) OR for healing | (Stage 2B) TTH | (Stage 2A) OR for healing | (Stage 2B) TTH |
| Type of regression model | Logit | OLS | IV, logit | IV, OLS |
| Adjustment | None | None | Adjustment set | Adjustment set |
| Area (> 25 cm2), x1 | 0.52 (95% CrI 0.24 to 0.99) | 27.7 (95% CrI 0.6 to 54.5) | ||
| Treatment location,a x2 | 0.52 (95% CrI 0.22 to 1.04) | 30.9 (95% CrI 9.2 to 52.3) | ||
| Tissue involvement,b x3 | 0.50 (95% CrI 0.25 to 0.91) | 11.4 (95% CrI –8.1 to 30.9) | ||
| BMI (kg/m3), v1 | 1.04 (95% CrI 1.00 to 1.09) | |||
| Location: foot, v2 | 0.34 (95% CrI 0.16 to 0.64) | 30.1 (95% CrI 9.2 to 52.3) | ||
| Surgery type, v3 | 13.1 (95% CrI –6.1 to 32.4) | |||
| SWHSI duration (days), v4 | 2.3 (95% CrI 0.6 to 4.0) | |||
| Diabetic feet, v5 | 42.8 (95% CrI 9.1 to 76.3) | |||
| NPWT, t | 0.35 (95% CrI 0.20 to 0.57) | 69.0 (95% CrI 48.2 to 89.1) | 0.59 (95% CrI 0.28 to 1.12) | 46.0 (95% CrI 19.6 to 72.5) |
| Constant, γ1 | 6.56 (95% CrI 4.70 to 9.37) | 82.0 (95% CrI 71.9 to 92.1) | 6.98 (95% CrI 1.62 to 29.55) | 36.9 (95% CrI 12.4 to 61.3) |
| Observations | 393 | 354 | ||
Effectiveness results when adjusting for observables and unobservables
Overall, conclusions were similar using modelling strategies A and B, and treatment effect estimates adjusted for unobservables were broadly consistent with the results we observed when adjusting only for observables. Participants treated with NPWT were predicted to take longer to heal, albeit with a high level of uncertainty in the estimates [e.g. modelling strategy B, second-stage regression: 46 days longer (95% CrI 19.6 to 72.5 days longer)].
Cost-effectiveness of negative-pressure wound therapy
The results for healing, utilities and costs were combined to generate cost-effectiveness estimates for NPWT. Evidence relating to the additional days until healing after censoring for unhealed patients within the follow-up period was elicited from three experts and used in an extended analysis, the results of which are shown in Table 14. Irrespective of the adjustment made (on either observables or unobservables), the results indicated that treatment with NPWT was expected to be less effective and a more costly use of NHS resources than treatment with standard dressings alone: associated incremental QALYs of –0.012 (SE 0.005) (model A, observables), –0.008 (SE 0.011) (model A, unobservables), –0.007 (SE 0.004) (model B, observables) and –0.027 (SE 0.017) (model B, unobservables). There was little decision uncertainty in this result. When losses arising from the displaced health-care resources were added to the direct health effects (by valuing 1 QALY at £20,000), the health consequences of using NPWT were evaluated at –0.111 (model A, observables), –0.095 (model A, unobservables), –0.087 (model B, observables) and –0.183 (model B, unobservables).
| Elicited | Fitted values for parameters of a gamma distribution | ||||||
|---|---|---|---|---|---|---|---|
| 50th percentile (days) | 90th percentile (days) | Shape, first and third quartiles | Scale, first and third quartiles | Mean (days), first and third quartiles | 50th percentile (days), first and third quartiles | 90th percentile, (days), first and third quartiles | |
| Expert 1 | 1044 | 3234 | 1.12 | 1290 | 1445 | 1031 | 3322 |
| Expert 2 | 496 | 679 | 15.3 | 33.1 | 506 | 495 | 683 |
| Expert 3 | 180 | 3650 | 0.23 | 5222 | 1201 | 176 | 3176 |
| Pooled using formal synthesis (95% CrI) | 1.79 (0.03 to 54.5) | 615 (241 to 1615) | 1038 (29 to 39,065) | 559 (0 to 28,011) | 1820 (15 to 37,161) | ||
Data and analyses of the cohort study indicate, with little uncertainty, that treatment with NPWT is less effective and more costly than treatment with standard dressings alone.
Value of a future randomised controlled trial
Although there is uncertainty in the incremental cost and QALY estimates, our analyses consistently indicate that there is little uncertainty in the optimal decision (i.e. the probability of treatment with NPWT being considered cost-effective to the UK NHS was close to zero). This means that conducting a future study in order to reduce the associated uncertainty (in the costs and QALYs) would not change the optimal decision and therefore there is no value in conducting such a study for this purpose. For this reason, the value of information analyses originally planned were redundant: the value of information is zero.
Negative-pressure wound therapy is currently widely used in clinical practice. The results from the observational study may contradict beliefs with regards the effectiveness of NPWT, but, given that our findings have been derived from observational research, there is the potential for unresolved confounding to affect the results and reduce confidence in them and the associated conclusions drawn. This lack of confidence may lead decision-makers and funders to justify calls for definitive evidence from a RCT. The value of conducting such research, to obtain definitive evidence, will be looked at next.
The value of a RCT to promote implementation was calculated here using a ‘value of implementation’ approach.38 This assumes that the results of our research provide the best available evidence on the clinical effectiveness and cost-effectiveness of NPWT, and that the adjustments made for confounding returned unbiased estimates. Data from the cross-sectional survey conducted within this research programme (workstream 1) indicated a prevalence of treated SWHSIs of 4.1 per 10,000 population (95% CI 3.5 to 4.7 per 10,000 population). Using the observed median duration of wounds (28 days), the daily incidence of SWHSIs was estimated to be approximately 0.0146. Therefore, across England and Wales (population of 57,408,654, as per the Office for National Statistics’ 2014 mid-year estimates),39 we would expect 305,000 patients with a SWHSI per year. Cohort study data indicated that 29.3% (115/393) of incident patients received NPWT and so we would expect > 89,000 patients with a SWHSI receiving NPWT per year. Using this information, the value of implementation approach indicated, should definitive RCT evidence correspond to the findings of the modelling of observational data, that successful cessation of NPWT in clinical practice would have direct health benefits that would sum to between 719 and 2427 QALYs. When displaced health-care benefits were included (by valuing 1 QALY at £20,000), the value of a definitive trial increased to approximately 8500–16,400 QALYs per year (conditional on the findings of a future trial and their successful implementation).
Conclusions
This work compared the clinical effectiveness and cost-effectiveness of NPWT in healing of SWHSIs, relative to dressings, using advanced methods of adjustment for the potential selection effects expected with observational data. The published paper from this stream of work (see Appendix 5) covers methodological aspects of this research; thus, here we aimed to discuss the clinical implications of our results.
Results show that NPWT users in the cohort took longer to heal and this result (which did not reach conventional levels of statistical significance) sustains even after adjustment for differences in potential prognostic factors between the groups. A number of limitations of the data and methods of analyses were also tested (e.g. alternative assumptions on censoring), with none leading to different conclusions. Across all analyses, treatment with NPWT for SWHSIs was not clinically effective or cost-effective and there was little uncertainty over this result. Our treatment effect findings are derived from modelling of observational data and our results (and our interpretation of these) assume that inferences obtained (after adjustment) are unbiased. However, even extensive adjustment of observational data does not negate the potential for unresolved confounding to affect the results and the risk of this can reduce confidence in conclusions drawn from observational data. Definitive evidence from a RCT may be the only way to overcome this lack of confidence.
It is also important to emphasise that we evaluated the causal effects of NPWT on the healing of SWHSIs. Wound infection at baseline was explored in our analyses and was not found to be associated with either treatment allocation or wound healing. However, NPWT may have further direct effects on HRQoL (independent of healing), for example in prevention of wound infection or assisting with wound management. These other effects are likely to be less significant in magnitude and in duration, as NPWT is typically used for only a short duration of time. If future research further explores the effectiveness of NPWT on infection (prevention or treatment), it is important that this research is designed to also explore effects on healing.
There are other important clinical conclusions to be drawn from our analyses. First, patients treated with NPWT in the cohort study differ from those who were not; that is, there was significant selection into treatment with NPWT in the cohort study. Trying to understand how clinicians allocate treatment with NPWT was, however, difficult. Wound area, level of tissue involvement and being an inpatient (rather than outpatient) seemed to have contributed to decisions on whether or not to use NPWT, but do not fully explain these treatment decisions.
Second, in adjusting for selection effects, it was important to consider relevant predictors of healing. However, given the absence of epidemiological research, there was no a priori evidence on what could predict healing in this patient group. The associations found within our work used a predefined process of model selection and indicated that being an inpatient (rather than outpatient) and having a history of a SWHSI may be relevant prognostic characteristics for healing. However, the sample of patients included in the cohort was very heterogeneous. For example, there was a range of anatomical sites for SWHSIs, with the most frequent being the abdomen (33.6%) and foot and leg wounds (30%); there was an approximately equal share of dehisced versus planned open wounds (60% vs. 40%, respectively) and an approximately equal share of wounds with superficial versus deeper tissue involvement (59% with skin involvement only vs. 41% with skin and subcutaneous tissue involvement). If determinants of healing differed across different subpopulations, then a more homogeneous sample would have meant that it was easier to identify these.
Third, our analyses of the cohort data suggest the existence of a subpopulation of ‘hard-to-heal’ patients: time to healing for those who healed (mean 99 days) was very different from the mean follow-up for those who did not heal (mean 416 days). Those who did not heal within follow-up also differed from those who did and our explorations indicate that BMI and foot wound location are associated with a lower probability of healing. Further research could continue to explore characteristics of ‘hard-to-heal’ patients who present with the biggest burden of disease and thus have a larger capacity to benefit from treatments.
Finally, our research also suggests that individuals with a history of SWHSIs treated with NPWT are expected to take even longer to heal than those who did not have a history of SWHSIs. Although we are not clear of the clinical plausibility of such an effect, future research on NPWT could explore the evidence for such treatment effect interaction.
The findings from this workstream have informed the final phase of work in our research programme: a pilot, feasibility RCT assessing NPWT compared with standard dressings.