High-risk closed incisions (ciNPT)

Cost-effectiveness (six economic studies, five indications including — new in pub7 — vascular surgery with inguinal incisions): results differ by indication. Moderate-certainty evidence that NPWT in surgery for lower limb fracture was NOT cost-effective at any willingness-to-pay threshold, and that it probably is cost-effective in obese women undergoing caesarean section. Other indications: low or very low certainty.

CAT: OCEBM 1 · The review applies GRADE itself and this appraisal adopts its per-outcome ratings rather than substituting a global one, because that is the whole point of a review that grades each outcome separately. SSI: MODERATE — start High as a systematic review of randomised trials, −1 risk of bias because most included trials had unclear or high risk in at least one key domain; NOT downgraded for inconsistency (I² 29%) or imprecision (RR 0.73, 95% CI 0.63–0.85 on 11,403 participants). Dehiscence: MODERATE for the finding of NO difference (RR 0.97, 0.82–1.16). Mortality, reoperation, seroma, skin blisters, pain: LOW. Haematoma: VERY LOW. The page-level rating of MODERATE reflects the primary outcome. NO upgrade is available and none is warranted: this is an unrestricted, independent, duplicate-assessed Cochrane review, and the ceiling on its certainty is set by the quality of the trials it pools, not by its own conduct.

Figures: Figures checked

• iNPWT reduced SSI versus standard dressing: RR 0.67 (95% CI 0.59–0.76).
• Statistical heterogeneity was low (I² = 21%).
• Certainty of evidence was rated HIGH on GRADE.
• On TSA, the cumulative Z-curve crossed the monitoring boundary for benefit, indicating the effect estimate is unlikely to be overturned by further trials.

CAT: OCEBM 1 · Start High (SR of RCTs) and NO downgrade: I²=21%, RR 0.67 (0.59–0.76) on 13,744 patients, and trial sequential analysis shows the cumulative Z-curve crossed the monitoring boundary for benefit — further trials are unlikely to overturn it. Agrees with the review's own GRADE of HIGH. Risk of bias was NOT downgraded here even though the dressing is unblindable, because with 57 trials, low heterogeneity and TSA confirmation, bias is not a plausible explanation for the effect; that reasoning does not transfer to the small 5-trial pools where this appraisal did downgrade. Applied to a VASCULAR GROIN motivation a −1 indirectness applies (57 trials across many specialties), giving Moderate.

Figures: Figures checked

Pooled incidence

• SSI with pNPWT 13.2% (95% CI 9.5–18.1, I² 75%; 1,185 patients).
• SSI with conventional dressing 26.1% (95% CI 22.1–30.5, I² 70%; 1,745 patients).

Whole body of evidence (21 studies, 2,930 patients)

• Risk difference −12% (95% CI −17% to −8%), I² 57%, p < 0.00001.
• Risk ratio 0.53 (95% CI 0.40–0.71), I² 56%, p < 0.0001.
• NNT 9 (6–13) to prevent one SSI.

Where the effect held and where it did not — the part that matters

• Restricting to the 5 RCTs (792 patients): RD −12% (−22% to −1%), p = 0.03 — significant. But RR 0.56 (0.30–1.03), p = 0.06 — not significant.
• High-quality RCTs (4 studies, 527 patients): RD −16% (−27% to −4%), p = 0.008; RR 0.40 (0.15–1.02), p = 0.05 — not significant. NNT 7 (4–25).
• High-quality observational studies (642 patients): RD −16% (−27% to −5%), p = 0.004; RR 0.50 (0.25–1.00), p = 0.05 — not significant. NNT 7 (4–20).
• Baseline-risk subgroups: low-risk RD +3% (−9% to +18%), p = 0.23 — no benefit, direction reversed. Medium-risk RD −17% (−21% to −12%). High-risk RD −19% (−29% to −9%). RR was significantly lower in high-risk than lower-risk pools (p = 0.049); the effect was more pronounced wherever control-arm SSI reached ≥20%.
• By procedure (RD, RR): colorectal −16%, 0.35 (0.16–0.76); abdominal wall reconstruction −11%, 0.65 (0.43–0.98); pancreaticoduodenectomy −15%, 0.42 (0.28–0.65). No statistical difference between procedures.
• By device (RD): PREVENA −12%, PICO −30%, VAC/VSD −12%; device difference significant for RD (p = 0.025), driven by PICO — on 149 patients in 3 studies.

Publication bias Funnel plots were asymmetric; trim-and-fill identified missing studies favouring control. Corrected estimates: RR 0.72 (0.53–0.96), p = 0.026; RD −10.1% (−14.6% to −5.7%), p < 0.001 — attenuated but still significant across all studies and in the cohort studies, not in the RCTs. The authors attribute part of the asymmetry to the large-effect PICO studies rather than to suppression alone.

CAT: OCEBM 2 · The pooled body is dominated by non-randomised studies (16 of 21), so the whole-body estimate starts LOW and stays LOW: no further downgrade because the direction is consistent across every procedure and device subgroup, the risk difference survives restriction to high-quality RCTs and to high-quality observational studies, and it survives trim-and-fill correction (RD −10.1%, −14.6 to −5.7). Assessed instead on the randomised subset alone, GRADE starts HIGH and takes −1 for imprecision (RR 0.56, 0.30–1.03, crosses 1) and −1 for inconsistency (I² 67%) → LOW, with a further concern that the dressing cannot be blinded and one of the five RCTs was rated medium quality. Both routes land on LOW, which is why the page is rated LOW rather than split. The finding that carries the HIGHEST certainty here is not the average effect but the interaction: benefit is absent at low baseline risk (RD +3%, −9 to +18) and present once control-arm SSI reaches 20% (RD −17% to −19%, device- and procedure-independent, p=0.049 across risk strata).

Figures: Figures checked

The single most important thing about this paper, for this wiki, is what is not in it. The word "Szilagyi" does not appear anywhere in Labib 2026 — not in the Methods, not in any forest plot, not in the GRADE table, not in the Discussion. There is no deep-versus-superficial split of any kind: no CDC superficial/deep/organ-space stratification either. The outcome is surgical site infection as each included trial defined it, pooled undivided.

So Labib does not disagree with Gombert. It is silent. The wiki's groin depth limit — that the demonstrated prophylactic benefit is superficial, on Gombert 2020's Szilagyi II p = 0.46 and Szilagyi III p = 0.07 — is not challenged by this paper and stands unchanged. An updated meta-analysis that adds four trials and doubles the sample would have been the natural place to resolve the depth question with more power; it was not attempted, and that absence should be recorded rather than glossed. Nothing here permits a claim that closed-incision NPWT prevents deep or graft-involving groin infection.

What it does report. Ten randomised trials, 1,449 patients, 1,681 inguinal incisions.

Every arm total and event total reconciles. The twelve forest-plot rows sum to 826 and 855 incisions and to 89 and 179 events, matching the printed totals and the printed percentages exactly. The subgroup plots reconcile too: diabetes 78 and 65, 12 and 22 events; overweight 116 and 113, 19 and 43 events. The forest plots carry no text layer in the published PDF and were recovered by optical character recognition of the figure images, then re-added by hand.

…

CAT: OCEBM 1 · START HIGH — a systematic review of 10 randomised trials in exactly the population this wiki is asked about, inguinal arterial surgery. For the overall SSI estimate: −1 risk of bias. The review's own RoB 2 grades two of ten trials at high risk and two with moderate concerns; four of the ten are open-label trials whose outcome is a clinician judgement of infection; and the pooled analysis is at INCISION level while randomisation in three of the trials was WITHIN PATIENT, so 1,681 incisions from 1,449 patients are treated as independent and the confidence interval is narrower than the data license. The authors concede the unit-of-analysis issue and state that a sensitivity analysis excluding clustered data did not change the result, but that analysis is not shown. NO downgrade for inconsistency (I² = 23%), indirectness (the population is the one being asked about) or imprecision (RR 0.53, 95% CI 0.40 to 0.71, 268 events). NET MODERATE — which is also the review's own grade, and the wiki agrees with it. For the two subgroups the wiki is more severe than the paper. Both rest on three trials only, and in both, ONE TRIAL — Gombert 2018 AIMS — carries 45.9% of the diabetic weight and 55.8% of the overweight weight. The diabetic subgroup has 143 patients and 34 events; the overweight subgroup 229 patients and 62 events. In the diabetic subgroup the AIMS arms are 42 against 22, a two-to-one imbalance inherited from a baseline diabetes prevalence of 43% against 24% in the parent trial — so the largest contributor to the diabetic estimate is a post-randomisation subset that is itself unbalanced. The overweight subgroup is not one population but three definitions pooled: Engelhardt contributed patients with BMI ≥ 30, the other two BMI ≥ 25, and the paper says so. −1 risk of bias, −1 imprecision. Net low for both, where the paper grades diabetes Moderate and overweight Low. The direction is not in doubt and should be said plainly: every one of the twelve overall forest-plot rows favours ciNPWT or is null, none favours the standard dressing, and the estimate sits on top of three independent predecessors. What is uncertain is the SIZE of the subgroup effects, not their existence. What this review cannot be graded on at all is depth: it reports no Szilagyi stratification and no deep-versus-superficial split of any kind.

Figures: Figures checked

Cost-utility analysis (the analysis the authors consider primary):

Cost-effectiveness analysis (cost per SSI avoided):

At a willingness-to-pay threshold of AUD$50,000 per QALY there is a 20% probability that ci-NPWT is cost-effective, and the authors state it remains unlikely to be cost-effective at thresholds up to at least AUD$150,000 per QALY. In the base case and the per-protocol analysis the standard dressing dominates: cheaper and, negligibly, better.

The abstract's conclusion, in full:

"ci-NPWT for the prevention of SSI in obese women undergoing CS is unlikely to be cost-effective in terms of health service resources and is currently unjustified for routine use for this purpose."

And the closing sentence of the paper:

"the results of this economic analysis assessed alongside the clinical outcomes from the large randomised controlled DRESSING trial, suggest that using ci-NPWT to prevent SSI in this population is unlikely to be cost-effective when compared to standard care, and should not be routinely implemented."

The authors also state directly which prior findings they overturn:

"Our findings based on an economic evaluation alongside the DRESSING trial are thus important, as they contradict the findings by Hyldig and colleagues, and Tuffaha and colleagues, and cast doubt on the value provided by routine prophylactic ci-NPWT for obese women following CS."

CAT: OCEBM 2 · The underlying clinical trial is OCEBM 1 - large, pragmatic, registered, multicentre, intention-to-treat. The ECONOMIC conclusion is graded one step lower for imprecision that the authors themselves foreground: the base-case incremental cost is AUD$310 with a 95% CI of minus $530 to $1,185, the incremental QALY is minus 0.0004 (minus 0.0032 to 0.0015), and both cross zero. What survives that imprecision is the decision-relevant quantity: At a willingness-to-pay threshold of AUD$50,000 per QALY there is a 20% probability that ci-NPWT IS COST-EFFECTIVE, and the authors state it stays unlikely up to at least $150,000 per QALY. In the base case and per-protocol analyses the standard dressing dominates - cheaper, and negligibly better. The result is stable across all three analysis populations, which is the strongest internal evidence that it is not an artefact of missing data or protocol deviation. No downgrade for indirectness of the economic question but a hard limit on its scope: this is obese women after caesarean section under Australian unit costs, not the vascular groin, not the sternum, not a contaminated laparotomy, and not South Africa. The most consequential feature for this library is that it reverses a standing position. Cochrane pub7 rates ci-NPWT probably cost-effective in obese women undergoing caesarean section at Moderate certainty on two studies - an analysis alongside an 87-patient pilot, and the Danish Hyldig evaluation which extrapolated a non-significant 30-day EQ-5D-5L difference across a full year. Whitty is larger than both together, uses within-trial data over its whole horizon, and names both as the findings it contradicts.

Figures: Figures checked