Wall-suction and improvised NPWT

• Time to 90% granulation: 18.8 ± 6 days with NPWT versus 32.3 ± 14 days with conventional dressing, p = 0.007.
• The authors state this as a 40% reduction in granulation time in the amputated diabetic foot.
• Hospital mortality 0 in both arms (peinemann 2011 iqwig npwt sr rcts, Table 4).

CAT: OCEBM 2 · Start HIGH (randomised trial). NO downgrade for risk of bias — and this is the only trial in the entire 21-trial pool of peinemann 2011 iqwig npwt sr rcts for which that can be said. IQWiG applied a five-criterion instrument requiring ALL FIVE to be met for a low grading — adequate sequence generation, concealed allocation, blinded endpoint assessment, reasons given for data loss, adequate intention-to-treat analysis — and Sepulveda met all five. Blinding was achievable here precisely BECAUSE the endpoint is a granulation percentage, which an assessor can score without knowing the allocation even though neither patient nor clinician can be blinded to the therapy. −1 imprecision: 24 patients, 12 per arm, and no secondary clinical outcome is assessable at that size. −1 indirectness: the primary outcome is time to 90% granulation, a wound-bed-preparation surrogate. It is not healing, not limb salvage, not discharge, and not amputation avoided — and the gap between a granulating bed and a closed limb is exactly where a motivation has to argue. Net low. The rating is the highest in its pool and still only low, which is the point: the best-conducted randomised trial in this literature reaches Low certainty on a surrogate endpoint in 24 patients. That is the state of the field, not a criticism of these investigators.

Figures: Figures checked

• Wound surface area fell 4.5%/day with GSUC and 4.9%/day with VAC (p < 0.001 for each within-group change); the between-group difference was not significant, p = 0.60, estimated VAC − GSUC 0.4% (95% CI −1.0 to 1.7).
• Wound volume fell 8.4%/day with GSUC and 9.8%/day with VAC; between-group p = 0.19, estimated difference 1.4% (95% CI −0.7 to 3.5).
• For both outcomes the 95% CI excluded the prespecified 5%/day margin, so GSUC was declared non-inferior to VAC.
• Cost of supplies: US$4.22 per day for GSUC versus US$96.51 per day for VAC, p < 0.01.
• Time per dressing change: 19 minutes GSUC versus 31 minutes VAC, p < 0.01. Labour cost per change was $25.16 versus $40.70, but because GSUC dressings were changed more often, daily labour cost did not differ ($21.18 versus $25.55, p = 0.11).
• Pain favoured GSUC: SPID 0.50 (95% CI −0.11 to 1.11) versus 1.73 (0.91 to 2.54), p = 0.02, with no difference in morphine-equivalent use (p = 0.63).
• Failures ran one way only. No patient discontinued GSUC. VAC was discontinued in 4 patients — three for a seal that could not be maintained (infected perineal wound, infected abdominal/flank wound with excessive drainage, upper-limb fasciotomy with excessive movement), all three successfully crossed over to GSUC, and one for a haematoma under the dressing in a fully heparinised patient after compartment release, who did not cross over.
• Patients in the GSUC arm underwent more dressing changes by design: mean 4.5 (range 2–15) versus 2.8 (2–6), p < 0.01.

CAT: OCEBM 2 · Start HIGH (randomised trial). −1 risk of bias: neither patients nor therapists could be blinded, which is unavoidable with a vacuum dressing, and — the part that is NOT unavoidable — wound area and volume were measured by the unblinded treating service using a geometric approximation (Xakellis and Frantz, area = length × width × 0.783) rather than planimetry or a blinded assessor. Against that, the trial's conduct is otherwise the best in this topic: registered, CONSORT-designed, prespecified margin, computer-generated sequence, complete follow-up, no protocol deviations, a genuine intention-to-treat analysis, and a linear mixed model that correctly handles multiple wounds per patient. −1 indirectness: the endpoints are rates of change in wound area and volume over the first seven days, not healing, closure, infection or limb salvage, and the analysis was truncated at day 7 post hoc because most patients had been discharged or closed operatively. NO downgrade for imprecision — the trial met its own power calculation and both confidence intervals excluded the prespecified margin. NO downgrade for inconsistency (single trial) or publication bias (single trial, no declared funding or interest). Net low. What the rating is about: this is a Low-certainty demonstration that two ways of applying negative pressure shrink an acute wound at indistinguishable rates over a week. It is not, and does not claim to be, evidence that either heals wounds. The comparator is what makes it valuable — this is the first randomised head-to-head of an improvised circuit against the marketed device held in this library, and it closes a gap that topics/improvised-npwt had recorded as open.

Figures: Figures checked

• Full (100%) graft take: 64 of 80 wounds with GSUC versus 60 of 77 with VAC, P = .80.
• Overall mean take 96.12% with GSUC versus 96.21% with VAC, P = .98.
• Of the wounds without complete take, mean non-take was 19.4% (GSUC) versus 17.1% (VAC), P = .89.
• Total graft loss in 4 wounds — 3 GSUC, 1 VAC — all four requiring reoperation, and in all four the underlying bed was found non-viable once the dressing came off.
• Three study failures, all in the VAC arm, all from loss of suction; all three were salvaged by conversion to wall suction. There were no study failures in the GSUC arm.
• The only statistically significant difference between the arms was duration of therapy: 4.58 versus 4.11 days (P < .05) — GSUC wounds received about half a day more.
• Costs are quoted in the discussion rather than analysed: about US$96/day to rent the VAC unit versus about US$4/day for GSUC supplies.
• Baseline imbalance worth noting: mean grafted wound size 165.8 cm² (GSUC) versus 104.9 cm² (VAC) — a 58% difference not tested for significance and not adjusted for, which cuts against GSUC rather than for it.

CAT: OCEBM 2 · Start HIGH (randomised trial). −1 risk of bias: the unit of randomisation was the WOUND, not the patient — 25 of 104 patients contributed more than one wound, up to more than three — and the analysis used independent-sample t-tests and chi-square on wounds with no adjustment for clustering, so the effective sample is smaller than 157 and the reported p-values are anticonservative. The same unit handled exactly this problem correctly three years earlier with a linear mixed model in dorafshar 2012 gsuc vs vac rct, which is why this counts as a defect rather than a limitation of the era. Added to that: blinding ended at allocation (the surgeon was blinded until the dressing was placed; the assessing therapist placed it and could not be), and mean grafted wound area differed by 58% between arms (165.8 vs 104.9 cm2) with no test and no adjustment. −1 imprecision AND indirectness taken together: No power calculation and no non-inferiority margin were set, so a p of .98 is an absence of evidence of difference rather than a demonstrated equivalence; and the endpoint is graft take by gross measurement at day 4-5, an early surrogate, with no 2-week or later healing outcome reported. NO downgrade for inconsistency (single trial). NO downgrade for publication bias, though no conflict-of-interest declaration is printed. Net low. The critical reading: the direction of the two unadjusted defects is INSTRUCTIVE. Both the clustering and the 58% larger mean wound area in the gauze-suction arm would, if corrected, tend to work AGAINST the gauze-suction arm, not for it — so the null result is unlikely to be an artefact favouring the cheap circuit. That is the strongest thing that can be said for this trial, and it is worth saying.

Figures: Figures checked

• Time to complete healing: 16 days with HM-VAC versus 25 days with saline gauze, P = 0.013.
• COST: HM-VAC US$360 per case versus WET US$271 per case, P = 0.008. The improvised vacuum system was significantly MORE expensive than gauze.
• A harm in the opposite direction, recorded in peinemann 2011 iqwig npwt sr rcts's Table 4: secondary surgery rate 7 versus 4, P = 0.038, favouring the comparator. That is one of only two results anywhere in Peinemann's nine new trials that runs against NPWT.
• Complete wound closure at 30 days after closure or skin transplantation: 18 (90%) versus 19 (95%), P = 0.302 — no difference on the primary endpoint of the systematic review that included it.

CAT: OCEBM 2 · Start HIGH — individually randomised parallel-group trial with a stated sample-size calculation. MINUS 1 for risk of bias: assignment by sealed envelope with no statement that the envelopes were opaque or sequentially numbered; No intention-to-treat analysis, with 9 of 49 randomised patients — 18 per cent — dropped for death or unknown cause and the survivors landing at a suspiciously tidy 20 against 20; and the authors state in terms that the wound healing results were judged by an unblinded observer. Minus 1 more for a confounder the paper denies: Antibiotic use ran 7 of 20 in the vacuum arm against 16 of 20 in the gauze arm, p = 0.005, printed in the trial's own baseline table, while the text asserts that the clinical and demographic characteristics were comparable in both groups. In a trial of infected and chronic wounds whose endpoint is time to healing, a more than two-fold imbalance in systemic antibiotics is not a minor baseline wobble and the assertion of comparability is contradicted by the table above it. MINUS 1 for very serious imprecision and analytical validity: 40 analysed, and the primary analysis is a paired t test applied to two independent parallel groups, stated in the statistics section and repeated in the footnotes of Tables 2, 3 and 4. Net VERY LOW. The direction of the healing result is plausible; every printed p-value in this trial rests on a test that does not fit the design.

Figures: Figures checked

peinemann 2011 iqwig npwt sr rcts extracts the following from this trial, and both entries carry caveats from the reviewers rather than the authors:

• Complete wound closure: 7 (48%) versus 16 (48%) — no difference. IQWiG notes the percentages relate to the number of patients treated, not the number randomised.
• Days to complete wound closure: 36 (SD 45) versus 28 (SD 19), P = 0.66 — numerically favouring the comparator, not significant. The standard deviation exceeding the mean indicates a highly skewed distribution.
• Randomised 19 versus 36. **CORRECTED 2026-08-02: this is a pre-specified 1:2 allocation ratio,

not an unexplained imbalance.** The page previously read it as a randomisation failure; appraising the trial for its CAT showed the ratio was intended. What remains a real problem is the dropout —, with dropouts of 5 versus 15 and 4 versus 3 by intention-to-treat.

• IQWiG graded it high potential for bias and noted few patient characteristics reported.

CAT: OCEBM 2 · Start HIGH — individually randomised trial with concealed block randomisation and an a priori sample-size calculation. MINUS 1 for risk of bias: the outcome that matters, days to satisfactory healing, was judged by the study investigator and the treating surgeon and was not blinded. The word blinded in the title refers to the pre-randomisation measurement of wound size from photographs, not to outcome assessment. MINUS 1 for very serious indirectness and inconsistency of the delivered intervention: Eight wounds began in the TNP arm and were switched to conventional dressings during follow-up, treatment lasted a mean of only 11.3 days against a mean 33.1 days of observation, and patients were moved off TNP when they no longer needed or could no longer afford inpatient care. More than half the treatment arm therefore spent most of its follow-up on the comparator. MINUS 1 for very serious imprecision: 15 against 33 patients across five heterogeneous wound types, 23 healing events, no confidence interval anywhere, and a one-sided t test. Net very low. The one positive result in the abstract — pressure ulcers healing in 10 days against 27.4 days, P = 0.05 — Is a subgroup of two TNP wounds and eight control wounds, a fact printed only in the limitations paragraph. It carries no certainty rating and should never be quoted as a result.

Figures: Figures checked