SWHSI-2: NPWT versus usual care for open surgical wounds healing by secondary intention — the largest negative trial in this file, in this practice's own population
Negative pressure wound therapy versus usual care in patients with surgical wound healing by secondary intention in the UK (SWHSI-2): an open-label, multicentre, parallel-group, randomised controlled trial — Arundel C, Mandefield L, Fairhurst C, Baird K, Gkekas A, Saramago P, et al · Lancet 2025;405:1689-99
What it found
- Primary outcome — no clear evidence of benefit: HR 1.08 (95% CI 0.88-1.32), p=0.47 (abstract p 1689; table 2 p 1694). Median time to healing 187 days (95% CI 169-226) with NPWT versus 195 days (158-213) with usual care — a difference of 8 days (p 1695).
- The confidence interval's upper bound is the trial's own target effect. The sample size was set "to detect a 25% reduction in median time to healing (from 86 days with usual care to 65 days with NPWT equating to a hazard ratio [HR] of 1·32 and a control group event proportion of 0·95), with 90% power" and 20% attrition, requiring 696 participants, 348 per group (p 1693). The paper says so itself: the HR of 1.08 "narrowly includes our target effect size" (p 1695). This is the strongest statement the result will bear: the trial did not exclude its own pre-specified clinically meaningful effect — it failed to demonstrate it. The authors put the same point in plain terms: the finding "is, however uncertain and NPWT might increase the time to healing by as much as 12% or decrease it by as much as 32%" (p 1695).
- Healing within 12 months occurred in 202 (57.9%) NPWT and 196 (58.2%) usual-care participants; 159 (23.2%) were censored for death, amputation, withdrawal or loss to follow-up (83 [23.8%] NPWT, 76 [22.6%] usual care) (p 1695).
- Masked photographic assessment agreed with the unmasked primary outcome: HR 1.13 (95% CI 0.87-1.47), p=0.36 (table 2 p 1694). Results were also consistent across all sensitivity analyses (competing risks by Fine and Gray, correction for stratification errors, adjustment for the baseline smoking and alcohol imbalance, a Complier Average Causal Effect instrumental-variable analysis) and the pre-specified subgroup analysis by previous SWHSI history (p 1695).
- No significant difference in any clinical secondary outcome (table 2 p 1694, denominators n=320 per arm): hospital admission 63 (19.7%) vs 58 (18.1%), OR 1.13 (0.76-1.69), p=0.54; reoperation 78 (24.4%) vs 69 (21.6%), OR 1.20 (0.82-1.74), p=0.35; amputation 35 (10.9%) vs 36 (11.2%), OR 0.98 (0.60-1.62), p=0.95; wound infection 102 (31.9%) vs 100 (31.2%), OR 1.05 (0.75-1.48), p=0.77; antibiotic use for the SWHSI 211 (65.9%) vs 210 (65.6%), OR 1.01 (0.70-1.45), p=0.96; death 40 (11.5%) vs 43 (12.8%), OR 0.89 (0.56-1.41), p=0.61.
- No significant difference in any patient-reported outcome at any timepoint — Bluebelle Wound Healing Questionnaire at 3, 6 and 12 months (p=0.66, 0.70, 0.22) and visual-analogue wound pain at 3, 6 and 12 months (p=0.85, 0.94, 0.80) (table 2 p 1694).
- Harms were similar. 448 adverse events, of which 14 were serious (nine participants in the NPWT group and five participants in the usual care group); 124 were deemed potentially related to treatment (abstract p 1689). At least one adverse event occurred in 150/349 (43.0%) NPWT and 139/337 (41.2%) usual care, risk difference -1.7% (95% CI -9.1 to 5.7); wound infection was the commonest event in both arms, and major wound infection the commonest serious event. Ten of the 14 serious events (71.4%) were judged unrelated or unlikely to be related to study treatment (table 3 p 1695; p 1696).
- NPWT was not cost-effective. Adjusted incremental QALYs +0.007 (95% CI -0.024 to 0.038) and adjusted incremental cost +£251.44 (95% CI -£2192.63 to £2695.52) over 12 months, neither significant. Incremental net monetary benefit -£93.22 at £20 000 and -£18.65 at £30 000 per QALY; probability of cost-effectiveness 47.2% at £20 000 and 49.7% at £30 000 (p 1696). Unadjusted mean per-participant cost was £5782.13 (95% CI £4151.32-£7412.95) for NPWT and £5801.99 (£3897.19-£7706.78) for usual care. Sensitivity analyses reduced the probability of cost-effectiveness further when participants who died were excluded and when cross-over was accounted for; a complete-case analysis suggested NPWT might be cost-effective but rests on 75.7% missing data and the authors call it likely biased (p 1696).
- The population is close to a South African vascular unit's. 549 of 686 (80.0%) were diabetic — 281 (85.4%) and 268 (84.5%) of those with the field recorded; peripheral vascular disease 181 (55.0%) and 168 (53.0%); wounds arose after vascular surgery in 619 (90.2%), with surgery type recorded as vascular in 314 (90.0%) and 305 (90.5%); wound location foot 279 (79.9%) and 272 (80.7%), leg 40 (11.5%) and 29 (8.6%), abdomen 11 (3.2%) and 13 (3.9%); 620 (90.4%) were on the foot or leg. Median wound area 18.30 cm² (IQR 8.10-35.00) and 18.00 cm² (7.26-33.75); the wound was classified dirty in 190 (54.4%) and 210 (62.3%) and the surgery was an emergency in 205 (58.7%) and 194 (57.6%). Median age 63 years (IQR 55-72), 74.8% male, 91.8% White (table 1 pp 1693-94; results p 1695).
- Funding and interests are clean. NIHR HTA Programme, project reference 17/42/94, and "The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report" (p 1695). Declared interests are minimal: CA reports support for attendance to present the study at Vascular Society of Great Britain and Ireland annual meetings, LM reports independent Data Monitoring and Ethics Committee membership for two unrelated studies (REACH-ASD and ACORN II), and "All other authors declare no competing interests" (p 1698). This matters — the trial cannot be dismissed as an interested party's work, in either direction.
- For context the paper cites the contrary aggregate evidence and explains it: the Cochrane review on this question found only two small trials at risk of bias, and "A separate meta-analysis of 48 low-quality studies at high risk of bias suggested a benefit in healing for SWHSI treated with NPWT (OR 1·56 [95% CI 1·15-2·13]; p=0·008)" (p 1696).
Limitations
- The assumed control-group healing rate was badly wrong, and this weakens the trial's own power. The sample size assumed a control-group event proportion of 0.95 and a usual-care median healing time of 86 days (p 1693). Observed: 58.2% of usual-care participants healed within 12 months and the observed median was 195 days (p 1695) — more than double the assumption, with 23.2% censored. Fewer healing events than planned means less information than 90% power implies, which is part of why the interval still contains HR 1.32.
- The primary outcome was unmasked. This is mitigated, not removed, by the masked photographic secondary analysis — which pointed the same way (HR 1.13) but which the authors say overestimated healing time because it censored wounds the masked assessors deemed unhealed, and should be read with caution (p 1697).
- Substantial crossover in both directions. 42 of 349 NPWT-allocated participants did not receive the intervention, and 45 of 337 usual-care participants received NPWT at some point during the study (figure 1 p 1692). Intention-to-treat with contamination of that size biases toward the null; the CACE analysis was pre-specified to address it and did not change the conclusion (p 1695).
- The exclusions bound the applicability, and two of them matter most. Of 658 ineligible patients, 249 were excluded because they were already receiving or had previously received NPWT on the SWHSI, and 114 because the wound was contraindicated to NPWT; also excluded were 113 whose wound was not considered ready for NPWT, 47 where a vacuum seal could not be obtained, 44 with active systemic infection, 32 malnourished, 25 with unclear undermining, 25 with tissue or eschar present, 15 with delayed primary closure planned, 10 at risk of bleeding, 8 with exposed vessels or organs and 6 with chronic non-surgical wounds (figure 1 p 1692). The 249 are the important number: patients whose clinicians had already committed them to NPWT were systematically removed, so the trial population is by construction the population in whom clinicians were uncertain. It says nothing about the patient in whom NPWT is already running and working.
- Two internal inconsistencies, both in the source. (i) The Results text reports "14 serious adverse events were reported (nine participants in the NPWT group and five participants in the usual care group)" (p 1696), but table 3 gives participants with at least one serious adverse event as 9 (2.6%) and 4 (1.2%) — the 9 and 5 are events, not participants, and the arm-level percentages in table 3 confirm 9 and 5 events. (ii) The marginal citation line printed on the first page reads "Lancet 2024; 405: 1689-99" while the running footer of every page reads "Vol 405 May 10, 2025" and the paper was published online 15 April 2025 — the year in that margin is a typesetting error; 2025 is correct.
- Recruitment ran through the COVID-19 pandemic; study activity was paused by the funder, sponsor and trial management group between March and July 2020, with site capacity effects afterwards, and from March 2020 weekly follow-up calls were made centrally rather than by site research nurses (pp 1691, 1697).
- Patient-reported outcome response rates were lower than anticipated; the WHQ and pain collection timepoints were revised in October 2022 to reduce participant burden, which helped only slightly (pp 1691, 1697). WHQ denominators fell from 195/190 at 3 months to 86/74 at 12 months (table 2 p 1694).
- The economic analysis is UK-specific and the authors say so — "the economic analysis conducted is specific to the UK and so the cost-effectiveness findings might not be generalisable to other countries" (p 1697). It cannot be transplanted to a South African scheme without local costing, in either direction.
- Non-lower-limb recruitment failed, so this is effectively a lower-limb trial despite an all-body protocol (p 1697).
- Gender and race were recorded by the investigator from medical records or discussion with the patient as necessary (p 1691), and the cohort was 91.8% White — a further limit on transfer to a South African population.
Appraisal and reference
CAT: OCEBM 2 · Start High (rct); no downgrade. RoB 2 overall Some concerns, from one domain only (unmasked primary outcome) — and that domain is mitigated by a pre-specified masked photographic assessment that pointed the same way (HR 1.13, 95% CI 0.87-1.47), while unmasking in an open-label trial biases towards the intervention and therefore cannot manufacture a null. Imprecision was considered and NOT taken: the 95% CI 0.88-1.32 narrowly includes the trial's own pre-specified target HR of 1.32, which is a real borderline, but with 686 randomised, 398 healing events, and concordance across every sensitivity, subgroup, competing-risks, CACE and masked analysis, the benefit-of-the-doubt tie-breaker (Dr Weir, 19 Jul 2026) holds this at High rather than Moderate. A null result from a well-conducted, publicly funded trial is not a weak result.
Figures: Figures checked
Arundel C, Mandefield L, Fairhurst C, Baird K, Gkekas A, Saramago P, Chetter I, on behalf of the SWHSI-2 Trial Investigators. Negative pressure wound therapy versus usual care in patients with surgical wound healing by secondary intention in the UK (SWHSI-2): an open-label, multicentre, parallel-group, randomised controlled trial. Lancet. 2025;405:1689-99 (Funded by the NIHR Health Technology Assessment Programme (17/42/94); the funder had no role in design, data collection, analysis, interpretation or writing)