How we appraise
Every source is read in full where the full text can be obtained, critically appraised with the instrument its design calls for, and filed under each wound type it bears on. Sources that do not support NPWT are kept and shown: an evidence library is most useful where it says what the evidence does not show.
Three different ratings
A card or a page can carry up to three ratings. They answer different questions and are never interchangeable.
| Rating | Question it answers | Scale |
|---|---|---|
| GRADE ⊕⊕⊕◯ Moderate | Evidence certainty. How sure can we be of this study's effect estimate? | GRADE: High ⊕⊕⊕⊕, Moderate ⊕⊕⊕◯, Low ⊕⊕◯◯, Very low ⊕◯◯◯. Where GRADE cannot apply, the OCEBM level of evidence or CERQual confidence is shown instead, or "Not graded" with the reason. |
| Guideline quality 4/7 | Guideline quality. How well was this guideline or consensus developed? | AGREE II overall score out of 7, from this library's own appraisal. |
| Evidence position: Moderate Evidence position: Guarded Evidence position: Not supported | Where the evidence stands. What does the current international evidence show for NPWT in this wound type? | Decided by the appraisal alone, irrespective of the position of any guideline, society or country, and of this library's own earlier positions. Only direct evidence counts: a source at GRADE High or Moderate certainty is counted for a wound type only if the patients studied had that wound type (a trial of preventing groin infection after vascular surgery is not evidence for treating an infected graft). High counts twice, Moderate once; papers reporting one trial count once. Moderate: direct evidence for NPWT outweighs direct evidence against it. Not supported: direct High or Moderate evidence against NPWT outweighs the evidence for it. Guarded: everything else, including wound types with no direct High or Moderate evidence either way. Moderate is the highest level used. Guidelines, including national ones, are appraised as sources and never set the position. It is not a guideline and not a GRADE rating. Each wound-type page lists the sources counted. |
Other labels on a card
| Study type | The design: randomised trial, systematic review, cohort, case series, guideline, and so on. |
| Direction | Supports NPWT, Challenges NPWT or Context, set by one mechanical rule applied to every source (4 October 2026). Supports: NPWT is compared with a treatment that is not NPWT and the primary outcome (or the pooled estimate) is significantly better with NPWT; a guideline or payer that recommends it; a cost analysis that finds it saves money. Challenges: the same comparison is null or worse with NPWT; a review, guideline or payer that finds no benefit; a cost analysis that finds it does not pay. Context: no comparison (case reports, case series, single-arm studies), a comparison of two NPWT systems, settings or modes, laboratory, animal and volunteer work, narrative reviews, surveys, technique papers and manufacturer material. Instillation is judged against standard NPWT. Abstract-only, flagged and disputed sources are shown as Context and not counted until the full text is read or the dispute is resolved. Not NPWT evidence marks a source with no negative pressure in it, held only to describe the comparator, the baseline infection rate or the cost an NPWT argument is measured against. |
| Figures | A separate check that the numbers quoted on the card appear in the source. It says nothing about the quality of the study: a paper can have its figures checked and still be GRADE Very low. Figures checked: the numbers on the card were compared line by line with the full text. Like the appraisals, these checks are carried out by Anthropic's Claude; they are not independent double extraction. (auto): an automated screen found them in the source PDF. Abstract only: appraised from the abstract; the full text has not been obtained. Figures unconfirmed or Flagged: not confirmed against a full text. Figures don't reconcile: the full text was read and the paper's own numbers contradict each other. |
| Year | The year of first publication (online-first where applicable); the full reference gives the issue year. |
| Reviewed | The date the appraisal was last reviewed. |
How a source is handled
Source identified → full text obtained → study design classified → critically appraised → numbers extracted and checked against the source → retraction status checked in PubMed → filed under each relevant wound type → cited on the wound-type page if it adds credible evidence.
Manufacturer-sponsored material is labelled as such and ranked lowest. Papers that are flagged, retracted or contain no data are listed but never counted as evidence.
Summary of findings and recomputed statistics
Each wound-type page carries a summary-of-findings table: for each outcome that matters to a patient, the number of trials and participants, the effect with its 95% confidence interval, the risk without and with NPWT, and the certainty of the evidence (GRADE). The table holds no numbers of its own. Every effect is read from the source's figures as stored on its appraisal, so a correction to a source changes the table on the next build.
Recomputed statistics. Where a paper prints its counts or means, this library stores them and recomputes the paper's effect and p value on every build, using standard methods: risk ratios with log intervals, Fisher's exact and chi-square tests, Welch and Student t tests. A published pooled estimate is checked against its own interval, and its p value against the p value the interval implies. A printed figure that does not reproduce is shown on the study card under "Statistics" and marked ⚠ in the table. It is a finding about the paper, recorded on its appraisal; the paper's numbers are never silently corrected.
Re-pooling held trials. Where this library holds several of the trials behind a review, it re-pools them (random effects) as a check, and says how many of the review's trials that covers. A difference is flagged only when at least half of the review's trials are held, because a re-pool of a few trials can differ from the review by chance.
Absolute risks ("without → with NPWT") use the control-group risk in the trial, the review, or the held trials, applied to the relative effect. Where the outcome is a time or a score, the column is left blank.
Direction of effects. Differences are stated as NPWT minus control. A hazard ratio for healing above 1, or a time ratio below 1, means faster healing with NPWT.
Harms register
Each wound-type page lists the harms reported by its sources in one table: the event, how often it happened, and the source. The events are death, bleeding, fistula, infection, ischaemia or tissue damage, skin injury, pain, retained foam or dressing, device problems, pressures outside the usual range, and other adverse events. Counts are entered as each source prints them; percentages and risk ratios are computed from those counts, and a printed effect or p value is re-tested in the same way as the summary of findings. Where two NPWT devices or two NPWT regimens are compared, the row says which arm is in the NPWT column.
Reading it. A randomised comparison or a review gives a rate against a comparator. A case series gives a rate with no comparator, and a report to a regulator gives a count with no denominator: a signal that a harm happens, not how often. Rows marked Serious are harms that threaten life or limb. Retracted and integrity-flagged sources are never entered. A pressure row records a setting outside the conventional −50 to −125 mmHg in use, or a comparison of pressures, with what it was associated with.
Borderline calls
Where a GRADE or AGREE II judgement falls genuinely between two levels, the higher level is given and the reason is stated in the appraisal (the "benefit-of-the-doubt tie-breaker"). The rule is the same whichever way the source points.
Retracted papers
Listed so they can be recognised; excluded from every count and hidden in the evidence lists unless "Show retracted papers" is ticked.
- Yuan S, Zhang T, Zhang D, He Q, Du M, Zeng F. [RETRACTED] Impact of negative pressure wound treatment on incidence of surgical site infection in varied orthopedic surgeries: a systematic review and meta-analysis. Int Wound J. 2023;20(6):2334-2345 (RETRACTED 2025 — retraction notice Int Wound J 2025;22(4):e70561, doi 10.1111/iwj.70561, PMID 40251145)
- Wang Z, Feng C, Wang XJ. [RETRACTED] Negative pressure wound therapy for patients with mediastinitis: a meta-analysis. Int Wound J. 2020;17(6):2019-2025 (RETRACTED April 2025 — retraction notice Int Wound J 2025;22(4):e70646, doi 10.1111/iwj.70646, PMID 40265215, PMC12015185. No funding statement and no conflict-of-interest statement appear in the original article.)
- Ali Z, Anjum A, Khurshid L, Ahad H, Maajid S, Dhar SA. [RETRACTED] Evaluation of low-cost custom made VAC therapy compared with conventional wound dressings in the treatment of non-healing lower limb ulcers in lower socio-economic group patients of Kashmir valley. J Orthop Surg Res. 2015;10:183 (RETRACTED 2017 — retraction notice J Orthop Surg Res 2017;12(1):62, doi 10.1186/s13018-017-0558-3, PMID 28427404, PMCID PMC5397907)
Changes
- 4 October 2026. The library moved from one long page to separate pages for each wound type, with the evidence in one minute, plain-language filters and downloadable reference lists. The "RSA clinical position" badge was renamed.
- 4 October 2026, after an adversarial audit. Every source's direction (Supports, Challenges, Context) was reset by one mechanical rule, which changed 274 of 638 labels. The position badge now counts only direct evidence and gained a third level, Not supported: open fractures, wounds in peripheral arterial disease and contaminated or dehisced surgical wounds became Not supported; vascular graft infection and sternal wounds moved from Moderate to Guarded. The same day Dr Weir set the rule that the evidence decides in both directions, overriding earlier positions: burns and NPWT with instillation, previously held at Guarded, became Moderate. The badge was then renamed the evidence position: it reflects the current international evidence, not South African or any other guidance, and Dr Weir takes no part in appraisal or rating. Each wound-type page now shows how the evidence points, by certainty.
- 27 September 2026. The position badge was made to follow one stated rule.
- 15 September 2026. Eight wound types previously marked "strong" on the strength of WHASA 2021 were reduced to Moderate, because this library appraises WHASA 2021 at AGREE II 4/7 (rigour of development 23%, editorial independence 25%). No evidence rating on any source changed.
About this library
Maintained by Dr Gregory Weir, vascular surgeon, Pretoria. Who decides what: Dr Weir adds sources to the library. The appraisals, figure checks, direction labels, certainty ratings and evidence positions are made by Anthropic's Claude under the rules on this page, without input from Dr Weir, so that the maintainer's own views and work cannot shape them.
Funding and conflicts of interest: the library receives no external or industry funding. Dr Weir has no financial ties to any NPWT manufacturer, supplier, rental company or distributor, and Dr Weir's practice does not bill for NPWT. Dr Weir is a co-author of the WHASA 2021 NPWT recommendations. That document stands on its own: this library appraises it as one guideline among others (AGREE II 4/7), does not speak for it, and does not use it to set any rating. Two included sources are Dr Weir's own work (Weir 2014, a groin pilot; Smart 2024, a Delphi consensus), and with WHASA 2021 their cards say so.
Privacy
The library collects no personal information through its pages. The site counts visits with Google Analytics and is served through Cloudflare; both process technical data such as IP address and browser type, and Google Analytics sets cookies. Nothing a visitor types into a search or filter is sent anywhere: filtering runs in the browser. Questions about personal information under the Protection of Personal Information Act: npwt@wondsorg.co.za.
Device names such as V.A.C., VeraFlo, PICO, Prevena, RENASYS and ABThera are trademarks of their owners and are used only to identify the devices studied. This library has no connection with any of them.
Corrections and suggestions are welcome: please email npwt@wondsorg.co.za.