Unverifiable numerical reporting and eligibility judgements in a meta-analysis of incisional negative pressure wound therapy in breast surgery

Correspondence to: Liwen Liu（email: liwenlmd01@163.com）

Department of Anesthesia,

The Reproductive Hospital of Guangxi Zhuang Autonomous Region

Nanning City

China

_____

BJS, https://doi.org/10.1093/bjs/znag056, published 05 May 2026

_____

We read with interest the meta-analysis by Kollatos et al.1 on incisional negative pressure wound therapy (iNPWT) in breast surgery. The authors concluded that current evidence does not establish a consistent clinical benefit and that routine iNPWT use is not justified. Our concern is not the acknowledged heterogeneity or predominance of observational studies, but several specific numerical inconsistencies that make parts of the analysis difficult to verify.

First, the study-selection flow is not arithmetically reconcilable. The PRISMA diagram reports 83 reports sought for retrieval and 0 reports not retrieved, but only 56 reports assessed for eligibility. The stated exclusion reasons then sum to 27 reports, yielding the final 29 included studies. Thus, 27 reports appear to disappear between retrieval and eligibility assessment, despite no report being unavailable. Because PRISMA 2020 requires transparent accounting of records and reports through each review stage2, this should be corrected or explained.

Second, the eligibility classification is internally inconsistent. The Methods state that eligible designs were randomized clinical trials (RCTs), prospective cohort studies, or retrospective cohort studies. However, the Results classify the included studies as five RCTs, eight prospective non-randomized studies, 13 retrospective studies, and one propensity-score-matched retrospective study, which totals 27 rather than 29 studies. Two included reports are described as having unreported study design. If their design was not reported, it is unclear how they satisfied the prespecified study-design criterion. The authors should clarify the eligibility basis for these two reports and provide a sensitivity analysis excluding studies whose design could not be classified.

Third, the haematoma analysis contains incompatible statistical results. The main article reports RR 1.017, 95% c.i. 1.002 to 1.032, P=0.545. For the absolute risk difference, it reports ARD 0.018, 95% c.i. 0.005 to 0.032, P=0.311. If these confidence intervals and P values come from the same models and null hypotheses, confidence intervals excluding the null are not compatible with highly non-significant P values. The supplementary material instead shows a haematoma estimate of approximately RR 1.02, 95% c.i. 0.99 to 1.04, P=0.13, and the certainty assessment states that the confidence interval “crosses null”. This contradicts the main article, where the RR confidence interval does not cross 1 and the ARD confidence interval does not cross 0. The reported haematoma leave-one-out range, RR 1.062–1.082, is also implausible relative to the pooled estimate of RR 1.017 and is identical to the leave-one-out range reported for wound dehiscence, suggesting a possible copy or statistical-output error.

Fourth, skin necrosis also requires clarification. The main article reports RR 1.073, 95% c.i. 1.005 to 1.147, P=0.001, whereas the corresponding supplementary forest plot presents P=0.05 for the same pooled effect. This difference is important because skin necrosis is one of the outcomes interpreted as showing benefit. If the correct P value is borderline rather than 0.001, the strength of statistical evidence and certainty interpretation should be revised.

Finally, the risk-of-bias assessment appears inconsistent. For the Galiano randomized trial, two domains for the wound dehiscence outcome are rated high risk of bias, yet the overall judgement is recorded as some concerns. Under the RoB 2 framework, a high-risk judgement in an important domain generally leads to an overall high-risk judgement unless a justified override is provided3. This matters because wound dehiscence was the primary outcome, and the certainty assessment treats the randomized evidence as not seriously limited by risk of bias.

We therefore ask the authors to provide a corrected study-selection flow, a complete classification of all 29 study designs, the eligibility rationale for studies with unreported design, corrected statistical outputs for haematoma and skin necrosis, and an updated risk-of-bias and GRADE assessment. If these corrections do not alter the conclusions, the revised analyses would strengthen confidence in the review. If they change effect estimates, statistical significance, or certainty ratings, the conclusions regarding the clinical benefit and selective use of iNPWT should be updated accordingly.