Sternal wounds

• Wound infection occurred in 3/75 (4%) NPWT patients versus 12/75 (16%) controls (P=.0266; OR 4.57; 95% CI 1.23-16.94) (p.1387).
• Infection by Gram-positive skin flora occurred in only 1 NPWT patient versus 10 controls (P=.0090; OR 11.39; 95% CI 1.42-91.36) (p.1387).
• When the foam dressing was removed at day 6-7, the incision was primarily closed in 71/75 (95%); no infection occurred thereafter, supporting closure as a barrier to skin flora (p.1390).
• 9/12 control infections occurred beyond the first postoperative week (up to day 35), indicating the sterile dressing established an adequate barrier considerably later (p.1391).
• All 3 control sternal osteomyelitis cases were skin-flora deep infections potentially preventable by prophylactic NPWT (p.1392).

CAT: OCEBM 3 · This is not a randomised trial and it should never be described as one in a motivation. Patients were allocated by ALTERNATION according to the time of operation, with diabetics deliberately split half and half between the arms 'with priority'. Alternation is fully predictable — the surgeon knows before the case which dressing the next patient will get — and the diabetic quota is an admission that the sequence could be steered. The effect is nevertheless large and the arithmetic is clean. Wound infection 3 of 75 (4.0%) with NPWT against 12 of 75 (16.0%) with conventional dressing. Recomputed here: relative risk 0.25, absolute risk reduction 12.0% (95% CI 2.6% to 21.4%), number needed to treat 8.3 (4.7 to 38.6). Every printed statistic reproduces exactly. Fisher exact on 3 of 75 against 12 of 75 returns p = 0.0266, matching the printed P = .0266 to four places; the odds ratio 4.57 (1.23 to 16.94) reproduces exactly on a Woolf interval. Gram-positive skin flora 1 of 75 against 10 of 75, Fisher p = 0.0090 against a printed .0090, odds ratio 11.39 (1.42 to 91.36) reproducing exactly. Sternal dehiscence 1 against 3, Fisher p = 0.6199 against a printed .6199. The mechanism is coherent: the effect sits almost entirely in skin flora, 9 of the 12 control infections arose after the first postoperative week, and once the foam came off on day 6 or 7 with the incision primarily closed in 71 of 75 (95%), no further infection occurred in that arm. GRADE therefore starts low, loses a step for risk of bias and regains one for a large effect, landing at low. It is not upgraded further for a dose-response gradient, because none was tested, and it is not downgraded for imprecision, because the interval for the absolute benefit excludes zero. What holds it at Low rather than Moderate is that a predictable allocation sequence plus an unadjusted analysis in a single centre can generate a 12-point difference on its own, and nothing in the paper rules that out.

Figures: Figures checked

CAT: OCEBM 3 · Start Low (predominantly observational). Prophylactic foam ciNPT: all 7 studies favoured NPWT (study RRs 0.25-0.94) — Low, and at risk of being overstated because the review's search closed before the two largest randomised nulls in this library (Scheier 2025, Pérez-Granda 2025). Canister-free ciNPT: one comparative study, RR 0.30 for wound failure — Very low. Open sternal infection: hospital stay 32.3 vs 57.3 days in observational comparisons with obvious severity confounding — Very low. Bridge to closure: recurrent SSI 2.9% vs 10.6% after secondary closure (P = .049) — Very low. The authors' own conclusion is appropriately cautious: prophylactic ciNPT 'may be reasonable to consider in selected high-risk patients pending randomized trial confirmation'.

Figures: Figures checked

Read this table with the adherence figure and the words "not powered" in view. Neither is a footnote.

The odds ratios reconcile with the raw counts. Re-derived by hand from 16/1,022 and 17/1,208, the crude odds ratio is 1.11 (0.56 to 2.22) against the model's 1.11 (0.56 to 2.20); for all sternal SSI, 1.13 (0.72 to 1.76) against 1.12 (0.71 to 1.75). The hierarchical model returned essentially the crude estimate, which is what a well-balanced two-period cross-over should do.

One printed denominator does not reconcile. The Results state the primary composite "occurred in 33/2107 (1.5%) diabetic or obese patients" and all sternal SSI in "80/2107 (3.6%)", but the trial's own arm totals are 1,022 + 1,208 = 2,230, the figure used everywhere else in the paper including Table 2. Against 2,230 the rates are 1.48% and 3.59%. This is a typographical defect in a denominator, not an analytical one — no effect estimate depends on it — but it should be quoted as 33/2,230.

Ten serious adverse device effects were reported and none was adjudicated as device-related.

CAT: OCEBM 2 · Start High — a randomised trial, prospective, blinded outcome adjudication, 90-day follow-up, 2,230 high-risk patients in the wound-management comparison, which makes it the largest prospective randomised sternal ciNPT dataset this library holds. −1 imprecision, and this is the downgrade that matters: the trial was explicitly not powered for a clinical outcome. It is a vanguard phase that was stopped 'due to lack of additional funding' before the main trial began, and the authors say so in their own conclusion — 'No firm conclusions should be drawn regarding the effectiveness of Prevena'. The primary composite produced 33 events in total. The confidence interval, OR 1.11 (0.56 to 2.20), is compatible with a 44% REDUCTION and with a 120% INCREASE in deep sternal infection. To detect the risk ratio of roughly 0.47 that the general cardiac syntheses report, from a 1.4% control rate at 80% power, requires about 2,900 patients per arm before any inflation for cluster randomisation; this trial had 1,022 and 1,208. −1 risk of bias from non-adherence: only 68.1% (696/1,022) of patients allocated to PREVENA actually received it, and the intention-to-treat estimate is therefore diluted toward the null by construction. Adherence was 93.0% at one site and 42.8% at the other, so more than half the intervention arm at one hospital received the control treatment. Not downgraded for indirectness — obese and diabetic sternotomy patients are exactly the population a ciNPT motivation concerns — and not downgraded for inconsistency, since the treatment-by-centre interaction was flat (p = 0.905). Net low. The inference this page exists to block: a null result from an underpowered trial in which a third of the intervention arm never received the intervention is not evidence that ciNPT does not work. It is evidence that this design could not answer the question, which is what a vanguard trial is built to find out. The as-treated analysis, which removes the dilution, gives OR 1.17 (0.54 to 2.53) — no closer to benefit, and no better powered. What this trial does legitimately establish is that the large treatment effects reported by retrospective sternal series are not visible in the one prospective randomised dataset of this size, and that any motivation quoting a 50% reduction should expect a reviewer to raise it.

Figures: Figures checked

Effect estimates, computed here because the paper reports none. Mediastinitis, NPWT versus gauze: RR 0.333 (0.068 to 1.638), risk difference −1.33% (−3.17 to +0.50), NNT 75, Fisher p = 0.286. NPWT versus the two other arms pooled: RR 0.333 (0.075 to 1.480), RD −1.33% (−2.78 to +0.12), Fisher p = 0.160. Superficial SSI, NPWT versus gauze: RR 1.286 (0.485 to 3.408). Mortality, NPWT versus foam: RR 0.529 (0.240 to 1.169).

CAT: OCEBM 2 · Start High (randomised, 900 patients). -1 risk of bias: No BLINDING of any clinical party is described; The analysis population is never declared - intention-to-treat, per-protocol and modified intention-to-treat appear nowhere and every denominator is fixed at 300; Losses to follow-up are reported nowhere while Figure 2's at-risk table implies 46 patients (16/19/11) left the risk set by day 90; Figure 1 contains no post-randomisation flow boxes at all; and six OR seven baseline variables are imbalanced at p < 0.05 - including DIABETES (21.0% / 32.0% / 26.7%, p = 0.010), which is the paper's OWN dominant risk factor for mediastinitis at IRR 5.420 (1.745-16.85) - while the text asserts 'no significant differences between the three groups in underlying conditions'. -1 imprecision: The sample size cannot be reproduced under any plausible baseline rate (an absolute 10-point reduction needs 195/arm from 20% to 10%, 146 from 16% to 6%, 132 from 15% to 5%; a relative 10% reduction from the observed 2.8% would need about 51,800/arm), and recomputed achieved power is 0.08 for superficial SSI and 0.31 to 0.40 for mediastinitis - the paper itself concedes 'barely more than 20% power'. Net Low for the null results, which is what this trial can support. For the subgroup claim, -2 further for indirectness and selective reporting: the CABG-with-mammary-graft finding is never labelled post hoc, the number of subgroups examined is never stated, it rests on 10 events in 301 patients with 2 in the NPWT arm, the paper's own two zero-inflated negative binomial models give the vacuum coefficient p = 0.923 and p = 0.378 in opposite directions, and the printed 13-FOLD estimate with CI 1.394 to 123.08 cannot be reproduced from any 2x2 table constructible from the paper's own published counts - its implied log-scale standard error is 1.143 against an actual 0.800 for the reconstructed 2/107 versus 8/184 table. Net Very low.

Figures: Figures checked