A cesarean surgical-site infection bundle: can a before-after improvement be trusted?

The crude infection rate fell from 8.1% to 4.9%, an absolute reduction of 3.2 percentage points. The crude relative risk is 4.9/8.1 = 0.60, suggesting a 40% reduction, and the number needed to treat is about 1/0.032 ≈ 31 cesareans. The reported p = 0.002 indicates that a difference this large is unlikely by chance alone, but statistical significance is not the same as causal effect. The design is a before-after comparison, so secular trends, regression to the mean, and co-interventions can explain the result. The unit performed more emergency cesareans after the bundle, which usually increases infection risk; that strengthens the signal somewhat but also means case mix changed. The new wound-closure protocol is a co-intervention, so the independent effect of the bundle cannot be isolated. Surveillance may have changed: staff aware of the project may classify wounds differently, follow patients less completely, or remove dressings earlier. There is no concurrent control group, no risk adjustment for BMI, diabetes, labour, or operative time, and no data on bundle compliance. The result is promising but not definitive.

Use a design that accounts for time trends and clustering. A stepped-wedge cluster randomised trial across several maternity units would be ideal: randomise the order in which sites introduce the bundle, with all sites eventually adopting it, and compare infection rates within sites over time. An interrupted time series is a pragmatic alternative if randomisation is not feasible. Collect at least 24 months of pre-implementation and 24 months of post-implementation data, with monthly infection rates plotted on a statistical process control chart, such as a u-chart. Pre-specify the primary outcome as superficial or deep surgical-site infection within 30 days using a standard definition such as the CDC/NHSN criteria. Risk-adjust for BMI, diabetes, labour, emergency status, operative time, and antibiotic timing. Measure bundle compliance as a process outcome, and include balancing measures such as cost, antibiotic choice, allergic reactions, and workflow delays. Sample size should account for clustering and the baseline rate; a stepped-wedge trial with 8–12 units and adequate observations per period would provide reasonable power. Include a health economist and a patient representative. Publish the protocol and analysis plan before rollout.

The bundle is a quality-improvement intervention, not a research experiment with individual randomisation, so the main ethical duties are transparency, privacy, and avoiding harm. A waiver of individual consent may be acceptable if the intervention is minimal-risk, is adopted as a unit standard, and uses routinely collected, de-identified data; however, the local research ethics committee should review the project, especially if the results will be published. Patients should receive information about the project through ward materials and hospital communication, with an opt-out for data use where required. Staff must be trained so that the bundle is delivered consistently, and compliance should be audited without blaming individuals. Avoid claiming that the bundle works before the evaluation is complete; the committee should describe the result as preliminary. If rollout occurs, monitor for unintended consequences, such as allergic reactions to chlorhexidine or delays in emergency cesareans. Report conflicts of interest, including any commercial support. Share the protocol and results publicly, including negative or null findings, and involve frontline staff and patients in design. Finally, ensure that the intervention does not widen inequalities; for example, check whether adherence is lower for women with limited English or those receiving care out of hours.