A Rise in Surgical Site Infection: Statistics, Confounding and Duty of Candour
A surgical directorate reports that colorectal surgical site infection (SSI) rates rose from 6.1% to 9.8% after a change from chlorhexidine-alcohol to povidone-iodine skin preparation, which followed a national supply shortage. There were 340 patients in the before period and 290 in the after period; a chi-square test gives p = 0.04. The change coincided with a reduction in theatre staffing and the arrival of a new cohort of trainees. Later that week, a woman who developed an SSI after a right hemicolectomy asks the on-call registrar whether the change in skin preparation "caused" her infection. The consultant is away and the registrar was not involved in her operation.
1. Interpret the result critically. What are the main threats to the validity of this conclusion, and what analysis or study design would you recommend to the directorate?
Read the worked answer
What the numbers do and do not show. A p-value of 0.04 means that, if there were truly no difference, a result at least this extreme would occur about 4 times in 100. It does not mean there is a 96% probability that the preparation caused the change, and it does not quantify the size or importance of the effect.
Threats to validity:
- Before–after design. This is an uncontrolled observational comparison; secular trends, seasonal variation and co-interventions cannot be separated from the exposure.
- Confounding. Reduced theatre staffing and a new trainee cohort are competing explanations, and both plausibly increase SSI through longer operating time, breaches in asepsis and technique. Case mix may also have changed - older or more comorbid patients in the second period would raise the rate independently.
- Multiple testing. Looking at many potential causes and highlighting the one that reached p < 0.05 inflates the false-positive rate. No correction or pre-specified hypothesis is mentioned.
- Small numbers and imprecision. The confidence interval around 9.8% with 290 patients is wide; the difference may be entirely compatible with chance or with a much smaller true effect.
- Measurement bias. Surveillance may have intensified after the change, increasing detection rather than incidence.
Better approach: use a run chart or statistical process control chart to see whether the rise is a genuine signal, adjust for confounders with multivariable regression (age, ASA, operative duration, conversion, trainee involvement, staffing levels), and if possible use an interrupted time-series or a difference-in-differences design comparing against a site that did not change preparation. Alternatively, restore chlorhexidine-alcohol as a planned quality improvement intervention and monitor with control charts.
2. How would you respond to the patient who asks whether the change caused her infection? Include what you would say, what you would not say, and what you would document.
Read the worked answer
Principles. The duty of candour requires an honest, prompt and compassionate explanation of what happened, an apology, and an account of what will be done. It does not require the registrar to speculate about causation or to admit liability, and it does not permit deflection or silence.
What I would say:
- Introduce myself and acknowledge that she has an infection that has caused her real harm and distress, and apologise that she has had this experience.
- Explain that there is a directorate-level review looking at infection rates and a change in skin preparation, that it is not yet complete, and that no individual patient's infection can currently be attributed to that change.
- Explain that infection after bowel surgery has several causes - the patient's own bowel flora, wound contamination, operative factors, diabetes, smoking, immunosuppression and technique - so a single case cannot establish cause.
- Tell her what will happen next: her infection will be treated, her case will be reviewed, and she will be given the outcome of the directorate review. Offer a follow-up conversation with the consultant.
What I would not do: state or imply that the change caused her infection; dismiss her concern as coincidental; or promise that "nothing went wrong".
Documentation: a contemporaneous note recording that the conversation took place, who was present, what was discussed, the apology offered, and the follow-up arranged. I would also inform the consultant and the clinical governance lead, because this is a patient-safety signal as well as an individual complaint - the same conversation may need to be repeated by a senior clinician with the full data.
3. The directorate asks you to design a quick study to settle the question. Outline the design, the primary outcome and the main statistical analyses you would plan, and identify the ethical and practical constraints.
Read the worked answer
Design. A prospective quality improvement evaluation is more feasible than a randomised trial, but a cluster or stepped-wedge design would answer the question far more securely. A pragmatic option: alternate skin preparation by operating list (cluster allocation by theatre session), with blinding of the SSI assessor, and a pre-specified primary endpoint. If that is impractical, a difference-in-differences analysis comparing the hospital with a similar unit that did not change preparation, using the same surveillance definitions, is the minimum acceptable standard.
Primary outcome: 30-day superficial, deep and organ-space SSI as defined by a recognised surveillance definition (for example CDC criteria), assessed independently and by a blinded reviewer. Secondary outcomes: length of stay, readmission, reoperation, and cost.
Analysis: baseline characteristics to demonstrate comparability; adjusted logistic regression or a mixed-effects model with the unit of clustering accounted for; report absolute risk difference with 95% confidence intervals rather than p-values alone; a pre-specified sample size calculation based on an expected reduction from about 10% to 6%, which requires several hundred patients per arm - a study with 290 patients cannot exclude a clinically important difference.
Constraints: obtaining consent where the preparation is a standard-of-care product may be difficult, but both agents are accepted practice, which makes equipoise defensible; surveillance requires trained and consistent assessors and adequate funding; and there is a practical risk of confusion on the ward with two products in use. The governance pathway is audit or service evaluation with ethics committee advice, and the protocol must be registered and pre-specified before data collection to avoid post-hoc cherry-picking.