Investigation into postoperative endophthalmitis and lessons learned.

Purpose: To describe an approach to the investigation of a series of endophthalmitis cases to determine whether there was a true outbreak.

Setting: Outpatient facility, Sunderland, England.

Methods: Different approaches to statistical analysis of the probability of infrequent events being due to chance occurrence are described. Potential factors leading to an outbreak were reviewed.

Results: Bayesian statistical analysis was shown to be appropriate in the determination of an endophthalmitis outbreak. The only factor found to account for the outbreak was the operating surgeon's recent abandonment of subconjunctival antibiotic prophylaxis. This decision was based on the absence of good evidence that subconjunctival antibiotic injection is effective in prophylaxis. These endophthalmitis cases demonstrate that the absence of evidence for effect is not the same as there being no effect.

Conclusions: Bayesian statistical analysis has a place in determining whether an outbreak has occurred. Withdrawing treatment simply on the basis of a lack of good evidence can lead to undesirable outcomes.