Confusion bias and seasonal variation of exogenous endophthalmitis incidence.

Abstract

We read with interest the study by Kim et al. in which they analyzed postoperative endophthalmitis incidences in South Korea from July 2014 to June 2017.1 A total of 951 (0.064%) of 1 498 548 cataract procedures were identified as presumed postoperative endophthalmitis cases. The researchers found that rates were higher in July, during the summer months and when relative humidity and the precipitation level were also higher. The differences in incidence among seasons were statistically significant. It has been known for decades, and more recently confirmed, that the microorganisms that cause postoperative endophthalmitis usually come from the patient's own conjunctival and eyelid flora that enters the eye through the incision during the procedure or within a short time span after the procedure.2–4 Therefore, as the authors commented, the reason as to why postoperative endophthalmitis showed seasonal variation is not straightforward. Many variables could play a role, but undoubtedly it could also have been the result of a confusion bias, leading to an overestimation of the actual association with the summer time. The difficulty lies in identifying that third variable (confounding factor) that was not controlled. Possibly pertinent questions in this regard would be as follows: do all clinics operate throughout the year, or some close during the summer? Is it possible that those that close during the summer time have different prophylaxis protocols (v.gr. intracameral antibiotics) and/or the incidence of endophthalmitis in those clinics that do not operate during the summer was much lower throughout the year?5 Do the same surgeons perform surgeries in summer, or are more procedures performed by surgeons in training during that time? Is the rate of intraoperative complications the same during the summer? Is there an influx of patients who are going to have surgery in South Korea, coming from other countries, during the summer? The exclusion of these possible confounding factors would serve to strengthen the finding of a real association between the incidence of acute postoperative endophthalmitis and the seasons of the year. Further studies would be required to confirm some of the possible explanations suggested by the researchers: for example, that the conjunctival flora is different. Preoperative cultures of patients undergoing cataract surgery could be performed throughout the year, which would generate valuable information. On another note, a false inference can be made when information at the collective or group level is incorrectly assumed at the individual level. This would imply a multilevel analysis, the environmental variables being the contextual ones. The association under study had to consider the hierarchical structure of the data, both individually and collectively to avoid the occurrence of an ecological fallacy. The huge sample size also attracted our attention. With almost 1.5 million observations, it is very easy to find statistically significant bivariate associations, which can be spurious (ie, lacking authenticity or validity). The association is not synonymous with causation. A rigorous multivariate analysis would be required, and the variables of the model should not only be included based on purely mathematical criteria but also based on causality criteria, to respect the principle of parsimony.