Letter to the Editor Regarding “Vancomycin Antibiotic Prophylaxis Compared to Cefazolin Increases Risk of Surgical Site Infection Following Spine Surgery”

Dear Editor,

We read with great interest the recent article titled “Vancomycin Antibiotic Prophylaxis Compared to Cefazolin Increases Risk of Surgical Site Infection Following Spine Surgery” written by Herrington et al. ¹ We commend the authors for addressing this important topic and wish to offer several observations to enrich the discussion:

First, the manuscript states that 32 patients (6% of the cohort) were excluded due to incomplete data. We would appreciate if the authors could provide more details regarding their assessment of the mechanism of missingness and possible differences between patients with complete and incomplete data. The careful evaluation of these factors will allow a better understanding of the reasoning behind their decision for a complete-case analysis, which risks losing statistical power and potentially introducing bias in the estimates.^2-4 Furthermore, these considerations would define the need for implementing appropriate imputation strategies and sensitivity analyses to handle missing data and test the robustness of the study’s conclusions.^4,5

Second, the authors describe using background knowledge to define an initial set of covariates, followed by backward elimination for model selection. However, aside from antibiotic regimen, the expertise-driven variables are not listed and events-per-variable for this global model are not reported. This is information is critical to understand whether variable selection should be applied and the variable selection method to be employed. Notably, with less than 10 events-per-variable, variable selection is not recommended as the resulting model may lead to biased estimates. ⁶ Moreover, the reported AUROC was derived from a logistic regression model that, without internal validation described in the methodology section, seems to be both trained and evaluated on the same 503 complete-case dataset. As this approach assesses performance on the training data, the AUROC is likely overestimated. ⁷ To provide a more realistic assessment of model performance, future studies could incorporate internal validation techniques such as cross-validation or bootstrapping.^4,7 As part of a comprehensive modelling process, these resampling approaches would allow for performance evaluation on unseen observations, thus producing more robust metrics and providing valuable information for the assessment of model stability.^6,7

We thank Herrington and colleagues for their important contribution to improving antibiotic prophylaxis in spine surgery. We believe their findings will encourage future researchers to refine antibiotic selection strategies, particularly for patients with cephalosporin hypersensitivity, and hope our observations foster a constructive dialogue and inform the design of subsequent studies aimed at preventing surgical‐site infections.