To the Editor—We read the paper by Walsh et alReference Walsh, Querry and McCool 1 in a recent issue of Infection Control & Hospital Epidemiology with great interest.Reference Walsh, Querry and McCool 1 They examined risk factors for the development of surgical-site infections (SSIs) in neurosurgery patients undergoing spinal fusion. They conducted as case-control study on 159 patients with SSIs and 161 controls. Previous methicillin-resistant Staphylococcus aureus (MRSA) carriage was associated with SSIs both in the univariate model (odds ratio [OR]=24.96; 95% confidence interval [CI], 5.90–105.52) and the multivariate model (OR=20.30; 95% CI, 4.64–88.78).Reference Walsh, Querry and McCool 1 Although this study makes a valuable contribution to the field, an important methodological issue needs to be noted.
The authors examined the association between previous MRSA carriage and SSIs. They reported large ORs with wide CIs in both the univariate and multivariate models. Several researchers have stated that a large measure of association with wide CI does not necessarily mean large effect; this result may be attributable to the lack of sufficient data for the different combinations between the independent and dependent variables.Reference Greenland, Mansournia and Altman 2 , Reference Greenland and Mansournia 3 Also, multivariate models are more susceptible to sparse data because the number of combinations between the independent and dependent variables is higher than in corresponding univariate models.Reference Greenland, Mansournia and Altman 2
We extracted the data provided by Walsh et al regarding the univariate association between previous MRSA carriage and SSIs (Table 1). The number of the events is low in one of the combinations and sparse data bias is expected. This bias can be removed or decreased in the analysis stage, and several statistical methods have been proposed to address this problem.Reference Greenland, Mansournia and Altman 2 – Reference Ayubi and Safiri 5 Penalization via data augmentation is an efficient method introduced in 2016.Reference Greenland, Mansournia and Altman 2 We used this method to re-estimate the crude association between previous MRSA carriage and SSIs. The OR and 95% CI shrank and narrowed considerably, which demonstrates the high statistical efficiency of this method (Table 1). Penalization can also be applied to more susceptible models to address data sparsity, such as multivariate models, but the individual data are needed. Hence, we suggest that Walsh et al reanalyze their adjusted association between previous MRSA carriage and SSIs using the efficient method introduced here to report a more valid and precise measure of association.
TABLE 1 The Crude Association Between the Previous MRSA Carriage and SSIs Through Ordinary and Penalized Logistic Regression
NOTE. MRSA, methicillin-resistant Staphylococcus aureus.
ACKNOWLEDGMENTS
Financial support: No financial support was provided relevant to this article.
Potential conflicts of interest: All authors report no conflicts of interest relevant to this article.
