Comments on incidence and risk factors for infection in spine surgery: A prospective multicenter study of 1,764 instrumented spine procedures

Abstract

We were interested in the recent article by Gu et al. 1 The authors provide interesting results; however, some methodologic and statistical issues should be noted. First, the authors point out that in univariate models, reason for surgery was the only significant risk factor of surgical site infection (SSI) after surgery (P < .05) and they did not attempt to estimate the effect size in univariate models. 1 It is crucial to emphasize that judgment about the strength of a risk factor based on only P value can be misleading because the value of a P value is a function of sample size, standard deviation, and effect size. As a general rule in biomedical research, the effect sizes will be estimated in univariate analysis and multivariable models 2,3 ; moreover, in univariate analysis a liberal P value such as < .10 or < .20 will be considered a criterion for statistical significance. 3 Second, the authors entered all the studied risk factors into the multivariable model. 1 However, the principle of parsimony states in statistical modeling, simpler models are preferred over more complex models because simpler models can provide better predictive abilities. 3 To derive simpler models, investigators and clinicians use stepwise methods and prior knowledge for selecting variables in the multivariable analysis. 3 Third, the design of the study was cross-sectional; however, in such designs the possibility of true association and causality is not possible because temporality (ie, exposure before outcome) may not be met. 4