Advertisement

Regarding “Epidemiology and risk factors for Clostridium difficile-associated diarrhea in adult inpatients in a university hospital in China: Methodologic issues”

  • Chenjie Tang
    Affiliations
    Department of Laboratory Medicine, The First Affiliated Hospital with Nanjing Medical University, Nanjing, China
    National Key Clinical Department of Laboratory Medicine, Nanjing, China
    Search for articles by this author
  • Yi Cui
    Affiliations
    Department of Epidemiology and Biostatistics, University at Albany-SUNY, Albany, NY
    Search for articles by this author
  • Genyan Liu
    Correspondence
    Address correspondence to Genyan Liu, MD, Department of Laboratory Medicine, The First Affiliated Hospital with Nanjing Medical University, 300 Guangzhou St, Nanjing 210029 China. (G. Liu).
    Affiliations
    Department of Laboratory Medicine, The First Affiliated Hospital with Nanjing Medical University, Nanjing, China
    National Key Clinical Department of Laboratory Medicine, Nanjing, China
    Search for articles by this author
Published:March 09, 2018DOI:https://doi.org/10.1016/j.ajic.2018.01.021
      To the Editor:
      We appreciate the interest in our study and would like to respond to the methodology-related issues raised by Safiri and Sullman in their letter to Editor. The letter concerned 2 statistics questions, mainly variable selection in the multivariable analysis and confounder identification.

      Variable selection in the multivariable analysis

      The forward selection method was used in our study and we apologize if not mentioning that in our article made Safiri and Sullman confused. We did mention that “some well-acknowledged factors had no significant difference in the study, such as age >64 years and nasogastric tube feeding, but they were also included for multivariate logistic regression alone with significant ones” in risk factors for Clostridium difficile-associated disease described as part of the Results section. Additionally, this was explained in the notes for Table S1. Specifically, blank cells in the multivariate analysis indicate P > .05, which has no statistical significance, and data not used for multivariate logistic regression.

      Confounder identification

      Confounding is usually referred to as a distortion of estimation of the true relationship between an exposure and a given outcome when the effect of primary exposure of interest are mixed in with the effects of an additional factor.
      • Ciccozzi M.
      • Menga R.
      • Ricci G.
      • Vitali M.A.
      • Angeletti S.
      • Sirignano A.
      • et al.
      Critical review of sham surgery clinical trials: confounding factors analysis.
      • Skelly A.C.
      • Dettori J.R.
      • Brodt E.D.
      Assessing bias: the importance of considering confounding.
      It usually has 3 major characteristics: a true confounding factor should be associated with the exposure of interest and the outcome simultaneously, the confounding factor is usually distributed unequally among the groups being compared, and a confounder cannot be involved in the causal pathway from the exposure of interest to the outcome. We used χ2 tests to exclude the possibility of diabetes as a confounding factor. Diabetes is not associated with any other of the 3 independent risk factors; that is, length of stay (P = .185), coloclysis (P = .563), and proton pump inhibitor (P = .224).
      To our knowledge, there are usually 3 methods used to identify confounding in clinical research. First of all, a clinically meaningful relationship among the variable, the risk factor, and the outcome is the key point of confounder identification, regardless of whether that relationship reaches statistical significance. Second, some formal tests of hypothesis, such as χ2 test, could be used to assess whether the variable is associated with the exposure of interest and with the outcome. Finally, a cut-off of 10% in the risk ratio, mentioned in the letter, is commonly used for the change-in-estimate criterion of confounder identification. In a previous clinical study,
      • Shakov R.
      • Salazar R.S.
      • Kagunye S.K.
      • Baddoura W.J.
      • DeBari V.A.
      Diabetes mellitus as a risk factor for recurrence of Clostridium difficile infection in the acute care hospital setting.
      diabetes mellitus was confirmed as a risk factor for recurrence of C difficile infection in an acute care hospital setting. Further analysis with the χ2 test mentioned above showed that diabetes mellitus was not a confounder. Last but not least, cut-off points for the change-in-estimate criterion varied according to the effect size of the exposure-outcome relationship, sample size, standard deviation of the regression error, and exposure-confounder correlation.
      • Lee P.H.
      Is a cutoff of 10% appropriate for the change-in-estimate criterion of confounder identification?.
      A 10% cut-off point is not the sole criterion. In many published articles, the differences in unadjusted odds ratios between the univariable and multivariable models is smaller than 10% but statistical significance was still recognized. For example, in a study about the risk factors for bronchiectasis in children with cystic fibrosis,
      • Sly P.D.
      • Gangell C.L.
      • Chen L.
      • Ware R.S.
      • Ranganathan S.
      • Mott L.S.
      • et al.
      Risk factors for bronchiectasis in children with cystic fibrosis.
      data in Table S2 draw a conclusion that neutrophil elastase activity in bronchoalveolar lavage fluid at age 3 months was the major predictor of persistent bronchiectasis at age 3 years, whereas the odds ratio in multivariate and univariate analysis was 4.21.

      Supplementary data

      The following is the supplementary data to this article:

      References

        • Ciccozzi M.
        • Menga R.
        • Ricci G.
        • Vitali M.A.
        • Angeletti S.
        • Sirignano A.
        • et al.
        Critical review of sham surgery clinical trials: confounding factors analysis.
        Ann Med Surg (Lond). 2016; 12: 21-26
        • Skelly A.C.
        • Dettori J.R.
        • Brodt E.D.
        Assessing bias: the importance of considering confounding.
        Evid Based Spine Care J. 2012; 3: 9-12
        • Shakov R.
        • Salazar R.S.
        • Kagunye S.K.
        • Baddoura W.J.
        • DeBari V.A.
        Diabetes mellitus as a risk factor for recurrence of Clostridium difficile infection in the acute care hospital setting.
        Am J Infect Control. 2011; 39: 194-198
        • Lee P.H.
        Is a cutoff of 10% appropriate for the change-in-estimate criterion of confounder identification?.
        J Epidemiol. 2014; 24: 161-167
        • Sly P.D.
        • Gangell C.L.
        • Chen L.
        • Ware R.S.
        • Ranganathan S.
        • Mott L.S.
        • et al.
        Risk factors for bronchiectasis in children with cystic fibrosis.
        NEJM. 2013; 368: 1963-1970

      Linked Article