Advertisement

Comments on decreased mortality in patients prescribed vancomycin after implementation of an antimicrobial stewardship program

  • Saeid Safiri
    Affiliations
    Managerial Epidemiology Research Center, Department of Public Health, School of Nursing and Midwifery, Maragheh University of Medical Sciences, Maragheh, Iran
    Search for articles by this author
  • Erfan Ayubi
    Correspondence
    Address correspondence to Erfan Ayubi, PhD, MSc, Department of Epidemiology, School of Public Health, Shahid Beheshti University of Medical Sciences, Velenjak St., Shahid Chamran Highway, Tehran, Iran. (E. Ayubi)
    Affiliations
    Department of Epidemiology, School of Public Health, Shahid Beheshti University of Medical Sciences, Tehran, Iran
    Department of Epidemiology & Biostatistics, School of Public Health, Tehran University of Medical Sciences, Tehran, Iran
    Search for articles by this author
Published:January 09, 2018DOI:https://doi.org/10.1016/j.ajic.2017.11.022
      To the Editor:
      We read the article by Conway et al
      • Conway E.L.
      • Sellick J.A.
      • Horey A.
      • Fodero K.
      • Ott M.C.
      • Krajewski M.P.
      • et al.
      Decreased mortality in patients prescribed vancomycin after implementation of antimicrobial stewardship program.
      with great interest; however, we think some methodologic and statistical issues should be taken into account to avoid misinterpretation.
      The authors
      • Conway E.L.
      • Sellick J.A.
      • Horey A.
      • Fodero K.
      • Ott M.C.
      • Krajewski M.P.
      • et al.
      Decreased mortality in patients prescribed vancomycin after implementation of antimicrobial stewardship program.
      stated that predictors with P values < .05 in univariate analysis were considered further for multivariable analysis. It is important to emphasize that such a univariate criterion can increase the risk of Testimation bias in the resulting effect estimates.
      • Steyerberg E.
      Clinical prediction models: a practical approach to development, validation, and updating.
      • Pakzad R.
      • Safiri S.
      Incidence and risk factors for surgical site infection posthysterectomy in a tertiary care center: Methodologic issues.
      Testimation bias will occur when a conservative univariate P value is considered in multivariable analysis.
      • Steyerberg E.
      Clinical prediction models: a practical approach to development, validation, and updating.
      The authors studied the effect of several baseline and clinical patient characteristics on death in univariate analysis.
      • Conway E.L.
      • Sellick J.A.
      • Horey A.
      • Fodero K.
      • Ott M.C.
      • Krajewski M.P.
      • et al.
      Decreased mortality in patients prescribed vancomycin after implementation of antimicrobial stewardship program.
      Some of them had a relatively large effect; for example, those with P < .05 and some others had a relatively small effect (eg, body mass index with P = .06 in univariate analysis).
      • Conway E.L.
      • Sellick J.A.
      • Horey A.
      • Fodero K.
      • Ott M.C.
      • Krajewski M.P.
      • et al.
      Decreased mortality in patients prescribed vancomycin after implementation of antimicrobial stewardship program.
      When predictors with relatively large effect only are included in multivariable analysis, the effect of such predictors will be overestimated.
      • Steyerberg E.
      Clinical prediction models: a practical approach to development, validation, and updating.
      Such overestimation of effects of strong predictors is known as Testimation bias.
      • Steyerberg E.
      Clinical prediction models: a practical approach to development, validation, and updating.
      To reduce this bias several strategies have been suggested. For example, a liberal univariate P value < .10 or < .20 can be considered in multivariable analysis or predictors can be selected based on both statistical test and clinical relevancy.
      • Steyerberg E.
      Clinical prediction models: a practical approach to development, validation, and updating.
      Moreover, the authors1 point out that external validity of the results may be low. We think internal validity of the results is still questionable. Cross-validation and bootstrapping are efficient strategies for checking internal validity of results.
      • Steyerberg E.
      Clinical prediction models: a practical approach to development, validation, and updating.
      A take-home message for readers is that the estimated odds ratios for the predictors of death in patients treated with intravenous vancomycin therapy should be considered along with the degree of Testimation bias.

      References

        • Conway E.L.
        • Sellick J.A.
        • Horey A.
        • Fodero K.
        • Ott M.C.
        • Krajewski M.P.
        • et al.
        Decreased mortality in patients prescribed vancomycin after implementation of antimicrobial stewardship program.
        Am J Infect Control. 2017; 45: 1194-1197
        • Steyerberg E.
        Clinical prediction models: a practical approach to development, validation, and updating.
        Springer Science & Business Media, 2008
        • Pakzad R.
        • Safiri S.
        Incidence and risk factors for surgical site infection posthysterectomy in a tertiary care center: Methodologic issues.
        Am J Infect Control. 2017; 45: 580-581

      Linked Article