Abstract

You have accessJournal of UrologyLetters to the Editor1 Mar 2023AUA-Recommended Antibiotic Prophylaxis for Primary Penile Implantation Results in a Higher, Not Lower, Risk for Postoperative Infection: A Multicenter Analysis. Reply.is a reply to letterAUA-Recommended Antibiotic Prophylaxis for Primary Penile Implantation Results in a Higher, Not Lower, Risk for Postoperative Infection: A Multicenter Analysis. Letter. David W. Barham, Nikolaos Pyrgidis, Martin S. Gross, Jay Simhan, and Faysal A. Yafi David W. BarhamDavid W. Barham * E-mail Address: [email protected] University of California, Irvine-Orange, California More articles by this author , Nikolaos PyrgidisNikolaos Pyrgidis University Hospital, LMU Munich-Munich, Germany More articles by this author , Martin S. GrossMartin S. Gross Dartmouth-Hitchcock Medical Center, Lebanon, New Hampshire More articles by this author , Jay SimhanJay Simhan University of California, Irvine-Orange, California More articles by this author , and Faysal A. YafiFaysal A. Yafi University of California, Irvine-Orange, California More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003150AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookTwitterLinked InEmail To the Editor: We appreciate the thought-provoking comment by Garman and Trock. As the authors correctly state, the findings of the present study should be interpreted with caution due to some important limitations relevant to its retrospective design, confounding factors, and inherent selection bias.1 Indeed, most baseline characteristics displayed statistically significant differences between patients undergoing treatment with vancomycin plus gentamicin vs other regimens. Nevertheless, we provide, to our knowledge, the largest study focusing on postoperative inflatable penile prosthesis infections in the published literature. Moreover, Garman and Trock raise some issues relevant to the applied statistical analysis and interpretation of the data which merit further clarification. Given that it is desirable to have a minimum of 10 events per variable (EPV) or even 20 EPV, they suggest that, with a total of 47 events, 5 or even fewer variables should have been included in the multivariable Cox proportional hazards model. However, it should be highlighted that this “1 in 10 rule” is only a rule of thumb that may be too conservative. Based on the previous notion, systematic discounting of results, in particular statistically significant associations, from any model with 5-9 EPV does not appear to be justified.2 In our study we included 8 variables, and thus a total of 5.9 EPV. As stated in our methods, these parameters were selected based on both clinical (age, diabetes, BMI, immunosuppression, prior pelvic radiotherapy) and statistical significance (perioperative antibiotic prophylaxis, antifungal prophylaxis, device manufacturer) in the univariate analysis. Of note, we also intended to control for other important variables such as race, Charlson comorbidity index and intraoperative approach, but these were not included in the multivariable Cox proportional hazards model to avoid overfitting and to remain over 5 EPV. Garman and Trock indicate that the variable selection for the multivariable Cox proportional hazards model should not have been based on clinical significance, but solely on statistical criteria to eliminate potential sources of confounding. They suggest putting variables “in a regression model and then add[ing] each of the other variables individually to identify the potential source of confounding.” In other words, they suggest performing automated selection methods (using backward elimination, forward selection, or both methods, ie, stepwise selection) to select the best model in many steps according to statistical criteria. These methods seem at first the “panacea” to all modeling problems. After all, it would be ideal to leave the selection of the most “correct” model to a statistical program. Nevertheless, automated selection methods have been criticized. By performing repeated testing and comparisons with automated selection methods, the probability of type I error increases substantially. Accordingly, the “correct” model after automated selection may not provide clinical utility, since, in this way, important risk factors may be excluded.3 It should be stressed that the current guidelines were not based on urological literature and recent work has questioned the efficacy of these recommendations.4-6 Although our study builds upon this knowledge, the limitations of our study should serve as a charge to the prosthetic urology community to improve our understanding of antimicrobial prophylaxis. There are certainly a number of barriers to performing prospective studies on penile prosthesis infections; however, if we want to overcome the limitations of our current literature, we must begin this process.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call