Discussion

This study retrospectively evaluated 354 individuals treated for infections at our institution and identified four factors that were independently associated with ESBLE (ie, cirrhosis, urinary catheter, central venous catheter, and history of ESBLE). Our study adds to a growing body of literature reporting risk factors for ESBLE infections. However, the lack of consistency in population, type of infection, and evaluated risk factors in previous studies make drawing conclusions difficult.

One of the largest variations in the literature of assessing ESBLE risk factors is the site of infection. For instance, some studies only evaluate risk factors in bloodstream infections, whereas others only evaluate risk factors in urinary tract infections (UTIs). ^{Reference Augustine, Testerman and Justo7,Reference Goodman, Lessler and Cosgrove8,Reference Elligsen, Pinto, Leis, Walker, Daneman and MacFadden10–Reference Goyal, Dean, Neill, Jones and Dascomb12} While site-specific risk factor data are valuable, use may be limited as it cannot be generalized to other infection types or when the site of infection is unknown at the time of treatment initiation. In contrast, there are two published studies that evaluated all infection types, which may be more generalizable. ^{Reference Tumbarello, Trecarichi and Bassetti3,Reference Ben-Ami, Rodríguez-Baño and Arslan11} We hypothesized evaluating all non-urine ESBLE-producing isolates would allow us to capture a generalizable sample, while not including an overly heterogenous population. Future studies are needed to evaluate whether all sites of infection should be included in one clinical decision support (CDS) tool or if different sites of infections should have their own unique CDS tools.

Another challenge to overcome when trying to use these findings in practice is the variation of risk factors evaluated throughout the literature; this includes both the list of potential risk factors collected as well as the definition of risk factors. For example, one study only analyzed ESBLE risk factors in UTIs and therefore evaluated risk factors relevant to that disease state not captured in previous studies, including neurogenic bladder and recurrent UTIs. ^{Reference Goyal, Dean, Neill, Jones and Dascomb12} Another example is the use of the Charlson comorbidity index, which was analyzed as a potential risk factor in three previous studies. ^{Reference Tumbarello, Trecarichi and Bassetti3,Reference Elligsen, Pinto, Leis, Walker, Daneman and MacFadden10,Reference Goyal, Dean, Neill, Jones and Dascomb12} One of those studies found that a score > 4 was independently associated with an ESBLE infection (aOR 3.8 [1.9–7.59]), whereas this was not found in the other two studies (aOR 0.83 [0.43–1.63]) and P = .507). ^{Reference Tumbarello, Trecarichi and Bassetti3,Reference Elligsen, Pinto, Leis, Walker, Daneman and MacFadden10,Reference Goyal, Dean, Neill, Jones and Dascomb12} Additionally, although most studies evaluated “prior antibiotic use” and the “duration of prior antibiotic use,” the definitions of these risk factors differed greatly among studies. Two studies analyzed antibiotics given 3 months preceding hospital admission, ^{Reference Tumbarello, Trecarichi and Bassetti3,Reference Goyal, Dean, Neill, Jones and Dascomb12} whereas another study analyzed antibiotics given 3 months preceding culture collection. ^{Reference Augustine, Testerman and Justo7} With these differences in mind, we opted to collect antibiotic data 3 months prior to the culture, as this was most consistent with most studies. Duration of prior antibiotic use is also defined differently throughout the literature. Some studies only included antibiotics if therapy was given for at least 24 hours. ^{Reference Augustine, Testerman and Justo7,Reference Goyal, Dean, Neill, Jones and Dascomb12} This ranged from > 48 hours to > 72 hours in other studies. ^{Reference Tumbarello, Trecarichi and Bassetti3,Reference Elligsen, Pinto, Leis, Walker, Daneman and MacFadden10} In contrast, our study evaluated antibiotic use regardless of duration. These variations highlight the importance of clear definitions and the need for appropriate education on an institution’s specific risk factors before rolling out an ESBLE risk score.

Our study showed that cirrhosis, presence of a urinary catheter, presence of a central venous catheter, and history of an ESBLE infection were independently associated with an ESBLE in the multivariate analysis. Previous studies also demonstrate that presence of a catheter may be a risk factor for ESBLE infections. ^{Reference Tumbarello, Trecarichi and Bassetti3,Reference Augustine, Testerman and Justo7,Reference Goyal, Dean, Neill, Jones and Dascomb12} The adjusted odds ratio for presence of urinary catheter in this study was 2.73 [1.62–4.62], which was very similar to that of the previous literature which ranged from 2.36 to 3.52. ^{Reference Tumbarello, Trecarichi and Bassetti3,Reference Augustine, Testerman and Justo7,Reference Goyal, Dean, Neill, Jones and Dascomb12} This finding has promising clinical implications as all four studies were conducted in vastly different populations; due to this finding, we believe that this factor should be evaluated by all institutions who are looking to develop an ESBLE risk score tool. Having a history of an ESBLE infection is another factor that both our study and multiple other studies identified an independent risk factor. ^{Reference Augustine, Testerman and Justo7,Reference Goodman, Lessler and Cosgrove8,Reference Ben-Ami, Rodríguez-Baño and Arslan11} While we acknowledge the adjusted odds ratios vary among studies (range 12.49 to 51.45), this was consistently the strongest predictor of ESBLE and should be included in future risk factor identification studies. ^{Reference Augustine, Testerman and Justo7,Reference Goodman, Lessler and Cosgrove8,Reference Ben-Ami, Rodríguez-Baño and Arslan11} While both presence of a urinary catheter and history of an ESBLE seem to be the most promising risk factors, more studies are still needed describing implementation and evaluation of these risk factors in a CDS tool.

Limitations from our analysis include its retrospective nature and the dependence on previously documented data. We were not able to capture all variables that impact ESBLE risk due to this limitation. For example, travel history is not frequently documented in our electronic health record; thus, we were not able to capture it in our study. It is also possible that unidentified errors were made as data were originally entered into patient charts. Additionally, our study is limited in the number of total patients. Over the course of the study period, only 177 ESBLE infections met inclusion criteria, therefore only 177 matched non-ESBLE patients were included. This small sample size may have impacted the precision of our results. For instance, the confidence interval for the factor prior ESBLE infection spanned from 4.71 to 33.15. However, this is consistent with findings from previous studies like, Augustine et al. who reported an aOR of 26.8 and 95% CI, 7.0–108.2 for prior infections or colonization with ESBLE. ^{Reference Augustine, Testerman and Justo7} We also acknowledge that during this study period, there were likely changes in the epidemiology of Gram-negative resistance both in the United States and worldwide. Furthermore, our study was not able to assess the difference between “colonization” with an organism versus “infection” with an organism. We tried to account for this by only including patients who received an antibiotic in hopes that the clinicians’ decisions to treat the organism helped represent a true infection. Another limitation is that we did not collect all social determinants of health; this should be noted when applying to different patient populations. Finally, the majority of our cultures were from the blood. Therefore, cultures from other sites were most likely not fairly represented in our analysis. Our study adds to the variable literature describing variable risk factors for Gram-negative infections. A future study validating previously published risk factors combined with our data will hope to offer a comprehensive CDS tool.