Condition-specific surveillance in health care-associated urinary tract infections as a strategy to improve empirical antibiotic treatment: an epidemiological modelling study

Health care-associated urinary tract infection (HAUTI) is a major concern with a prevalence of 7.7% (4.6–17.3%) in urology departments, which is higher compared to other departments (0.5–1.7%) [1‐4]. HAUTIs consist of three main clinical conditions that include cystitis, pyelonephritis and urosepsis. In severe HAUTIs, timely administration of antibiotics is crucial with delays increasing the risk of morbidity and mortality [5‐9]. Lack of rapid microbiological diagnostics means that first-line antibiotic selection is carried out in the absence of individual patient-specific antibiograms (empirical antibiotics). Empirical antibiotic selection process requires judicious assessment of prior information such as the patient-specific conditions and local antibiotic resistance information [9]. Otherwise, selected empirical antibiotics for infections can be in discordance with the causative pathogen in up to 50% of cases leading to inappropriate antibiotic usage [10, 11].

Selection of empirical antibiotics can be improved with infection surveillance data [12]. As part of routine surveillance studies, HAUTIs are reported as an individual clinical diagnostic entity [12, 13]. This can be problematic because HAUTI is an umbrella term covering different infectious conditions of the urinary tract (cystitis, pyelonephritis and urosepsis), and, therefore, a condition-specific discrepancy of antibiotic resistance (AMR) can be expected [1]. In our previous work, AMR rates were higher in urosepsis for specific antibiotics compared to pyelonephritis and cystitis [1]. Additionally, pathogen diversity between conditions can also be different. The combination of difference in pathogen diversity with AMR can impact the empirical choices between conditions.

Comparison of empirical choices between conditions can be achieved using a composite index called weighted incidence syndromic combination antibiogram (WISCA) [14, 15]. WISCA measures the chances of coverage of the possible causative pathogens for each antibiotic choice (single or combination). This allows a direct comparison between different HAUTI conditions. Pathogen and AMR data derived from sequential surveillance studies include analytical challenges that can be overcome using Bayesian methods [16, 17]. In a recent study, we demonstrated that the Bayesian WISCA alongside probabilistic methods were useful in determining the impact of infection control policies in HAUTIs [15].

In the present study, we hypothesized that due to discrepancies of pathogens and AMR between HAUTI conditions, the chances of coverage of empirical antibiotic choices (WISCA) will be different. The aim of this study is to compare the estimated chances of coverage of empirical antibiotics between conditions (cystitis, pyelonephritis and urosepsis) in urology departments from Europe based on large sequential surveillance data. If the hypothesis is supported with strong evidence amongst a range of antibiotics, it will create an opportunity to improve empirical antibiotic selection for future patients with HAUTIs. In other words, an individualized approach for empirical antibiotics will be justified.

During the 12-year surveillance, 18,447 patients were screened in Europe and HAUTI was documented in 9.6% (n 1767) of them, which 7.6% (n 1398) were confirmed with culture tests. Summary of patient acquisition according to STROBE guidance is provided in Supplement V.

Mean age was 63.4 (SD 16.6) and mean Charlson comorbidity score was 2.4 (SD 2.7) (Supplement VI). Frequency of clinical conditions was: 38.0% cystitis, 31.6% pyelonephritis and 30.3% urosepsis. Mean Charlson score was similar in clinical conditions (p = 0.09). A Charlson score above 1 was measured in 62.2% of patients with urosepsis (significantly higher than other conditions p = 0.000). Nephrostomies, ureteral stents and ureteric stones were more common in pyelonephritis and urosepsis compared to cystitis (p = 0.000, Supplement VI).

In this study, we hypothesized that distinct patterns of pathogen diversity and AMR between conditions would lead to differences in empirical antibiotic choices. Chances of coverage of 21 antibiotic choices were studied and condition-specific differences were supported in 81% of them. We achieved this by obtaining the Bayesian WISCA of antibiotics per year and compared each antibiotic choice between conditions using the Bayesian factor. We also accounted for geographical variation by separating the European countries into high and low resistance areas using the ECDC surveillance data.

Frequencies of pathogens both in our previous and current (Fig. 1a) study were similar between the conditions [1]. Due to the large variety and heterogeneity of causative pathogens in HAUTI conditions, frequencies can be inadequate for comparison. We, therefore, employed diversity indices to quantify the pathogen diversity (Shannon index) within conditions and dissimilarity (Bray–Curtis index) between conditions [19]. The Bray–Curtis index revealed a dissimilarity of pathogen diversity between conditions (Fig. 1c). This dissimilarity could reflect the different responses elicited by bacteria and/or host leading to different clinical manifestations. Studies investigating the genetic or phylogenetic classifications alongside the host response can improve our understanding of the impact of pathogen diversity amongst conditions. This can help explore the host–pathogen relationship and individualize managements.

In our analysis, the AMR frequency of E. coli showed differences between conditions in 6 (28%) out of 21 studied antibiotics. These findings were similar to our previous published results in 2016 and illustrate that AMR can be different for pathogens between conditions but it does not explain the coverage of an empirical antibiotic [1, 15]. We, therefore, used the Bayesian WISCA that estimated a difference in 81% of studied antibiotics. There was decisive evidence to support the difference between conditions for the following single agents: cefuroxime, cefotaxime, levofloxacin and trimethoprim + sulfamethoxazole. Overall, a lower Bayesian WISCA for urosepsis was noted for all antibiotics compared to other conditions. The only antibiotic that did not follow this trend was gentamicin. The estimates obtained from this study support the concept that chances of coverage of antibiotics are different between conditions. It is likely that condition-specific surveillance can improve antibiotic stewardship and contribute as a strategy to tackle AMR in HAUTIs. A study by Bielicki et al. illustrated that continental level surveillance information would not be accurate to guide local empirical antibiotic selection [17]. Therefore, antibiotic stewardship programs for HAUTIs conditions that utilize surveillance data should test these findings at a local level. A recent study from The Netherlands reported WISCA values of urosepsis and HAUTIs, for which they attempted to provide recommendations for empirical antibiotic treatments [24]. However, due to lack of clinical information and retrospective nature, the recommendations would still be inaccurate. Therefore, we also suggest that WISCA estimates should be used for empirical recommendations if only local protocol-driven clinical data derived through validated definitions for clinical conditions of HAUTIs are available. We expect that HAUTI condition-specific Bayesian WISCA estimates can help in improving empirical antibiotic treatment of HAUTI conditions. Thus, the risk of morbidity and mortality decreases.

The strength of the study is based on two main factors. First, the GPIU surveillance to our knowledge is the only study that has a sequential measure of HAUTIs in urology departments using validated definitions. This has generated data comparable over time. Second, analysis of the data utilizing the Bayesian WISCA approach has allowed us to unmask previously unknown findings. The WISCA measure has been used before [14, 24] and inclusion of the Bayesian approach has only been of note in two other studies [15, 17]. In our recent study, we established the methodology for the Bayesian approach in WISCA estimates [15]. Accounting for the prior information improved the estimates. In addition, the Bayesian factor was used to test the hypothesis that chances of coverage of an antibiotic will differ amongst conditions. This was compared against the competing hypothesis of no such relation and allowed us to judge the strength of the evidence obtained from the GPIU study to support or refute the hypothesis.

Although this study has demonstrated that condition-specific surveillance information in HAUTIs can improve the information obtained for the empirical antibiotic selection, there are several limitations. First, the data are derived from an annual surveillance study at a continental level and participation from each country was varied. Hence, the accuracy of the results is questionable but previous studies have identified that the inaccuracy is more likely to be gauged towards an underestimate of AMR [25, 26]. Second, we used a 2-year prior information to calculate the Bayesian WISCA and the value of a longer duration for prior information should be further explored.

Our estimates illustrate that antibiotic choices can be different between HAUTI conditions. The Bayesian WISCA approach was instrumental in highlighting these differences. The findings can improve empirical antibiotic selection towards a personalized approach but need to be validated in local surveillance studies.