Multiple antimicrobial resistance and outcomes among hospitalized patients with complicated urinary tract infections in the US, 2013–2018: a retrospective cohort study

Complicated urinary tract infections (cUTI) are a leading infection-related reason for acute hospitalization. A cUTI can also arise as a nosocomial complication that represents a key focus for prevention among hospitalized patients. Frequently occurring in the presence of an indwelling catheter, catheter-associated urinary tract infections (CAUTIs) are the publicly reportable subset of cUTI, and are included in the Hospital Compare metrics by the United States Centers for Medicare and Medicaid Services [1, 2]. Irrespective of whether community- (CO) or hospital-onset (HO), cUTIs place a significant burden on the healthcare system.

Rising rates of antimicrobial resistance to commonly used antibiotics are adding hurdles to patient care [3]. Particularly troubling is the growing prevalence of extended-spectrum beta-lactamase producing Enterobacteraciae (ESBLs) at the time that the in vitro susceptibility rates to fluroquinolones and other routinely employed antimicrobials for cUTI, such as 3rd generation cephalosporins and trimethoprim-sulfamethoxazole, are diminishing [4‐7]. These shifts are making it difficult for clinicians to target empiric therapy.

In order to weigh the risks of therapeutic failure when administering standard treatment regimens, we examined the prevalence of overlapping resistance to commonly utilized antibiotics in cUTI and its impact on the outcomes. We also estimated the rate of inappropriate empiric therapy (IET) as a function of compound resistance in cUTI, and of its impact on morbidity and mortality.

Among 23,331 patients with cUTI who met the enrollment criteria 28,192 organisms were isolated [9]. The pathogens, TR prevalence over the study timeframe, and resistance rates to individual antimicrobials are in Table 1, Supplemental Table 1, as well as in the Supplemental Table 4 in citation 10. The prevalence of TR in the cohort was 13.0% (n = 3040, Table 2). Compared to patients with non-TR, those with TR were more likely male (57.6% vs. 47.7%, p < 0.001), black (17.9% vs. 13.6%, p < 0.001), and in the South (46.3% vs. 41.5%, p < 0.001), and had a higher chronic disease burden (median [IQR] Charlson score 3 [2, 4] vs. 2 [1, 4], p < 0.001). Patients with TR were hospitalized at centers with higher median prevalence of both C3R and TR (Table 2).

TR was associated with increase in some illness severity markers relative to non-TR (need for MV and ICU), though others were similar between the two groups (Table 3). Though the majority of all infections were monomicrobial (71.8% TR vs. 84.4% non-TR), patients with TR were more likely to have a polymicrobial cUTI. Similarly, while > 95% of all cUTI was CO, HCA cUTI was more common in TR patients than among non-TR (49.5% vs. 36.2%, p < 0.001). Consequently, more patients with TR had experienced a hospitalization and antimicrobial treatment within 90 days prior to the index hospitalization, and had within the same time period an isolate exhibiting resistance to one of the antibiotics or classes (Table 3). Despite more frequent use of such broad-spectrum empiric coverage as piperacillin-tazobactam and carbapenems, the group with TR, compared to non-TR, were more likely to receive IET (19.6% vs. 5.4%, p < 0.001).

The unadjusted outcomes are depicted in Supplemental Table 3. As for adjusted outcomes, although TR was not associated with a rise in hospital mortality or 30-day readmission rate, it was associated with greater hospital LOS and costs (Table 4). Testing for interactions revealed that TR affects both LOS and costs differently among patients with CAUTI versus non-CAUTI cUTI. Namely, though the entire cUTI cohort’s mean TR-attributable cost excess was $754 (95% CI $406, $1103, p < 0.001), it was $125 (95% CI -$275, $525, p = 0.540) for CAUTI and $1637 (95% CI $1045, $2229, p < 0.001) for non-CAUTI cUTI (Supplemental Fig. 1). Similarly, though non-CAUTI-cUTI patients had a TR-attributable increase in post-infection onset LOS of 0.62 days (95% CI 0.35, 0.88, p < 0.001), CAUTI patients stayed longer regardless of their TR status, driving the overall cohort’s LOS excess related to TR to 0.34 days (95% CI 0.18, 0.49, p < 0.001). The results were similar for the total LOS (data not shown).

Examining the impact of IET on the outcomes of cUTI revealed similar interaction with the type of cUTI in the mortality estimate. That is, in patients with non-CAUTI cUTI, IET raised the risk of mortality (OR 2.44; 95% CI 1.30, 4.56, p = 0.005), while this effect was absent in the CAUTI group (1.26; 95% CI 0.77, 2.04, p = 0.355). Additionally, IET was associated with increases in marginal hospital costs ($1364 in total costs; 95% CI $923, $1805, p < 0.001), overall LOS (0.66 days; 95% CI 0.46, 0.86, p < 0.001), and post-infection LOS (0.73 days; 95% CI 0.52, 0.94, p < 0.001) in the cUTI cohort overall.

Although incident CDI occurred in only 0.5% of the cohort, TR increased the risk of its occurrence, but did not reach statistical significance (OR 1.49, 95% CI 0.95, 2.32, p = 0.08, Table 4). TR was not associated with an increased risk of cUTI recurrence. Notably, the development of TR was rare in the non-TR group (0.45%, Supplemental Table 3).

In this large multicenter retrospective analysis of US hospitals TR is present in approximately 13% of patients with a cUTI. That is, nearly 1 in 8 patients with a cUTI, the vast majority of which are community-onset, are infected with a pathogen that is resistant to at least 3 of the following antimicrobials/classes: 3rd generation cephalosporins, fluoroquinolones, trimethoprim-sulfamethoxazole, fosfomycin, and nitrofurantoin. Not surprisingly, the presence of a TR pathogen increases the risk for IET. Interestingly, both TR and IET impact the outcomes differentially, depending on the type of cUTI. TR and IET are not significantly associated with higher costs or LOS in CAUTI, which are already high in this subset of cUTI. In contrast, non-CAUTI cUTI incur both higher costs and LOS with TR and IET, and higher mortality with IET. Indeed, IET results in an increase in the LOS of nearly 1 day, and excess costs of over $1300. TR and IET were important drivers in non-CAUTI cUTI of these outcomes even in the face of overall high severity of illness, with nearly ¼ requiring ICU, and over 10% with septic shock. Finally, though its overall incidence was low, CDI was associated more frequently with TR than non-TR, though this difference failed to reach statistical significance likely due to its low incidence.

We have specifically avoided the language of “multidrug resistant” (MDR) in our analysis. The US Centers for Disease Control and Prevention defines as MDR an isolate that is resistant to at least one antibiotic in three or more drug classes [14]. If those classes are not routinely employed as treatment for a specific syndrome, however, the term has little practical application to front line physicians while they make treatment choices. Hence, instead of enumerating the frequency of in vitro non-susceptibility to choices one might never consider in cUTI, our study addressed agents regularly given in this specific setting, so as to identify the characteristics of patients in whom the “standard” regimens might fail.

In contrast to some other populations, where both resistance and IET contribute to a rise in mortality we did not find that to be the case in the overall cUTI cohort [15‐18]. This is likely due to the low (2%) baseline mortality rate compared to infections such as healthcare- and ventilator-associated pneumonia and sepsis, where crude case fatality ranges from 10 to 40% [11, 12, 15‐22]. This lack of a mortality effect in our study mirrors that in other low-risk of death infections, including severe skin infection [23]. However, we identified a differential effect for both TR and IET on the outcomes, depending on the type of cUTI – CAUTI vs. non-CAUTI cUTI. Namely, while TR and IET alter the outcomes in the latter, they do not appear to have an impact in the former, suggesting that these two groups should be analyzed separately vis-à-vis their treatment outcomes.

The relationship between TR and IET and excess costs in our population is also consistent with the findings of others. Work in multiple other infections, including lower mortality syndromes, regularly illustrates that resistance contributes to IET, and delayed or inappropriate empiric therapy increases the LOS and, in turn, adds to costs [23]. Our findings specific to cUTI are novel and add to the present literature. Importantly, although an absolute increase in LOS of less than one day may appear trivial at first glance, our estimate of IET-attributable excess of 0.7 days represents approximately 10% of the entire LOS for the average patient with a cUTI. Similarly, even though TR- and IET-attributable excess costs of $754 and $1364, respectively, may seem modest, from a hospital perspective these costs can quickly become substantial, given the combination of frequency of admissions with a cUTI and the already strained reimbursement rates.

One additional novel aspect of our study is that we quantified incident CDI in cUTI. Although the overall rate was lower than in other hospitalized populations, TR did increase the risk of this infection [13, 24‐26]. Though we did not specifically examine the implications of CDI on 30-day readmission, it is likely another potential source of LOS and cost rise.

What are the practical implications of our observations? First, certain exposures remain associated with TR. Some of these relate to prior interaction with the healthcare system and suggest that physicians must strive to determine a patient’s prior healthcare and antibiotic exposures when making prescribing decisions. Though the concept of “healthcare-associated infection” may currently be out of favor in select formal guidelines, the evidence indicates that not all community-onset infections pose the same risk for resistance and the accompanying concern for IET [27]. Second, pre-test probability matters. That is, readers should not interpret our findings as a call to abandon current practices and move to selecting broader-acting antimicrobials for first line therapy in cUTI; that would only serve to foster more resistance. Rather, our observations stress the imperative for physicians to have granular data on local microbiology as a function of the syndromes they treat.

Our study has a number of limitations and strengths. The observational nature of the study predisposes it to multiple threats to validity, particularly a selection bias. By defining the cohort prospectively, we attempted to mitigate the magnitude of this bias. Misclassification is an issue, particularly when using administrative data. To deal with this, we used a previously published, though not clinically validated, algorithm, excluded other potential sources of infection, and included microbiology specimens, pharmacy data, and dates of cultures and treatments to minimize its magnitude. If present, however, this type of misclassification would drive the differences between groups toward null. At the same time, we could not differentiate between infection and colonization. Confounding is a potential problem in all observational studies. We performed multivariable modeling to minimize its impact using many confounders. However, some residual confounding may remain. Because this is a large multicenter geographically representative database, generalizability is of minimal concern, though we must caution that our results apply only to hospitalized patients, and not those treated in the community. Despite these limitations, this is the largest multicenter cohort study to examine the prevalence, time trends and outcomes of antimicrobial resistance in cUTI.

In summary, we have demonstrated that resistance to combinations of regularly used antimicrobials is prevalent and on the rise in the most common cUTI organisms in the US hospitals. Though increasing resistance alone does not impact hospital mortality, it does expose patients to an elevated risk of worsened outcomes through increasing the likelihood of inappropriate empiric therapy.