Comparative effectiveness of oral antibiotics to treat uncomplicated urinary tract infections in male outpatients

A national retrospective cohort study of outpatient male veterans diagnosed with UTI during calendar years 2019–2022 was conducted. Inclusion criteria required that patients received in-person care in an emergency department (ED), urgent care (UC) or primary care (PC) setting; that an International Classification of Diseases-Clinical Modification 10th revision (ICD-CM 10) for cystitis, related codes or UTI-NOS (Supplement 1) was coded for the index visit [11], and that a prescription for an oral antibiotic of interest was released to the patient on days 0–3 after the index visit. Antibiotics of interest included: β-lactams (amoxicillin, amoxicillin/clavulanate, cephalexin, cefuroxime, cefpodoxime, cefdinir), fluoroquinolones (levofloxacin, ciprofloxacin), nitrofurantoin and TMP/SMX. Fosfomycin was excluded due to sparse prescription for index visits. A single dose of intravenous or intramuscular antibiotics prior to initiation of oral antibiotics was allowed. Patients were excluded based on the following criteria (Supplement 1): UTI diagnosis within 30 days prior to the index visit; hospitalization within 7 days prior to the index visit; hospitalization or death within 2 days after the index visit; active prescription indicating treatment with antibiotics at the time of index visit; a temperature of >99.9 F or presence of more than one systemic inflammatory response criteria ± 1 day of visit; a co-diagnosis of infection (including acute prostatitis or pyelonephritis) for which antibiotics may have been indicated on the day of index visit; an antibiotic fill of >15 days as these likely represented intent to treat complicated infection or given for prophylaxis; release of multiple antibiotics within 2 days of index visit; a scheduled urological procedure 1 day before to 7 days after their index visit; or a patient with urinalyses that indicated collection via catheter [4,10–14].

Data were extracted from the VA Corporate Data Warehouse (CDW) [15]: patient demographics; vital signs; laboratory values including urine culture and susceptibility results; acute co-diagnoses; chronic comorbidities; antibiotic resistance risk factor history (prior hospitalizations/skilled nursing facility residence, antibiotic exposures, antibiotic resistant bacteria reported from prior cultures); the antibiotic dosage and duration prescribed; visit provider, practice setting, VA facility; and clinical outcomes of interest.

The primary end point was a new outpatient acute care visit that occurred between 3 and 30 days after the initial antibiotic prescription release (i.e., dispensed) date in which the subsequent visit had ICD-10 documentation consistent with a UTI diagnosis (including pyelonephritis or prostatitis) and a new oral antibiotic prescription was released (i.e., UTI return visit). Return visits with documented competing tier 1 or 2 diagnosis codes where antibiotics may have been indicated were not classified as UTI return visits [12]. A secondary clinical end point included hospitalization with antibiotics administered within the same 3–30 day time-frame in which there was ICD-10 documentation of a UTI or a condition that could have arisen as a complicating genitourinary condition with a suspected urinary source (e.g., sepsis ICD-10 code PLUS UTI code) with an antibiotic prescribed (Supplement 1) [13,14].

Manual chart review was performed on 150 randomly selected index visit records from the CDW extracted cohort to validate that no exclusion criteria were present and that the clinician intent was to treat a UTI with antibiotics. Chart review was also performed on an additional 100 records defined as having a UTI outpatient return visit, but was not performed on UTI-related inpatient hospitalization as these were previously assessed [10,13]. Chart review was performed by the primary author using a standardized assessment process in which cases were first reviewed for prior/current exclusion criteria, followed by evaluation of the visit note assessment/plan where treatment allocation became apparent. Results were reported as the positive predictive value (PPV) along with the primary reasons for non-inclusion.

Regression models of treatment allocation were developed separately for each antibiotic class, (β-lactams, nitrofurantoin and TMP/SMX) relative to fluoroquinolones. For each model, the first patient-visit with either prescribed antibiotic class was selected and a logistic regression model was developed from factors deemed likely to be associated with antibiotic selection. These included race, ethnicity, age, indicators of chronic comorbidity (Charlson index) [16], vital signs, ICD-10 coded UTI diagnosis (cystitis versus UTI-NOS), antibiotic-resistant risk factors identified within the prior 90 days (prescription of antibiotic class of interest, recovery of study antibiotic class resistant organism from culture, hospitalization or nursing home residence) [10,13], time represented as a binary variable indicating pre/post COVID-19 onset (i.e., mid-March, 2020), and setting (ED/UC or PC) To account for potential correlation within the same facility, a generalized estimating equation model with clustering of patients by facility using an exchangeable correlation structure was fit [17,18]. Overlap weights [19,20] were calculated from model outputs and balance was checked (standardized mean difference and variance ratio) across variables included in both propensity and outcome models [21]. VA facility complexity, as an indicator of size and structure of the treating facility [22], and region of the country (VA VISN code) were not used in propensity models but balance on both of these variables was confirmed.

The overlap weights were applied in each antibiotic class comparison with UTI return visit as the outcome, using a Poisson regression with a log link to facilitate calculation of relative risk [23,24]. The three antibiotic classes were modeled relative to fluoroquinolones. An additional variable included the impact of treatment duration for each antibiotic class, which was categorized into 0–7 versus 8–15 day increments. The duration variable was not balanced in all models, so it was included in an interaction with each antibiotic class. A generalized estimating equation framework with clustered patients by facility and an exchangeable correlation structure was again employed for the outcome models. The same approach was used to assess relative risk for hospitalization, both for propensity and outcome models, for the same three antibiotic class comparisons.

Two different sensitivity analyses were performed. For all sets of models, E-values were calculated to address the magnitude of unidentified confounders which could potentially impact conclusions regarding treatment [25]. Simulations were also conducted to evaluate the change in the relative risk for an antibiotic that could potentially occur due to the inclusion of errant cases in the dataset (based on chart validation). For each antibiotic, based on chart review findings, 20% of the outcome (outpatient return visits) were randomly changed to treatment success. The outcome models were refit and the relative risk for the antibiotic was recorded. This was repeated 1000-times, and the 2.5th and 97.5th percentiles of relative risk were identified. These intervals were compared with the original relative risk for each antibiotic to evaluate whether inclusion of errant cases could have affected the magnitude and statistical significance of relative risk outcome. This sensitivity simulation was repeated for the hospitalization outcome. Additional post hoc analyses were conducted to better understand the findings. Specifically, we compared characteristics of patients prescribed nitrofurantoin across hospitalizations, return visits and treatment success, using one-way ANOVA or chi-square tests. We further compared these characteristics among hospitalizations and return visits using t-tests or chi-square tests.

All models were built using R version 4.3.1 [26] with RStudio [27], and using the libraries tidyverse [28], geepack [29], cobalt [30], gt summary [31], gt [32] and openxlsx [33].

This research was determined as exempt by the VA Puget Sound Investigational Review Board and complied with all federal guidelines and Department of Veterans Affairs policies relative to human subjects research.

A total of 103,190 UTI patient-visits within 130 VA medical centers were identified, of which 45,442 patient-visits were analyzed after applying exclusions (Table 1 & Figure 1). Antibiotic treatments of interest included: β-lactams 17,655 (38.9%), nitrofurantoin 8394 (18.5%), TMP/SMX 9709 (21.4%) and fluoroquinolones 9684 (21.3%). Most patients were elderly, male, Caucasian, had normal vital signs and modest comorbidity. The aggregate 30-day UTI return visit rate was 5453 (12.0%) and the frequency of hospitalization was 1431 (3.1%).

After development of overlap weighting models, potential confounders in treatment groups were well balanced with the exception of duration of therapy (Supplement 2 & 3).

The adjusted relative risk (aRR [±95% CI]) of 30-day outpatient UTI return visits for antibiotics of interest compared with fluoroquinolones was higher for β-lactams (1.22, [1.02, 1.48]) and nitrofurantoin (1.47, [1.23, 1.74]) but not for TMP/SMX (0.99, [0.80,1.21]). Relative risk for hospitalization associated with β-lactams (1.06, [0.80, 1.40]) and TMP/SMX (0.80, [0.56, 1.15]) was not different than the fluoroquinolones; whereas, relative risk of hospitalization was lower with nitrofurantoin (0.60, [0.41, 0.89]) (Tables 2 & 3 & Figure 2 & Supplement 4).

An analysis of interactions in UTI return visits models indicated that neither treatment duration, practice setting, or visit time dichotomized by pre/post COVID-19 onset were associated with the outcome for any antibiotic comparison. Significant interactions between nitrofurantoin and treatment duration [1.54 (1.09, 2.22)] and pre/post COVID-19 onset [1.90 (1.3, 2.77)] were identified in the hospitalization model (Table 3 & Supplement 4). The interactions indicate that nitrofurantoin had large increase in aRR for hospitalization after COVID-19 onset [1.14 (0.89, 1.47)], (Figure 2) and with a longer nitrofurantoin treatment duration compared with the fluoroquinolones.

Manual chart review of index visits indicated a PPV of 88.0%; 132/150 cases had no exclusion criteria present and the clinician intent was to treat a UTI. The primary reasons for exclusion were that patients were treated for UTI external to the VA in the previous 30 days or had an un-coded concurrent tier 1 or 2 diagnosis where antibiotics may have been indicated. Manual chart review of the outpatient UTI-related return visit cases indicated that the PPV was 79.0%; 79/100 cases had a return visit in which the clinician prescribed further antibiotics for a UTI-related condition. The primary reason for exclusion was an un-coded concurrent tier 1 or 2 diagnosis where antibiotics may have been indicated. The possibility that as many as 21% of return visits could have been for something other than UTI was the basis of the sensitivity analysis performed for this study.

Sensitivity analyses indicated that E-values and lower confidence levels (LCLs) for outpatient UTI return visits were higher than observed covariate relative risks for the outpatient UTI-return visits reported in the β-lactam and nitrofurantoin models but not for the TMP/SMX models (Table 4). These findings indicate that a very large (aRR >1.75–2.25) unobserved confounding variable could alter results for β -lactam or nitrofurantoin outpatient revisit models. For inpatient admissions, point estimates for the E-values were higher than observed covariate relative risks for the nitrofurantoin and TMP/SMX models, but not for the β-lactam model. The point estimates of the E-values for the β-lactam and TMP/SMX are larger than most, but not all, of the model coefficients, suggesting relatively large unidentified covariates could potentially alter results.

The simulations utilizing altered outcome data indicated that even with a 20% drop in outpatient UTI-related return visits, the original aRR for each antibiotic relative to fluoroquinolone was within the simulated 95% confidence interval for aRR (Table 4). However, while the modelled aRR estimates for longer treatment duration lie within the 2.5th and 97.5th percentiles of the simulated coefficients, the results indicate that the aRR for longer treatment duration is higher for all comparators. Finally, a 20% drop in hospitalizations indicates that TMP/SMX could result in a significant lower aRR whereas a significant aRR for longer treatment is possible.

An assessment of positive culture results indicated that treatment discordance with nitrofurantoin was significantly higher (adjusted p < 0.05) compared with the other three antibiotic classes (Table 1). Discordance for patients with E. coli recovered was highest for fluoroquinolones and TMP/SMX and lowest for nitrofurantoin and β-lactams (adjusted p < 0.05). While E. coli was the most common organism recovered overall it was only identified in 40.8% of positive cultures. In contrast, other Gram-negative organisms were identified in 39.8% of positive cultures, and treatment discordance with β-lactams was higher (adjusted p < 0.001) compared with the other three antibiotic classes, primarily due to treatment with oral cephalosporins and the recovery of Enterococcus spp.

Comparison of the patient-level characteristics for fluoroquinolone or nitrofurantoin recipients who experienced treatment success or failure with an outpatient revisit or inpatient admission are presented in (Supplement 5). As expected patients experiencing treatment success were less acutely ill and had less chronic comorbidity than treatment failures. They were more likely to have been initially treated in primary care, less likely to have antibiotic resistance risk factors, have a UC collected, or receive discordant treatment. Differences in characteristics between inpatient and outpatient failures were few and sample size for some characteristics for inpatient admissions was limited. Patients who experienced inpatient admission had more comorbidity, more likely to have been initially seen in the ED, and were more likely to have been recently hospitalized than patients experiencing outpatient failure. Differences in microbiology or treatment discordance were not apparent between inpatient and outpatient failures.

There are several notable findings from this comparative effectiveness analysis of oral antibiotics for the treatment of uUTI in males. First, modestly higher outpatient return visit rates were observed for treatment with β-lactams and nitrofurantoin relative to fluoroquinolones. Overall, UTI-related hospitalizations were low; and while no differences between β-lactam or TMP/SMX treatments and fluoroquinolones were observed, nitrofurantoin recipients had a reduced aRR for hospitalization. Assessment of the interaction between nitrofurantoin and post COVID-19 onset indicated a large increase in aRR for hospitalization, consistent with the estimated aRR for outpatient return visits. The interactions between antibiotic and treatment duration did not identify a modifying effect with a shorter versus longer treatment duration for either end point.

Plausible explanations for the increased outpatient return visits include treatment culture discordance, undiagnosed or mis-coded complicated infection, potentially combined with unfavorable pharmacokinetic properties of β-lactams and nitrofurantoin. Antibiotic resistance risk factors, particularly recent prior antibiotic exposure exhibited large aRR for both end points. In other VHA work we have demonstrated a protective effect for UC in patients with recent prior antibiotic exposures [13]. While positive UC were obtained in approximately 1/3rd of patients; E. coli was only identified in 40% of positive cultures. Additionally, discordance between nitrofurantoin and cultured Gram-negative organisms other than E. coli was high. A possible explanation for the higher return visit rate observed with nitrofurantoin includes the potential for non-E. coli pathogens to be intrinsically resistant to this antibiotic [34]. A second possibility is that clinicians have difficulty distinguishing between uUTI and acute prostatitis. In other work we have observed that documentation of acute prostatitis assessment and/or diagnosis in the eHR occurs in approximately 2% of ED/UC visits for UTI related presentations where an antibiotic is prescribed [35]. Both nitrofurantoin and oral β-lactam antibiotics possess unfavorable pharmacokinetics properties that contribute to low concentrations in prostate tissue or in systemic infections. While the study criteria attempted to restrict visits consistent with the new UTI guideline definition for uUTI it is plausible that not all patients with cUTI were identified during data extraction or that clinicians underdiagnosed cUTI. The outpatient return visit findings are also consistent with observations and guideline recommendations that nitrofurantoin should not be used to treat cUTI and that β-lactam therapy is associated with a higher relapse rate in men [4,6,10,36]. The observation that nitrofurantoin had a lower aRR for hospitalization was unanticipated, especially as the unadjusted rate was similar to fluoroquinolones. It is important to recognize that the baseline reported point estimate for nitrofurantoin represented the Pre-COVID-19 period, and when the interaction with COVID-19 is considered the point estimate for inpatient failures is non-significant [1.14 (0.89, 1.47)], Other relative changes in antibiotic prescription changed over time with reductions in fluoroquinolone use and increased for alternative antibiotics. For example, nitrofurantoin prescription increased from an average of 150/month prior to COVID-19 onset to 270/month post COVID-19 onset. One possibility is that as clinicians sought alternatives to fluoroquinolones they broadened patient selection for empirical nitrofurantoin use. Comparison of inpatient and outpatient failures indicated that treatment discordance was similar between end points, and admitted patients on average had more comorbidity, ED presentation, and antibiotic resistance risk factors; however, sample sizes for inpatient failures were limited in these strata. Finally, the potential for unmeasured confounding as represented by E-values was the highest for the nitrofurantoin hospitalization model and cannot be excluded.

Study strengths include the use of the nationwide VHA data warehouse for conducting the study, which allowed development of a detailed analysis including integration of outpatient visits and hospital admissions. Index visits were limited to those conducted in person with prescription of a single antibiotic to facilitate risk adjustment based on complete data. Another strength includes adjustment using overlap weighting which balances the mean of measured covariates [19,20]. Finally, the cohort criteria generally aligns with the new guideline definitions for uUTI in men.

Limitations are consistent with other retrospective database studies conducted within the VHA and elsewhere. Patients could have received diagnosis or treatment in non-VHA settings that would not be completely captured within the CDW, which may have biased risk adjustment or outcome. Second, UTI index visits were identified utilizing administrative data and in other work we observed that the specificity for UTI type (e.g., cystitis, pyelonephritis, prostatitis) within the VHA is poor as >85% of cases are classified as UTI-NOS and frequently other genitourinary diagnostic codes are used despite chart documentation of UTI treatment intent [10,13,37]. While most cases of cUTI were removed by applying other exclusion criteria including fever, we were unable to fully exclude catheterized patients with administrative data. In prior work we identified that ASB treatment occurs and it is likely that some patients were treated irrespective of symptoms [10]. In the current study, chart review demonstrated that in 88% of index cases, clinician prescribed an antibiotic with the intent to treat a UTI, irrespective of true diagnosis. The difference in return visits as an end point is meaningful in that the assumption on follow-up is that these patients were initially diagnosed with a UTI and were prescribed another antibiotic. An additional limitation was that chart review indicated that 78% of UTI return visits received antibiotics for a UTI-related condition. However, the sensitivity analysis with random replacement of cases did not identify an issue with inclusion of errant UTI return visits. Hospitalization was uncommon and the power to detect differences in this outcome and for select interactions was low. VHA patients could also have a higher portion of non- E. coli pathogens compared with other healthcare systems. However, A recent nationwide (n = 33,113) retrospective cohort study of UTIs in males (mean age = 70.7 years) in 15 French EDs reported a similar E. coli prevalence of 40.0%, and similar rates of non-E coli GNB and GPC [38]. Culture and susceptibility results were reported in less than half of the eligible visits, and each facility applies their own selective susceptibility reporting criteria. While not all susceptibility results performed are entered into the electronic health record (e.g., selective reporting) they are available in the VHA CDW. While discordant treatment with nitrofurantoin is a plausible explanation, caution is advised when extrapolating these results to other patient populations. Finally, we did not consider the downstream effects of the index visit therapy on subsequent antibiotic resistance or adverse events.

Historically there has been a lack of consensus on the definition of cUTI in men with some guidelines recommending that all UTIs in males are complicated versus others that take a more selective approach [39,40]. The recently published IDSA guidelines for cUTI suggest that men without systemic signs of illness or signs or symptoms of prostatitis or pyelonephritis can be managed as uUTI [4]. Our findings are consistent with cUTI guideline recommendation that list fluoroquinolones or TMP/SMX as preferred oral therapy; reservation or β-lactams as alternatives, and avoidance of nitrofurantoin. Consistent with guideline recommendation, we did not observe additional benefit with therapy >7 days. These recommendations are supported in part by the findings of Drekonja et al. that found no difference between 7 versus 14 days of therapy for afebrile male veterans treated with ciprofloxacin or TMP/SMX [7]. In contrast, a recent multi-centered non-inferiority study of 7 versus 14 days of antibiotic therapy in febrile men determined that 7 days of treatment was inferior to 14 days of therapy [41]. Further, a recent analysis of male UTIs in Europe observed a higher clinical failure rate of UTI treatment with nitrofurantoin and β-lactams relative to fluoroquinolones which is similar to our findings [35].

Future work should include additional adequately powered comparative studies of outpatient antibiotic regimens in men for uUTI. Improved diagnostic strategies are also needed to assist in differentiating complicated from uUTI in men and assessment of the new IDSA guideline UTI definitions. Given the increasing frequency of discordant empirical antibiotic selection for UTI, additional diagnostic studies are also needed to aid clinicians in identifying patients at risk for therapeutic failure and selection of effective antibiotics. Finally, outpatient diagnostic and antimicrobial stewardship studies evaluating surrogate markers of improvement in UTI management should also evaluate patient outcomes.

In conclusion, in this large comparative effectiveness analysis of oral antibiotics for the treatment of outpatient uUTI in men a modestly higher risk of clinical failure with β-lactam and nitrofurantoin compared with fluoroquinolones was observed. Clinicians should carefully differentiate between complicated and uUTI and carefully select appropriate antibiotic treatment in male patients.

K Madaras-Kelly: Conceptualization, obtaining funding, development of methods, data validation, initial draft of manuscript. J Boyd: Development of methods, data extraction, analysis, manuscript editing. L Bond: Development of methods, statistical analysis, manuscript editing

The findings and conclusions in this manuscript are those of the authors and do not necessarily represent the official position of the Department of Veterans Affairs.

This work was supported in part with resources and use of the Department of Veterans Affairs and was funded in part by an investigator-initiated research grant from BioMerieux, France (contract #E0358) to Idaho State University (Madaras-Kelly) with subcontracts awarded to the Idaho Veterans Research and Education Foundation (Boyd), and Boise State University (Bond). K Madaras-Kelly and L Bond also receive general support from the Institutional Development Awards (IDeA) from the National Institute of General Medical Sciences of the National Institutes of Health under grants #P20GM103408 and #P20GM155898, and L Bond from the Biomolecular Research Center at Boise State, RRID: SCR_019174.

The authors have no competing interests or relevant affiliations with any organization or entity with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No funded writing assistance was utilized in the production of this manuscript.

This research was determined as exempt by the VA Puget Sound Investigational Review Board and complied with all federal guidelines and Department of Veterans Affairs policies relative to human subjects research.

This manuscript reports the results of a real-world evidence study. The protocol was not publicly registered, and the data are not publicly available due to contractual restrictions. Select data and materials may be available from the authors upon reasonable request with the permission of Department of Veterans Affairs.

This work is licensed under the Attribution-NonCommercial-NoDerivatives 4.0 Unported License. To view a copy of this license, visit https://creativecommons.org/licenses/by-nc-nd/4.0/