Michal Chowers

Background: Enterococcal bloodstream infection (BSI) is common with 30-day mortality rates approaching 25%–30%, yet evidence guiding optimal antibiotic treatment duration, particularly for patients with uncomplicated infection, is limited. Methods: We conducted a retrospective cohort study using a target trial emulation framework across 8 hospitals in Israel (2013–2023) of adults hospitalized with uncomplicated monomicrobial enterococcal BSI. Treatment strategies were defined as short (7 ± 2 days) versus long (14 ± 2 days) antibiotic therapy. A cloning-censoring-weighting framework with inverse probability of censoring weights was applied to account for immortal time bias and potential confounders. The primary outcome was 30-day all-cause mortality; secondary outcomes included 90-day all-cause mortality, recurrence, length of stay, readmission and Clostridioides difficile infection. Analyses were performed using pooled logistic regression and negative binomial regression models, with prespecified subgroup and sensitivity analyses. Results: Of 3450 patients with enterococcal bacteremia screened, 870 were eligible, and 485 received treatment durations consistent with the study treatment strategies (short n = 283, long n = 202). Mean age was 74 years (±15.0); 87/485 (18%) had pre-existing cardiac hardware, 164/485 (34%) a urinary source and 180/485 (37%) had nosocomial BSI. Adjusted analyses demonstrated no difference in 30-day mortality between short and long therapy (21.6% vs 21.3%; adjusted risk ratio 1.01, 95% CI 0.84 to 1.22). There was no evidence of differences across the secondary outcomes. Results were consistent across subgroup and sensitivity analyses. Conclusions: Short-course antibiotic treatment was associated with comparable clinical outcomes to long-course treatment in uncomplicated monomicrobial enterococcal BSI, supporting the use of short treatment durations in appropriately selected patients.

Antibiotic resistance is influenced by prior antibiotic use, but precise causal estimates are limited. This study uses penicillin allergy as an instrumental variable (IV) to estimate the causal effect of antibiotics on resistance. A retrospective cohort of 36,351 individuals with E. coli positive urine cultures and prior outpatient antibiotic use, with outcomes assessed up to one year post-exposure, was analyzed using data from Maccabi Healthcare Services (MHS), the second-largest non-profit health fund in Israel. IV methods estimated risk differences (RD) and numbers needed to harm (NNH) for penicillin versus other antibiotics. The RD for resistance was 11.4% (95% CI: 7.6%, 15.4%) for amoxicillin/clavulanic acid, 14.1% (95% CI: 9.0%, 19.4%) for ampicillin, and 0.8% (95% CI: 0.2%, 1.4%) for piperacillin/tazobactam, with NNHs of 8.8, 7.1, and 122.0, respectively. Risks declined over time since exposure. Gentamicin, used as a negative control, showed no effect (95% CI: −2.4%, 1.8%). When directly comparing penicillin and quinolone effects on their respective AMR, penicillin use within 180 days increased resistance to amoxicillin/clavulanic acid by an RD of 17.8% (95% CI: 2.1%, 35.2%; NNH: 5.6), while quinolones raised ciprofloxacin resistance by 43.9% (95% CI: 29.9%, 59.4%; NNH: 2.3). These findings provide quantitative evidence of the impact of prior penicillin use on resistance, with implications for clinical practice and prescription policies.

Antibiotic resistance is influenced by prior antibiotic use, but precise causal estimates are limited. This study uses penicillin allergy as an instrumental variable (IV) to estimate the causal effect of antibiotics on resistance. A retrospective cohort of 36,351 individuals with E. coli positive urine cultures and prior outpatient antibiotic use, with outcomes assessed up to one year post-exposure, was analyzed using data from Maccabi Healthcare Services (MHS), the second-largest non-profit health fund in Israel. IV methods estimated risk differences (RD) and numbers needed to harm (NNH) for penicillin versus other antibiotics. The RD for resistance was 11.4% (95% CI: 7.6%, 15.4%) for amoxicillin/clavulanic acid, 14.1% (95% CI: 9.0%, 19.4%) for ampicillin, and 0.8% (95% CI: 0.2%, 1.4%) for piperacillin/tazobactam, with NNHs of 8.8, 7.1, and 122.0, respectively. Risks declined over time since exposure. Gentamicin, used as a negative control, showed no effect (95% CI: -2.4%, 1.8%). When directly comparing penicillin and quinolone effects on their respective AMR, penicillin use within 180 days increased resistance to amoxicillin/clavulanic acid by an RD of 17.8% (95% CI: 2.1%, 35.2%; NNH: 5.6), while quinolones raised ciprofloxacin resistance by 43.7% (95% CI: 29.9%, 59.4%; NNH: 2.3). These findings provide quantitative evidence of the impact of prior penicillin use on resistance, with implications for clinical practice and prescription policies.

Despite the progress in contemporary antiretroviral therapy (ART) and the continuous changes in treatment guidelines, virological failure (VF) is still an ongoing concern. The goal of this study was to assess factors related to VF after first-line ART. A longitudinal cohort retrospective study of individuals on first-line ART diagnosed with HIV-1 in 2010–2018 and followed-up for a median of two years was conducted. Demographics, baseline and longitudinal CD4 counts, treatment regimens, adherence and VF were recorded. The Cox proportional hazards regression and mixed models were used. A cohort of 1130 patients were included. Overall, 80% were males and 62% were Israeli-born individuals. Compared to individuals diagnosed in 2010–2014, when treatment was initiated according to CD4 levels, those diagnosed in 2015–2018 were older and had lower baseline CD4 counts. VF was recorded in 66 (5.8%) patients. Diagnosis with CD4 p < 0.001) and non-integrase strand transfer inhibitor regimens (non-INSTI, HR = 1.80, 95%CI:1.01–3.24, p = 0.047) increased VF risk. No impact of baseline resistance was observed. We concluded that the early detection of HIV-1 infection and usage of INSTI-based regimens are recommended to reduce VF.

Background: Antimicrobial resistance is a major healthcare burden, aggravated when it extends to multiple drugs. While cross-resistance is well-studied experimentally, it is not the case in clinical settings, and especially not while considering confounding. Here, we estimated patterns of cross-resistance from clinical samples, while controlling for multiple clinical confounders and stratifying by sample sources. Methods: We employed additive Bayesian network (ABN) modelling to examine antibiotic cross- resistance in five major bacterial species, obtained from different sources (urine, wound, blood, and sputum) in a clinical setting, collected in a large hospital in Israel over a 4-year period. Overall, the number of samples available were 3525 for E coli, 1125 for K pneumoniae, 1828 for P aeruginosa, 701 for P mirabilis, and 835 for S aureus. Results: Patterns of cross-resistance differ across sample sources. All identified links between resistance to different antibiotics are positive. However, in 15 of 18 instances, the magnitudes of the links are significantly different between sources. For example, E coli exhibits adjusted odds ratios of gentamicin-ofloxacin cross-resistance ranging from 3.0 (95%CI [2.3,4.0]) in urine samples to 11.0 (95%CI [5.2,26.1]) in blood samples. Furthermore, we found that for P mirabilis, the magnitude of cross-resistance among linked antibiotics is higher in urine than in wound samples, whereas the opposite is true for K pneumoniae and P aeruginosa. Conclusions: Our results highlight the importance of considering sample sources when assessing likelihood of antibiotic cross-resistance. The information and methods described in our study can refine future estimation of cross-resistance patterns and facilitate determination of antibiotic treatment regimens.

Background: Ciprofloxacin is a widely used antibiotic that has lost efficiency due to extensive resistance. We developed machine learning (ML) models that predict the probability of ciprofloxacin resistance in hospitalized patients. Methods: Data were collected from electronic records of hospitalized patients with positive bacterial cultures, during 2016-2019. Susceptibility results to ciprofloxacin (n=10,053 cultures) were obtained for E. coli, K. pneumoniae, M. morganii, P . aeruginosa, P. mirabilis and S. aureus . An ensemble model, combining several base models, was developed to predict ciprofloxacin resistant cultures, either with (gnostic) or without (agnostic) information on the infecting bacterial species. Results: The ensemble models’ predictions were well-calibrated, and yielded ROC-AUCs (area under the receiver operating characteristic curve) of 0.763 (95%CI 0.634-0.785) and 0.849 (95%CI 0.799-0.921) on independent test-sets for the agnostic and gnostic datasets, respectively. Shapley additive explanations analysis identified that influential variables were related to resistance of previous infections, where patients arrived from (hospital, nursing home, etc.), sex, and recent resistance frequencies in the hospital. A decision curve analysis revealed that implementing our models can be beneficial in a wide range of cost-benefits considerations of ciprofloxacin administration. Conclusions: This study develops ML models to predict ciprofloxacin resistance in hospitalized patients. The models achieved high predictive ability, were well calibrated, had substantial net-benefit across a wide range of conditions, and relied on predictors consistent with the literature. This is a further step on the way to inclusion of ML decision support systems into clinical practice.

Objectives: The objective of the study was to estimate how the time elapsed from previous antibiotic use is associated with antibiotic resistance. Methods: Data comprised electronic medical records of all patients in an Israeli hospital who had a positive bacterial culture between 2016-19. These included susceptibility testing results, and clinical and demographic data. Mixed-effects time-varying logistic models were fitted to estimate the association between time elapsed since last use of aminoglycosides and gentamicin resistance ( n = 13,095), cephalosporins and ceftazidime resistance ( n = 13,051), and fluoroquinolones and ciprofloxacin resistance ( n = 15,364), while adjusting for multiple covariates. Results: For all examined antibiotics, previous antibiotic use had a statistically significant association with resistance ( p < 0.001). These associations exhibited a clear decreasing pattern over time, which we present as a flexible function of time. Nonetheless, previous antibiotic use remained a significant risk factor for resistance for at least 180 days, with adjusted odds ratios of 1.94 (95% CI 1.40-2.69), 1.33 (95% CI 1.10-1.61), and 2.25 (95% CI 1.49-3.41), for gentamicin, ceftazidime and ciprofloxacin, respectively. Conclusions: The association between prior antibiotic use and resistance decreases over time. Commonly used cut-offs for prior antibiotic use can either misclassify patients still at higher risk, when too recent; or provide a diluted estimate of the effects of antibiotic use on future resistance, when too distant. Hence, prior antibiotic use should be considered as a time-dependent risk factor for resistance, both in epidemiological research and clinical practice.

To the Editor—Mical Paul and Leonard Leibovici raised concern that our recently published study [ 1] may be harmful to antibiotic stewardship attempts. As they have noted, current guidelines recommend adding macrolide or fluoroquinolone, therapy for atypical pathogens, to beta-lactam in all patients with severe community-acquired pneumonia (CAP) [ 2]. Their assumption is that all patients who did not receive macrolides were likely to be elderly or critically ill patients from nursing homes, and thus at higher risk for mortality. Notably, in our study, more than 30% of the patients who did not receive macrolides received quinolones ( Supplementary Table 1 ). Although quinolones are given for the same indication to patients with similar characteristics, we did not detect an association of quinolones with survival as observed with the macrolides [ 1], thereby highlighting the macrolide-specific effect in this patient population. They have also questioned the validity of the study results due to lack of a dose effect of macrolide use. However, the macrolide effect in our cohort was most likely immunomodulatory and such effect may not be related to conventional expected dose effects—that is, longer treatment duration may not be necessarily better.