Study design

This retrospective cohort study was conducted across 4 outpatient cancer clinics within Henry Ford Health in southeastern Michigan. The study received institutional review board approval (no. 15246-01) with waiver of consent.

Patients were included in the study if they received an antibiotic prescription for an upper respiratory tract infection (URTI), lower respiratory tract infection (LRTI), urinary tract infection (UTI), or acute bacterial skin and skin structure infection (ABSSSI) from an ambulatory oncology clinic between May 1, 2021, and December 31, 2021. Additional inclusion criteria were age ≥18 years and active follow-up with a hematologist or oncologist for a cancer diagnosis at the time of care. Patients were excluded if they had multiple infections, were pregnant, were breastfeeding, were currently enrolled in a clinical trial, had been diagnosed with febrile neutropenia, or had had a previous treatment for infection of interest in the prior 30 days. Patients with a history of solid-organ or bone marrow transplant were excluded due to complexity in comparison to hematology-oncology populations and to prevent heterogeneity.

Data source

Patient data were manually reviewed and extracted from the electronic health record (EHR) using the Epic Slicer Dicer data extraction tool (Epic Systems, Verona, WI). Data extraction parameters were set to identify patients who received an antibiotic prescription in an ambulatory oncology clinic during the prespecified timeframe. The identified patients were then randomized and screened for inclusion. Data collection included patient demographics, comorbid conditions including cancer diagnosis, infection characteristics (ie, risk factors for multidrug-resistant organisms), and antibiotic treatment. Patient outcomes included treatment failure, delayed chemotherapy, and adverse drug events. All data were captured by a single investigator using a standardized electronic case report form.

Definitions

The primary end point of optimal antibiotic therapy was defined in accordance with the National Comprehensive Cancer Network (NCCN) guidelines and institutional antimicrobial stewardship guidelines was defined in accordance with the standard of care. Optimal treatment was defined as a composite of optimal drug, dose, and duration (Supplementary Table S1). ¹¹ The cohort was divided according to receipt of suboptimal or optimal treatment. Infection types were categorized according to medical record documentation; LRTI included chronic pulmonary obstructive disorder (COPD) exacerbation and community-acquired pneumonia (CAP); URTI included pharyngitis, bronchitis, and sinusitis; ABSSSI included purulent and nonpurulent cellulitis, abscess, and infected bite injuries; UTI included lower urinary tract infections (ie, cystitis in the absence of fever or flank pain suggestive of an upper UTI) and pyelonephritis. The full definitions of optimal regimens by disease state are listed in Supplementary Table 1. An analysis of prolonged antibiotic durations distinct from NCCN duration of therapy recommendations was performed and was defined as days of therapy exceeding institutional guideline recommended short-course durations (Supplementary Table S1). ^{Reference Spellberg12}

Secondary end points included the proportion of optimal components of the composite outcome, treatment failure, delay in chemotherapy due to active infection, and adverse drug events (ADEs) while on antibiotic therapy. We defined 30-day treatment failure as a documented change in antibiotic due to a lack of clinical improvement or due to a documented antibiotic adverse drug event, hospital admission due to worsening index infection, or development of recurrent infection. Only ADEs that were suspected to be due to the antibiotics the patient received were captured, as documented by the treating provider in the electronic health record (EHR).

All types of encounters including clinic visits, telephone encounters, and Epic MyChart EHR messages were reviewed for up to 30 days after antibiotics were initiated to find a documented secondary end point. Patients were presumed to be end point free at 30 days if they had no documented encounters.

Statistical analysis

This study was designed to detect a 20% difference between any 2 variables associated with receipt of optimal antibiotics and assumed a baseline prevalence of 30% suboptimal antibiotic use. Assuming a type 1 error frequency of 5% and a power of 80%, a sample size of 200 patients was targeted. The assumption of 30% suboptimal antibiotic use was derived from prior internal data related to outpatient antibiotic prescribing at Henry Ford Health. ^{Reference Vuong, Faison, Thomson, Kenney and Davis13}

Descriptive statistics (ie, proportion [%] and median [IQR]), were used to describe the cohorts of optimal and suboptimal antibiotic use. In bivariate analyses, categorical variables were compared using the Pearson χ² or Fisher exact test, and continuous variables were compared using the Mann-Whitney U test. To determine variables independently associated with optimal antibiotic prescribing, variables associated with the outcome (P < .25) from bivariate analysis were entered into a multivariable regression model using a backward, stepwise approach. Variables included in the model were restricted to an event-to-variable ratio of 10:1; model fit was assessed using the Hosmer-Lemeshow goodness of fit test. A separate a priori analysis of optimal versus prolonged antibiotic duration of therapy was performed to assess associations with patient harm. All statistics were calculated using SPSS Statistics for Windows version 26.0 software (IBM, Armonk, NY).