Study design and approval

This was a retrospective, observational cohort study with propensity score matching, comparing outcomes between patients treated with IV versus PO antibiotics for bone and joint infections. Patients were identified by searching the University of Utah Health Enterprise Data Warehouse (EDW) for patient encounters associated with surgical procedures performed by the Orthopedic service, or with the ICD-10 diagnosis codes M86XXX (osteomyelitis), T845XXX (infection of prosthetic devices), or M462X (vertebral osteomyelitis), from September 30, 2020, to April 30, 2022. Patients were included if they received at least two consecutive calendar days of IV antibiotics while inpatient, were discharged on an IV or oral antibiotic regimen for a planned total treatment duration ≥14 days for a bone or joint infection, were seen by the ID consult service while inpatient, and had a scheduled ID clinic follow-up visit after discharge. Those inclusion criteria were then confirmed by manual chart review. Patients were excluded if they: were <18 years old; pregnant; discharged to an outside hospital, long-term acute care hospital (LTACH), prison, or hospice; left against medical advice; required chronic renal replacement therapy; or if their planned antibiotic treatment duration was <7 days after discharge. Inclusion was limited to the first eligible encounter for patients with more than one encounter meeting criteria for inclusion during the period of interest. Patients were assigned to their cohort (IV or PO) based on their antibiotic regimen at hospital discharge. Since combination antimicrobial regimens are commonly used for bone and joint infections (eg, adjunctive rifampin), patients treated with more than one agent were included. Patients receiving fully oral combination regimens (eg, levofloxacin and rifampin) were assigned to the PO treatment cohort, while patients receiving partially oral combination regimens (eg, cefazolin and rifampin) were assigned to the IV treatment cohort.

Patient consent was not required for this retrospective chart review study. The study was approved by the University of Utah Health Institutional Review Board under IRB Protocol 00111238.

Outcomes

The primary outcome was the occurrence of treatment failure within one year of hospital discharge. Treatment failure was defined as a worsening or recurrent infection that required unplanned surgical intervention or antibiotic therapy escalation and was identified by chart review of progress notes. Therapy escalation was defined as extending the duration of therapy beyond the original intended duration, changing to a different antibiotic regimen due to concerns for treatment failure, or reinitiating therapy after a first treatment course, with accompanying progress note documentation from an ID provider of suspected bone or joint infection. Secondary outcomes, assessed within 60 days of hospital discharge, included the occurrence of treatment failure, adverse drug events (ADE), hospital readmission, emergency department (ED) encounters that were not associated with readmission, and ID clinic “no show” visits. ADEs were defined as provider documentation of a suspected medication event combined with discontinuation of the suspected agent. We also collected data on patients who switched from IV to PO regimens prior to the planned end of therapy, and vice versa, along with the reasons for the switch. Patients without a full year of postdischarge follow-up within our medical record were categorized as “lost-to-follow-up.”

Data collection

Data were collected from the EDW and by chart review by study personnel (JG, RB, HC) using a standard data collection form. Data gathered from the EDW included hospital length of stay, primary hospital service, age, sex assigned at birth, patient-reported race and ethnicity, primary language, discharge disposition, Charlson comorbidity index, primary payor, zip code of primary residence, and hospital readmission or ED encounter within 60 days of discharge.

Data gathered by chart review included discharge antibiotic regimen, antibiotic indication, microbiologic indication, surgical intervention for source control, unhoused status, history of IV drug use, the presence of a chronic infection (defined as an infection lasting ≥6 months, categorized as chronic based on imaging, or a patient previously on suppressive antibiotics), bacteremia, total treatment duration, time to first ID clinic follow-up visit, and the occurrence and time-to-event for the outcomes of treatment failure, ADE, and ID clinic no-show visits. Blinding to treatment assignment was not possible for this retrospective chart review.

Statistical analysis

Given considerable selection bias in the decision to prescribe oral versus IV treatment for bone and joint infections, a propensity score matching approach was utilized to better estimate the treatment effect between cohorts. Variables expected to be associated with treatment selection and treatment failure, or treatment failure alone, were included in the propensity score model (Table 1). Propensity scores were estimated by logistic regression using the MatchIt package in R. ^{Reference Ho, Imai, King and Stuart21} At first, propensity score matching was performed using a 1:1 nearest neighbor greedy matching algorithm without replacement (ie, each PO-treated patient was matched with the most similar IV-treated patient and patients could not be paired more than once). This led to a suboptimal match result (Supplementary Figure 1), so a caliper term was incorporated into the algorithm to limit the range of acceptable differences in propensity scores in matched patient pairs. An acceptable match quality was found with a caliper distance of 0.25 standard deviations (Supplementary Figure 2), and smaller caliper distances did not appreciably improve the quality of the matched cohorts.

Table 1.

Baseline characteristics

Abbreviations: IV, intravenous; SD, standard deviation; IQR, interquartile range; MRSA, methicillin-resistant Staphylococcus aureus; MSSA, methicillin-susceptible Staphylococcus aureus.

^a Variable included in propensity score matching.

Total counts and percentages, median and interquartile ranges, and means and standard deviations were used to summarize data as appropriate. Inferential tests of significance were avoided for descriptive data tables. ^{Reference Vandenbroucke, von Elm and Altman22} Patients without a documented substance abuse history were considered to have no history of IV drug use; no other adjustments were necessary to account for missing data as all other data were gathered completely. No formal power calculation was determined a priori; a convenience sample of all eligible patients was included. Differences in proportions were determined for each outcome and Wald-based 95% confidence intervals that considered the IV and PO cohorts as paired were determined using the Misty package in R (Yanagida T, “Package ‘misty’” Version 0.5.3. 2023-09-17). Differences in outcomes between IV- and PO-treated patients were considered statistically significant if the 95% confidence intervals did not contain zero. P-values for univariate comparisons were determined using McNemar’s test.

Survival functions for time-to-treatment failure by IV versus PO treatment were estimated using the Kaplan-Meier method and assessed for significance using the log-rank test. Time zero was defined as the date of hospital discharge, and patients were assessed from the time of discharge until they experienced treatment failure or up to one year following discharge. Secondary outcomes were assessed up to 60 days after discharge. Patients were censored at the date of last observation in our health system if less than one year postdischarge.

Statistical analyses were completed using JMP Pro version 16.1.0 (SAS Institute, Cary, NC) and R Statistical Software version 4.2.2 (R Core Team 2022).