Data source and study design

We performed a retrospective observational cohort study using the Premier PINC AI Healthcare Database (PHD). ¹⁷ The PHD is an all-payer hospital administrative database for geographically diverse inpatient and outpatient visits from more than 1,300 hospitals. Inpatient discharges in PHD represent approximately 25% of all inpatient admissions in the US since 2000. PHD patients are tracked within the same hospital or hospital system using a unique identifier. ^{17,Reference Moon, Bleak and Rosenthal18} The standard hospital discharge files include demographic characteristics, disease states, and a time-stamped log of billed items (e.g., procedures, medications, laboratory services, and diagnostic services) at the patient level, and geographic location, rural/urban populations served, teaching status, and bed capacity at the hospital-level. A subset of hospitals (∼25%) submit microbiology laboratory data to the PHD. Microbiology data include specimen collection date, type of specimen, types of tests performed, and observations for the specimens. All data are statistically deidentified and compliant with the Health Insurance Portability and Accountability Act. Based on US Title 45 Code of Federal Regulations, Part 46, the study was exempted from institutional review board approval. We did not pursue informed consent of study participants due to the nature of deidentified data. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) ^{Reference von Elm, Altman and Egger19} reporting guideline.

Study population

Adults (aged ≥ 18 years) who had an inpatient visit, had ≥ 1 blood culture isolate belonging to E. coli, K. oxytoca, K. pneumoniae, or P. mirabilis, and received empiric IV or oral β-lactam antimicrobial therapy (within 2 days of BCC) between January 1, 2017, and June 30, 2022, were included in the study. Patients were excluded if they: (1) had polymicrobial infection defined as positive cultures belonging to > 1 pathogens from blood or any other site within 30 days before and 5 days after the index BCC date, (2) did not have antimicrobial susceptibility testing (AST) results available by BCC Day 7, (3) had insufficient AST results to define susceptibility profile, (4) did not receive systemic antimicrobial treatment of β-lactam or other oral (fluoroquinolone [FQ] or trimethoprim-sulfamethoxazole [TMP-SMX]) agents for at least 3 consecutive days on or after BCC Day 0, (5) had an intra-abdominal infection (to limit polymicrobial infections and because treatment for intra-abdominal infections may require coverage for additional organisms) ^{Reference Perez, Colindres and Wilson20} , (6) were transferred from another acute care facility, (7) expired, discharged, or transferred to another hospital within 2 days following BCC Day 0, and (8) were from hospitals without continuous data submission during the 3 months before and 30 days after the visit. If a patient had multiple hospitalizations meeting the selection criteria, the earliest admission meeting all inclusion and exclusion criteria was considered as “index admission”.

Definitions of DOOR-MAT and PDAT

PDAT was the primary variable of interest in this study and defined as the β-lactam antibiotic with the narrowest spectrum that is in vitro active against the final isolated pathogen. This is operationalized and denoted using the Desirability of Outcome Ranking for the Management of Antimicrobial Therapy (DOOR-MAT) framework (Figure 1). The definition of DOOR-MAT categories was adapted from the naïve approach in the paper by Perez et al. ^{Reference Perez, Colindres and Wilson20} Antimicrobials were classified by their spectrum of activity (Supplement eTable 1) with a focus on β-lactams as these are preferred for BSI. Use of an antimicrobial inactive against a bacterial phenotype was considered as “undertreatment.” Use of an Intermediate II antimicrobial (e.g., piperacillin-tazobactam) for isolates that were resistant to expanded-spectrum cephalosporins (e.g., ceftriaxone) was defined as “potentially appropriate”, but not as receiving desirable treatment (PDAT), given recent evidence that carbapenems may be most desirable for this phenotype. ^{Reference Harris, Tambyah and Lye21} Among patients receiving PDAT, patients were identified as ‘receiving early PDAT’ if desirable treatment was given within 2 days of BCC (Day 0−2), and ‘receiving delayed PDAT’ otherwise.

Figure 1.

Grid illustrating the desirability of outcome ranking for the management of antimicrobial therapy (DOOR-MAT) for the treatment of bloodstream infections caused by Escherichia coli, Klebsiella oxytoca, Klebsiella pneumoniae, or Proteus mirabilis.

In a subgroup of patients who received oral antimicrobials, a secondary analysis was performed wherein oral fluoroquinolone (FQ), trimethoprim-sulfamethoxazole (TMP-SMX), or β-lactams within 4 days of BCC (Day 0–4) were defined as effective oral therapy if the pathogen was in vitro susceptible. Oral agents were not included in defining PDAT by DOOR-MAT because (1) these could not be ranked by spectrum of activity within context of β-lactams, (2) there remains controversy regarding the appropriate use of oral β-lactams for GN-BSI, and (3) for oral TMP-SMX and FQ, these are often deemed preferred to transition to oral therapy in GN-BSI.

Patient, hospital, and visit characteristics

Baseline patient characteristics and hospital characteristics including geographical region (i.e., Midwest, Northeast, South, or West), hospital size, urbanicity of served population (rural vs urban), and teaching status were provided by the hospitals. Charlson-Deyo comorbidities were identified during index visit and any visit to the same hospital within 6 months prior to the index visit using International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) and International Classification of Diseases, 10th Revision, Procedural Coding System (ICD-10-PCS) codes (Supplement eTable 2) and Charlson-Deyo Comorbidity Index (CCI) score was calculated using a previously validated method. ^{Reference Rosenthal, Cao and Chung22,Reference Deyo, Cherkin and Ciol23} The 3M All Patient Refined Diagnosis Related Groups (APR-DRG) classification of Severity of Illness (SOI) and Risk of Mortality (ROM) measures were used. APR-DRG SOI and ROM are a previously validated method of estimating the extent of physiologic decompensation and the likelihood of in-hospital death using four subclass levels (minor, moderate, major, and extreme). ^{Reference Averill, Goldfield, Muldoon, Steinbeck and Grant24,Reference Romano and Chan25}

DOOR-MAT agents and oral agents were identified using hospital chargemaster descriptions. Time to first AST results was calculated by subtracting the time of BCC from the time of result observation. All AST results were available as S (susceptible), I (intermediate), or R (resistant).

Statistical analysis

Descriptive statistics were used to present baseline patient and hospital characteristics of GN-BSI patients and their outcomes. Continuous variables were reported as mean (standard deviation) or median (1^st quartile, 3^rd quartile), and categorical variables were reported as counts and percentages. For statistical difference between two groups, the Student’s t-test or Mann–Whitney test was used for continuous variables, as indicated, and Pearson’s χ ^{Reference Schoneweck, Schmitz and Rissner2} test or Fisher’s exact test for categorical variables. All analyses and figures were performed and generated using R v.3.6.3 or higher (R Foundation for Statistical Computing, Vienna, Austria).