Data source
All data were extracted from electronic health record system. General and demographic patients characteristics were collected: age, sex, admitted departments, comorbidities (respiratory disease, hypertension, coronary heart disease, congestive heart failure, diabetes mellitus, renal failure, HBsAg positive, fatty liver, HIV, liver cirrhosis, autoimmune disease, transplant recipient, cancer, long-term use of hormones or immunosuppressants), smoking, alcohol drinking, previous bloodstream infection, fever, suspected site of infection (respiratory, urinary tract, skin or soft tissue, blood, abdomen, multisite, catheter-related bloodstream infection, other). To account for severity of disease, we performed Sequential [Sepsis-related] Organ Failure Assessment (SOFA) score for each subject, clinically assessed whether they met the diagnostic criteria for sepsis/septic shock, and recorded the use of vasopressor, the use of colloid, mechanical ventilation, white blood cell count, neutrophil count (proportion of neutrophils), lymphocyte count, platelet count, hemoglobin, procalcitonin (PCT), C-reactive protein, erythrocyte sedimentation rate, prothrombin time, activated partial thromboplastin time, D-dimer, urea, serum creatinine, cystatin-C, alanine aminotransferase, aspartate aminotransferase, total bilirubin, albumin, troponin, brain natriuretic peptide, creatine kinase, lactate dehydrogenase, serum lactate, and Glasgow Coma Scale. ICU care, ICU length of stay (LOS), hospital LOS, hospital costs and patient outcomes were also included. Patient outcomes included cure/improvement, death, and discharge against medical advice. In a previous large national study, 1–2% inpatients discharge against medical advice, and increasing their risk of hospital readmission, morbidity, and mortality15. Therefore, we included discharge against medical advice and death in the analysis of poor prognosis of patients.
Microbiology data contained pathogens and antimicrobial susceptibility testing report. Bacteria were classified by Gram staining. We collected the empirical use of antibiotics during the hospitalization of each patient and evaluated the rationality of the antibiotic strategy based on the drug sensitivity test report, of which intermediate susceptibilities were treated as non-susceptible. The types of bacterial sensitivity could be classified into full sensitivity, single antibiotic resistance, two kinds of antibiotic resistance, multidrug-resistant organism (MDRO), extensively drug resistant (XDR), and pandrug resistant (PDR). Many bacteria have intrinsic antibiotic resistance (e.g., ceftriaxone and Pseudomonas aeruginosa) and were considered insensitive. When not reported for the antibiotic or antibiotics received, in-vitro susceptibility or resistance was imputed from interpretations that were reported within the same antibiotic category. For example, a Gram-negative organism susceptible to ceftriaxone may not have susceptibilities reported to all higher-generation cephalosporins (i.e., cefepime), but these agents can be safely used9.
Definitions
EAT was defined as antibiotic treatment prior to the drug sensitivity tests report was obtained (at least 24 hours after sampling)16. Appropriate and necessarily empirical antibiotic therapy (ANEAT) was defined as patient was treat with empiric antibiotics, and the antibiotic regimen was active against the identified pathogen based on in susceptibility testing17. AUEAT was defined as the patient received empiric antibiotics and anti-methicillin-resistant Staphylococcus aureus (MRSA) antibiotics, anti-vancomycin-resistant enterococci (VRE) antibiotics, anti-Pseudomonas β-lactam, or carbapenem, but none of these drug-resistant bacteria (MRSA, VRE, ceftriaxone-resistant Gram-negative organisms (CTX-RO), or extended-spectrum β-lactamase (ESBL) Gram-negative organism) was cultured at any infected site9. IEAT was defined as the patient was treated with empiric antibiotics, but at least one pathogen isolated from any clinical culture site was not sensitive to all antibiotics used6,9. MDRO is defined as non-susceptibility to at least one agent in three or more antimicrobial categories18, including MRSA, VRE and certain Gram-negative bacilli19. XDR is defined as non-susceptibility to at least one agent in all but two or fewer antimicrobial categories (i.e., bacterial isolates remain susceptible to only one or two categories) 18. PDR is defined as non-susceptibility to all agents in all antimicrobial categories (i.e., no agents tested as susceptible for that organism) 18. Sepsis was defined according to Sepsis-3 criteria20 as life-threatening organ dysfunction, which can be represented by an increase in the SOFA score of 2 points or more, caused by a dysregulated host response to infection. Septic shock was defined as a subset of sepsis, which used vasopressor requirement to maintain a mean arterial pressure of 65 mmHg or greater and serum lactate level greater than 2 mmol/L (> 18 mg/dL) in the absence of hypovolemia.20.
Data statistics
We calculated the overall and categorical prevalence of EAT. Based on our electronic health record system, we collected a high quality of data, with < 1% missing data across demographic variables. We did not conduct formal sample size calculations, and all available data were used to maximize the power. Univariate analysis was performed on the collected baseline characteristics of patients (ANEAT vs AUEAT; ANEAT vs IEAT). Bivariate associations were assessed using the Pearson chi-square test or Fisher exact test for categorical variables and the students t-test or the Mann-Whitney U test for continuous variables. After the initial descriptive analysis of the cohort, we estimated the proportion of each covariate in different groups. To assess the relative factors of EAT with adjusted odds ratios (OR), all variables in the univariate analysis with P < 0.1 and priori variables were included in a enter multivariable logistic regression model. A priori we chose the following variables: sepsis or septic shock. These variables were included in each model irrespective of their statistical significance since they were considered having an important impact on outcome indicators in the clinical. In addition, we tried to include variables with P < 0.1, variables with P < 0.2 and re-screening variables without considering their statistical significance into the logistic regression model, and we finally obtained similar statistical results as before.
All tests of significance used a 2-sided P < 0.05. Analyses were conducted using IBM SPSS Statistic version 25.0.