Study design, setting, and participants
This was a retrospective cohort study, conducted in a university-affiliated medical center (Cathay General Hospital, Taipei) in Northern Taiwan with 40 ED beds and 800 ward beds and approximately 55,000 patients visiting annually. The study period was between January 01, 2014 and December 31, 2017. Patients who fulfilled the following inclusion criteria were included: (1) age above 18 years, (2) suspected bacterial infection, and (3) bacterial culture from blood, sputum or urine, ordered and obtained in the ED. Patients who were transferred from other hospitals, had out-of-hospital cardiac arrest, were pregnant, or had mixed infections were excluded.
Definition of variables and primary outcome
Suspected bacterial infection is identified by: (1) physician’s clinical judgment through chart review and infection related disease codes, or (2) ED clinical parameters that indicate infection, such as severe inflammatory response syndrome (SIRS) and qSOFA score. SIRS is defined as a heart rate > 90 beats per minute, respiratory rate > 20 breaths per minute, temperature < 36°C or > 38°C, white blood cell count < 4000 / mm3 or > 12 000 / mm3, and band form > 10% . qSOFA score is defined as systolic blood pressure ≤ 100 mmHg, respiratory rate ≥ 22 breaths per minute, and Glasgow Coma Scale < 15 . Positive cultures were defined as: (1) Positive blood culture: at least two bottles of blood culture yielding the same pathogen . Two sets of blood cultures (two aerobic bottles, two anaerobic bottles), were collected from each patient via peripheral venipuncture at two sites, with a 30-minute interval between sample collections. (2) Positive sputum culture: pathogen growth in sputum specimens with fewer than 25 squamous epithelial cells per low-power field . Sputum specimen was indicated if a patient had clinical signs, such as productive cough, purulent sputum, dyspnea, or desaturation. Sputum culture was collected by either patient’s expectorated sputum in a sterile container after deep coughs, nasotracheal or orotracheal aspiration, or endotracheal tube aspiration. (3) Positive urine culture: pathogen growth > 105 colony-forming units (CFU) per milliliter in clean-catch midstream urine specimens . Patients with symptoms of dysuria, frequency, urgency, hematuria, suprapubic pain, fever, chills, or flank pain, with or without costovertebral tenderness, were indicated for urine specimen collection. The methods of urine culture collection included self-collection techniques, urethral catheterization, and suprapubic puncture.
Contaminants in cultures were determined as follows: a single positive blood culture with growth of pathogens that represent contamination, including coagulase-negative staphylococci, Corynebacterium species, Bacillus species, Propionibacterium acnes, Micrococcus species, viridans group streptococci, enterococci, and Clostridium perfringens ; urine culture with more than two isolates at greater than or equal to 10 000 CFU/mL ; positive sputum cultures in which the isolates consisted of normal oral flora, including Neisseria catarrhalis, Candida albicans, diphtheroids, alpha-hemolytic streptococci, and some staphylococci.
Sepsis is defined as life-threatening organ dysfunction caused by a dysregulated host response to infection, where organ dysfunction can be identified by sepsis-related organ failure assessment (SOFA) score ≥2. However, a quick sepsis-related organ failure assessment (qSOFA) score does not require laboratory tests and can be assessed quickly, therefore, in the setting of an emergency department. We used a qSOFA score ≥ 2 to define sepsis 
The diagnosis of pneumonia was based on both clinical presentation of productive cough, chest pain, fever, and dyspnea, and pulmonary infiltration visible on the chest image. The indications for chest radiographies were based on both physician’s clinical judgement and patient’s clinical presentation, including productive cough, chest pain, fever, and dyspnea.
Data collection and assignment to case and control groups
The retrospective chart review method was used to acquire data of patients who fulfilled the inclusion criteria. Demographic characteristics, including vital signs (obtained at the ED triage), laboratory data, infection sites, cultured microorganisms, qSOFA scores, SIRS criteria, and clinical outcomes, were obtained by an emergency physician (Table 1). In total, 903 bacteria-infected ED patients were initially recruited, with a total of 797 patients included in the study. Exclusions (106 patients) were made for insufficient data, presence of mixed infections, occurrence of an out-of-hospital cardiac arrest, transferal of patients treated at other hospitals, or pregnant patients (Figure 1). The recruited patients were further divided into two groups based on the culture result, with 278 patients assigned to the GNB group and 519 patients to the non-GNB group. All variables were compared between the two groups, and the accuracy of clinical parameters to predict GNB infection were also analyzed.
This study was approved by the institutional Review Board of the Cathay General Hospital and was conducted according to the Declaration of Helsinki. This was an observational study, and the patients’ data were fully anonymized; the need for informed consent from the patients was waived.
We used SPSS 23.0 for Mac (SPSS Inc., Chicago, IL, USA) to perform statistical analysis. Continuous normally distributed data are presented as mean ± standard deviation (SD), while continuous data that were not normally distributed are presented as median +/- interquartile range (IQR). Categorical variables were presented as percentages. Independent samples t-test, the Mann–Whitney, or Wilcoxon test were used to analyze continuous variables. Pearson’s chi-square test or Fisher’s exact test was used for categorical variables. Univariate analysis was performed to evaluate the prediction of GNB infection among the four clinical parameters that showed significant difference (p < 0.05) between the GNB and the non-GNB group (Table 2). We calculated the odds ratio of GNB infection in patients with suspected bacterial infection, using 2, 3, or 4 positive clinical parameters. The optimal cut-off point of each clinical parameter used to predict GNB infection was calculated via Youden index. The area under the receiver operating characteristic curve (AUROC) was then used to evaluate GNB prediction discrimination ability (Table 3). Sensitivity, specificity, positive predictive value, and negative predictive value were also calculated to determine the diagnostic accuracy of the four clinical parameters in predicting GNB infection (Supplementary table 3).