Abstract
Background
Bacteremia is a life-threatening condition with a high mortality rate in critical care and emergency settings. The current study investigated the trend of mortality and developed predictive models of mortality for adults with bacteremia at emergency departments (EDs).
Methods
We conducted a retrospective cohort study of adults with bacteremia at the ED of China Medical University Hospital. Patient data were obtained from the Clinical Research Data Repository, and mortality information was obtained from the National Death Registry. We developed a new model to predict 7-day mortality in the derivation population and compared the model performance of the new model with Pitt Bacteremia Score (PBS) and Bloodstream Infection Mortality Risk Score (BSIMRS) in the validation population.
Results
We identified 14625 adult patients with first-time bacteremia at the ED, of whom 8.4% died within 7 days. From 2003 to 2016, both the cumulative incidence and 7-day mortality rate of bacteremia decreased significantly. The ED bacteremia mortality (ED-BM) model included PBS parameters, age, infection source, baseline steroid use, and biochemical profiles (estimated glomerular filtration rate, platelet, blood urea nitrogen, potassium, and hemoglobin) for predicting 7-day mortality. The discrimination performance of the ED-BM model (area under curve [AUC], 0.903) was significantly better than that of PBS (AUC, 0.848) or BSIMRS (AUC, 0.885).
Conclusions
Although the cumulative incidence and mortality of ED bacteremia decreased, its mortality burden remains critical. The proposed ED-BM model had significantly better model performance than other scoring systems in predicting short-term mortality for adult patients with bacteremia at EDs.
Keywords: bacteremia, emergency department, epidemiology, mortality, prediction model
Bloodstream infection (BSI) or bacteremia is a life-threatening condition with a high mortality rate (14%–37%), particularly among patients in critical care units [1]. Similarly, a significantly high mortality rate (5.3%–14.4%) has been observed in emergency department (ED) patients with BSIs or bacteremia [2–5]. Therefore, early identification of the mortality risk of bacteremia is crucial for patients at EDs, a common entry point to hospitals for nearly all patients with acute illnesses.
In the past decade, several studies have investigated the burden of bacteremia among adult ED patients receiving blood culture (BC) [6–9]. In a 1-year study conducted by a medical center in the United States, Chase et al. reported that 12.4% (409/3310) of ED adult patients receiving BC had positive culture results [6]. Similarly, a 1-year study by Kao et al. in Taiwan reported a bacteremia rate of 13.5% (831/6137) [7], and another study in a German university hospital reported the 5-year cumulative incidence of bacteremia to be 14.3% (740/5191) [9]. However, the timelines covered by these studies and other studies reporting the mortality of ED bacteremia [2–5] have often been too short to understand the long-term trend and mortality of ED bacteremia.
Subjective variables such as infection symptoms or terminal illness [2] and objective variables such as biochemical measurements and vitals have been used to predict the mortality risk of ED patients with suspected sepsis or bacteremia [2–4, 10]. The Pitt Bacteremia Score (PBS), consisting of 5 variables (mental status, presence of fever, hypotension, requirement for respiratory support, and cardiac arrest), and BSI mortality risk score (BSIMRS), consisting of PBS and 3 additional clinical conditions (malignancy, cirrhosis, and BSI source), are the 2 scoring systems that use objective variables to predict mortality risk in patients with bacteremia or nonbacteremia conditions [11–14]. Most studies on ED patients with bacteremia or sepsis have evaluated the 28-day mortality rate, a rather long-term mortality evaluation for such an acute event [2–4, 15]. During the 28-day period, other medical conditions, such as health care–associated infections, may influence the outcomes of these patients [16]. Short-term mortality, such as 7-day mortality, represents the direct effect of bacteremia on the outcomes of ED patients. To address this research lacuna, we conducted a 14-year retrospective cohort study in the ED setting of a tertiary medical center to determine the trend in cumulative incidence and mortality rate of bacteremia, and we proposed a new predictive algorithm for the 7-day mortality rate. The performance of our model and that of the existing scoring systems (PBS and BSIMRS) were compared for predicting short-term mortality among adult ED patients with bacteremia.
METHODS
Source Population
The Big Data Center and the Office of Information Technology of China Medical University Hospital (CMUH) established the CMUH Clinical Research Data Repository (CMUH-CRDR), which contains the medical records of 2988912 patients who sought medical care at CMUH between January 1, 2003, and December 31, 2019. Details of the data quality and interoperability have been published elsewhere [17–19]. This study was approved by the Big Data Center and Research Ethical Committee/Institutional Review Board of CMUH (CMUH105-REC3–068), and informed consent has been waived.
Definition of Bacteremia at the ED
The index ED admission was ED admission in which BCs were obtained and subsequently yielded positive results during the study period. We treated consecutive ED visits that were ≤3 days apart as a single ED visit. Clinically significant bacteremia was defined as at least 1 blood specimen that grew noncommensal microorganisms. For BCs that grew commensal microorganisms, at least 2 positive culture results on separate occasions during the index ED admission were required to be counted as clinically significant bacteremia (Supplementary Table 1) [20].
Study Population
From the CMUH-CRDR, we analyzed all bacterial BC reports from 2003 and 2016 except those of (1) cultures obtained during outpatient or inpatient visits, (2) cultures obtained >48 hours after the ED admission, (3) culture results that did not meet the definition of clinically significant bacteremia, and (4) patients aged <18 years (Figure 1). The included patients were then randomly divided into derivation and validation populations at a ratio of 8:2 for the subsequent analysis. The clinical characteristics were similar between the derivation and validation populations (Supplementary Table 2). Detailed information on the selection process is summarized in Figure 1.
Figure 1.
Selection process of the study population of patients admitted to the emergency department (ED) with bacteremia from 2003 to 2016 at China Medical University Hospital (CMUH).
Covariates and Mortality Outcome
Baseline demographic information, comorbidities, vital signs, biochemical profiles, and medication history were verified from the CMUH-CRDR. Comorbidities were defined by the International Classification of Diseases (ICD) diagnosis codes recorded within 1 year before the index ED admission, if not otherwise defined (Supplementary Table 3). Furthermore, medication history of the use of systemic antibiotics, steroids, and nonsteroidal anti-inflammatory drugs (NSAIDs) within 14 days before the index ED admission was extracted. Baseline vital signs and biochemical profiles were the first measurements conducted within 48 hours following the index ED visit date. Infection source was defined by ICD diagnosis during the index ED admission or the following hospitalization (Supplementary Table 4). Empirical antibiotic treatment was defined as “appropriate” if the minimal inhibitory concentration tests revealed that the bacterial isolate was susceptible to the given antimicrobial agent(s) during the index ED visit (Supplementary Figure 1); otherwise, they were regarded as “inappropriate.”
The primary outcome of interest was all-cause mortality within 7 days following the index ED visit date. We traced each patient’s mortality outcome by linking the CMUH-CRDR to Taiwan’s National Death Registry Database. All patients enrolled in the CMUH-CRDR were followed up until December 31, 2019, or death, whichever occurred earlier. We used PBS and BSIMRS as the reference prediction scoring systems (Supplementary Table 5) [12, 21]. We obtained data for each component recorded within 48 hours following the ED admission and calculated PBS and BSIMRS.
Statistical Analysis
Values for continuous and categorical variables are expressed as median (interquartile range) and frequency (%), respectively. Annual cumulative incidence of ED bacteremia was calculated by dividing the number of ED bacteremia episodes per year by the total number of ED visits with BC obtained per year. The yearly proportion of 7-day mortality was calculated by dividing the number of deaths per year by the total number of ED bacteremia episodes per year. The trend during the study period was tested using the Cochran-Armitage trend test, and the age-adjusted trend was analyzed using logistic regression.
To establish a multivariable prediction model for the 7-day mortality rate, we used the derivation population to identify predictors and build the model. We then tested the predictive performance of the model in the validation population. We first chose the input variables for multivariable variable selection based on clinical relevance. We then identified variables with a P value <.25 in the univariable analysis and input them into backward, forward, and stepwise selection methods in multiple logistic regression models to identify variables that were associated with the 7-day mortality rate. From the 17 variables in common, we excluded aspartate aminotransferase due to data unavailability (missing rate >20%) and excluded implants and red blood cell (RBC) because implants information was recorded in infection source variable and because RBC and hemoglobin were similar indicators. We added age and liver cirrhosis back to the model because of clinical relevance and because this variable is included in the BSIMRS model. Finally, the ED bacteremia mortality (ED-BM) model was developed, and it included age, temperature, hypotension, mechanical ventilation, cardiac arrest, mental status, cancer, liver cirrhosis, infection source, steroid use, estimated glomerular filtration rate (eGFR), platelet, blood urea nitrogen (BUN), red blood cell volume distribution width (RDW), serum potassium, and hemoglobin (Supplementary Figure 2). The variable types and the beta estimates for each variable in the ED-BM model are specified in Supplementary Table 6.
The cutoffs for eGFR, platelet, BUN, RDW, serum potassium, and hemoglobin were determined based on clinical practice and statistical evidence of sensitivity, specificity, and Youden’s index (Supplementary Table 7). We also performed multicollinearity diagnostics of variables included in the ED-BM model by variance inflation factor (VIF) and tolerance. All variables included in the ED-BM model had VIF <10 and tolerance >0.1, showing no evidence of multicollinearity (Supplementary Table 8).
We compared the discrimination and calibration performance of the ED-BM model with those of PBS and BSIMRS in the validation population using the area under curve (AUC) and calibration plots [22]. Furthermore, we estimated the net reclassification index to determine the improvement in prediction of 7-day mortality risk of ED bacteremia when using our ED-BM model compared with PBS and BSIMRS. All statistical analyses were performed using SAS, version 9.4 (SAS Institute Inc., Cary, NC, USA), and R, version 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria). A 2-sided statistical significance level of α was set at .05.
RESULTS
Trends and Characteristics of Adult Patients With Bacteremia
During 2003 and 2016, a declining trend of cumulative incidence of ED bacteremia was observed, with a peak of 13.48% in 2006 and a nadir of 8.84% in 2015. A similar trend was observed for both males and females (Ptrend < .0001) (Figure 2A). The trend of the 7-day mortality rate peaked in 2004 (11.25%) and then significantly decreased in 2016 (6.83%; Ptrend = .009; age-adjusted Ptrend = .002) (Figure 2B). Overall, 14625 adult patients with first-time bacteremia at CMUH ED were included in the present study, and 80% of them (11700 patients) were randomly selected as the derivation population for model establishment; the remaining patients (2925 patients) were included as the validation population for model performance evaluation. In the derivation population, the median age was 65.5 years and 48.6% were men. Renal insufficiency (55.6%), hypertension (24.1%), cancer (18.7%), and diabetes mellitus (17.2%) were the most common comorbidities (Table 1). Genitourinary system (36.0%) was the most common source of bacteremia, followed by the respiratory (18.0%) and digestive (16.2%) systems (Supplementary Figure 3). Gram-negative bacteria (73.4%) were the predominant pathogens. The most common bacteremia pathogens were Escherichia coli (38.3%), Klebsiella pneumoniae (12.5%), Staphylococcus aureus (11.7%), polymicrobial (7.48%), and Pseudomonas aeruginosa (2.34%) (Supplementary Table 9). The clinical and microbiological characteristics of the validation population did not differ significantly from those of the derivation population (Supplementary Tables 2 and 9).
Figure 2.
Trends of the (A) cumulative incidence of bacteremia and (B) 7-day mortality rate of bacteremia in the emergency department (ED) from 2003 to 2016 at China Medical University Hospital (CMUH).
Table 1.
Demographic and Clinical Characteristics of 11700 Patients With First-Time Bacteremia Identified at the Emergency Department in the Derivation Population
| Characteristics | Available No. | Patients With Bacteremiaa | 7-Day Mortality | ||
|---|---|---|---|---|---|
| Survivors | Nonsurvivors | ||||
| n = 11700 | n = 10702 (91.5%) | n = 998 (8.53%) | P Value | ||
| Age, median (Q1, Q3), y | 65.5 (52.6, 77.0) | 65.1 (52.4, 76.8) | 68.0 (54.6, 79.1) | <.0001 | |
| Male | 11 700 | 5681 (48.6) | 5066 (47.3) | 615 (61.6) | <.0001 |
| Comorbidityb | |||||
| Hypertension | 11 700 | 2819 (24.1) | 2569 (24.0) | 250 (25.1) | .46 |
| Congestive heart failure | 11 700 | 798 (6.82) | 723 (6.76) | 75 (7.52) | .36 |
| Coronary artery disease | 11 700 | 979 (8.37) | 906 (8.47) | 73 (7.31) | .21 |
| Chronic lung disease | 11 700 | 921 (7.87) | 811 (7.58) | 110 (11.0) | .0001 |
| Diabetes mellitus | 11 700 | 2015 (17.2) | 1838 (17.2) | 177 (17.7) | .65 |
| Liver cirrhosis | 11 700 | 723 (6.18) | 619 (5.78) | 104 (10.4) | <.0001 |
| Renal insufficiency (eGFR <60mL/min/1.73 m2) | 11 693 | 6506 (55.6) | 5658 (52.9) | 848 (85.0) | <.0001 |
| Rheumatic disease | 11 700 | 158 (1.35) | 141 (1.32) | 17 (1.70) | .31 |
| Transplantation | 11 700 | 74 (0.63) | 72 (0.67) | 2 (0.20) | .07 |
| Cancer | 11 700 | 2187 (18.7) | 1834 (17.1) | 353 (35.4) | <.0001 |
| AIDS/HIV | 11 700 | 45 (0.38) | 40 (0.37) | 5 (0.50) | .53 |
| Medication use within 14 d before index ED admission | |||||
| Recent systemic antibiotic use | 11 700 | 1005 (8.59) | 909 (8.49) | 96 (9.62) | .22 |
| Recent steroid use | 11 700 | 1200 (10.3) | 974 (9.10) | 226 (22.7) | <.0001 |
| Recent NSAID | 11 700 | 823 (7.03) | 732 (6.84) | 91 (9.12) | .007 |
| Medical history before index ED admission | |||||
| Dialysis within 30 d prior | 11 700 | 289 (2.47) | 267 (2.49) | 22 (2.20) | .57 |
| Hospitalization within 90 d prior | 11 700 | 2700 (23.1) | 2354 (22.0) | 346 (34.7) | <.0001 |
| Implants in placed within 90 d prior | 11 700 | 481 (4.11) | 448 (4.19) | 33 (3.31) | .18 |
| Central venous catheter in placed within 14 d prior | 11 700 | 27 (0.23) | 22 (0.21) | 5 (0.5) | .06 |
| Physiological measurement within 48h after the index ED admission (first) | |||||
| Systolic blood pressure, mmHg | 11 146 | 127 (108, 149) | 128 (110, 150) | 108 (88, 131) | <.0001 |
| Heart rate, /min | 11 179 | 108 (92, 122) | 107 (92, 121) | 112 (92, 132) | <.0001 |
| Respiratory rate, /min | 9150 | 20 (20, 22) | 20 (20, 22) | 22 (20, 26) | <.0001 |
| Ear temperature, °C | 11 343 | 38.2 (37.1, 39.1) | 38.2 (37.2, 39.1) | 36.9 (36, 38.1) | <.0001 |
| qSOFA, median (Q1, Q3) | 8926 | 1 (0, 1) | 0 (0, 1) | 1.5 (1, 2) | <.0001 |
| 0 | 4266 (47.8) | 4186 (50.7) | 80 (11.8) | ||
| 1 | 3137 (35.1) | 2879 (34.9) | 258 (38.2) | ||
| 2 | 1267 (14.2) | 1018 (12.3) | 249 (36.8) | ||
| 3 | 256 (2.87) | 167 (2.02) | 89 (13.2) | ||
| Components of Pitt Bacteremia Scorec | |||||
| Ear temperature | 11 343 | .70 | |||
| 36.1°C–38.9°C | 7247 (61.9) | 6692 (64.0) | 555 (62.6) | ||
| 35.1°C–36.0°C or 39.0°C–39.9°C | 2985 (25.5) | 2743 (26.2) | 242 (27.3) | ||
| ≤35°C or ≥ 40°C | 1111 (9.50) | 1021 (9.76) | 90 (10.2) | ||
| Hypotension | 11 700 | 3429 (29.3) | 2663 (24.9) | 766 (76.8) | <.0001 |
| Mechanical ventilation | 11 700 | 1865 (15.9) | 1308 (12.2) | 557 (55.8) | <.0001 |
| Cardiac arrest | 11 700 | 569 (4.86) | 233 (2.18) | 336 (33.7) | <.0001 |
| Mental status | 11 688 | <.0001 | |||
| Alert (GCS 15) | 9137 (78.1) | 8693 (81.3) | 444 (44.6) | ||
| Disoriented (GCS 9–14) | 1554 (13.3) | 1302 (12.2) | 252 (25.3) | ||
| Stuporous (GCS 4–8) | 657 (5.62) | 520 (4.86) | 137 (13.8) | ||
| Comatose (GCS 3) | 340 (2.91) | 178 (1.66) | 162 (16.3) | ||
| Pitt Bacteremia Score | 11 343 | 1 (0, 2) | 1 (0, 2) | 5 (2, 8) | <.0001 |
| Bloodstream Infection Mortality Risk Score | 11 343 | 4 (2, 7) | 4 (2, 6) | 9 (7, 9) | <.0001 |
| Biochemical profiles within 48h after the index ED admission (first) | |||||
| hs-CRP, mg/dL | 11 522 | 11.0 (3.35, 21.1) | 10.3 (3.02, 20.2) | 18.3 (8.51, 28.0) | <.0001 |
| White blood cell count, ×103/μL | 11 687 | 11.9 (7.89, 16.7) | 11.9 (8.10, 16.5) | 10.6 (3.97, 18.5) | <.0001 |
| Neutrophil-to-lymphocyte ratio | 11 611 | 13.2 (7.37, 22.8) | 13.3 (7.56, 22.7) | 11.9 (4.83, 24.1) | <.0001 |
| Neutrophil band | 410 | 7.0 (3.0, 16.0) | 6.0 (2.5, 15.0) | 8.0 (3.0, 18.0) | .14 |
| Platelet, ×103/μL | 11597 | 171 (115, 234) | 174 (121, 236) | 113 (47.7, 211) | <.0001 |
| Red blood cell, ×106/μL | 9732 | 4.02 (3.40, 4.52) | 4.06 (3.47, 4.55) | 3.44 (2.85, 4.00) | <.0001 |
| Blood urine nitrogen, mg/dL | 10 947 | 20 (13, 37) | 19 (13, 33) | 45 (27, 72) | <.0001 |
| Serum creatine, mg/dL | 11 651 | 1.20 (0.88, 2.00) | 1.16 (0.85, 1.84) | 2.10 (1.37, 3.55) | <.0001 |
| eGFR, mL/min/1.73 m2 | 11 651 | 54.1 (28.9, 82.4) | 56.9 (31.5, 84.0) | 28.1 (14.8, 47.0) | <.0001 |
| Serum sodium, mmol/L | 11 593 | 134 (131, 137) | 134 (131, 137) | 133 (128, 138) | .0008 |
| Serum potassium, mmol/L | 11 623 | 3.7 (3.4, 4.2) | 3.7 (3.3, 4.1) | 4.2 (3.6, 5.0) | <.0001 |
| AST, IU/L | 8703 | 40 (27, 79) | 38 (26, 72) | 66 (38, 158) | <.0001 |
| ALT, IU/L | 9367 | 30 (18, 57) | 29 (18, 56) | 38 (22, 73) | <.0001 |
| RDW, % | 9680 | 14.1 (13.2, 15.7) | 14.0 (13.1, 15.5) | 15.8 (14.3, 18.1) | <.0001 |
| Glucose, mg/dL | 11 603 | 145 (117, 212) | 145 (118, 210) | 143 (103, 228) | .0002 |
| Lactate, mg/dL | 4433 | 26.2 (16.1, 45.9) | 23.8 (14.9, 39.5) | 54.5 (28.5, 99.8) | <.0001 |
| Hemoglobin, g/dL | 11 641 | 11.9 (10.1, 13.4) | 12.1 (10.2, 13.5) | 10.4 (8.6, 12.2) | <.0001 |
| Infection sourced | 11 700 | <.0001 | |||
| Genitourinary | 4214 (36.0) | 4100 (38.3) | 114 (11.4) | ||
| Respiratory | 2108 (18.0) | 1727 (16.1) | 381 (38.2) | ||
| Digestive | 1891 (16.2) | 1796 (16.8) | 95 (9.52) | ||
| Skin and soft tissue | 838 (7.16) | 782 (7.31) | 56 (5.61) | ||
| Nervous system | 149 (1.27) | 137 (1.28) | 12 (1.20) | ||
| Musculoskeletal | 112 (0.96) | 109 (1.02) | 3 (0.30) | ||
| Circulatory | 126 (1.08) | 116 (1.08) | 10 (1.00) | ||
| Primary bacteremia | 2222 (19.0) | 1896 (17.7) | 326 (32.7) | ||
| Implant | 34 (0.29) | 33 (0.31) | 1 (0.10) | ||
| Unclassified | 6 (0.05) | 6 (0.06) | 0 (0) | ||
| Pathogens causing bacteremia | 11 700 | <.0001 | |||
| Gram-negative | 8589 (73.4) | 7903 (73.9) | 686 (68.7) | ||
| Gram-positive | 2849 (24.4) | 2587 (24.2) | 262 (26.3) | ||
| Both | 262 (2.24) | 212 (1.98) | 50 (5.01) | ||
| Resistant pathogens causing bacteremia | |||||
| Methicillin-resistant Staphylococcus aureus | 1470 | 574 (39.1) | 493 (37.9) | 81 (47.9) | .01 |
| Vancomycin-resistant Enterococcus | 354 | 34 (9.60) | 30 (9.49) | 4 (10.5) | .84 |
| Carbapenem-resistant Enterobacteriaceae | 7245 | 81 (1.12) | 76 (1.13) | 5 (0.96) | .71 |
| Ceftriaxone-resistant Enterobacteriaceae | 7245 | 809 (11.2) | 737 (11.0) | 72 (13.8) | .05 |
| Antibiotic treatment during episodee | 11 700 | ||||
| Inappropriate use | 4318 (36.9) | 3924 (36.67) | 394 (39.48) | .08 | |
| Time to antibiotic use, h | 9946 | 1.70 (1.12, 2.60) | 1.75 (1.17, 2.65) | 1.28 (0.80, 1.93) | <.0001 |
Abbreviations: AST, aspartate aminotransferase; ALT, alanine transaminase; ED, emergency department; eGFR, estimated glomerular filtration rate; GCS, Glasgow Coma Scale; hs-CRP, high-sensitive C-reactive protein; ICD, International Classification of Diseases; NSAID; nonsteroid anti-inflammatory drug; qSOFA, quick Sepsis-Related Organ Failure Assessment; RDW, red blood cell volume distribution width.
Values for the continuous variables are expressed as median (interquartile range), and values for the categorical variables are expressed as frequency (%).
Comorbidities were defined as having ICD diagnosis within 1 year before the index ED admission, if not otherwise defined. Rheumatic diseases, transplantation, and cancer were defined as having the catastrophic illness diagnosis any time before the index ED admission (Supplementary Table 3).
We obtained data for each component recorded within 48 hours after the ED admission and calculated the Pitt Bacteremia Score.
Infection source was defined as the ICD diagnosis during the index ED episode or the following hospital admission (Supplementary Table 4).
Inappropriate antibiotic was defined as when the susceptibility results did not prove that bacteria were susceptible to the given antibiotic (Supplementary Figure 1).
Prediction Model for 7-Day Mortality (Derivation Population)
The pooled 7-day mortality of ED bacteremia was 8.53% in the derivation population. Table 1 summarizes the differences between patients who survived ≥7 days (survivors) and those who died within 7 days (nonsurvivors) after ED admission. Compared with the survivors, nonsurvivors were more likely to be older and men; to have chronic lung disease, liver cirrhosis, renal insufficiency, and cancer; and to have a history of steroid use, NSAID use, or recent hospitalization. Furthermore, the quick Sepsis-Related Organ Failure Assessment (qSOFA) score, PBS, and BSIMRS were more frequently higher among nonsurvivors than among survivors. Similarly, the levels of high-sensitive C-reactive protein (hs-CRP), BUN, serum creatinine, RDW, serum potassium, and liver enzymes were more likely to be higher in the nonsurvivors than in the survivors. Contrarily, the values of white blood cell count, neutrophil-to-lymphocyte ratio, platelet count, and hemoglobin were significantly lower among the survivors than among the nonsurvivors. The nonsurvivors were marginally more likely to receive inappropriate antibiotic treatment than the survivors (39.5% vs 36.7%; P = .08) (Table 1).
We then established the ED-BM model for predicting 7-day mortality of ED bacteremia. This model contained the following variables: variables derived from PBS and BSIMRS (Supplementary Table 5), steroid use, eGFR, platelet, BUN, RDW, potassium, and hemoglobin (Table 2). The variable with the most significant effect size was cardiac arrest (odds ratio [OR], 7.29), followed by comatose (Glasgow Coma Scale of 3; OR, 3.62), hypotension (OR, 2.74), infection source of nongenitourinary or central venous catheter (OR, 2.62), cancer (OR, 2.58), BUN >30mg/dL (OR, 2.29), platelet <100 × 103/μL (OR, 2.14), steroid use (OR, 1.62), mechanical ventilation (OR, 1.58), age ≥85 (OR, 1.55), eGFR <45mL/min/1.73 m2 (OR, 1.53), and RDW >14.5% (OR, 1.47).
Table 2.
Multivariable Logistic Regression Models of 7-Day All-Cause Mortality in the Derivation Population (n = 8808)
| Variables | PBSa | BSIMRSb | ED-BMc |
|---|---|---|---|
| OR (95% CI) | OR (95% CI) | OR (95% CI) | |
| Age | |||
| <65 y | Reference | ||
| 65–84 y | 0.95 (0.78–1.17) | ||
| ≥85 y | 1.55 (1.13–2.14) | ||
| Ear temperature | |||
| 36.1°C–38.9°C | Reference | Reference | Reference |
| 35.1°C–36.0°C or 39.0°C–39.9°C | 0.91 (0.74–1.11) | 0.95 (0.77–1.17) | 0.92 (0.74–1.14) |
| ≤35.0°C or ≥ 40.0°C | 0.82 (0.61–1.11) | 0.94 (0.68–1.28) | 1.02 (0.73–1.42) |
| Hypotension | 4.00 (3.23–4.95) | 3.76 (3.03–4.68) | 2.74 (2.18–3.44) |
| Mechanical ventilation | 1.92 (1.57–2.36) | 1.90 (1.54–2.36) | 1.58 (1.26–1.98) |
| Cardiac arrest | 6.43 (5.04–8.20) | 6.53 (5.07–8.40) | 7.29 (5.57–9.56) |
| Mental status | |||
| Alert (GCS 15) | Reference | Reference | Reference |
| Disoriented (GCS 9–14) | 2.46 (1.98–3.04) | 2.45 (1.97–3.06) | 1.81 (1.43–2.29) |
| Stuporous (GCS 4–8) | 2.35 (1.76–3.14) | 2.48 (1.83–3.34) | 1.90 (1.37–2.62) |
| Comatose (GCS 3) | 4.75 (3.37–6.70) | 4.98 (3.50–7.07) | 3.62 (2.50–5.24) |
| Cancer | 3.24 (2.66–3.94) | 2.58 (2.07–3.23) | |
| Liver cirrhosis | 1.55 (1.15–2.09) | 1.11 (0.81–1.53) | |
| Infection source | |||
| Genitourinary | Reference | Reference | |
| Nongenitourinary or CVC | 2.61 (2.03–3.36) | 2.62 (2.01–3.41) | |
| Steroid use | 1.62 (1.26–2.10) | ||
| eGFR <45mL/min/1.73 m2 | 1.53 (1.19–1.97) | ||
| Platelet <100 ×103/μL | 2.14 (1.75–2.61) | ||
| BUN >30mg/dL | 2.29 (1.78–2.94) | ||
| RDW >14.5% | 1.47 (1.19–1.81) | ||
| Serum potassium <4.0 mmol/L | 0.70 (0.58–0.86) | ||
| Hemoglobin <11.5g/dL | 1.17 (0.95–1.44) | ||
| Model performance | |||
| AIC | 3579.948 | 3533.314 | 3228.458 |
| AUC (95% CI) | 0.850 (0.835–0.865) | 0.883 (0.871–0.895) | 0.909 (0.900–0.919) |
| P difference in AUC | <.0001 | <.0001 | Reference |
Abbreviations: AIC, Akaike information criterion; AUC, area under curve; BSIMRS, Bloodstream Infection Mortality Risk Score; BUN, blood urea nitrogen; CVC, central venous catheter; ED-BM, emergency department bacteremia mortality model; eGFR, estimated glomerular filtration rate; GCS, Glasgow Coma Scale; OR, odds ratio; PBS, Pitt Bacteremia Score; RDW, red cell distribution width.
PBS model included temperature, hypotension, mechanical ventilation, cardiac arrest, and mental status.
BSIMRS model included components of PBS, cancer, liver cirrhosis, and infection source.
ED-BM model included components of PBS and BSIRMS, age, cancer, liver cirrhosis, infection source, eGFR, platelet, BUN, RDW, serum potassium, and hemoglobin.
Discrimination and Calibration Performance of the ED-BM Model
The ED-BM model had good discrimination and calibration performance in both the derivation and validation populations, with AUCs of 0.909 (95% CI, 0.900–0.919) and 0.903 (95% CI, 0.882–0.924), respectively (Figure 3; Supplementary Table 10). In predicting the 7-day mortality rate of ED bacteremia, the discrimination performance of the ED-BM model was significantly better than that of PBS (AUC, 0.848; 95% CI, 0.817–0.879; P < .0001) and BSIMRS (AUC, 0.885; 95% CI, 0.860–0.910; P = .01) in the validation population. This improved discrimination performance of our model was retained in predicting the 28-day mortality rate of ED bacteremia and was better than that of PBS and BSIMRS (Supplementary Table 10). Furthermore, the ED-BM model had satisfactory additive reclassification performance because it could correctly reclassify 9.49% and 1.53% of patients compared with PBS and BSIMRS, respectively (Supplementary Table 11).
Figure 3.
Discrimination and calibration performance of the Pitt Bacteremia Score (PBS), Bloodstream Infection Mortality Risk Score (BSIMRS), and emergency department bacteremia mortality (ED-BM) model in the validation population (n = 2172): (A) receiver operating characteristic curves and (B) calibration curves.
Sensitivity Analysis
We conducted a few sensitivity analyses to test the robustness of our ED-BM model. First, because PBS and BSIMRS were initially developed to predict the 14-day mortality and 28-day mortality related to gram-negative bacteremia, we examined our model performance in predicting the 14-day and 28-day mortality in all ED adults with bacteremia and in those with GN bacteremia. In the validation set, our ED-BM model had an AUC of 0.891 (95% CI, 0.782–0.909) for 14-day mortality and 0.879 (95% CI, 0.862–0.896) for 28-day mortality in all bacteremia. The AUC was 0.909 (95% CI, 0.889–0.928) for the 14-day mortality and 0.899 (95% CI, 0.881–0.916) for the 28-day mortality related to GN bacteremia (Supplementary Table 10). Our ED-BM model remained significantly superior to PBS and BSIMRS in predicting 14-day mortality and 28-day mortality in all bacteremia and in GN bacteremia (Supplementary Table 10). Second, we created an additional model with fewer variables (ED-BMminus) by excluding mechanical ventilation and cardiac arrest because they present only in critically ill patients (Supplementary Table 12). In the derivation population, the Akaike information criterion increased to 3493.66 with a minor drop in AUC (ED-BM vs ED-BMminus, 0.909 vs 0.893; P < .0001). In the validation population, the AUC of the ED-BM model was also better than that of the ED-BMminus model (0.903 vs 0.884; P < .0001). Third, we tested our model performance by patients’ severity as indicating by intensive care unit (ICU) admission status and qSOFA level (Supplementary Table 13). Overall, the model discrimination performance in the validation set remained the highest for the ED-BM model and the lowest for the PBS model or ED-BMminus. The model performance decreased in patients with more severe conditions (ie, ICU admission, qSOFA ≥2). However, almost all AUCs of ED-BM or ED-BMminus were around 0.80, indicating good discrimination performance. To conclude, the above-mentioned evidence indicates that the original ED-BM model is robust and that overfitting is not of concern.
DISCUSSION
The current study demonstrated the decreasing trends of the cumulative incidence of ED bacteremia (from 13.48% to 8.84%) and ED bacteremia–associated 7-day mortality (from 11.25% to 6.83%) from 2003 to 2016 in a tertiary medical center in central Taiwan. In addition, the novel ED-BM model proposed in this study demonstrated better discrimination, calibration, and reclassification performance than PBS and BSIMRS in predicting 7- and 28-day mortality rates among adult ED patients with bacteremia.
Depending on patient populations or bacterial species studied, the positive rate (ie, diagnostic yields) of ED BCs have varied widely (range, 9.8%–31.7%) [7, 23–26]. Similar to prior studies [5, 7], the following findings were observed in the present study: Only 1 of 10 patients receiving BCs had clinically significant bacteremia; gram-negative microorganisms (particularly E. coli) were the most common isolates; and urinary tract infection was the most common infection source (Table 1). The consistently low positive rate of BC indicates a clinical practice gap in identifying the patients who require BC exam in the ED [1]. In our study, the trend of bacteremia declined slightly from 2003 to 2016, which was similar to the observation in a Denmark study of community-acquired bacteremia from 2000 to 2008 (3.7% decrease per year) [27]. The decreasing trend may be partially due to the increased ED patient volume and the liberal performance of BCs. In our study, the number of ED visits with BC obtained increased by 3.03-fold, and the number of bacteremia events increased by only 2.7-fold (Figure 1).
In the literature, the 28-day mortality rate of ED bacteremia has been commonly reported to range from 5.3% to 14.4% [2–5]. An earlier study found that the 7-day mortality rate could more accurately reflect the clinical impact of bacteremia itself rather than underling diseases of patients [28]. In addition to comorbidities, other in-hospital factors, such as nosocomial infections, can have an unfavorable effect on a patient’s outcome. The 7-day mortality rate of our study was 8.53%, which was comparable with the results reported by Al-Hasan and Rannikko [11, 28]. Furthermore, we observed a declining trend of short-term mortality, which has rarely been mentioned in previous studies [7, 23–26], among adult ED patients with bacteremia. A nationwide study that evaluated the sepsis trend between 2009 and 2014 found a significantly decreasing trend of in-hospital mortality using either the claims-based (6.6% decrease per year) or electronic health record–based criterion (3.3% decrease per year) [29]. The timing of decline in the mortality of ED bacteremia coincided with the first release of the Surviving Sepsis Campaign guidelines [30]. The implementation of these guidelines at the ED may partially explain the reduction in short-term mortality of bacteremia. Additionally, a significant decrease in antibiotic inappropriateness from 52.7% in 2003 to 26.5% in 2016 was observed in our study, which could be another ecological explanation for the decreasing 7-day mortality rate due to precise empirical antibiotic selection (Supplementary Figure 4). Furthermore, in studies of organism-specific bacteremia (carbapenem-resistant Acinetobacter baumannii or S. aureus), a decreasing mortality rate was observed, and appropriateness of antibiotic use was found to be associated with decreased mortality [31–33]. Further studies are required to investigate guideline adherence or antibiotic treatment behaviors and their association with mortality reduction in patients with bacteremia at EDs.
Numerous scoring systems have been used to assess the disease severity and death risk of non-ED patients with bacteremia [12, 34–39]. However, these models typically require information that is not readily available in ED settings. Thus, these scoring systems may not be practical for ED physicians to predict the mortality risk of patients with bacteremia. Several scoring systems to predict the mortality of ED bacteremia or sepsis have been proposed [2–5, 10, 40]. Each of these scoring systems, however, has its own limitations. For example, some prediction models used poorly defined predictors, such as the status of terminal illness and absence of chill, which may lead to nonstandardized assessment [2, 4]. Yeh [3] used S. aureus bacteremia and bacteremia with an unknown focus as predictors to stratify the mortality rate of ED patients with BSIs; however, as in real-world practice, clinicians usually need to make disposition decisions before BC results. In other studies, certain clinical conditions of the patient population (eg, cardiac arrest) have posed a substantial mortality risk [10, 40]. Using variables that tightly link to an extremely high risk of death makes these equations unsuitable for most ED patients with bacteremia.
Compared with the aforementioned models [2–4, 10, 40], the ED-BM model has the following strengths: The included variables are objective and readily available during ED admission, the process of variable selection is robust, and the evaluation method of model performance is comprehensive and is validated in the validation population. In addition to the variables proposed by PBS and BSIMRS, we found that impaired renal function, low platelet count, recent steroid use, and wide RDW have a significant effect on the mortality of ED bacteremia, which has been confirmed in studies of mortality for bacteremia or sepsis [2–4, 41]. For instance, a study reported that baseline steroid use is associated with an increased risk of infection and sepsis [42]. Moreover, steroid use per se can compromise the host’s immune response to infections and thus influence the clinical prognosis [43].
Compared with PBS and BSIMRS [12, 14], our model had a better discrimination performance in predicting 7-, 14-, and 28-day mortality among adult ED patients with bacteremia and among adult ED patients with gram-negative bacteremia (Supplementary Table 10). Our model could correctly reclassify high-risk patients (Figure 3; Supplementary Table 11). Although the ED-BM predicts lower mortality than the true mortality, the calibration curve of our model aligns closer to the diagonal line of an ideal prediction model than PBS and BSIMRS. Although model overfitting is a potential concern for the ED-BM model with 16 variables, this concern seems to be negligible considering its good performance in the validation population. In the modern Health Information System (HIS), the ED-BM model could be embedded in the ED setting and assist in risk stratification for adult patients with bacteremia. We expect that the ED-BM can be used to predict patient outcomes when physicians highly suspect a bacteremia etiology or when the initial reports of gram stain culture results are available (within 24–48 hours after blood culture collection). However, further efforts should be allocated in the external validation and implementation of different digital health care systems worldwide.
This study has several limitations. First, the retrospective design could not take into account the trend of changing practice patterns. Further research is required to address potential practice changes, such as the pattern of antimicrobial prescription or the proportion of responsiveness of ED physicians to guidelines. Second, the derivation data set was from a single institution, which may limit its generalizability to other patient populations such as regional or community hospitals. External databases with the data from emergency services, such as the Medical Information Mart for Intensive Care (MIMIC)–IV database [44], should be used to validate our mortality prediction model for ED bacteremia.
In conclusion, this study showed a decreasing trend of short-term mortality of ED bacteremia at a tertiary center in central Taiwan. Moreover, we derived and validated a new prediction tool, the ED-BM model, with better discrimination and calibration performance than PBS and BSIMRS in predicting 7-day mortality among adult ED patients with bacteremia. Further studies are required to validate the ED-BM model and evaluate its effectiveness in the real-world settings of other health care systems.
Supplementary Data
Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.
Acknowledgments
We appreciate the support from the Big Data Center of CMUH for data exploration, statistical analysis, manuscript preparation, and support of the iHi Clinical Research Platform. We would like to thank the Health and Welfare Data Science Center, Ministry of Health Welfare, and Health Data Science Center, China Medical University Hospital, for providing administrative, technical, and funding support.
Financial support. This study was supported by grants from the Ministry of Science and Technology, Taiwan (MOST 108-2314-B-039-038-MY3 and MOST 110-2321-B-468-001), and from China Medical University Hospital (DMR-110-001 and DMR-HHC-110-2). This study was not sponsored by industry.
Potential conflicts of interest. All authors: no reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.
Patient consent. This study was approved by the Big Data Center and Research Ethical Committee/Institutional Review Board of CMUH (CMUH105-REC3-068), and informed consent has been waived.
References
- 1. Long B, Koyfman A.. Best clinical practice: blood culture utility in the emergency department. J Emerg Med 2016; 51:529–39. [DOI] [PubMed] [Google Scholar]
- 2. Shapiro NI, Wolfe RE, Moore RB, et al. Mortality in Emergency Department Sepsis (MEDS) score: a prospectively derived and validated clinical prediction rule. Crit Care Med 2003; 31:670–5. [DOI] [PubMed] [Google Scholar]
- 3. Yeh CF, Chen KF, Ye JJ, Huang CT.. Derivation of a clinical prediction rule for bloodstream infection mortality of patients visiting the emergency department based on predisposition, infection, response, and organ dysfunction concept. J Microbiol Immunol Infect 2014; 47:469–77. [DOI] [PubMed] [Google Scholar]
- 4. Chen KF, Liu SH, Li CH, et al. Development and validation of a parsimonious and pragmatic CHARM score to predict mortality in patients with suspected sepsis. Am J Emerg Med 2017; 35:640–6. [DOI] [PubMed] [Google Scholar]
- 5. Hsieh CC, Yang CY, Lee CH, et al. Validation of MEDS score in predicting short-term mortality of adults with community-onset bacteremia. Am J Emerg Med 2020; 38:282–7. [DOI] [PubMed] [Google Scholar]
- 6. Chase M, Klasco RS, Joyce NR, et al. Predictors of bacteremia in emergency department patients with suspected infection. Am J Emerg Med 2012; 30:1691–7. [DOI] [PubMed] [Google Scholar]
- 7. Kao CH, Kuo YC, Chen CC, et al. Isolated pathogens and clinical outcomes of adult bacteremia in the emergency department: a retrospective study in a tertiary referral center. J Microbiol Immunol Infect 2011; 44:215–21. [DOI] [PubMed] [Google Scholar]
- 8. Lee CC, Wu CJ, Chi CH, et al. Prediction of community-onset bacteremia among febrile adults visiting an emergency department: rigor matters. Diagn Microbiol Infect Dis 2012; 73:168–73. [DOI] [PubMed] [Google Scholar]
- 9. Rothe K, Wantia N, Spinner CD, et al. Antimicrobial resistance of bacteraemia in the emergency department of a German university hospital (2013-2018): potential carbapenem-sparing empiric treatment options in light of the new EUCAST recommendations. BMC Infect Dis 2019; 19:1091. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Sivayoham N, Rhodes A, Cecconi M.. The MISSED score, a new scoring system to predict mortality in severe sepsis in the emergency department: a derivation and validation study. Eur J Emerg Med 2014; 21:30–6. [DOI] [PubMed] [Google Scholar]
- 11. Al-Hasan MN, Juhn YJ, Bang DW, et al. External validation of bloodstream infection mortality risk score in a population-based cohort. Clin Microbiol Infect 2014; 20:886–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Al-Hasan MN, Lahr BD, Eckel-Passow JE, Baddour LM.. Predictive scoring model of mortality in gram-negative bloodstream infection. Clin Microbiol Infect 2013; 19:948–54. [DOI] [PubMed] [Google Scholar]
- 13. Henderson H, Luterbach CL, Cober E, et al. The Pitt Bacteremia Score predicts mortality in nonbacteremic infections. Clin Infect Dis 2020; 70:1826–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Korvick JA, Bryan CS, Farber B, et al. Prospective observational study of Klebsiella bacteremia in 230 patients: outcome for antibiotic combinations versus monotherapy. Antimicrob Agents Chemother 1992; 36:2639–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Vorwerk C, Loryman B, Coats TJ, et al. Prediction of mortality in adult emergency department patients with sepsis. Emerg Med J 2009; 26:254–8. [DOI] [PubMed] [Google Scholar]
- 16. Haque M, Sartelli M, McKimm J, Abu Bakar M.. Health care-associated infections - an overview. Infect Drug Resist 2018; 11:2321–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Liang HY, Lo YC, Chiang HY, et al. Validation and comparison of the 2003 and 2016 diastolic functional assessments for cardiovascular mortality in a large single-center cohort. J Am Soc Echocardiogr 2020; 33:469–80. [DOI] [PubMed] [Google Scholar]
- 18. Chiang HY, Lin KR, Hsiao YL, et al. Association between preoperative blood glucose level and hospital length of stay for patients undergoing appendectomy or laparoscopic cholecystectomy. Diabetes Care 2021; 44:107–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Chiang H-Y, Liang L-Y, Lin C-C, et al. Electronic medical record-based deep data cleaning and phenotyping improve the diagnostic validity and mortality assessment of infective endocarditis: medical big data initiative of CMUH. BioMedicine 2021; 11:59–67. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Centers for Disease Control and Prevention. National healthcare safety network manual for bloodstream infections. Available at: https://www.cdc.gov/nhsn/pdfs/pscmanual/4psc_clabscurrent.pdf. Accessed 9 June 2021.
- 21. Paterson DL, Ko WC, Von Gottberg A, et al. International prospective study of Klebsiella pneumoniae bacteremia: implications of extended-spectrum beta-lactamase production in nosocomial infections. Ann Intern Med 2004; 140:26–32. [DOI] [PubMed] [Google Scholar]
- 22. Alba AC, Agoritsas T, Walsh M, et al. Discrimination and calibration of clinical prediction models: users’ guides to the medical literature. JAMA 2017; 318:1377–84. [DOI] [PubMed] [Google Scholar]
- 23. Sinclair D, Svendsen A, Marrie T.. Bacteremia in nursing home patients. Prevalence among patients presenting to an emergency department. Can Fam Physician 1998; 44:317–22. [PMC free article] [PubMed] [Google Scholar]
- 24. Liao CH, Chen SY, Chang SC, et al. Characteristics of community-acquired and health care-associated Staphylococcus aureus bacteremia in patients treated at the emergency department of a teaching hospital. Diagn Microbiol Infect Dis 2005; 53:85–92. [DOI] [PubMed] [Google Scholar]
- 25. Pan SC, Chen SY, Wang JT, et al. Changing epidemiology of community-onset Staphylococcus aureus bacteremia over nine years in an emergency department in Taiwan. J Infect 2013; 66:187–9. [DOI] [PubMed] [Google Scholar]
- 26. Villalon N, Farzan N, Freeman K.. Rate of bacteremia in the hemodialysis patient presenting to the emergency department with fever: a retrospective chart review. Int J Emerg Med 2018; 11:29. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Nielsen SL, Pedersen C, Jensen TG, et al. Decreasing incidence rates of bacteremia: a 9-year population-based study. J Infect 2014; 69:51–9. [DOI] [PubMed] [Google Scholar]
- 28. Rannikko J, Syrjänen J, Seiskari T, et al. Sepsis-related mortality in 497 cases with blood culture-positive sepsis in an emergency department. Int J Infect Dis 2017; 58:52–7. [DOI] [PubMed] [Google Scholar]
- 29. Rhee C, Dantes R, Epstein L, et al. ; CDC Prevention Epicenter Program. Incidence and trends of sepsis in US hospitals using clinical vs claims data, 2009-2014. JAMA 2017; 318:1241–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Rhodes A, Evans LE, Alhazzani W, et al. Surviving Sepsis campaign: international guidelines for management of sepsis and septic shock: 2016. Intensive Care Med 2017; 43:304–77. [DOI] [PubMed] [Google Scholar]
- 31. Choi SH, Cho EB, Chung JW, Lee MK.. Changes in the early mortality of adult patients with carbapenem-resistant Acinetobacter baumannii bacteremia during 11 years at an academic medical center. J Infect Chemother 2019; 25:6–11. [DOI] [PubMed] [Google Scholar]
- 32. Benfield T, Espersen F, Frimodt-Møller N, et al. Increasing incidence but decreasing in-hospital mortality of adult Staphylococcus aureus bacteraemia between 1981 and 2000. Clin Microbiol Infect 2007; 13:257–63. [DOI] [PubMed] [Google Scholar]
- 33. van Hal SJ, Jensen SO, Vaska VL, et al. Predictors of mortality in Staphylococcus aureus bacteremia. Clin Microbiol Rev 2012; 25:362–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Knaus WA, Draper EA, Wagner DP, Zimmerman JE.. APACHE II: a severity of disease classification system. Crit Care Med 1985; 13:818–29. [PubMed] [Google Scholar]
- 35. Knaus WA, Zimmerman JE, Wagner DP, et al. APACHE—Acute Physiology And Chronic Health Evaluation: a physiologically based classification system. Crit Care Med 1981; 9:591–7. [DOI] [PubMed] [Google Scholar]
- 36. Le Gall JR, Loirat P, Alperovitch A, et al. A simplified acute physiology score for ICU patients. Crit Care Med 1984; 12:975–7. [DOI] [PubMed] [Google Scholar]
- 37. Lemeshow S, Teres D, Klar J, et al. Mortality probability models (MPM II) based on an international cohort of intensive care unit patients. JAMA 1993; 270:2478–86. [PubMed] [Google Scholar]
- 38. Teres D, Lemeshow S.. The APACHE III prognostic system. Chest 1992; 102:1919–20. [DOI] [PubMed] [Google Scholar]
- 39. Vincent JL, Moreno R, Takala J, et al. The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure. On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine. Intensive Care Med 1996; 22:707–10. [DOI] [PubMed] [Google Scholar]
- 40. Askim Å, Moser F, Gustad LT, et al. Poor performance of quick-SOFA (qSOFA) score in predicting severe sepsis and mortality - a prospective study of patients admitted with infection to the emergency department. Scand J Trauma Resusc Emerg Med 2017; 25:56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Kim YC, Song JE, Kim EJ, et al. A simple scoring system using the red blood cell distribution width, delta neutrophil index, and platelet count to predict mortality in patients with severe sepsis and septic shock. J Intensive Care Med 2019; 34:133–9. [DOI] [PubMed] [Google Scholar]
- 42. Chaudhary NS, Donnelly JP, Moore JX, et al. Association of baseline steroid use with long-term rates of infection and sepsis in the REGARDS cohort. Crit Care 2017; 21:185. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Coutinho AE, Chapman KE.. The anti-inflammatory and immunosuppressive effects of glucocorticoids, recent developments and mechanistic insights. Mol Cell Endocrinol 2011; 335:2–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Johnson A, Bulgarelli, Pollard L, et al. MIMIC-IV (version 0.4). 2020. Available at: https://physionet.org/content/mimiciv/0.4/. Accessed 9 June 2021.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.