Sepsis has been defined as a dysregulated host immune response to infections, leading to a life-threatening organ dysfunction [1
]. Escherichia coli
) as one major kind of gram-negative bacilli may cause intra-abdominal infections, urinary tract infections, and sepsis [2
]. An early-onset neonatal sepsis research with 235 cases showed that the most frequent pathogen was E. coli
(86 [36.6%]) with higher incidence of mortality [3
]. In China, a recent study clarified that E. coli
infection accounted for nearly 30% in neonatal sepsis with more than a 10% death rate [4
]. The immature immune systems of neonates may lead to a higher mortality in E. coli
infection. Hence, early identification of those sepsis patients with poor prognosis was significant.
However, for E. coli infection, most previous studies focused on neonatal sepsis and few studies have been done for investigating the clinical characteristics of adult patients. Moreover, little has been known about the predictive values of different laboratory variables in adult sepsis with E. coli infection. Therefore, in our study, we aimed to explore the factors which were associated with in-hospital mortality in adult sepsis with E. coli infection based on a public database.
All sepsis patients with E. coli
infection in MIMIC-III were included in this study. MIMIC-III database as an US-based critical care public database includes data linked with 53,423 adult patients (aged 16 years or above) from 2001 to 2012 and 7870 neonates from 2001 to 2008 admitted to a intensive care unit (ICU) [5
]. Data including vital signs, medications, laboratory measurements, observations and notes charted by care providers, fluid balance, procedure codes, diagnostic codes, imaging reports, hospital length of stay and survival data were comprehensively recorded. The following tables in MIMIC III dataset were utilized in our study: ADMISSIONS, CHARTEVENTS, D_ICD DIAGNOSIS, D_ITEMS, D_LABIEVENTS, DIAGNOSIS_ICD, ICUSTAYS, LABEVENTS, NOTEEVENTS, PATIENTS, INPUTEVENTS_CV, INPUTEVENTS _MV and OUTPUTEVENTS [5
All patients with a diagnosis relevant to sepsis with E. coli infection in the database were initially screened. The diagnosis of sepsis with E. coli infection in the database was confirmed by the lab findings when the pathogen culture in blood was positive in E. coli. Only the data of each patient in the first admission were utilized in this study. Exclusion criteria included as follows: patients with missing > 5% individual data and age less than 18.
Data extraction was performed by using structure query language (SQL). The data of demographic characteristics, clinical variables, laboratory variables and scoring systems were extracted for further analysis. The baseline characteristics used were those recorded within 24 h after admission. When one variable was recorded at a different time compared to the initial 24 h, the first one was enrolled in the study. Demographic characteristics included age, gender, marital status, ethnicity, ICU department, admission type, and comorbidities (renal disease, coronary artery disease (CAD), diabetes, hypertension). Clinical and laboratory variables included systolic blood pressure (SBP), diastolic blood pressure (DBP), heart rate (HR), respiratory rate (RR), white blood cells (WBC), neutrophils, lymphocytes, basophils, platelet (PLT), red cell volume distribution width (RDW), hematocrit (HCT), glucose, prothrombin time (PT), thrombin time (TT), albumin, alanine aminotransferase (ALT), aspartate aminotransferase (AST), mean corpuscular volume (MCV), total bilirubin, creatinine, lactate, total calcium and anion gap. Clinical outcomes including length of stay (LOS) in ICU and in-hospital mortality and scoring systems including sequential organ failure assessment (SOFA) and acute physiology and chronic health evaluation (APACHEII) were also extracted.
Characteristics are expressed as mean ± standard deviation or median (IQR) for continuous variables and a percentage or frequency for categorical variables. Continuous variables were compared using Student’s t-test (normal distribution) or Mann–Whitney U-test (Skewed distribution), and categorical variables were compared using Fisher’s exact test or Chi-square analysis. Stepwise logistic regression for variables selection in multivariable logistic regression was performed. Variables with P < 0.2 which were compared between the survivor and non-survivor groups were further enrolled in multivariable logistic regression. Then, factors associated with in-hospital mortality was identified by multivariate logistic regression. Finally, the receiver-operator characteristic (ROC) analysis of different factors for predicting in-hospital mortality were performed. The cut-off values of variables were confirmed by the Youden Index (sensitivity + specificity-1). The value of each variable with the maximum Youden Index was the cut-off value.
SPSS software (version 26) was implemented for statistical analysis. Two-sided P values < 0.05 were considered statistically significant.
In our retrospective study, RDW and HCT were identified as factors associated with in-hospital mortality in adult sepsis patients with E. coli infection. To the best of our knowledge, this was the first study to explore the association of the factors with clinical prognosis in adult sepsis with E. coli infection based on MIMIC-III public database.
RDW as a parameter which could measure the range of variation of red blood cell size has been proved to be a common and inexpensive biomarker in critical illness [6
]. Elevated RDW levels implicated higher variation in size, which has been usually applied for differentiation in anemia due to nutritional deficiency [7
There is an accumulation of studies that have explored the association between RDW and clinical outcomes in sepsis. Recently, two modified and simple scores both including RDW have been proved to be useful tools for predicting short-term outcomes in sepsis or septic shock [8
]. One study focusing on neonatal sepsis elucidated that RDW to platelet ratio as a useful systemic inflammatory marker could be an indicator for sepsis occurrence in early stage [10
]. In adult sepsis patients, the combination of three parameters including RDW, platelet distribution width and the neutrophil‑lymphocyte count ratio which were easily acquired from whole blood cell count analysis had a good diagnostic performance [11
]. A nomogram including RDW provided a relatively accurate prediction for the early identification of septic patients at high risk of mortality in the emergency department [12
]. One meta-analysis including 17,961 sepsis patients from 11 studies demonstrated that RDW was a significantly useful predictor of mortality in sepsis and patients with elevated RDW were more likely to have higher mortality [13
Studies on RDW in different cohorts had different cut-off values. In a study with a total of 103 patients with community-acquired intra-abdominal sepsis, RDW ≥ 16 had an AUC of 0.867 for predicting in-hospital mortality [14
]. Another study with 1046 patients concluded that for 30-day mortality and early clinical deterioration, an optimal cut-off value of RDW were 12.95 and 14.48, respectively [15
]. One recent study on sepsis patients demonstrated that Youden Index was maximum (37%) at RDW value 14.75, which was good at predicting mortality within 28-days of emergency admission [16
]. In our study, the best threshold value of RDW for predicting in-hospital mortality was 15.45.
The underlying mechanisms as to why increased RDW was associated with adverse prognosis in sepsis remained largely unknown, but several explanations have been illuminated in some studies. First, elevated inflammatory markers due to systemic inflammation response in sepsis may affect the erythrocytes maturation and lead to the migration of reticulocytes into the peripheral circulation, thereby resulting in RDW being elevated [17
]. Second, reactive erythropoiesis was stimulated under oxidative stress which was one of the pathophysiologic entities of sepsis. Then, large immature red cells with poor oxygen-binding capacity were released, causing an increase in the RDW [18
]. Third, sepsis can interrupt the iron steady state, trigger bone marrow suppression, and downregulate the expression of the erythropoietin receptor, which all contribute and cause more production of ineffective red blood cell and RDW increased [19
In our research, HCT was another factor which was associated with in-hospital mortality in sepsis. One recent study based on machine learning for early detection of late-onset neonatal sepsis showed that HCT was one of top three predictive variables [20
]. Another study in Brazil found that as a predictor of mortality risk in the sepsis, the level of HCT decreased with worse outcomes [21
]. However, a positive relationship between HCT and mortality was found in our study, which was not consistent with some previous studies [22
]. The differences could be partly explained by two reasons. First, sepsis patients with poor outcomes were more likely to be suffering from hypovolemia due to increased capillary permeability [24
], which resulted in higher levels of HCT. Second, the general characteristics of sepsis patients in different studies were not the same.
Limitations should also be clarified in our study. First, the study was on the basis of a publicly single-center database in US. While applying to other nations, concerns regarding the generalizability of the conclusions and the confounding bias caused by the missing data should be considered. Second, the new definition of Sepsis-3 was not included in this study because the patients in MIMIC-III were enrolled before 2012, which may lead to some limitations in applying our results. Third, RDW is always related to the underlying condition, especially chronic anemia, while anemia is one of the most common complications in patients with sepsis in the ICU [25
]. Sepsis-related anemia can be caused by some factors including fluid loading-related hemodilution, iatrogenic blood loss, and inflammation-associated abnormalities in erythropoiesis [26
]. Due to lack of some data in MIMIC-III, the anemia which involved past medical history or caused by sepsis couldn’t be defined clearly. Further research should be done for exploring the differences between sepsis with anemia and without anemia in order to validate our results. Fourth, samples in our study were relatively small and subgroups were not divided for further analysis. Due to lack of some data in MIMIC-III, not all the variables which may affect the association between RDW and prognosis were enrolled. Hence, more samples with more variables and multiple centers should be explored for validating our results.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.