DOI: https://doi.org/10.21203/rs.3.rs-2551943/v1
Background: Both in-hospital cardiac arrest (IHCA) and out-of-hospital cardiac arrest (OHCA) have higher incidence and lower survival rates. Predictors of in-hospital mortality for intensive care unit (ICU) admitted cardiac arrest (CA) patients remain unclear.
Methods: The Medical Information Mart for Intensive Care Ⅳ (MIMIC-Ⅳ) database was used to perform a retrospective study. Patients meeting the inclusion criteria were identified from the MIMIC-Ⅳ database and randomly divided into training set (n=1206, 70%) and validation set (n=516, 30%). Candidate predictors consisted of the demographics, comorbidity, vital signs, laboratory test results, scoring systems, and treatment information on the first day of ICU admission. Independent risk factors for in-hospital mortality were screened using the least absolute shrinkage and selection operator (LASSO) regression model and the extreme gradient boosting (XGBoost) in the training set. Multivariate logistic regression analysis was used to build prediction models in training set, and then validated in validation set. Discrimination, calibration and clinical utility of these models were compared using the area under the curve (AUC) of the receiver operating characteristic (ROC) curves, calibration curves and decision curve analysis (DCA). After pairwise comparison, the best performing model was chosen to build a nomogram.
Results: Among the 1722 patients, in-hospital mortality was 52.43%. In both sets, the LASSO, XGBoost and The National Early Warning Score 2 (NEWS 2) models showed acceptable discrimination. In pairwise comparison, the prediction effectiveness was higher with the LASSO and XGBoost models than with the NEWS 2 model (p<0.001). The LASSO and XGBoost models also showed good calibration. The LASSO model was chosen as our final model for its higher net benefit and was presented as the nomogram.
Conclusions: The LASSO model enabled good prediction of in-hospital mortality in ICU admission CA patients, which may be widely used in clinical decision-making.
The morbidity and mortality surrounding CA remained a global challenge. In the United States, more than 350000 people experience an OHCA every year, and about 10.8% of them survive to hospital discharge. (1) The in-hospital survival rate of OHCA patients was 26.4%. (2) IHCA occurs in over 290000 adults each year in the United States. (3) The 30day survival rate of patients with IHCA was 27.8%, and the one-year survival rate was 20%. (4)
Although most CA patients die in acute events, a large proportion of CA deaths occur in patients after successful resuscitation, which can be attributed to the development of post CA syndrome, including neurological dysfunction and other types of organ dysfunction. After return of spontaneous circulation of patients with CA, they will generally be sent to the intensive care unit (ICU) for advanced life support treatment. About one third of patients admitted to the ICU survive to discharge, but there are considerable differences in the treatment and prognosis of patients after CA between different institutions. (5–7)
Machine learning (ML) belongs to the category of artificial intelligence, ML systems have been developed in many fields of medicine, including radiology, with the interpretation of images from chest X-ray or magnetic resonance imaging for diagnostic purposes. (8–11)
Different from the traditional prediction model that uses selected variables for calculation, ML can not only easily combine a large number of variables with computers to improve the accuracy of prediction, but also screen variables by a variety of methods when selecting variables to improve the accuracy and efficiency of the model. (12)
By using ML to screen variables and establish prediction models, adverse factors for patients with CA can be identified at the early stage of admission to the ICU, and corrected as soon as possible to improve the prognosis of patients. The purpose of this study was to develop and validate a predictive model for in-hospital mortality of patients with CA admitted to ICU using data from MIMIC-Ⅳ database.
Study Design and Data source
We conducted a retrospective analysis using all the relevant data extracted from the MIMIC-Ⅳ database.
The MIMIC-Ⅳ database is an open and publicly available database that contains high-quality data between 2008 and 2019, constructed by Institutional Review Boards of the Massachusetts Institute of Technology (MIT, Cambridge, MA, America) and Beth Israel Deaconess Medical Center. To access the database, we passed the National Institutes of Health Protecting Human Research Participants web-based training course and we obtained approval to extract data from the MIMIC-Ⅳ for research purposes (Certification Number: 50778029).
Study patients
Patients with a diagnosis of CA, defined as ICD-9 codes of 4275 or ICD-10 codes of I46, I462, I468 and I469. Patients were ≥ 18 years old at the time of ICU admission were included in the study; Patients without an ICU record were excluded from the study. The flow chart showed the selection of patients into the study. (Fig. 1A) Patients with a diagnosis of CA were screened and 1722 adult patients were included in this study.
Data Extraction and Processing
Demographics, vital signs, laboratory tests, scoring systems, relevant treatment information, and others were extracted from the MIMIC-Ⅳ database using structured query language with PostgreSQL (version 14, www.postgresql.org). The prediction model only included the clinical and laboratory variables on the first day of ICU admission. If the patient received more than one vital sign measurement or laboratory tests on the first day of admission, the average values were used for subsequent analysis. Comorbidities were identified using ICD-9/10 code. Based on previous research and clinical applicability, the following data were extracted: demographic characteristics (age at the time of hospital admission, sex); vital signs (heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), mean blood pressure (MBP), respiratory rate (RR), body temperature, saturation pulse oxygen (SPO2); comorbidities (hypertension, congestive heart failure (CHF), myocardial infarction, diabetes mellitus (DM), and chronic obstructive pulmonary disease (COPD)) and laboratory variables (hematocrit (HCT), hemoglobin (HB), platelet count, white blood cells (WBC), prothrombin time (PT), international normalized ratio (INR), creatinine, blood urea nitrogen (BUN), glucose, potassium, sodium, calcium, chloride, the anion gap, bicarbonate, lactate, hydrogen ion concentration (pH); treatment information (ventilation, epinephrine, dopamine) ; score system: sequential organ failure assessment (SOFA), simplified acute physiology score Ⅲ (SAPS Ⅲ), Glasgow coma scale (GCS). The primary outcome of the study was in-hospital mortality, defined as the vital status of the patient at discharge.
Missing data handling
Variables with missing data are common in the MIMIC-Ⅳ, and directly eliminating patients with missing values or analyzing variables with missing values will cause bias. We excluded variables with more than 25% of missing values. For variables with missing proportion < 5%, the normal distribution with continuous variables were replaced by the mean for the patient group and the skewed distributions with continuous variables were replaced by their median. For variables with missing proportion > 5%, we applied multiple imputation to handle the data. (Missing value details in Supplementary Table 1)
Values were presented as total numbers with percentages for categorical variables and the means with standard deviations (SD) (if normal) or medians with inter quartile ranges (IQR) (if non normal) for continuous variables. Proportions were compared using χ² test or Fisher exact tests. For all continuous variables, we used a two-sided one-way analysis of variance or Wilcoxon rank-sum tests when comparing the two groups. P < 0.05 was statistically significant.
The flow chart demonstrated the methodology for developing the predictive model (Fig. 1B). In this study, the data were divided at random, with 70% utilized for training and 30% for validating. Table 1 summarized the predictor variables and statistics.
Two methods were used to select the most important predictors of the in-hospital mortality prediction model from the training set. First, we used the LASSO method, which was conducted via a continuous shrinking operation and minimizing regression coefficients, in order to reduce the likelihood of overfitting. And, LASSO can shrink the sum of the absolute value of regression coefficients, forcing and producing coefficients that are exactly 0. (13, 14) Variables with non-zero coefficients were selected for the next logical regression analysis. This enhanced the prediction accuracy and interpretation ability of the prediction model, and which was suitable for high-dimensional data processing.
Second, we used XGBoost, which was an efficient and scalable ML classifier. XGBoost can achieve high prediction accuracy and low computational costs in various practical applications. (15) Gradient boosting decision tree is the original model of XGBoost, which combines multiple decision trees in boosting way. XGBoost used the number of boosts, learning rate, subsampling ratio, and maximum tree depth to control overfitting and enhance the better performance. Moreover, XGBoost optimized the target of the function, the size of the tree, and the size of the weight through regularization. (16).
To investigate the independent risk factors of in-hospital mortality, the variables screened by two methods were used in the training set of univariate logistic regression analysis to evaluate the significance of variables. In the univariate logistic regression analysis, the variables significantly related to the in-hospital mortality will be further analyzed by multivariate logistic regression analysis. Prediction models were evaluated in terms of discrimination and calibration. Discrimination was assessed by calculating the area under the curve (AUC) of the receiver operating characteristic (ROC) curve and C-statistic testing. Decision curve analysis (DCA) was used to compare the clinical net benefit associated with the use of these models. (17)
The model with the highest AUC and the highest clinical net benefit was identified as the final model, and a nomogram was built to predict in-hospital mortality. The nomogram was a visualization of the results of the regression equation by calculating the probability of occurrence by using some disjoint segments.
The National Early Warning Score 2 (NEWS 2) was a modification of NEWS and a simple aggregate scoring system. In the early stage of disease deterioration, the NEWS 2 can detect the potential disease changes in patients early and provide preventive measures for disease deterioration, which has important clinical implications to improve the survival rate of patients with CA in ICU. (18–22) Because the final calculation was a score, we performed subsequent analysis after translating it into a univariate predict model, to analyze C-index and estimate differences in the discrimination between the models.
All analyses were performed by the statistical software packages R version 4.2.2 (http://www.R-project.org, The R Foundation). P < 0.05 (two-sided test) were considered statistically significant.
Baseline Characteristics
As shown in Fig. 1A, a total of 1722 patients diagnosed as CA were included in our study. According to the research method of the previous study, we randomly divided all patients into training set (1206 people, 70%) and validation group (516 people, 30%). In the whole study population, the in-hospital mortality rate of CA patients was 52.43% (793 survivors and 929 non-survivors). Table 1 showed the comparison of demographics and variables between the training set and the validation set, as well as the comparison of dead patients and survivors during hospitalization. SBP, DBP, MBP values were lower in the training set. The proportion of dopamine use and in-hospital mortality were lower in the validation set. There were no significant differences in other selected variables between the training set and validation set. Patients who died also had lower SBP, DBP, MBP, temperature, SPO2, HCT, HB, platelet, bicarbonate, calcium, pH, GCS score, the proportion of man, CHF and myocardial infarction. However, age, SOFA score and SAPS Ⅲ score, epinephrine use, dopamine use, HR, RR, WBC, anion gap, BUN, creatinine, glucose, sodium, potassium, INR, PT and lactate levels in patients who died during their hospital stay were significantly increased. There was no significant difference in terms of whether they had DM, hypertension, ventilation and chloride levels, between the surviving and non-surviving patients.
Selected variables
In the training set, we conducted the regularization process of LASSO. The binomial deviance was computed for the test data as measures of the predictive performance of the fitted models. The binomial deviance curve was plotted versus log (λ) using 10-fold cross-validation via minimum criteria, where λ was a tuning hyperparameter. The dotted vertical lines were drawn at optimal values by using the minimum criteria and within one standard error range of the minimum criteria. We chose the latter criteria (λ = 0.01944) as it results in stricter penalty allowing us to reduce the number of covariates even further than the minimum criteria(λ = 0.00332). (Fig. 2A, Fig. 2B) Finally, 17 nonzero coefficients were resulted in LASSO regression. (Fig. 3A) Meanwhile, XGboost was also used to analyze the patients who died in the training set, ranked the predictive importance of all included variables, and selected the top 17 variables. (Fig. 3B)
Model development
In the training set, 17 variables respectively screened by LASSO and XGboost, were used to conduct univariate logistic regression with the in-hospital mortality, and variables with statistically significance in univariate logistic regression were used to conduct multivariate logistic regression. Table 2 and Table 3 showed the variables selected in the univariate and multivariate analysis by LASSO and XGBoost. Among the variables screened using LASSO, multivariate logistic regression identified age, SAPS Ⅲ, HR, MBP, RR, temperature, SPO2, GCS, man, bicarbonate, PT as the most significant mortality risk predictors. Among the XGBoost selected variables, SAPS Ⅲ, RR, bicarbonate, SPO2, temperature, age, HR, GCS, HB as the most significant mortality risk predictors.
We established an in-hospital mortality prediction algorithm using LASSO selected variables as follows: log odds of mortality = 18.746877 + 0.013344× age + 0.010997×SAPS Ⅲ+ 0.019006×HR- 0.017839×MBP + 0.048912×RR- 0.286264×temperature- 0.080727×SPO2- 0.142085×GCS- 0.258837×man- 0.064604×bicarbonate + 0.021723×PT.
The variance inflation factors for these variables were 1.1, 1.2, 1.3, 1.1, 1.3, 1.2, 1.1, 1.0, 1.0, 1.1 and 1.0, respectively.
Based on XGBoost, the selected variables for the in-hospital mortality prediction algorithm were as follows: log odds of mortality = 20.258476 + 0.011845×SAPS Ⅲ+ 0.052959×RR- 0.071588×bicarbonate- 0.099384×SPO2- 0.279351×temperature + 0.013794×age + 0.018419×HR- 0.138026×GCS- 0.102837×HB.
The variance inflation factors for these variables were 1.2, 1.3, 1.1, 1.1, 1.2, 1.1, 1.3, 1.0, and 1.1, respectively.
Model validation
The discrimination and calibration of the LASSO model and the XGBoost model in the training set and validation set were shown in Fig. 4A-C and Fig. 5A-C, respectively. In the training set, the AUC of LASSO model and XGBoost model were 0.7879 (0.7627–0.8132) and 0.7854 (0.7599–0.8109) respectively. In the validation set, the AUC of LASSO model and XGBoost model were 0.7994 (0.7618–0.8369) and 0.7941 (0.7560–0.8321), respectively. As shown in Fig. 4B and 4C, Fig. 5B and 5C, in the calibration curves of the training set and the validation set of the same model, it can be seen that the prediction models had a strong concordance performance in both sets.
The NEWS 2 based on RR, SPO2, SBP, pulse rate, level of consciousness or new confusion, temperature to predict the risk of in-hospital mortality for patients in ICU with CA. We calculated the NEWS 2 for all study patients. The ROC curve and calibration curve of NEWS 2 in the training set and verification set were shown in Fig. 6A-C. The AUC of the training set and verification set were 0.6944 (0.6651–0.7237) and 0.7030 (0.6588–0.7472), respectively.
The DCA for the LASSO model, the XGBoost model and the NEWS 2 model were presented in Fig. 7A. It can be seen that when the threshold probability was 0.18 to 0.88 in the three models, the models added more net benefit than the ‘All’ or ‘None’ scheme.
Model comparison
We compared the AUC of the LASSO, XGBoost and NEWS 2 model in our total study population to assess the predictive effectiveness of the three models. Figure 7B showed that the AUC for the LASSO model, XGBoost model and NEWS 2 model were 0.7912(0.7703–0.8122), 0.7875(0.7663–0.8088) and 0.6969 (0.6725–0.7212), respectively, which were confirmed to be 0.7845,0.7873, and 0.6969 via bootstrapping validation(repeat = 1000). By comparing the AUC, the predictive effectiveness of the LASSO model and the XGBoost model were both significantly better than the NEWS 2 model (p < 0.001). And there was no statistical significance difference between the LASSO model and the XGBoost model (p = 0.4605) (Table 4).
In the LASSO model, 11 variables were included, while 9 variables were included in the XGBoost model. Although the XGboost model was more concise, the net benefit of the LASSO model was higher than the XGBoost model within the threshold range of 0.6-1.0. We believed that higher net benefit was more beneficial for patients with CA. Therefore, we chose the LASSO model as the final model, and represented by the nomogram in Fig. 8. The nomogram used some parallel lines with scales to estimate the probability of occurrence of each risk factor. The score of each risk factor can be calculated, and then the probability of occurrence to the total score of all risk factors can be calculated, which is the probability of occurrence of this event.
CA is a major public health event with high mortality rate. Early and accurate prediction of in-hospital mortality of patients with CA can give clinicians more time to provide individualized treatment strategies, and play an important role in rational planning of medical resources and personnel scheduling. Many studies had analyzed the short-term and long-term survival rates of OHCA and IHCA respectively. Recently, more and more studies established prediction models for in-hospital patients with CA. (23–25)
With the advancement of electronic medical records and artificial intelligence, ML algorithms have become more widely utilized in individualized medicine to assist clinical decision-making. Therefore, we used the data from MIMIC-Ⅳ database to screen the independent risk factors of in-hospital mortality with CA patients by ML algorithm, and then obtained a predictive nomogram by logistic regression analysis.
In our study, the in-hospital mortality is 52.4%, which is significantly lower than other study population. (2, 4, 26) The lower mortality rate may be related to the small sample size and the location of CA. The age, HR, RR, bicarbonate, SPO2, temperature, SAPS Ⅲ score and GCS score were included in the XGBoost and LASSO multiple regression equations. The HB were included in the XGBoost multiple regression equation; Meanwhile, the man, PT and MBP were included in the LASSO multiple regression equation. Consistent with previous studies, our study also took age, man, HR, SPO2, and lower MBP values as independent predictors.(24, 27, 28)
After CA, the occurrence of tissue ischemia and hypoxia will lead to the decrease of pH value and increase of lactate value. In OHCA patients, the pH value was related to the neurological state of the patient at discharge, which may help predict the adverse neurological state of the patient at discharge. (29, 30) However, some studies had shown that dynamic monitoring of pH may be more meaningful for the prognosis of the nervous system. (31) Moreover, the prognosis of the nervous system was not consistent with the in-hospital survival rate. Lactate was also an indicator that can affect the prognosis of CA patients. (32) However, some studies had pointed out that in addition to the absolute value of lactate, the change rate of early lactate value also had important prognostic significance. (33, 34) In our study, pH and lactate were not included in the final model due to the inconsistent sample collection time of all patients and the possible bias caused by the longtime span of the database.
Several prognostic scoring systems had been developed to predict the in-hospital mortality of ICU patients.(35) SAPS Ⅲ was a more modern mortality prediction model that used a larger cohort that included the first from outside North America and Europe and utilized new computer-intensive analysis methods.(36) Although some studies showed that the SAPS III did not predict mortality in patients admitted to ICU after CA. (37, 38) SAPS III was also considered as a reliable, simple and easy to use prognostic model in clinical practice.(39) In our study, SAPS Ⅲ was included in the final model as an independent predictor.
Targeted temperature management is the sole intervention for improving neurological outcomes in the post return of spontaneous circulation phase of care, which has been considered a standard of care treatment for over almost two decades and has been included in the International Post-Cardiac Arrest Guidelines. (40, 41) However, in our study, the effect of temperature was opposite to targeted temperature management, which may be because low body temperature was due to poor peripheral circulation, rather than hypothermia treatment.
During CA, metabolic acidosis is caused by hypoxia-induced anaerobic metabolism and decreased metabolic acid excretion due to renal insufficiency. (42) The lower the bicarbonate, the more serious the metabolic acidosis. In our study, bicarbonate was included in our model as an independent predictor.
PT shows the status of exogenous coagulation system. PT prolongation was prevalent in critically ill patients and was independently associated with higher ICU mortality. (43) In other studies, prolonged PT was also associated with increased mortality. (44, 45) In our study, PT was an independent predictor of in-hospital mortality in patients with CA.
NEWS 2 has clinically significant in predicting the incidence of CA within 24h, ICU occupancy and mortality. (46, 47) Compared with the NEWS 2 model, XGBoost model and LASSO model showed the advantages of prediction effect in our study. Both models showed good discrimination and calibration abilities in the training set and validation set. To obtain a greater net benefit, we selected the LASSO model to develop our predict nomogram.
A lot of variables were reported to correlate with mortality in CA patients, such as hypertension, DM, the use of vasopressor. (32, 48–50) Nevertheless, our study showed that hypertension, DM and the use of epinephrine and dopamine were not predictors of in-hospital mortality of ICU admitted CA patients, which may be attribute to our relatively small sample size, and the difference in amount and duration of use vasopressors.
Our study had several limitations. Firstly, as this was a retrospective study, we were unable to avoid selection bias. Secondly, as a single center study with the earliest cases from almost 20 years ago, the treatment and care of CA had been inconsistent with current standards, which requires a multicenter registration and prospective study to verify. Thirdly, our cases were all adult CA patients in ICU. Whether the results of this study can be applied to other populations needs further research on more patients in different clinical environments to confirm our results. Fourthly, limited by the contents of the MIMIC-Ⅳ database, incomplete recorded data was not included in the analysis.
We developed a predictive nomogram for in-hospital mortality of CA patients in ICU, including variables that can be routinely collected during ICU admission. With a high AUC of 0.7879 (95% CI 0.7627–0.8132), a wide net benefit threshold range (0.2-1.0) and high net benefit, this nomogram may be widely used in clinical decision-making.
Area under curves (AUC)
Blood urea nitrogen (BUN)
Cardiac arrest (CA)
Chronic obstructive pulmonary disease (COPD)
Congestive heart failure (CHF)
Decision curve analysis (DCA)
Diabetes mellitus (DM)
Diastolic blood pressure (DBP)
Extreme gradient boosting (XGBoost)
Glasgow Coma Scale (GCS)
Heart rate (HR)
Hematocrit (HCT)
Hemoglobin (HB)
Hydrogen ion concentration (pH)
In-hospital cardiac arrest (IHCA)
Intensive care unit (ICU)
Inter quartile ranges (IQR)
International normalized ratio (INR)
Least absolute shrinkage and selection operator (LASSO)
Machine learning (ML)
Mean blood pressure (MBP)
Medical Information Mart for Intensive Care Ⅳ (MIMIC-Ⅳ)
National Early Warning Score 2 (NEWS 2)
Out-of-hospital cardiac arrest (OHCA)
Prothrombin time (PT)
Respiratory rate (RR)
Receiver operating characteristic (ROC)
Saturation pulse oxygen (SPO2)
Sequential Organ Failure Assessment (SOFA)
Simplified Acute Physiology Score Ⅲ (SAPS Ⅲ)
Systolic blood pressure (SBP)
White blood cells (WBC)
Ethics approval and consent to participate
The Institutional Review Boards (IRBs) at both Massachusetts Institute of Technology and the Beth Israel Deaconess Medical Center approved the use of the data for research. The MIMIC-Ⅳ (https://mimic-iv.mit.edu) database can be accessed by certified researchers, so no additional informed consent of the patient and ethic approvement are required. The certified researcher of this study is Yiwu Sun (Certification Number: 50778029). All methods of this article were performed in accordance with the relevant guidelines and regulations.
Consent for publication
Not applicable.
Availability of data and material
The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.
Competing Interests
The authors declare that they have no conflict of interest.
Funding
Not applicable.
Author contributions
Conception and design: YWS, ZYH, JR; Data collection and analysis: YWS, ZYH, YFW; Drafting the manuscript: YWS. All authors have read, edited, and approved the manuscript.
Acknowledgements
Not applicable.
Tables 1 is available in the Supplementary Files section.
Univariate analysis |
Multivariate analysis |
|||||
---|---|---|---|---|---|---|
OR |
95%CI |
P |
OR |
95%CI |
||
Age (years old) |
1.01 |
1.01 to 1.02 |
< 0.001 |
1.01 |
1.00 to 1.02 |
0.013 |
Congestive heart failure |
0.79 |
0.61 to 1.04 |
0.0928 |
NA |
||
SAPS Ⅲ |
1.02 |
1.02 to 1.03 |
< 0.0001 |
1.01 |
1.00 to 1.01 |
0.002 |
Heart rate (beats/minute) |
1.03 |
1.02 to 1.03 |
< 0.001 |
1.02 |
1.01 to 1.03 |
< 0.001 |
MBP (mmHg) |
0.97 |
0.96 to 0.98 |
< 0.001 |
0.99 |
0.98 to 1.00 |
0.049 |
Respiratory rate (beats/minute) |
1.13 |
1.09 to 1.16 |
< 0.001 |
1.05 |
1.01 to 1.09 |
0.007 |
Temperature (°C) |
0.72 |
0.64 to 0.80 |
< 0.001 |
0.76 |
0.65 to 0.87 |
< 0.001 |
SPO2 (%) |
0.86 |
0.82 to 0.90 |
< 0.001 |
0.93 |
0.88 to 0.97 |
0.002 |
GCS |
0.85 |
0.80 to 0.90 |
< 0.001 |
0.86 |
0.80 to 0.91 |
< 0.001 |
Man (n, %) |
0.73 |
0.58 to 0.92 |
0.0084 |
0.76 |
0.57 to 1.00 |
0.048 |
Hemoglobin (g/dL) |
0.89 |
0.85 to 0.94 |
< 0.001 |
0.95 |
0.89 to 1.01 |
0.100 |
Anion gap (mmol/L) |
1.14 |
1.11 to 1.18 |
< 0.001 |
1.02 |
0.97 to 1.06 |
0.466 |
Bicarbonate (mmol/L) |
0.89 |
0.86 to 0.91 |
< 0.001 |
0.95 |
0.92 to 0.98 |
0.003 |
BUN (mg/dL) |
1.02 |
1.01 to 1.02 |
< 0.001 |
1.01 |
1.00 to 1.01 |
0.087 |
Sodium (mmol/L) |
1.02 |
1.00 to 1.04 |
0.0702 |
NA |
||
PT (s) |
1.05 |
1.03 to 1.07 |
< 0.001 |
1.01 |
1.00 to 1.03 |
0.049 |
Lactate (mmol/L) |
1.25 |
1.19 to 1.31 |
< 0.001 |
1.06 |
0.99 to 1.13 |
0.103 |
BUN, blood urea nitrogen; GCS, Glasgow coma scale; MBP, mean blood pressure; PT, prothrombin time; SAPS Ⅲ, simplified acute physiology score; SPO2, saturation pulse oxygen. P < 0.05 was statistically significant.
Univariate analysis |
Multivariate analysis |
|||||
---|---|---|---|---|---|---|
OR |
95%CI |
P |
OR |
95%CI |
P |
|
SAPS Ⅲ |
1.02 |
1.02 to 1.03 |
< 0.001 |
1.01 |
1.00 to 1.02 |
0 |
SPO2 (%) |
0.86 |
0.82 to 0.90 |
< 0.001 |
0.92 |
0.88 to 0.96 |
0.001 |
Age (years old) |
1.01 |
1.01 to 1.02 |
< 0.001 |
1.01 |
1.00 to 1.02 |
0.005 |
Respiratory rate (beats/minute) |
1.13 |
1.09 to 1.16 |
< 0.001 |
1.05 |
1.01 to 1.09 |
0.009 |
Lactate (mmol/L) |
1.25 |
1.19 to1.31 |
< 0.001 |
1.05 |
0.99 to 1.12 |
0.096 |
Bicarbonate (mmol/L) |
0.89 |
0.86 to 0.91 |
< 0.001 |
0.96 |
0.93 to 0.99 |
0.023 |
Heart rate (beats/minute) |
1.03 |
1.02 to 1.03 |
< 0.001 |
1.02 |
1.01 to 1.03 |
< 0.001 |
Platelet (109 /L) |
1.00 |
1.00 to 1.00 |
0.1729 |
NA |
||
Temperature (°C) |
0.72 |
0.64 to 0.80 |
< 0.001 |
0.77 |
0.66 to 0.88 |
< 0.001 |
Calcium (mmol/L) |
0.90 |
0.80 to 0.99 |
0.0347 |
0.93 |
0.82 to 1.07 |
0.308 |
GCS (mmol/L) |
0.85 |
0.80 to 0.90 |
< 0.001 |
0.86 |
0.81 to 0.92 |
< 0.001 |
Potassium (mmol/L) |
1.22 |
1.04 to 1.43 |
0.0157 |
0.93 |
0.76 to 1.13 |
0.441 |
Anion gap (mmol/L) |
1.14 |
1.11 to 1.18 |
< 0.001 |
1.04 |
1.00 to 1.08 |
0.051 |
MBP (mmHg) |
0.97 |
0.96 to 0.98 |
< 0.001 |
0.98 |
0.96 to 1.01 |
0.217 |
Hemoglobin (g/dL) |
0.89 |
0.85 to 0.94 |
< 0.001 |
0.92 |
0.86 to 0.98 |
0.008 |
WBC (109 /L) |
1.02 |
1.01 to 1.04 |
0.0022 |
1.00 |
0.99 to 1.02 |
0.665 |
DBP (mmHg) |
0.98 |
0.97 to 0.99 |
< 0.001 |
1.00 |
0.98 to 1.03 |
0.903 |
DBP, diastolic blood pressure; GCS, Glasgow coma scale; MBP, mean blood pressure; SAPS Ⅲ, simplified acute physiology score; SPO2, saturation pulse oxygen; WBC, white blood cells. P < 0.05 was statistically significant.
Comparison between models |
Z statistic |
95% Confidence interval |
Significance level (p value) |
---|---|---|---|
LASSO model vs XGBoost model |
0.73806 |
-0.01 to 0.02 |
0.4605 |
NEWS 2 model vs LASSO model |
-5.5442 |
-0.13 to -0.06 |
< 0.001 |
NEWS 2 model vs XGBoost model |
-5.1227 |
-0.13 to -0.06 |
< 0.001 |
AUC, area under curve; LASSO, least absolute shrinkage and selection operator; NEWS 2, the national early warning score 2; XGBoost, extreme gradient boosting. P < 0.05 was statistically significant.