BACKGROUND

Pediatric Early Warning Scores (PEWS) help identify children at risk of clinical deterioration, but their accuracy across diverse settings, populations, interventions, and outcomes remains unexplored.

To determine the accuracy of PEWS in detecting clinical deterioration events (CDE) among pediatric patients seen at the emergency department (ED).

Retrospective cohort study.

Pediatric patients aged 1 month to 18 years seen at the ED.

Southern Philippines Medical Center Emergency Department, Davao City, Philippines from January 2021 to December 2022.

Area under the curve (AUC) of PEWS in detecting CDE; Brighton PEWS optimal cut-off and its sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (+LR), and negative likelihood ratio (-LR).

Among the 345 patients, 56 experienced CDE and 289 did not. Patients with CDE had significantly lower median age (1.00 year vs 5.00 years; p < 0.001), oxygen saturation (93.00% vs 98.00%; p < 0.001), and pediatric Glasgow Coma Scale scores (8.00 vs 15.00; p < 0.0001) compared to those without CDE. Heart rate (135.00 vs 111.00 beats per minute; p < 0.001), and respiratory rate (32.50 vs 24.00 breaths per minute; p < 0.001) were significantly higher in patients with CDE. The two groups also differed significantly in terms of comorbidity distribution (p < 0.001) and diagnosis (p < 0.001). The AUC of Brighton PEWS was 0.9064 (95% CI 0.8716 to 0.9357), with an optimal cut-off score of ≥4.00. This threshold yielded 76.79% sensitivity, 88.58% specificity, 56.60% PPV, 95.20% NPV, 6.72 LR+, and 0.26 LR-.

The Brighton PEWS demonstrates strong diagnostic accuracy in predicting CDE among pediatric patients. A cut-off score of ≥4.00 offers a balanced combination of sensitivity, specificity, and likelihood ratios for ED application.

BACKGROUND AND OBJECTIVE

Patients with liver cirrhosis have an increased risk for poor postoperative outcomes after non-hepatic surgery, with liver dysfunction being the most important predictor of poor outcomes. This study aims to determine the association of the albumin-bilirubin (ALBI) grade, aspartate aminotransferase-platelet ratio index (APRI) score, and ALBI-APRI score with postoperative outcomes among cirrhotic patients who have undergone non-hepatic surgery.

This was a retrospective cohort study involving 34 patients. Age, ASA class, urgency of surgery, etiology of liver cirrhosis, preoperative Child-Turcotte-Pugh (CTP) score, Model for End-Stage Liver Disease (MELD) score, ALBI grade, APRI score, and ALBI-APRI score were documented. The outcomes analyzed were postoperative hepatic decompensation (POHD) and in-hospital mortality. Bivariate analysis using the Mann-Whitney U test and Fisher’s exact test was performed. Receiver operating characteristic (ROC) curve analysis was performed to compare the ability of the liver scoring systems to predict the occurrence of study outcomes. Binary logistic regression was performed to measure the odds ratio.

The ALBI grade and ALBI-APRI score were significantly associated with both POHD and in-hospital mortality. Both scores were non-inferior to the CTP and MELD scores in predicting study outcomes. Compared to CTP and MELD scores, the ALBI grade was more sensitive but less specific in predicting POHD and as sensitive but more specific in predicting in-hospital mortality. The ALBI-APRI score was less sensitive but more specific than the ALBI grade in predicting both POHD and in-hospital mortality.

The ALBI grade and ALBI-APRI score were both associated with postoperative hepatic decompensation and in-hospital mortality and were noninferior to the CTP score and MELD score in predicting short-term in-hospital outcomes among cirrhotic patients after non-hepatic surgery.

INTRODUCTION: Out-of-hospital-cardiac-arrest (OHCA) is a major public health challenge and post-return-of-spontaneous-circulation (ROSC) goals have shifted from just survival to survival with intact neurology. Although post-ROSC care is crucial for survival with intact neurology, there are insufficient well-established protocols for post-resuscitation care. We aimed to evaluate post-resuscitation care in the emergency department (ED) of adult (aged ≥16 years) OHCA patients with sustained ROSC and its associated neurologically intact survival. METHODS: A retrospective review of electronic medical records was conducted for OHCA patients with sustained ROSC at the ED. Data including demographics, pre-hospital resuscitation, ED resuscitation, post-resuscitation care and eventual outcomes were analysed. RESULTS: Among 921 OHCA patients, 85 (9.2%) had sustained ROSC at the ED. Nineteen patients (19/85, 22.4%) survived, with 13 (13/85, 15.3%) having intact neurology at discharge. Electrocardiogram and chest X-ray were performed in all OHCA patients, whereas computed tomography (CT) was performed inconsistently, with CT brain being most common (74/85, 87.1%), while CT pulmonary angiogram (6/85, 7.1%), abdomen and pelvis (4/85, 4.7%) and aortogram (2/85, 2.4%) were done infrequently. Only four patients (4.7%) had all five neuroprotective goals of normoxia, normocarbia, normotension, normothermia and normoglycaemia achieved in the ED. The proportion of all five neuroprotective goals being met was significantly higher ( P = 0.01) among those with neurologically intact survival (3/13, 23.1%) than those without (1/72, 1.4%). CONCLUSION Post-resuscitation care at the ED showed great variability, indicating gaps between recommended guidelines and clinical practice. Good quality post-resuscitation care, centred around neuroprotection goals, must be initiated promptly to achieve meaningful survival with intact neurology.
Humans ; Out-of-Hospital Cardiac Arrest/mortality* ; Retrospective Studies ; Male ; Female ; Middle Aged ; Emergency Service, Hospital ; Cardiopulmonary Resuscitation/methods* ; Return of Spontaneous Circulation ; Aged ; Adult ; Treatment Outcome ; Electrocardiography ; Tomography, X-Ray Computed ; Aged, 80 and over

OBJECTIVES: The Charlson comorbidity index reflects overall comorbidity burden and has been applied in cardiovascular medicine. However, its role in predicting in-hospital mortality in patients with acute myocardial infarction (AMI) complicated by ventricular arrhythmias (VA) remains unclear. This study aims to evaluate the predictive value of the Charlson comorbidity index in this setting and to construct a nomogram model for early risk identification and individualized management to improve outcomes. METHODS: Using the open-access critical care database MIMIC-IV (Medical Information Mart for Intensive Care IV), we identified intensive care unit (ICU) patients diagnosed with AMI complicated by VA. Patients were grouped according to in-hospital survival. The predictive performance of the Charlson comorbidity index and other clinical variables for in-hospital mortality was analyzed. Key predictors were selected using the least absolute shrinkage and selection operator (LASSO) regression, followed by multivariable Logistic regression. A nomogram model was constructed based on the regression results. Model performance was assessed using receiver operating characteristic (ROC) curves and calibration plots. RESULTS: A total of 1 492 patients with AMI and VA were included, of whom 340 died and 1 152 survived during hospitalization. Significant differences were observed between survivors and non-survivors in sex distribution, vital signs, comorbidity burden, organ function, and laboratory parameters (all P<0.05). The area under the curve (AUC) of the Charlson comorbidity index for predicting in-hospital mortality was 0.712 (95% CI 0.681 to 0.742), significantly higher than albumin, international normalized ratio (INR), hemoglobin, body temperature, and platelet count (all P<0.001), but comparable to Sequential Organ Failure Assessment (SOFA) score (P>0.05). LASSO regression identified seven key predictors: the Charlson comorbidity index (quartile groups: T1, <6; T2, ≥6-<7; T3, ≥7-<9; T4, ≥9), ventricular fibrillation, age, systolic blood pressure, respiratory rate, body temperature, and SOFA score. Multivariate Logistic regression showed that compared with T1, mortality risk increased significantly in T2 (OR=1.996, 95% CI 1.135 to 3.486, P=0.016), T3 (OR=3.386, 95% CI 2.192 to 5.302, P<0.001), and T4 (OR=5.679, 95% CI 3.711 to 8.842, P<0.001). Age (OR=1.056, P<0.001), respiratory rate (OR=1.069, P<0.001), SOFA score (OR=1.223, P<0.001), and ventricular fibrillation (OR=2.174, P<0.001) were independent risk factors, while systolic blood pressure (OR=0.984, P<0.001) and body temperature (OR=0.648, P<0.001) were protective factors. The nomogram incorporating these predictors achieved an AUC of 0.849 (95% CI 0.826 to 0.871) with high discrimination and good calibration (mean absolute error=0.014). CONCLUSIONS The Charlson comorbidity index is an independent predictor of in-hospital mortality in AMI patients complicated by VA, with performance comparable to the SOFA score. The nomogram model based on the Charlson comorbidity index and additional clinical variables effectively estimates mortality risk and provides a valuable reference for clinical decision-making.
Humans ; Nomograms ; Hospital Mortality ; Myocardial Infarction/complications* ; Male ; Female ; Comorbidity ; Middle Aged ; Aged ; Arrhythmias, Cardiac/complications* ; ROC Curve ; Intensive Care Units

OBJECTIVES: Global epidemiological data indicate that 20% to 30% of intensive care unit (ICU) sepsis patients progress to deep vein thrombosis (DVT) due to coagulopathy, with an associated mortality rate of 25% to 40%. Existing prognostic tools have limitations. This study aims to develop and validate nomogram and machine learning models to predict in-hospital mortality in sepsis patients with DVT and assess their clinical applicability. METHODS: This multicenter retrospective study drew on data from the Medical Information Mart for Intensive Care IV (MIMIC-IV; n=2 235), the eICU Collaborative Research Database (eICU-CRD; n=1 274), and the Patient Admission Dataset from the ICU of Third Xiangya Hospital, Central South University (CSU-XYS-ICU; n=107). MIMIC-IV was split into a training set (n=1 584) and internal validation set (n=651), with the remaining datasets used for external validation. Predictors were selected via least absolute shrinkage and selection operator (LASSO) regression and Bayesian Information Criterion (BIC), and a nomogram model was constructed. An extreme gradient boosting (XGBoost) algorithm was used to build the machine learning model. Model performance was assessed by the concordance index (C-index), calibration curves, Brier score, decision curve analysis (DCA), and net reclassification improvement index (NRI). RESULTS: Five key predictors, age [odds ratio (OR)=1.02, 95% CI 1.01 to 1.03, P<0.001], minimum activated partial thromboplastin (APTT; OR=1.09, 95% CI 1.08 to 1.11, P<0.001), maximum APTT (OR=1.01, 95% CI 1.00 to 1.01, P<0.001), maximum lactate (OR=1.56, 95% CI 1.39 to 1.75, P<0.001), and maximum serum creatinine (OR=2.03, 95% CI 1.79 to 2.30, P<0.001), were included in the nomogram. The model showed robust performance in internal validation (C-index=0.845, 95% CI 0.811 to 0.879) and external validation (eICU-CRD: C-index=0.827, 95% CI 0.800 to 0.854; CSU-XYS-ICU: C-index=0.779, 95% CI 0.687 to 0.871). Calibration curves indicated good agreement between predicted and observed outcomes (Brier score<0.25), and DCA confirmed clinical benefit. The XGBoost model achieved an area under the receiver operating characteristic curve (AUC) of 0.982 (95% CI 0.969 to 0.985) in the training set, but performance declined in external validation (eICU-CRD, AUC=0.825, 95% CI 0.817 to 0.861; CSU-XYS-ICU, AUC=0.766, 95% CI 0.700 to 0.873), though it remained above clinical thresholds. Net reclassification improvement was slightly lower for XGBoost compared with the nomogram (NRI=0.58). CONCLUSIONS Both the nomogram and XGBoost models effectively predict in-hospital mortality in sepsis patients with DVT. However, the nomogram offers superior generalizability and clinical usability. Its visual scoring system provides a quantitative tool for identifying high-risk patients and implementing individualized interventions.
Humans ; Sepsis/complications* ; Machine Learning ; Nomograms ; Venous Thrombosis/complications* ; Retrospective Studies ; Hospital Mortality ; Male ; Female ; Middle Aged ; Aged ; Intensive Care Units ; Prognosis ; Bayes Theorem

OBJECTIVES: To evaluate the correlation of mean arterial oxygen tension (PaO₂) during the first 24 h following intensive care unit (ICU) admission with mortality in critically ill patients with acute kidney injury (AKI) and determine the optimal PaO₂ threshold for devising oxygen therapy strategies for these patients. METHODS: We collected the clinical data of ICU patients with AKI from the MIMIC-IV database. Based on the optimal first 24-h PaO₂ threshold determined by receiver operating characteristic (ROC) curve analysis and the Youden index maximization principle, we classified the patients into hyperoxia group (with PaO₂ ≥137.029 mmHg) and hypoxemia group (PaO₂<137.029 mm Hg). Multivariable logistic regression and propensity score matching were used to evaluate the correlation of first 24-h PaO₂ levels with in-hospital mortality of the patients. RESULTS: Among the 18 335 patients, 46.7% were in the hyperoxia group, who had an overall mortality rate of 16.9%. The optimal PaO₂ threshold (137.029 mm Hg) had a sensitivity of 78.3%, a specificity of 63.7%, and an AUC of 0.76 (95% CI: 0.74=0.78). Hyperoxia within the first 24 h after ICU admission was associated with a significantly lower in-hospital mortality (OR=0.78) and 90-day mortality (OR=0.77), particularly in stage 1 AKI patients. A non-linear relationship was identified between PaO₂ and mortality of the patients (P<0.001). Kaplan-Meier survival curves indicated a significantly increased 90-day survival rate in the patients in hyperoxia group (P<0.001), who also had shorter durations of mechanical ventilation, less vasopressor use, and shorter lengths of hospital/ICU stay. CONCLUSIONS Maintenance of a PaO₂ level ≥137.029 mmHg within 24 h after ICU admission may improve clinical outcomes of critically ill AKI patients, which underscores the importance of targeted oxygen delivery in ICU care.
Humans ; Acute Kidney Injury/blood* ; Male ; Female ; Middle Aged ; Intensive Care Units ; Aged ; Oxygen/blood* ; Hospital Mortality ; Partial Pressure ; Adult ; Databases, Factual

OBJECTIVE: To investigate the predictive value of oxygenation index (PaO₂/FiO₂) at intensive care unit (ICU) admission on 30-day mortality in patients with sepsis. METHODS: A retrospective study was conducted. Patients with sepsis who were hospitalized in the ICU of the Affiliated Hospital of Jining Medical University from April 2015 to October 2023 were enrolled. The demographic information, comorbidities, sites of infection, vital signs and laboratory test indicators at the time of admission to the ICU, disease severity scores within 24 hours of admission to the ICU, treatment process and prognostic indicators were collected. According to the PaO₂/FiO₂ at ICU admission, patients were divided into Q1 group (PaO₂/FiO₂ of 4.1-16.4 cmHg, 1 cmHg ≈ 1.33 kPa), Q2 group (PaO₂/FiO₂ of 16.5-22.6 cmHg), Q3 group (PaO₂/FiO₂ of 22.7-32.9 cmHg), and Q4 group (PaO₂/FiO₂ of 33.0-94.8 cmHg). Differences in the indicators across the four groups were compared. Multifactorial Cox regression analysis was used to assess the relationship between PaO₂/FiO₂ and 30-day mortality of patients with sepsis. The predictive value of PaO₂/FiO₂, sequential organ failure assessment (SOFA) and acute physiology and chronic health evaluation II (APACHE II) on 30-day prognosis of patients with sepsis was analyzed by receiver operator characteristic curve (ROC curve). RESULTS: A total of 1 711 patients with sepsis were enrolled, including 428 patients in Q1 group, 424 patients in Q2 group, 425 patients in Q3 group, and 434 patients in Q4 group. 622 patients died at 30-day, the overall 30-day mortality was 36.35%. There were statistically significant differences in age, body mass index (BMI), history of smoking, history of alcohol consumption, admission heart rate, respiratory rate, APACHE II score, SOFA score, Glasgow coma score (GCS), site of infection, Combined chronic obstructive pulmonary disease (COPD), blood lactic acid (Lac), prothrombin time (PT), albumin (Alb), total bilirubin (TBil), pH, proportion of mechanical ventilation, duration of mechanical ventilation, proportion of vasoactive medication used, and maximal concentration, length of ICU stay, hospital stay, incidence of acute kidney injury, in-hospital mortality, 30-day mortality among the four groups. Multivariate Cox regression analysis showed that after adjusting for confounding factors, for every 1 cmHg increase in PaO₂/FiO₂ at ICU admission, the 30-day mortality risk decreased by 2% [hazard ratio (HR) = 0.98, 95% confidence interval (95%CI) was 0.98-0.99, P < 0.001]. The 30-day mortality risk in the Q4 group was reduced compared with the Q1 group by 41% (HR = 0.59, 95%CI was 0.46-0.76, P < 0.001). The fitted curve showed that a curvilinear relationship between PaO₂/FiO₂ and 30-day mortality after adjustment for confounders. In the inflection point analysis, for every 1 cmHg increase in PaO₂/FiO₂ at PaO₂/FiO₂ < 28.55 cmHg, the risk of 30-day death in sepsis patients was reduced by 5% (HR = 0.95, 95%CI was 0.94-0.97, P < 0.001); when PaO₂/FiO₂ ≥ 28.55 cmHg, there was no statistically significant association between PaO2/FiO2 and the increase in the risk of 30-day death in sepsis (HR = 1.01, 95%CI was 0.99-1.02, P = 0.512). ROC curve analysis showed that the area under the curve (AUC) for the prediction of 30-day mortality by admission PaO₂/FiO₂ in ICU sepsis patients was 0.650, which was lower than the predictive ability of the SOFA score (AUC = 0.698) and APACHE II score (AUC = 0.723). CONCLUSION In patients with sepsis, PaO₂/FiO₂ at ICU admission is strongly associated with 30-day mortality risk, alerting healthcare professionals to pay attention to patients with low PaO₂/FiO₂ for timely interventions.
Humans ; Sepsis/mortality* ; Intensive Care Units ; Retrospective Studies ; Prognosis ; Hospital Mortality ; Oxygen ; Male ; Predictive Value of Tests ; Female ; Middle Aged ; Aged

OBJECTIVE: To investigate the predictive value of early lactate/albumin ratio (LAR) combined with quick sequential organ failure assessment (qSOFA) for the 28-day prognosis of patients with sepsis caused by emergency community-acquired pneumonia (CAP). METHODS: The clinical data of patients with sepsis caused by CAP admitted to the department of emergency of Beijing Haidian Hospital from June 2021 to August 2022 were retrospectively analyzed, including gender, age, comorbidities, lactic acid (Lac), serum albumin (Alb), LAR, procalcitonin (PCT) within 1 hour, and 28-day prognosis. Patients were divided into two groups based on 28-day prognosis, and risk factors affecting patients' prognosis were analyzed using univariate and multivariate Cox regression methods. Patients were divided into two groups according to the best cut-off value of LAR, and Kaplan-Meier survival curves were used to analyze the 28-day cumulative survival of patients in each group. Time-dependent receiver operator characteristic curve (ROC curve) were plotted to analyze the predictive value of sequential organ failure assessment (SOFA), acute physiology and chronic health evaluation II (APACHE II), and qSOFA+LAR score on the prognosis of patients with sepsis caused by CAP at 28 days. The area under the curve (AUC) was calculated and compared. RESULTS: A total of 116 patients with sepsis caused by CAP were included, of whom 80 survived at 28 days and 36 died, 28-day mortality of 31.0%. There were no statistically significant differences in age, gender, comorbidities, pH, platelet count, and fibrinogen between the survival and death groups, and there were significantly differences in blood urea nitrogen (BUN), white blood cell count (WBC), hemoglobin, Lac, Alb, PCT, D-dimer, LAR, as well as qSOFA score, SOFA score, and APACHE II score. Univariate Cox regression analyses showed that BUN, WBC, pH, Lac, Alb, PCT, LAR, qSOFA score, SOFA score, and APACHE II score were associated with mortality outcome. Multifactorial Cox regression analysis of the above variables showed that BUN, WBC, PCT, and APACHE II score were independent risk factors for 28-day death in the emergency department in patients with sepsis caused by CAP [hazard ratio (HR) were 1.081, 0.892, 1.034, and 1.135, respectively, all P < 0.05]. The best cut-off value of early LAR for predicting the 28-day prognosis of sepsis patients was 0.088, the Kaplan-Meier survival curve showed that the 28-day cumulative survival rate of sepsis patients in the LAR ≤ 0.088 group was significantly higher than that in the LAR > 0.088 group [82.9% (63/76) vs. 42.5% (17/40), Log-Rank test: χ² = 22.51, P < 0.001]. The qSOFA+LAR score was calculated based on the LAR cut-off value and qSOFA score, and ROC curve analysis showed that the AUCs of SOFA score, APACHE II score, and qSOFA+LAR score for predicting 28-day death of patients with sepsis caued by CAP were 0.741, 0.774, and 0.709, respectively, with the AUC of qSOFA+LAR score slightly lower than those of SOFA score and APACHE II score, but there were no significantly differences. When the best cut-off value of qSOFA+LAR score was 1, the sensitivity was 63.9% and the specificity was 80.0%. CONCLUSION The qSOFA+LAR score has predictive value for the 28-day prognosis of patients with sepsis caused by CAP in the emergency department, its predictive value is comparable to the SOFA score and the APACHE II score, and it is more convenient for early use in the emergency department.
Emergency Service, Hospital/statistics & numerical data* ; Sepsis/etiology* ; Prognosis ; Community-Acquired Pneumonia/mortality* ; Organ Dysfunction Scores ; Predictive Value of Tests ; Lactic Acid/blood* ; Serum Albumin, Human/analysis* ; Biomarkers/blood* ; Retrospective Studies ; Hospital Mortality ; Kaplan-Meier Estimate ; APACHE ; Procalcitonin/blood* ; ROC Curve ; Area Under Curve ; Humans

OBJECTIVE: To explore the risk factors affecting the prognosis of patients with sepsis complicated with acute pulmonary embolism, and to construct and validate a nomogram predictive model for in-hospital mortality risk. METHODS: Based on the American Medical Information Mart for Intensive Care (MIMIC-III, MIMIC-IV) databases, the data were collected on patients with sepsis complicated with acute pulmonary embolism from 2001 to 2019, including baseline characteristics, and vital signs, disease scores, laboratory tests within 24 hours of admission to the intensive care unit (ICU), and interventions. In-hospital mortality was the outcome event. The total samples were divided into training and testing sets in a 7:3 ratio by random sampling. Univariate Cox regression analysis was used to verify the impact of all variables on the risk of in-hospital mortality, thereby screen potential influencing factors. Subsequently, a stepwise bi-directional regression method was applied to select factors one by one, leading to the construction of a nomogram prediction model. Collinearity testing was used to demonstrate the absence of strong multicollinearity among the influencing factors in the nomogram prediction model. The discrimination of the nomogram model, sequential organ failure assessment (SOFA), and simplified pulmonary embolism severity index (sPESI) was evaluated using C-index in the test set. Receiver operator characteristic curve (ROC curve) was drawn to evaluate the predictive value of various models for in-hospital mortality in patients with sepsis complicated with acute pulmonary embolism. RESULTS: A total of 562 patients with sepsis complicated with acute pulmonary embolism were included, including 393 in the training set and 169 in the testing set. Univariate Cox regression analysis showed that 30 factors associated with in-hospital mortality in patients with sepsis complicated with acute pulmonary embolism. Through stepwise bi-directional regression, 12 variables were ultimately selected, including gender, presence of malignant tumors, body temperature, red cell distribution width (RDW), blood urea nitrogen (BUN), serum potassium, prothrombin time (PT), 24-hour urine output, mechanical ventilation, vasoactive drugs, warfarin use, and sepsis-induced coagulopathy (SIC). Collinearity testing indicated no strong multicollinearity among the influencing factors [all variance inflation factor (VIF) > 10]. A nomogram model was constructed using the 12 variables mentioned above. The nomogram model predicted the C-index and its 95% confidence interval (95%CI) of in-hospital mortality in patients with sepsis complicated with acute pulmonary embolism better than SOFA score and sPESI [0.771 (0.725-0.816) vs. 0.579 (0.519-0.639), 0.608 (0.554-0.663)]. The ROC curve showed that the area under the curve (AUC) and its 95%CI of the nomogram model were higher than those of the SOFA score and sPESI [0.811 (0.766-0.857) vs. 0.630 (0.568-0.691), 0.623 (0.566-0.680)]. These findings were consistently replicated in the internal validation of the testing set. In both the training and testing sets, Delong's test showed that the AUC of the nomogram model was significantly higher than the SOFA score and sPESI (both P < 0.05). CONCLUSION The nomogram model demonstrated good predictive effectiveness for the risk of in-hospital mortality in patients with sepsis complicated with acute pulmonary embolism, enabling clinicians to predict mortality risk in advance and take timely interventions to reduce mortality.
Humans ; Pulmonary Embolism/mortality* ; Hospital Mortality ; Nomograms ; Sepsis/complications* ; Prognosis ; Risk Factors ; Intensive Care Units ; Male ; Female ; Middle Aged ; Aged

OBJECTIVE: To explore the feasibility of incorporating simple bedside indicators into death predictive model for elderly critically ill patients based on interpretability machine learning algorithms, providing a new scheme for clinical disease assessment. METHODS: Elderly critically ill patients aged ≥ 65 years who were hospitalized in the intensive care unit (ICU) of Tacheng People's Hospital of Ili Kazak Autonomous Prefecture from June 2017 to May 2020 were retrospectively selected. Basic parameters including demographic characteristics, basic vital signs and fluid intake and output within 24 hours after admission, as well acute physiology and chronic health evaluation II (APACHE II), Glasgow coma score (GCS) and sequential organ failure assessment (SOFA) were also collected. According to outcomes in hospital, patients were divided into survival group and death group. Four datasets were constructed respectively, namely baseline dataset (B), including age, body temperature, heart rate, pulse oxygen saturation, respiratory rate, mean arterial pressure, urine output volume, infusion volume, and crystal solution volume; B+APACHE II dataset (BA), B+GCS dataset (BG), and B+SOFA dataset (BS). Then three machine learning algorithms, Logistic regression (LR), extreme gradient boosting (XGboost) and gradient boosting decision tree (GBDT) were used to develop the corresponding mortality predictive models within four datasets. The feature importance histogram of each prediction model was drawn by SHapley additive explanation (SHAP) method. The area under curve (AUC), accuracy and F1 score of each model were compared to determine the optimal prediction model and then illuminate the nomogram. RESULTS: A total of 392 patients were collected, including 341 in the survival group and 51 in the death group. There were statistically significant differences in heart rate, pulse oxygen saturation, mean arterial pressure, infusion volume, crystal solution volume, and etiological distribution between the two groups. The top three causes of death were shock, cerebral hemorrhage, and chronic obstructive pulmonary disease. Among the 12 prognostic models trained by three machine learning algorithms, overall performance of prognostic models based on B dataset was behind, whereas the LR model trained by BA dataset achieved the best performance than others with AUC of 0.767 [95% confidence interval (95%CI) was 0.692-0.836], accuracy of 0.875 (95%CI was 0.837-0.903) and F1 score of 0.190. The top 3 variables in this model were crystal solution volume with first 24 hours, heart rate and mean arterial pressure. The nomogram of the model showed that the total score between 150 and 230 were advisable. CONCLUSION The interpretable machine learning model including simple bedside parameters combined with APACHE II score could effectively identify the risk of death in elderly patients with critically illness.
Humans ; Critical Illness ; Machine Learning ; Aged ; Algorithms ; Intensive Care Units ; Retrospective Studies ; APACHE ; Prognosis ; Organ Dysfunction Scores ; Hospital Mortality ; Male ; Female