ANN Model Excels in Predicting Aspiration During Gastric Lavage for Acute Poisoning: A Comparison with RF, XGBoost, and GBDT

ANN Model Excels in Predicting Aspiration During Gastric Lavage for Acute Poisoning: A Comparison with RF, XGBoost, and GBDT | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article ANN Model Excels in Predicting Aspiration During Gastric Lavage for Acute Poisoning: A Comparison with RF, XGBoost, and GBDT Shuo Ni Zhang, Bo Zhang, Xuelan Liu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9227317/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Background: Aspiration during gastric lavage is a severe complication in acute poisoning patients, associated with high morbidity and mortality. Early identification of high-risk individuals is critical for improving patient safety. This study aimed to develop and validate machine learning models to predict aspiration risk during gastric lavage and compare their performance. Methods: A retrospective cohort study was conducted involving 843 adult patients who underwent gastric lavage for acute poisoning at a tertiary hospital in Ningbo, China, from January 2020 to July 2024. After excluding features with > 50% missing values, 25 variables were included. Four machine learning models—artificial neural network (ANN), random forest (RF), extreme gradient boosting (XGBoost), and gradient boosting decision tree (GBDT)—were trained and validated using five-fold cross-validation. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), calibration curves, Hosmer-Lemeshow test, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) were used to interpret the best-performing interpretable model. Results: Among 843 patients, 304 (36%) experienced aspiration. In both training and validation sets, the ANN model demonstrated the highest discriminative performance (training AUC = 0.911; validation AUC = 0.894), followed by XGBoost (0.893 and 0.885), GBDT (0.852 and 0.766), and RF (0.846 and 0.805). ANN also showed superior calibration and clinical utility. SHAP analysis of the XGBoost model identified consciousness status, age, time since toxin ingestion, and neutrophil-to-lymphocyte ratio (NLR) as the most influential predictors. Conclusion: The ANN model outperformed RF, XGBoost, and GBDT in predicting aspiration risk during gastric lavage in acute poisoning patients. Key risk factors included impaired consciousness, advanced age, delayed lavage, and elevated NLR. Integrating the ANN model into clinical decision-making may enhance early risk identification and improve patient outcomes. Future studies should focus on prospective validation and real-time implementation. Aspiration Gastric lavage Machine learning Artificial neural network Risk prediction Acute poisoning Figures Figure 1 Figure 2 Figure 3 Figure 4 1. Introduction In recent years, China has witnessed an increasing number of acute poisoning cases presenting to emergency departments, with a particularly notable rise in drug overdoses and food poisoning incidents, thereby placing higher demands on emergency care capabilities [ 1 ] . Data from the *China Health Statistics Yearbook 2022* [ 2 ] indicate that from 2005 to 2021, injuries and poisonings consistently ranked as the fifth leading cause of death among both urban and rural residents in China. Ingestion via the digestive tract was identified as the most common route of toxic exposure. Acute poisoning is a systemic disorder characterized by the absorption of a substantial amount of toxins within a short period, leading to functional or structural damage of tissues and organs. It is marked by acute onset, rapid progression, and high fatality rates. As a common critical condition in the emergency department, it poses a severe threat to patient life and health [ 3 ] . Therefore, implementing prompt and effective therapeutic interventions is crucial for reducing mortality. Gastric lavage, employed as a pivotal method for toxin elimination in China for nearly two centuries, plays a significant role in the management of acute poisoning [ 4 ] . However, the procedure itself can lead to complications such as electrolyte disturbances and aspiration [ 4 ] . Research [ 5 ] has demonstrated that aspiration is a critical risk factor for poor patient outcomes; greater aspiration volume correlates with more severe injury, worse prognosis, and can result in an in-hospital mortality rate as high as 70%. In conclusion, within the context of acute poisoning treatment involving gastric lavage, the identification of prognostic risk factors and the development of an early and accurate predictive model for stratifying aspiration risk are of paramount clinical importance and practical value for improving patient outcomes and reducing mortality. ML algorithms can capture nonlinear relationships in data and complexity among multiple predictor variables, accurately identify risk factors for diseases, and make early predictions and diagnosis [ 6 ] .Artificial neural networks(ANNs) are a subset of traditional machine learning methods, belonging to the field of AI. Their structure and function are designed to resemble biologic nervous systems, with powerful learning algorithms and training capabilities to perform simulations with high accuracy [ 7 ] . The weights are randomly chosen at the beginning of a training process and are then calculated to minimize errors between predicted values from the output layers and actual values [ 8 ] .Fujita et al. [ 9 ] developed Random Forest(RF) model to evaluate the carcinogenicity of 230 chemicals. The balanced accuracy from the evaluation for the RF model was 0.755. Although RF is less interpretable, it is computationally efficient and has been very successful in developing classification models for a wide range of toxicity types.The eXtreme Gradient Boosting(XGBoost) model is a collection of weak prediction trees capable of capturing complex relationships in the data without the need for higher-order interactions and nonlinear functions that are explicitly specified. A study had shown that in accordance with the predictive factors, XGBoost model, characterized by excellent discrimination and high accuracy, can be used to clinically identify severe acute pancreatitis patients with a high risk of enteral nutrition aspiration [ 10 ] . The Gradient Boosting Decision Tree(GBDT) algorithm is less sensitive to hyperparameters, less prone to overfitting, and easy to implement. And it has been applied to diagnose and predict the outcomes of several diseases [ 11 ] . The purpose of this study is to develop and validate three prediction models, RF, XGBoost and GBDT, and compare their effectiveness with the traditional ANN model to predict the risk of aspiration in patients with acute poisoning after gastric lavage. 2. Patients and methods 2.1 Inclusion and exclusion criteria A retrospective study was conducted, selecting patients who underwent gastric lavage for acute poisoning in the emergency department of a tertiary general hospital in Ningbo City from January 2020 to July 2024 as the research subjects. Inclusion criteria: ① Age ≥ 18 years; ② Ingested toxic substances orally with indications for gastric lavage; ③ Comatose patients who could undergo gastric lavage smoothly under airway protection; ④ All patients or their families have signed the gastric lavage informed consent form. Exclusion criteria: ① Presence of contraindications for gastric lavage; ② Poor cooperation, making smooth gastric lavage impossible; ③ Accompanied by severe organ diseases. 2.2 Data collection The collected data is categorized into three major groups: patient general information, clinical indicators, and laboratory indicators. Patient general information includes age, gender, consciousness status, history of hypertension, past medical history, types of ingested toxins, and time of toxin ingestion. Clinical indicators include resuscitation drugs, endotracheal intubation status, and imaging results. Laboratory indicators include WBC, neutrophil count, lymphocyte count, NLR, CRP, creatinine, K+, Na+, CK, CK-MB, AST, ALT, AST/ALT ratio, D-dimer, fibrin (ogen) degradation products, cholinesterase, serum amylase, and Brain Natriuretic Peptide (BNP). 2.3 The definition of aspiration 《The Chinese Expert Consensus on Swallowing Disorder Assessment and Treatment》 [ 12 ] states that aspiration refers to the phenomenon where contents from the oropharynx or stomach are inhaled into the lower respiratory tract below the vocal cords. After aspiration occurs, patients immediately exhibit symptoms such as irritative coughing, shortness of breath, or even asthma, which is termed overt aspiration. On the other hand, covert aspiration occurs when patients do not exhibit external signs such as coughing within 1 minute of aspiration and do not experience symptoms like irritative coughing or shortness of breath. Covert aspiration is often misdiagnosed and can be identified based on the following criteria [ 13 ] : ① No vomiting before gastric lavage, but nausea and vomiting occur during lavage, with gastric contents spewing out through the mouth and nose, accompanied by reduced breath sounds or crackles in the lungs, and possibly difficulty breathing or cyanosis; ② Gastric contents are suctioned from the airway after lavage; ③ A chest X-ray or CT scan performed after lavage shows new patchy shadows in the lung fields, possibly accompanied by localized atelectasis, confirming aspiration pneumonia. Meeting any one of these criteria indicates aspiration. 2.4 Sample size and missing values We included 29 baseline characteristic data as study variables in this study. Based on the rule of thumb, the sample size was estimated to be 10 to 15 times the number of variables [ 14 ] . Considering a 10% data loss, the estimated sample size was 484 cases. Ultimately, this study included 843 acute poisoning patients undergoing gastric lavage who met the inclusion and exclusion criteria. Additionally, we removed features with missing values exceeding 50%, a total of three, namely BNP, Cholinesterase, and Blood_amylase. 2.5 Ethics statement The research protocol (KY2023R270) was approved by the Ethics Committee of Ningbo Medical Center Lihuili Hospital. Written informed consent from participants is not required as their data are analyzed retrospectively and anonymously. 2.6 Statistical analysis Using the Scikit-Learn database in Python 3.11 software, count data are represented by frequency, percentage, or proportion. Statistical inference uses chi-square tests and non-parametric rank sum tests; continuous variables are expressed as M(P25, P75), and intergroup comparisons use the Mann-Whitney U test. Differences are considered statistically significant at P < 0.05.Three prediction models, including RF, XGBoost, and GBDT, were developed and validated. Their performance was compared with traditional artificial neural network (ANN) models to select the optimal algorithm and model for better predicting aspiration risk in patients undergoing gastric lavage for acute poisoning.The performance evaluation metrics of predictive models include receiver operating characteristic (ROC) curve, area under curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value(NPV), decision curve analysis (DCA) curve, Hosmer-Lemeshow (HL) curve, and inter-group difference analysis. Select the optimal model among RF, XGBoost, and GBDT, and use the SHapley Additive exPlanation (SHAP) tool to visualize and explain the decision-making role of features in the optimal model. Assess the feature importance in the decision-making process and demonstrate the contribution of each feature from a group-level perspective. Data processing is conducted under the guidance of professionals to ensure the objectivity and scientific nature of the analysis. 3. Results 3.1 Baseline characteristics This study collected a total of 28 research variables, excluding imaging results. Through data preprocessing, 3 samples with missing values > 50% were removed, and ultimately, 25 feature variables were included, comprising aspiration and 24 risk factors for aspiration. Clinical data of patients in the training set and validation set were compared, with T-tests performed for continuous variables and chi-square tests for categorical variables. No statistically significant differences were found between groups for independent variables (P > 0.05), indicating meaningful data partitioning.(Table 1 ) Table 1 Comparison of Factors Affecting Aspiration During Gastric Lavage in Acute Poisoning Patients Using Training and Testing Set Data( n = 843) Variables Total ( n = 843) Train ( n = 591) Validation (n = 252) P Statistic Method aspiration, n (%) 1 0 Chi-squared yes 304 (36) 213 (36) 91 (36) no 539 (64) 378 (64) 161 (64) sex, n (%) 0.11 2.548 Chi-squared male 262 (31) 194 (33) 68 (27) female 581 (69) 397 (67) 184 (73) age, Median (Q1,Q3) 28 (19, 45.5) 28 (18, 44.5) 28 (19, 51) 0.183 -1.446 T-test category, n (%) 0.703 0.706 Chi-squared antipsychotics 517 (61) 366 (62) 151 (60) pesticide 144 (17) 102 (17) 42 (17) other 182 (22) 123 (21) 59 (23) time_h, Median (Q1,Q3) 2 (1, 3) 2 (1, 3) 2 (1, 3) 0.608 1.292 T-test history, n (%) 0.475 1.487 Chi-squared mental illness 420 (50) 297 (50) 123 (49) other 97 (12) 72 (12) 25 (10) no 326 (39) 222 (38) 104 (41) hypertension, n (%) 0.671 0.943 Chi-squared yes 84 (10) 56 (9) 28 (11) no 759 (90) 535 (91) 224 (89) Rescue_medication, n (%) 0.672 1.588 Chi-squared 1 360 (43) 260 (44) 100 (40) 2 8 (1) 6 (1) 2 (1) 3 471 (56) 322 (54) 149 (59) 4 4 (0) 3 (1) 1 (0) consciousness, n (%) 0.938 0.143 Chi-squared sober 608 (72) 424 (72) 184 (73) sleepy 221 (26) 157 (27) 64 (25) coma 14 (2) 10 (2) 4 (2) Endotracheal_intubation, n (%) 0.755 0.097 Chi-squared yes 17 (2) 13 (2) 4 (2) no 826 (98) 578 (98) 248 (98) WBC, Median (Q1,Q3) 7.3 (6.2, 8.85) 7.3 (6.25, 8.8) 7.3 (6.1, 8.9) 0.635 -0.069 T-test Neutrophils, Median (Q1,Q3) 4.7 (3.65, 6.05) 4.7 (3.7, 6.25) 4.7 (3.5, 5.73) 0.248 0.409 T-test lymphocyte, Median (Q1,Q3) 1.8 (1.3, 2.3) 1.8 (1.3, 2.2) 1.8 (1.37, 2.4) 0.09 -1.218 T-test Neutrophil to Lymphocyte Ratio, Median (Q1,Q3) 2.58 (1.89, 3.87) 2.58 (1.96, 4) 2.58 (1.81, 3.55) 0.209 0.719 T-test CRP, Median (Q1,Q3) 0.8 (0.25, 1.5) 0.8 (0.25, 1.4) 0.8 (0.25, 1.62) 0.511 0.204 T-test creatinine, Median (Q1,Q3) 56.65 (50.3, 65) 56.65 (50.75, 64.9) 56.65 (49.77, 66.25) 0.735 0.061 T-test K+, Median (Q1,Q3) 3.61 (3.4, 3.8) 3.61 (3.42, 3.8) 3.61 (3.37, 3.79) 0.363 0.114 T-test Na+, Median (Q1,Q3) 139.8 (138.3, 141.1) 139.8 (138.5, 141.15) 139.8 (138.1, 141) 0.147 0.373 T-test CK, Median (Q1,Q3) 81 (63, 106) 81 (64, 105) 81 (62, 113) 0.872 1.312 T-test CKMB, Median (Q1,Q3) 11 (8.5, 14.65) 11 (8.4, 14.75) 11 (8.9, 14.43) 0.812 -0.736 T-test Aspartate_aminotransferase, Median (Q1,Q3) 25 (23, 28) 25 (23, 28) 25 (22.75, 29) 0.611 -0.382 T-test Alanine_aminotransferase, Median (Q1,Q3) 17 (14, 21.5) 17 (14, 21) 17 (14, 22) 0.757 0.439 T-test AST_ALT, Median (Q1,Q3) 1.48 (1.31, 1.66) 1.48 (1.32, 1.65) 1.48 (1.3, 1.67) 0.634 0.53 T-test D_dimer, Median (Q1,Q3) 89.5 (57, 143.5) 89.5 (55, 152) 89.5 (59.75, 126.5) 0.509 -0.541 T-test Fibrinogen_degradation, Median (Q1,Q3) 0.82 (0.57, 1.16) 0.82 (0.57, 1.17) 0.82 (0.57, 1.13) 0.962 0.08 T-test 3.2 Univariable analysis of the training sample(Table 2 ) A baseline data table was created for the training set, and features were selected. The results (Table 2 ) show that the following 19 predictors have significant differences: sex, age, category, time_h, history, hypertension, Rescue_medication, consciousness, Neutrophils, CK, Alanine_aminotransferase, lymphocyte, Neutrophil to Lymphocyte Ratio (NLR), CRP, creatinine, CKMB, Aspartate_aminotransferase, D_dimer, and Fibrinogen_degradation. Table 2 Analysis of associated factors of aspiration during gastric lavage in acute poisoning patients using training samples ( n = 591) Variables Total ( n = 591) Negative ( n = 213) Positive (n = 378) P Statistic Method sex, n (%) 0.013 6.14 Chi-squared male 194 (33) 84 (39) 110 (29) female 397 (67) 129 (61) 268 (71) age, Median (Q1,Q3) 28 (18, 44.5) 41 (24, 65) 24 (17, 34) < 0.001 -10.15 T-test category, n (%) 0.023 7.527 Chi-squared antipsychotics 366 (62) 135 (63) 231 (61) pesticide 102 (17) 45 (21) 57 (15) other 123 (21) 33 (15) 90 (24) time_h, Median (Q1,Q3) 2 (1, 3) 3 (1, 7) 1 (1, 2) < 0.001 -6.302 T-test history, n (%) < 0.001 29.328 Chi-squared mental illness 297 (50) 111 (52) 186 (49) other 72 (12) 44 (21) 28 (7) no 222 (38) 58 (27) 164 (43) hypertension, n (%) < 0.001 46.576 Chi-squared yes 56 (9) 43 (20) 13 (3) no 535(91) 170 (80) 365 (97) Rescue_medication, n (%) 0.002 12.868 Chi-squared 1 260 (44) 113 (53) 147 (39) 2 6 (1) 2 (1) 4 (1) 3 322 (54) 96 (45) 226 (60) 4 3 (1) 2 (1) 1 (0) consciousness, n (%) < 0.001 129.551 Chi-squared sober 424 (72) 93 (44) 331 (88) sleepy 157 (27) 113 (53) 44 (12) coma 10 (2) 7 (3) 3 (1) Endotracheal_intubation, n (%) 0.077 2.703 Chi-squared yes 13 (2) 8 (4) 5 (1) no 578 (98) 205 (96) 373 (99) WBC, Median (Q1,Q3) 7.3 (6.25, 8.8) 7.3 (6.2, 9.4) 7.3 (6.3, 8.6) 0.285 -2.17 T-test Neutrophils, Median (Q1,Q3) 4.7 (3.7, 6.25) 4.7 (3.9, 7) 4.7 (3.52, 5.97) 0.038 -3.041 T-test lymphocyte, Median (Q1,Q3) 1.8 (1.3, 2.2) 1.7 (1.2, 2) 1.8 (1.5, 2.28) < 0.001 3.146 T-test Neutrophil to Lymphocyte Ratio, Median (Q1,Q3) 2.58 (1.96, 4) 2.62 (2.15, 5.1) 2.58 (1.83, 3.38) < 0.001 -3.808 T-test CRP, Median (Q1,Q3) 0.8 (0.25, 1.4) 0.8 (0.6, 3.1) 0.8 (0.25, 1) < 0.001 -4.262 T-test creatinine, Median (Q1,Q3) 56.65 (50.75, 64.9) 56.65 (52.5, 67.8) 56.65 (49.2, 63.7) 0.012 -2.679 T-test K+, Median (Q1,Q3) 3.61 (3.42, 3.8) 3.61 (3.36, 3.79) 3.61 (3.44, 3.81) 0.134 1.666 T-test Na+, Median (Q1,Q3) 139.8 (138.5, 141.15) 139.8 (138.1, 141.45) 139.8 (138.6, 141.1) 0.399 1.063 T-test CK, Median (Q1,Q3) 81 (64, 105) 81 (67, 127) 81 (62, 98) 0.011 -2.152 T-test CKMB, Median (Q1,Q3) 11 (8.4, 14.75) 11 (8.6, 15.8) 11 (8, 13.95) 0.023 -1.893 T-test Aspartate_aminotransferase, Median (Q1,Q3) 25 (23, 28) 25 (24, 30) 25 (23, 26) 0.006 -1.944 T-test Alanine_aminotransferase, Median (Q1,Q3) 17 (14, 21) 17 (15, 23) 17 (14, 20) 0.003 -1.927 T-test AST_ALT, Median (Q1,Q3) 1.48 (1.32, 1.65) 1.48 (1.27, 1.58) 1.48 (1.41, 1.67) 0.055 1.244 T-test D_dimer, Median (Q1,Q3) 89.5 (55, 152) 123 (83, 270) 89.5 (49, 107) < 0.001 -4.65 T-test Fibrinogen_degradation, Median (Q1,Q3) 0.82 (0.57, 1.17) 1 (0.74, 1.76) 0.82 (0.51, 0.99) < 0.001 -4.76 T-test 3.3 Develop, compare, and calibrate predictive models for training samples Using the occurrence of patient aspiration as the dependent variable and 19 factors identified as statistically significant in univariate analysis as independent variables, we developed four prediction models: Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Gradient Boosting Decision Tree (GBDT), and Artificial Neural Network (ANN). The performance metrics—including AUC, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV)—were reported as the average values derived from five-fold cross-validation for each algorithm. The RF model achieved an AUC of 0.846 (95% CI: 0.814–0.877), accuracy of 0.790, sensitivity of 0.942, specificity of 0.521, PPV of 0.777, and NPV of 0.835. The XGBoost model yielded an AUC of 0.893 (95% CI: 0.868–0.921), accuracy of 0.816, sensitivity of 0.921, specificity of 0.629, PPV of 0.815, and NPV of 0.817. The GBDT model showed an AUC of 0.852 (95% CI: 0.820–0.883), accuracy of 0.787, sensitivity of 0.944, specificity of 0.507, PPV of 0.773, and NPV of 0.837. The ANN model demonstrated an AUC of 0.911 (95% CI: 0.885–0.935), accuracy of 0.841, sensitivity of 0.931, specificity of 0.681, PPV of 0.838, and NPV of 0.848. Furthermore, we plotted the receiver operating characteristic (ROC) curves, decision curve analysis (DCA) curves, and Hosmer-Lemeshow (HL) calibration curves, and conducted between-group difference analysis (Fig. 1 ). The ANN-based prediction model for aspiration risk achieved the largest area under the ROC curve (Fig. 1 A), indicating its superior predictive performance. The calibration curve and Hosmer-Lemeshow goodness-of-fit test (Fig. 1 C) suggested excellent agreement between the predicted probabilities of aspiration risk and the observed outcomes in the ANN model. DCA (Fig. 1 B) further indicated that the ANN model had high clinical utility. Additionally, the risk scores were significantly higher in the aspiration group compared to the non-aspiration group, with a statistically significant intergroup difference (p < 0.05). In conclusion, based on a comprehensive comparison of all evaluation metrics, the ANN algorithm constructed the most effective prediction model for aspiration risk. 3.4 Validate, compare, and calibrate the predictive model on test samples The predictive performance of the models was evaluated using the same performance metrics as those applied during model development, with the model exhibiting the highest AUC value selected as the optimal one. The RF model achieved an AUC of 0.805 (95% CI: 0.746–0.858), accuracy of 0.762, sensitivity of 0.938, specificity of 0.451, PPV of 0.751, and NPV of 0.804. The XGBoost model yielded an AUC of 0.885 (95% CI: 0.844–0.923), accuracy of 0.798, sensitivity of 0.969, specificity of 0.495, PPV of 0.772, and NPV of 0.900. The GBDT model showed an AUC of 0.766 (95% CI: 0.704–0.826), accuracy of 0.714, sensitivity of 0.925, specificity of 0.341, PPV of 0.713, and NPV of 0.721. The ANN model demonstrated an AUC of 0.894 (95% CI: 0.843–0.936), accuracy of 0.837, sensitivity of 0.901, specificity of 0.725, PPV of 0.853, and NPV of 0.805. Furthermore, receiver operating characteristic (ROC) curves, decision curve analysis (DCA) curves, Hosmer–Lemeshow (HL) calibration curves, and between-group difference analysis were plotted (Fig. 2 ). The ANN-based aspiration risk prediction model achieved the largest area under the ROC curve (Fig. 2 A). The calibration curve and Hosmer–Lemeshow goodness-of-fit test (Fig. 2 C) indicated excellent agreement between the predicted probabilities of aspiration risk and the observed outcomes in the ANN model. DCA (Fig. 2 B) further demonstrated the high clinical utility of the ANN model. Additionally, the risk scores were significantly higher in the aspiration group than in the non-aspiration group, with a statistically significant intergroup difference (p < 0.05). Therefore, based on a comprehensive comparison of all evaluation metrics, the ANN-based aspiration risk prediction model demonstrated the best overall performance. 3.5 The explainability of predictive models The optimal model among RF, XGBoost, and GBDT was determined to be XGBoost (AUC = 0.885). The top ten features in terms of importance were visualized and ranked as follows: age, consciousness, time_h, NLR, fibrinogen degradation, lymphocyte, D-dimer, CRP, neutrophils, and creatinine (Fig. 3 ). To enhance interpretability, the SHapley Additive exPlanation (SHAP) tool was employed to visualize the contribution of each feature to the decision-making process of the XGBoost model. This approach illustrates the average impact of individual features on model output, helping clinicians understand the rationale behind specific predictions and thereby increasing trust in the model's decision process. The SHAP summary plot (Fig. 4 ) displays the direction and magnitude of each feature's influence on the model’s prediction of aspiration risk. Each row represents the distribution of SHAP values for a specific feature. SHAP values greater than 0 indicate that the feature contributes to a positive prediction (aspiration risk), while values less than 0 suggest a tendency toward a negative prediction. In Fig. 4 , the color of each point represents the corresponding feature value: red indicates a higher value, and blue indicates a lower value. Features were sorted based on the sum of the absolute SHAP values across all samples in the dataset. The four most important predictors were consciousness, age, time to toxin absorption (time_h), and NLR. Among these, consciousness showed a wide distribution of SHAP values with large absolute magnitudes, indicating it is the most influential predictor—a finding highly consistent with clinical experience, which significantly enhances the model’s credibility. Through SHAP value analysis, the model provides clear and interpretable evidence for predictions, assisting clinicians in comprehending the decision logic, optimizing clinical decision-making, and improving doctor-patient communication. Therefore, the model not only demonstrates strong predictive performance but also offers excellent interpretability and practical clinical applicability. 4. Discussion In this retrospective study, we developed and compared four machine learning models—ANN, RF, XGBoost, and GBDT—to predict the risk of aspiration during gastric lavage in patients with acute oral poisoning. Among them, the ANN model demonstrated the best discriminative performance with an AUC of 0.911 in the training set and 0.894 in the validation set. Calibration curves and DCA further confirmed its clinical utility and reliability. To enhance interpretability, SHAP analysis was applied to the XGBoost model, revealing that consciousness status, age, time since toxin ingestion, and NLR were the most influential predictors of aspiration risk. Our findings align with existing clinical evidence that impaired consciousness is the strongest predictor of aspiration. Patients with altered mental status are more likely to lose protective airway reflexes, increasing the likelihood of gastric content reflux into the respiratory tract. This observation is concordant with the work of Chi-Syuan Pan et al. [ 15 ] Therefore, during gastric lavage for patients with impaired consciousness, nurses should exercise greater caution and assist the patient in turning their head to the side to prevent complications such as aspiration. Age was also identified as a significant risk factor. This finding is consistent with the results reported by Conzelmann et al [ 16 ] . Older patients often exhibit decreased gag reflexes, delayed gastric emptying, and comorbidities that compromise respiratory defense mechanisms. This suggests that age-stratified protocols should be considered when deciding on gastric lavage, particularly in elderly patients with diminished physiological reserves. When performing gastric lavage on elderly patients, nursing procedures must be more gentle and meticulous. It is crucial to closely monitor the balance of irrigation fluid inflow and outflow to prevent gastric intolerance. A longer time interval between toxin ingestion and gastric lavage was associated with an increased risk of aspiration. A study by Akram et al. [ 17 ] demonstrated that a shorter time interval from toxic substance ingestion to hospital admission is associated with improved patient recovery and reduced mortality.Early lavage may reduce gastric volume and toxin absorption, thereby lowering the chance of regurgitation. This reinforces the importance of timely intervention in acute poisoning cases, ideally within the first hour post-ingestion. Interestingly, neutrophil-to-lymphocyte ratio (NLR), a marker of systemic inflammation, was found to be significantly elevated in patients who aspirated. Several investigators [ 18 ] have reported a significant correlation between an increased NLR and the presence of infection.This may reflect a stress response to toxin exposure, which could indirectly impair swallowing coordination or lower the threshold for vomiting. While NLR is not traditionally used in aspiration risk assessment, its inclusion in predictive models may offer additional prognostic value. Although ANN outperformed other models, its “black-box” nature limits direct clinical interpretability. To address this, we employed SHAP analysis on the XGBoost model, which ranked features in a clinically meaningful way. This hybrid approach—using ANN for prediction and XGBoost for explanation—may offer a balanced solution between accuracy and transparency in future clinical decision support systems. Despite the strengths, this study has several limitations. First, its retrospective design may introduce selection bias. Second, the definition of aspiration was based on clinical and radiological criteria, which may underdiagnose covert aspiration. Third, external validation in multicenter cohorts is needed to confirm generalizability. Lastly, the ANN model, while accurate, requires further simplification or visualization tools to facilitate bedside application. 5. Conclusion In conclusion, we developed a machine learning-based prediction model to assess aspiration risk during gastric lavage in patients with acute oral poisoning. The ANN model demonstrated superior predictive performance and clinical utility compared to RF, XGBoost, and GBDT. Key risk factors included impaired consciousness, advanced age, delayed lavage, and elevated NLR. Integrating this model into clinical workflows may support early identification of high-risk patients, guide preventive strategies, and ultimately improve patient safety. Future studies should focus on prospective validation, model simplification, and real-time implementation in emergency settings. Ethics Statement The study was conducted in accordance with the Declaration of Helsinki. Declarations Competing interests All authors declare no conflict of interest. Author contribution statement Shuoni Zhang: literature retrieval and analysis, data collection and collation, statistical analysis, thesis writing; Bo Zhang : data collection and collation; Xuelan Liu: thesis guidance, statistics guidance, financial support. Funding This work was partially supported by Zhejiang medical and health project(2023KY249).But the funds were not used. Author Contribution Shuoni Zhang: literature retrieval and analysis, data collection and collation, statistical analysis, thesis writing; Bo Zhang : data collection and collation; Xuelan Liu: thesis guidance, statistics guidance, financial support. Data Availability All data supporting the findings of this study are available within the paper and its Supplementary Information. References Gong F, Wang Y. Application effect of fully automatic gastric lavage machine in emergency poisoned patients[J]. China Med Device Inform. 2025;31(06):72–4. National Health Commission of the People’s Republic of China. China Health Statistics Yearbook 2022[M]: Beijing. Peking Union Medical College Pres; 2022. Zhang YT, Jiang SF, Lang N et al. [epidemiological characteristics and toxicant type of acute poisoning cases in china, 2016–2022][J].Zhonghua Liu Xing Bing Xue Za Zhi,2024,45(10):1376–82. Xu C, Shi YT, Zhu Y, et al. Analysis of gastric lavage status and indications in 330 poisoned patients at primary hospitals[J]. Chin J Crit Care Med. 2025;45(1):63–6. Košutova P. Mikolka P.Aspiration syndromes and associated lung injury: Incidence, pathophysiology and management[J]. Physiol Res 2021,70(Suppl4):S567–83. Bacchi S, Tan Y, Oakden-Rayner L, et al. Machine learning in the prediction of medical inpatient length of stay[J]. Intern Med J. 2022;52(2):176–85. Hong WD, Chen XR, Jin SQ et al. Use of an artificial neural network to predict persistent organ failure in patients with acute pancreatitis[J].Clinics (Sao Paulo),2013,68(1):27–31. Guo W, Liu J, Dong F et al. Review of machine learning and deep learning models for toxicity prediction[J]. Exp Biol Med (Maywood) 2023,248(21):1952–73. Fujita Y, Honda H,Yamane M et al. A decision tree-based integrated testing strategy for tailor-made carcinogenicity evaluation of test substances using genotoxicity test results and chemical spaces[J].Mutagenesis,2019,34(1):101–9. Zhang B, Xu H, Xiao Q et al. Machine learning predictive model for aspiration risk in early enteral nutrition patients with severe acute pancreatitis[J].Heliyon,2024,10(23):e40236. Liu F, Yao J, Liu C, et al. Construction and validation of machine learning models for sepsis prediction in patients with acute pancreatitis[J]. BMC Surg. 2023;23(1):267. Chinese Expert Consensus Group on Rehabilitation Assessment and Treatment of Dysphagia. Chinese expert consensus on assessment and treatment of dysphagia (2017 edition) Part 2: Treatment and rehabilitation management[J]. Chin J Phys Med Rehabilitation. 2018;40(1):1–10. Liu CJ, Chang N. Prevention and nursing of enteral nutrition complicated with aspiration in mechanically ventilated patients[J]. Int J Nurs. 2019;38(15):2380–3. Gao YX, Zhang JX. Sample size determination in logistic regression analysis[J]. J Evidence-Based Med. 2018;18(2):122–4. Pan CS, Lee CC, Yu JH, et al. Reassessing clinical presentations of emamectin benzoate poisoning: A comprehensive study[J]. Hum Exp Toxicol. 2024;43:9603271241249965. Conzelmann M, Hoidis A,Bruckner T et al. Aspiration risk in relation to glasgow coma scale score and clinical parameters in patients with severe acute alcohol intoxication: A single-centre, retrospective study[J].BMJ Open,2021,11(10):e053619. Akram S. Fazil M,Ullah K.Poppy intoxication in infants and children: Hazards of a folk remedy[J]. J Coll Physicians Surg Pak 2021,30(5):576–81. Burchette DT, Dasci MF, Fernandez Maza B, et al. Neutrophil-lymphocyte ratio and lymphocyte-monocyte ratio correlate with chronic prosthetic joint infection but are not useful markers for diagnosis[J]. Arch Orthop Trauma Surg. 2024;144(1):297–305. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 04 May, 2026 Reviewers agreed at journal 02 May, 2026 Reviewers invited by journal 30 Apr, 2026 Editor invited by journal 03 Apr, 2026 Editor assigned by journal 31 Mar, 2026 Submission checks completed at journal 31 Mar, 2026 First submitted to journal 25 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9227317","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":635973025,"identity":"ae64f714-3298-48ee-8a7c-1bd503440f83","order_by":0,"name":"Shuo Ni Zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0UlEQVRIiWNgGAWjYBACNmb+jw8+/rFhtm9vPkCcFj52BmPDmQ1p7AY8xxKI0yLHz2AmzNtwiN9AIseAWIcxpDHO3HFA2pwh5+ONNwx2croNhLUce/DxzB1jy4azmy3nMCQbmx0gqIWx3XAG27NkhoO926R5GA4kbiOshZlNmoftcH3DYZ5nxGphY5PmbTvMbHCMh41YLTzMhjPOpDFL9rAZW84xIMIv8v1nGB98qLBh5pd//PDGmwo7OYJaUIAED5FRg6yFVB2jYBSMglEwIgAAvtg+921EGPcAAAAASUVORK5CYII=","orcid":"","institution":"Ningbo Medical Center Lihuili Hospital","correspondingAuthor":true,"prefix":"","firstName":"Shuo","middleName":"Ni","lastName":"Zhang","suffix":""},{"id":635973028,"identity":"fc7a5c03-90cb-497f-b879-a0b3c5ee0991","order_by":1,"name":"Bo Zhang","email":"","orcid":"","institution":"Ningbo Medical Center Lihuili Hospital","correspondingAuthor":false,"prefix":"","firstName":"Bo","middleName":"","lastName":"Zhang","suffix":""},{"id":635973029,"identity":"12432a2f-1367-47ca-8308-6d575edaca07","order_by":2,"name":"Xuelan Liu","email":"","orcid":"","institution":"Ningbo Medical Center Lihuili Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xuelan","middleName":"","lastName":"Liu","suffix":""}],"badges":[],"createdAt":"2026-03-25 22:23:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9227317/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9227317/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108968969,"identity":"bffa410e-b442-4c2f-bf11-8775e8fbd0ef","added_by":"auto","created_at":"2026-05-11 10:07:13","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":155865,"visible":true,"origin":"","legend":"\u003cp\u003eEvaluation of Predictive Models on Training Samples：(A) ROC curve; (B) DCA curve; (C) HL curve; (D) Analysis of Variance between groups\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-9227317/v1/c54e426dd56bb994b72a21a8.png"},{"id":108968970,"identity":"e01f47cf-5ef7-4c7f-825d-73cd49b8b34e","added_by":"auto","created_at":"2026-05-11 10:07:13","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":184221,"visible":true,"origin":"","legend":"\u003cp\u003ePredictive Model Evaluation on Test Samples：(A) ROC curve; (B) DCA curve; (C) HL curve; (D) Analysis of Variance between groups\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-9227317/v1/446558ae727f51b464c70a75.png"},{"id":108978181,"identity":"5dc1d66b-2d50-4a8f-a57d-e8adebcff615","added_by":"auto","created_at":"2026-05-11 11:34:41","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":48447,"visible":true,"origin":"","legend":"\u003cp\u003eFeature importance ranking of XGBoost model\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-9227317/v1/1e70c228222380fa06f88bbd.png"},{"id":108968971,"identity":"f2256392-652f-4e12-89fe-4fcf4ff43e36","added_by":"auto","created_at":"2026-05-11 10:07:13","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":181097,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP tree view of XGBoost model features\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-9227317/v1/f9aae88a9f1c8f11b2bb742d.png"},{"id":108979827,"identity":"0cf63a66-0f62-43fd-b608-16caf4dc792e","added_by":"auto","created_at":"2026-05-11 12:01:44","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1060822,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9227317/v1/f36b8ada-7837-4983-9a89-f42e3509d3c3.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"ANN Model Excels in Predicting Aspiration During Gastric Lavage for Acute Poisoning: A Comparison with RF, XGBoost, and GBDT","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eIn recent years, China has witnessed an increasing number of acute poisoning cases presenting to emergency departments, with a particularly notable rise in drug overdoses and food poisoning incidents, thereby placing higher demands on emergency care capabilities\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. Data from the *China Health Statistics Yearbook 2022* \u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e indicate that from 2005 to 2021, injuries and poisonings consistently ranked as the fifth leading cause of death among both urban and rural residents in China. Ingestion via the digestive tract was identified as the most common route of toxic exposure. Acute poisoning is a systemic disorder characterized by the absorption of a substantial amount of toxins within a short period, leading to functional or structural damage of tissues and organs. It is marked by acute onset, rapid progression, and high fatality rates. As a common critical condition in the emergency department, it poses a severe threat to patient life and health\u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e. Therefore, implementing prompt and effective therapeutic interventions is crucial for reducing mortality. Gastric lavage, employed as a pivotal method for toxin elimination in China for nearly two centuries, plays a significant role in the management of acute poisoning\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e. However, the procedure itself can lead to complications such as electrolyte disturbances and aspiration\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e. Research\u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e has demonstrated that aspiration is a critical risk factor for poor patient outcomes; greater aspiration volume correlates with more severe injury, worse prognosis, and can result in an in-hospital mortality rate as high as 70%.\u003c/p\u003e \u003cp\u003eIn conclusion, within the context of acute poisoning treatment involving gastric lavage, the identification of prognostic risk factors and the development of an early and accurate predictive model for stratifying aspiration risk are of paramount clinical importance and practical value for improving patient outcomes and reducing mortality.\u003c/p\u003e \u003cp\u003eML algorithms can capture nonlinear relationships in data and complexity among multiple predictor variables, accurately identify risk factors for diseases, and make early predictions and diagnosis\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e.Artificial neural networks(ANNs) are a subset of traditional machine learning methods, belonging to the field of AI. Their structure and function are designed to resemble biologic nervous systems, with powerful learning algorithms and training capabilities to perform simulations with high accuracy\u003csup\u003e[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e. The weights are randomly chosen at the beginning of a training process and are then calculated to minimize errors between predicted values from the output layers and actual values\u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e.Fujita et al.\u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e developed Random Forest(RF) model to evaluate the carcinogenicity of 230 chemicals. The balanced accuracy from the evaluation for the RF model was 0.755. Although RF is less interpretable, it is computationally efficient and has been very successful in developing classification models for a wide range of toxicity types.The eXtreme Gradient Boosting(XGBoost) model is a collection of weak prediction trees capable of capturing complex relationships in the data without the need for higher-order interactions and nonlinear functions that are explicitly specified. A study had shown that in accordance with the predictive factors, XGBoost model, characterized by excellent discrimination and high accuracy, can be used to clinically identify severe acute pancreatitis patients with a high risk of enteral nutrition aspiration\u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e. The Gradient Boosting Decision Tree(GBDT) algorithm is less sensitive to hyperparameters, less prone to overfitting, and easy to implement. And it has been applied to diagnose and predict the outcomes of several diseases\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe purpose of this study is to develop and validate three prediction models, RF, XGBoost and GBDT, and compare their effectiveness with the traditional ANN model to predict the risk of aspiration in patients with acute poisoning after gastric lavage.\u003c/p\u003e"},{"header":"2. Patients and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Inclusion and exclusion criteria\u003c/h2\u003e \u003cp\u003eA retrospective study was conducted, selecting patients who underwent gastric lavage for acute poisoning in the emergency department of a tertiary general hospital in Ningbo City from January 2020 to July 2024 as the research subjects. Inclusion criteria: ① Age\u0026thinsp;\u0026ge;\u0026thinsp;18 years; ② Ingested toxic substances orally with indications for gastric lavage; ③ Comatose patients who could undergo gastric lavage smoothly under airway protection; ④ All patients or their families have signed the gastric lavage informed consent form. Exclusion criteria: ① Presence of contraindications for gastric lavage; ② Poor cooperation, making smooth gastric lavage impossible; ③ Accompanied by severe organ diseases.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Data collection\u003c/h2\u003e \u003cp\u003eThe collected data is categorized into three major groups: patient general information, clinical indicators, and laboratory indicators. Patient general information includes age, gender, consciousness status, history of hypertension, past medical history, types of ingested toxins, and time of toxin ingestion. Clinical indicators include resuscitation drugs, endotracheal intubation status, and imaging results. Laboratory indicators include WBC, neutrophil count, lymphocyte count, NLR, CRP, creatinine, K+, Na+, CK, CK-MB, AST, ALT, AST/ALT ratio, D-dimer, fibrin (ogen) degradation products, cholinesterase, serum amylase, and Brain Natriuretic Peptide (BNP).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 The definition of aspiration\u003c/h2\u003e \u003cp\u003e《The Chinese Expert Consensus on Swallowing Disorder Assessment and Treatment》\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003estates that aspiration refers to the phenomenon where contents from the oropharynx or stomach are inhaled into the lower respiratory tract below the vocal cords. After aspiration occurs, patients immediately exhibit symptoms such as irritative coughing, shortness of breath, or even asthma, which is termed overt aspiration. On the other hand, covert aspiration occurs when patients do not exhibit external signs such as coughing within 1 minute of aspiration and do not experience symptoms like irritative coughing or shortness of breath. Covert aspiration is often misdiagnosed and can be identified based on the following criteria\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e: ① No vomiting before gastric lavage, but nausea and vomiting occur during lavage, with gastric contents spewing out through the mouth and nose, accompanied by reduced breath sounds or crackles in the lungs, and possibly difficulty breathing or cyanosis; ② Gastric contents are suctioned from the airway after lavage; ③ A chest X-ray or CT scan performed after lavage shows new patchy shadows in the lung fields, possibly accompanied by localized atelectasis, confirming aspiration pneumonia. Meeting any one of these criteria indicates aspiration.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Sample size and missing values\u003c/h2\u003e \u003cp\u003eWe included 29 baseline characteristic data as study variables in this study. Based on the rule of thumb, the sample size was estimated to be 10 to 15 times the number of variables\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. Considering a 10% data loss, the estimated sample size was 484 cases. Ultimately, this study included 843 acute poisoning patients undergoing gastric lavage who met the inclusion and exclusion criteria. Additionally, we removed features with missing values exceeding 50%, a total of three, namely BNP, Cholinesterase, and Blood_amylase.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Ethics statement\u003c/h2\u003e \u003cp\u003e The research protocol (KY2023R270) was approved by the Ethics Committee of Ningbo Medical Center Lihuili Hospital. Written informed consent from participants is not required as their data are analyzed retrospectively and anonymously.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Statistical analysis\u003c/h2\u003e \u003cp\u003eUsing the Scikit-Learn database in Python 3.11 software, count data are represented by frequency, percentage, or proportion. Statistical inference uses chi-square tests and non-parametric rank sum tests; continuous variables are expressed as M(P25, P75), and intergroup comparisons use the Mann-Whitney U test. Differences are considered statistically significant at P\u0026thinsp;\u0026lt;\u0026thinsp;0.05.Three prediction models, including RF, XGBoost, and GBDT, were developed and validated. Their performance was compared with traditional artificial neural network (ANN) models to select the optimal algorithm and model for better predicting aspiration risk in patients undergoing gastric lavage for acute poisoning.The performance evaluation metrics of predictive models include receiver operating characteristic (ROC) curve, area under curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value(NPV), decision curve analysis (DCA) curve, Hosmer-Lemeshow (HL) curve, and inter-group difference analysis.\u003c/p\u003e \u003cp\u003eSelect the optimal model among RF, XGBoost, and GBDT, and use the SHapley Additive exPlanation (SHAP) tool to visualize and explain the decision-making role of features in the optimal model. Assess the feature importance in the decision-making process and demonstrate the contribution of each feature from a group-level perspective. Data processing is conducted under the guidance of professionals to ensure the objectivity and scientific nature of the analysis.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Baseline characteristics\u003c/h2\u003e\n \u003cp\u003eThis study collected a total of 28 research variables, excluding imaging results. Through data preprocessing, 3 samples with missing values\u0026thinsp;\u0026gt;\u0026thinsp;50% were removed, and ultimately, 25 feature variables were included, comprising aspiration and 24 risk factors for aspiration. Clinical data of patients in the training set and validation set were compared, with T-tests performed for continuous variables and chi-square tests for categorical variables. No statistically significant differences were found between groups for independent variables (P\u0026thinsp;\u0026gt;\u0026thinsp;0.05), indicating meaningful data partitioning.(Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e)\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of Factors Affecting Aspiration During Gastric Lavage in Acute Poisoning Patients Using Training and Testing Set Data(\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;843)\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eVariables\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTotal (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;843)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTrain (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;591)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eValidation (n\u0026thinsp;=\u0026thinsp;252)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eStatistic\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMethod\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003easpiration, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eyes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e304 (36)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e213 (36)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e91 (36)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eno\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e539 (64)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e378 (64)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e161 (64)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003esex, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.548\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e262 (31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e194 (33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e68 (27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003efemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e581 (69)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e397 (67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e184 (73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eage, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28 (19, 45.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28 (18, 44.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28 (19, 51)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.183\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-1.446\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ecategory, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.703\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.706\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eantipsychotics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e517 (61)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e366 (62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e151 (60)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003epesticide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e144 (17)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e102 (17)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e42 (17)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eother\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e182 (22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123 (21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e59 (23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003etime_h, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2 (1, 3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2 (1, 3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2 (1, 3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.608\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.292\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ehistory, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.475\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.487\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emental illness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e420 (50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e297 (50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123 (49)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eother\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e97 (12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e72 (12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25 (10)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eno\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e326 (39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e222 (38)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e104 (41)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ehypertension, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.671\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.943\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eyes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e84 (10)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e56 (9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28 (11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eno\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e759 (90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e535 (91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e224 (89)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRescue_medication, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.672\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.588\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e360 (43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e260 (44)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e100 (40)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e471 (56)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e322 (54)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e149 (59)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4 (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1 (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003econsciousness, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.938\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.143\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003esober\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e608 (72)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e424 (72)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e184 (73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003esleepy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e221 (26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e157 (27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e64 (25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ecoma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14 (2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10 (2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4 (2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEndotracheal_intubation, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.755\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.097\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eyes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17 (2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13 (2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4 (2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eno\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e826 (98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e578 (98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e248 (98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWBC, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7.3 (6.2, 8.85)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7.3 (6.25, 8.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7.3 (6.1, 8.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.635\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-0.069\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNeutrophils, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.7 (3.65, 6.05)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.7 (3.7, 6.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.7 (3.5, 5.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.248\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.409\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003elymphocyte, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.8 (1.3, 2.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.8 (1.3, 2.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.8 (1.37, 2.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-1.218\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNeutrophil to Lymphocyte Ratio, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.58 (1.89, 3.87)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.58 (1.96, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.58 (1.81, 3.55)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.209\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.719\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCRP, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.8 (0.25, 1.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.8 (0.25, 1.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.8 (0.25, 1.62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.511\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.204\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ecreatinine, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e56.65 (50.3, 65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e56.65 (50.75, 64.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e56.65 (49.77, 66.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.735\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.061\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eK+, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.61 (3.4, 3.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.61 (3.42, 3.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.61 (3.37, 3.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.363\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.114\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNa+, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e139.8 (138.3, 141.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e139.8 (138.5, 141.15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e139.8 (138.1, 141)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.147\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.373\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCK, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e81 (63, 106)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e81 (64, 105)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e81 (62, 113)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.872\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.312\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCKMB, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11 (8.5, 14.65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11 (8.4, 14.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11 (8.9, 14.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.812\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-0.736\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAspartate_aminotransferase, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25 (23, 28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25 (23, 28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25 (22.75, 29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.611\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-0.382\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAlanine_aminotransferase, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17 (14, 21.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17 (14, 21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17 (14, 22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.757\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.439\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAST_ALT, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.48 (1.31, 1.66)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.48 (1.32, 1.65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.48 (1.3, 1.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.634\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eD_dimer, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e89.5 (57, 143.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e89.5 (55, 152)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e89.5 (59.75, 126.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.509\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-0.541\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFibrinogen_degradation, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.82 (0.57, 1.16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.82 (0.57, 1.17)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.82 (0.57, 1.13)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.962\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003e3.2 Univariable analysis of the training sample(Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e)\u003c/h2\u003e\n \u003cp\u003eA baseline data table was created for the training set, and features were selected. The results (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e) show that the following 19 predictors have significant differences: sex, age, category, time_h, history, hypertension, Rescue_medication, consciousness, Neutrophils, CK, Alanine_aminotransferase, lymphocyte, Neutrophil to Lymphocyte Ratio (NLR), CRP, creatinine, CKMB, Aspartate_aminotransferase, D_dimer, and Fibrinogen_degradation.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eAnalysis of associated factors of aspiration during gastric lavage in acute poisoning patients using training samples (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;591)\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eVariables\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTotal (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;591)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNegative (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;213)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePositive (n\u0026thinsp;=\u0026thinsp;378)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eStatistic\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMethod\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003esex, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.013\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e6.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e194 (33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e84 (39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e110 (29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003efemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e397 (67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e129 (61)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e268 (71)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eage, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28 (18, 44.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e41 (24, 65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e24 (17, 34)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-10.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ecategory, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e7.527\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eantipsychotics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e366 (62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e135 (63)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e231 (61)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003epesticide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e102 (17)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e45 (21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e57 (15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eother\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123 (21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e33 (15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e90 (24)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003etime_h, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2 (1, 3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3 (1, 7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1 (1, 2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-6.302\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ehistory, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e29.328\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emental illness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e297 (50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e111 (52)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e186 (49)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eother\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e72 (12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e44 (21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28 (7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eno\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e222 (38)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e58 (27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e164 (43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ehypertension, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e46.576\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eyes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e56 (9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e43 (20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13 (3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eno\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e535(91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e170 (80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e365 (97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRescue_medication, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e12.868\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e260 (44)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e113 (53)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e147 (39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e322 (54)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e96 (45)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e226 (60)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1 (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003econsciousness, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e129.551\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003esober\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e424 (72)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e93 (44)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e331 (88)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003esleepy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e157 (27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e113 (53)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e44 (12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ecoma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10 (2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7 (3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEndotracheal_intubation, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.077\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e2.703\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChi-squared\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eyes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13 (2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8 (4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eno\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e578 (98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e205 (96)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e373 (99)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWBC, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7.3 (6.25, 8.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7.3 (6.2, 9.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7.3 (6.3, 8.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.285\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-2.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNeutrophils, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.7 (3.7, 6.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.7 (3.9, 7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.7 (3.52, 5.97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.038\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-3.041\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003elymphocyte, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.8 (1.3, 2.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.7 (1.2, 2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.8 (1.5, 2.28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e3.146\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNeutrophil to Lymphocyte Ratio, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.58 (1.96, 4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.62 (2.15, 5.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.58 (1.83, 3.38)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-3.808\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCRP, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.8 (0.25, 1.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.8 (0.6, 3.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.8 (0.25, 1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-4.262\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ecreatinine, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e56.65 (50.75, 64.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e56.65 (52.5, 67.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e56.65 (49.2, 63.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.012\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-2.679\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eK+, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.61 (3.42, 3.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.61 (3.36, 3.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.61 (3.44, 3.81)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.134\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e1.666\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNa+, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e139.8 (138.5, 141.15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e139.8 (138.1, 141.45)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e139.8 (138.6, 141.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.399\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e1.063\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCK, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e81 (64, 105)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e81 (67, 127)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e81 (62, 98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-2.152\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCKMB, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11 (8.4, 14.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11 (8.6, 15.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11 (8, 13.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-1.893\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAspartate_aminotransferase, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25 (23, 28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25 (24, 30)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25 (23, 26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.006\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-1.944\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAlanine_aminotransferase, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17 (14, 21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17 (15, 23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17 (14, 20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.003\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-1.927\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAST_ALT, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.48 (1.32, 1.65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.48 (1.27, 1.58)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.48 (1.41, 1.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e0.055\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e1.244\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eD_dimer, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e89.5 (55, 152)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123 (83, 270)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e89.5 (49, 107)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-4.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFibrinogen_degradation, Median (Q1,Q3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.82 (0.57, 1.17)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1 (0.74, 1.76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.82 (0.51, 0.99)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\"\u003e\n \u003cp\u003e-4.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eT-test\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003e3.3 Develop, compare, and calibrate predictive models for training samples\u003c/h2\u003e\n \u003cp\u003eUsing the occurrence of patient aspiration as the dependent variable and 19 factors identified as statistically significant in univariate analysis as independent variables, we developed four prediction models: Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Gradient Boosting Decision Tree (GBDT), and Artificial Neural Network (ANN). The performance metrics\u0026mdash;including AUC, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV)\u0026mdash;were reported as the average values derived from five-fold cross-validation for each algorithm.\u003c/p\u003e\n \u003cp\u003eThe RF model achieved an AUC of 0.846 (95% CI: 0.814\u0026ndash;0.877), accuracy of 0.790, sensitivity of 0.942, specificity of 0.521, PPV of 0.777, and NPV of 0.835. The XGBoost model yielded an AUC of 0.893 (95% CI: 0.868\u0026ndash;0.921), accuracy of 0.816, sensitivity of 0.921, specificity of 0.629, PPV of 0.815, and NPV of 0.817. The GBDT model showed an AUC of 0.852 (95% CI: 0.820\u0026ndash;0.883), accuracy of 0.787, sensitivity of 0.944, specificity of 0.507, PPV of 0.773, and NPV of 0.837. The ANN model demonstrated an AUC of 0.911 (95% CI: 0.885\u0026ndash;0.935), accuracy of 0.841, sensitivity of 0.931, specificity of 0.681, PPV of 0.838, and NPV of 0.848.\u003c/p\u003e\n \u003cp\u003eFurthermore, we plotted the receiver operating characteristic (ROC) curves, decision curve analysis (DCA) curves, and Hosmer-Lemeshow (HL) calibration curves, and conducted between-group difference analysis (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). The ANN-based prediction model for aspiration risk achieved the largest area under the ROC curve (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eA), indicating its superior predictive performance.\u003c/p\u003e\n \u003cp\u003eThe calibration curve and Hosmer-Lemeshow goodness-of-fit test (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eC) suggested excellent agreement between the predicted probabilities of aspiration risk and the observed outcomes in the ANN model. DCA (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eB) further indicated that the ANN model had high clinical utility. Additionally, the risk scores were significantly higher in the aspiration group compared to the non-aspiration group, with a statistically significant intergroup difference (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\n \u003cp\u003eIn conclusion, based on a comprehensive comparison of all evaluation metrics, the ANN algorithm constructed the most effective prediction model for aspiration risk.\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003e3.4 Validate, compare, and calibrate the predictive model on test samples\u003c/h2\u003e\n \u003cp\u003eThe predictive performance of the models was evaluated using the same performance metrics as those applied during model development, with the model exhibiting the highest AUC value selected as the optimal one. The RF model achieved an AUC of 0.805 (95% CI: 0.746\u0026ndash;0.858), accuracy of 0.762, sensitivity of 0.938, specificity of 0.451, PPV of 0.751, and NPV of 0.804. The XGBoost model yielded an AUC of 0.885 (95% CI: 0.844\u0026ndash;0.923), accuracy of 0.798, sensitivity of 0.969, specificity of 0.495, PPV of 0.772, and NPV of 0.900. The GBDT model showed an AUC of 0.766 (95% CI: 0.704\u0026ndash;0.826), accuracy of 0.714, sensitivity of 0.925, specificity of 0.341, PPV of 0.713, and NPV of 0.721. The ANN model demonstrated an AUC of 0.894 (95% CI: 0.843\u0026ndash;0.936), accuracy of 0.837, sensitivity of 0.901, specificity of 0.725, PPV of 0.853, and NPV of 0.805.\u003c/p\u003e\n \u003cp\u003eFurthermore, receiver operating characteristic (ROC) curves, decision curve analysis (DCA) curves, Hosmer\u0026ndash;Lemeshow (HL) calibration curves, and between-group difference analysis were plotted (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). The ANN-based aspiration risk prediction model achieved the largest area under the ROC curve (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eA). The calibration curve and Hosmer\u0026ndash;Lemeshow goodness-of-fit test (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eC) indicated excellent agreement between the predicted probabilities of aspiration risk and the observed outcomes in the ANN model. DCA (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eB) further demonstrated the high clinical utility of the ANN model. Additionally, the risk scores were significantly higher in the aspiration group than in the non-aspiration group, with a statistically significant intergroup difference (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\n \u003cp\u003eTherefore, based on a comprehensive comparison of all evaluation metrics, the ANN-based aspiration risk prediction model demonstrated the best overall performance.\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003e3.5 The explainability of predictive models\u003c/h2\u003e\n \u003cp\u003eThe optimal model among RF, XGBoost, and GBDT was determined to be XGBoost (AUC\u0026thinsp;=\u0026thinsp;0.885). The top ten features in terms of importance were visualized and ranked as follows: age, consciousness, time_h, NLR, fibrinogen degradation, lymphocyte, D-dimer, CRP, neutrophils, and creatinine (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). To enhance interpretability, the SHapley Additive exPlanation (SHAP) tool was employed to visualize the contribution of each feature to the decision-making process of the XGBoost model. This approach illustrates the average impact of individual features on model output, helping clinicians understand the rationale behind specific predictions and thereby increasing trust in the model\u0026apos;s decision process.\u003c/p\u003e\n \u003cp\u003eThe SHAP summary plot (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e) displays the direction and magnitude of each feature\u0026apos;s influence on the model\u0026rsquo;s prediction of aspiration risk. Each row represents the distribution of SHAP values for a specific feature. SHAP values greater than 0 indicate that the feature contributes to a positive prediction (aspiration risk), while values less than 0 suggest a tendency toward a negative prediction. In Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e, the color of each point represents the corresponding feature value: red indicates a higher value, and blue indicates a lower value.\u003c/p\u003e\n \u003cp\u003eFeatures were sorted based on the sum of the absolute SHAP values across all samples in the dataset. The four most important predictors were consciousness, age, time to toxin absorption (time_h), and NLR. Among these, consciousness showed a wide distribution of SHAP values with large absolute magnitudes, indicating it is the most influential predictor\u0026mdash;a finding highly consistent with clinical experience, which significantly enhances the model\u0026rsquo;s credibility.\u003c/p\u003e\n \u003cp\u003eThrough SHAP value analysis, the model provides clear and interpretable evidence for predictions, assisting clinicians in comprehending the decision logic, optimizing clinical decision-making, and improving doctor-patient communication. Therefore, the model not only demonstrates strong predictive performance but also offers excellent interpretability and practical clinical applicability.\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003e In this retrospective study, we developed and compared four machine learning models\u0026mdash;ANN, RF, XGBoost, and GBDT\u0026mdash;to predict the risk of aspiration during gastric lavage in patients with acute oral poisoning. Among them, the ANN model demonstrated the best discriminative performance with an AUC of 0.911 in the training set and 0.894 in the validation set. Calibration curves and DCA further confirmed its clinical utility and reliability. To enhance interpretability, SHAP analysis was applied to the XGBoost model, revealing that consciousness status, age, time since toxin ingestion, and NLR were the most influential predictors of aspiration risk.\u003c/p\u003e \u003cp\u003eOur findings align with existing clinical evidence that impaired consciousness is the strongest predictor of aspiration. Patients with altered mental status are more likely to lose protective airway reflexes, increasing the likelihood of gastric content reflux into the respiratory tract. This observation is concordant with the work of Chi-Syuan Pan et al.\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e Therefore, during gastric lavage for patients with impaired consciousness, nurses should exercise greater caution and assist the patient in turning their head to the side to prevent complications such as aspiration.\u003c/p\u003e \u003cp\u003eAge was also identified as a significant risk factor. This finding is consistent with the results reported by Conzelmann et al\u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e. Older patients often exhibit decreased gag reflexes, delayed gastric emptying, and comorbidities that compromise respiratory defense mechanisms. This suggests that age-stratified protocols should be considered when deciding on gastric lavage, particularly in elderly patients with diminished physiological reserves. When performing gastric lavage on elderly patients, nursing procedures must be more gentle and meticulous. It is crucial to closely monitor the balance of irrigation fluid inflow and outflow to prevent gastric intolerance.\u003c/p\u003e \u003cp\u003eA longer time interval between toxin ingestion and gastric lavage was associated with an increased risk of aspiration. A study by Akram et al.\u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e demonstrated that a shorter time interval from toxic substance ingestion to hospital admission is associated with improved patient recovery and reduced mortality.Early lavage may reduce gastric volume and toxin absorption, thereby lowering the chance of regurgitation. This reinforces the importance of timely intervention in acute poisoning cases, ideally within the first hour post-ingestion.\u003c/p\u003e \u003cp\u003eInterestingly, neutrophil-to-lymphocyte ratio (NLR), a marker of systemic inflammation, was found to be significantly elevated in patients who aspirated. Several investigators\u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e have reported a significant correlation between an increased NLR and the presence of infection.This may reflect a stress response to toxin exposure, which could indirectly impair swallowing coordination or lower the threshold for vomiting. While NLR is not traditionally used in aspiration risk assessment, its inclusion in predictive models may offer additional prognostic value.\u003c/p\u003e \u003cp\u003eAlthough ANN outperformed other models, its \u0026ldquo;black-box\u0026rdquo; nature limits direct clinical interpretability. To address this, we employed SHAP analysis on the XGBoost model, which ranked features in a clinically meaningful way. This hybrid approach\u0026mdash;using ANN for prediction and XGBoost for explanation\u0026mdash;may offer a balanced solution between accuracy and transparency in future clinical decision support systems.\u003c/p\u003e \u003cp\u003eDespite the strengths, this study has several limitations. First, its retrospective design may introduce selection bias. Second, the definition of aspiration was based on clinical and radiological criteria, which may underdiagnose covert aspiration. Third, external validation in multicenter cohorts is needed to confirm generalizability. Lastly, the ANN model, while accurate, requires further simplification or visualization tools to facilitate bedside application.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003e In conclusion, we developed a machine learning-based prediction model to assess aspiration risk during gastric lavage in patients with acute oral poisoning. The ANN model demonstrated superior predictive performance and clinical utility compared to RF, XGBoost, and GBDT. Key risk factors included impaired consciousness, advanced age, delayed lavage, and elevated NLR. Integrating this model into clinical workflows may support early identification of high-risk patients, guide preventive strategies, and ultimately improve patient safety. Future studies should focus on prospective validation, model simplification, and real-time implementation in emergency settings.\u003c/p\u003e \u003cp\u003e \u003cb\u003eEthics Statement\u003c/b\u003e \u003c/p\u003e \u003cp\u003e The study was conducted in accordance with the Declaration of Helsinki.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eAll authors declare no conflict of interest.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eAuthor contribution statement\u003c/h2\u003e \u003cp\u003eShuoni Zhang: literature retrieval and analysis, data collection and collation, statistical analysis, thesis writing; Bo Zhang : data collection and collation; Xuelan Liu: thesis guidance, statistics guidance, financial support.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis work was partially supported by Zhejiang medical and health project(2023KY249).But the funds were not used.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eShuoni Zhang: literature retrieval and analysis, data collection and collation, statistical analysis, thesis writing; Bo Zhang : data collection and collation; Xuelan Liu: thesis guidance, statistics guidance, financial support.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eAll data supporting the findings of this study are available within the paper and its Supplementary Information.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eGong F, Wang Y. Application effect of fully automatic gastric lavage machine in emergency poisoned patients[J]. China Med Device Inform. 2025;31(06):72\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNational Health Commission of the People\u0026rsquo;s Republic of China. China Health Statistics Yearbook 2022[M]: Beijing. Peking Union Medical College Pres; 2022.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang YT, Jiang SF, Lang N et al. [epidemiological characteristics and toxicant type of acute poisoning cases in china, 2016\u0026ndash;2022][J].Zhonghua Liu Xing Bing Xue Za Zhi,2024,45(10):1376\u0026ndash;82.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu C, Shi YT, Zhu Y, et al. Analysis of gastric lavage status and indications in 330 poisoned patients at primary hospitals[J]. Chin J Crit Care Med. 2025;45(1):63\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKošutova P. Mikolka P.Aspiration syndromes and associated lung injury: Incidence, pathophysiology and management[J]. Physiol Res 2021,70(Suppl4):S567\u0026ndash;83.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBacchi S, Tan Y, Oakden-Rayner L, et al. Machine learning in the prediction of medical inpatient length of stay[J]. Intern Med J. 2022;52(2):176\u0026ndash;85.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHong WD, Chen XR, Jin SQ et al. Use of an artificial neural network to predict persistent organ failure in patients with acute pancreatitis[J].Clinics (Sao Paulo),2013,68(1):27\u0026ndash;31.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo W, Liu J, Dong F et al. Review of machine learning and deep learning models for toxicity prediction[J]. Exp Biol Med (Maywood) 2023,248(21):1952\u0026ndash;73.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFujita Y, Honda H,Yamane M et al. A decision tree-based integrated testing strategy for tailor-made carcinogenicity evaluation of test substances using genotoxicity test results and chemical spaces[J].Mutagenesis,2019,34(1):101\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang B, Xu H, Xiao Q et al. Machine learning predictive model for aspiration risk in early enteral nutrition patients with severe acute pancreatitis[J].Heliyon,2024,10(23):e40236.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu F, Yao J, Liu C, et al. Construction and validation of machine learning models for sepsis prediction in patients with acute pancreatitis[J]. BMC Surg. 2023;23(1):267.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChinese Expert Consensus Group on Rehabilitation Assessment and Treatment of Dysphagia. Chinese expert consensus on assessment and treatment of dysphagia (2017 edition) Part 2: Treatment and rehabilitation management[J]. Chin J Phys Med Rehabilitation. 2018;40(1):1\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu CJ, Chang N. Prevention and nursing of enteral nutrition complicated with aspiration in mechanically ventilated patients[J]. Int J Nurs. 2019;38(15):2380\u0026ndash;3.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGao YX, Zhang JX. Sample size determination in logistic regression analysis[J]. J Evidence-Based Med. 2018;18(2):122\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePan CS, Lee CC, Yu JH, et al. Reassessing clinical presentations of emamectin benzoate poisoning: A comprehensive study[J]. Hum Exp Toxicol. 2024;43:9603271241249965.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eConzelmann M, Hoidis A,Bruckner T et al. Aspiration risk in relation to glasgow coma scale score and clinical parameters in patients with severe acute alcohol intoxication: A single-centre, retrospective study[J].BMJ Open,2021,11(10):e053619.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAkram S. Fazil M,Ullah K.Poppy intoxication in infants and children: Hazards of a folk remedy[J]. J Coll Physicians Surg Pak 2021,30(5):576\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBurchette DT, Dasci MF, Fernandez Maza B, et al. Neutrophil-lymphocyte ratio and lymphocyte-monocyte ratio correlate with chronic prosthetic joint infection but are not useful markers for diagnosis[J]. Arch Orthop Trauma Surg. 2024;144(1):297\u0026ndash;305.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-informatics-and-decision-making","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"midm","sideBox":"Learn more about [BMC Medical Informatics and Decision Making](http://bmcmedinformdecismak.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/midm/default.aspx","title":"BMC Medical Informatics and Decision Making","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Aspiration, Gastric lavage, Machine learning, Artificial neural network, Risk prediction, Acute poisoning","lastPublishedDoi":"10.21203/rs.3.rs-9227317/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9227317/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground:\u003c/h2\u003e \u003cp\u003eAspiration during gastric lavage is a severe complication in acute poisoning patients, associated with high morbidity and mortality. Early identification of high-risk individuals is critical for improving patient safety. This study aimed to develop and validate machine learning models to predict aspiration risk during gastric lavage and compare their performance.\u003c/p\u003e\u003ch2\u003eMethods:\u003c/h2\u003e \u003cp\u003eA retrospective cohort study was conducted involving 843 adult patients who underwent gastric lavage for acute poisoning at a tertiary hospital in Ningbo, China, from January 2020 to July 2024. After excluding features with \u0026gt;\u0026thinsp;50% missing values, 25 variables were included. Four machine learning models\u0026mdash;artificial neural network (ANN), random forest (RF), extreme gradient boosting (XGBoost), and gradient boosting decision tree (GBDT)\u0026mdash;were trained and validated using five-fold cross-validation. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), calibration curves, Hosmer-Lemeshow test, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) were used to interpret the best-performing interpretable model.\u003c/p\u003e\u003ch2\u003eResults:\u003c/h2\u003e \u003cp\u003eAmong 843 patients, 304 (36%) experienced aspiration. In both training and validation sets, the ANN model demonstrated the highest discriminative performance (training AUC\u0026thinsp;=\u0026thinsp;0.911; validation AUC\u0026thinsp;=\u0026thinsp;0.894), followed by XGBoost (0.893 and 0.885), GBDT (0.852 and 0.766), and RF (0.846 and 0.805). ANN also showed superior calibration and clinical utility. SHAP analysis of the XGBoost model identified consciousness status, age, time since toxin ingestion, and neutrophil-to-lymphocyte ratio (NLR) as the most influential predictors.\u003c/p\u003e\u003ch2\u003eConclusion:\u003c/h2\u003e \u003cp\u003eThe ANN model outperformed RF, XGBoost, and GBDT in predicting aspiration risk during gastric lavage in acute poisoning patients. Key risk factors included impaired consciousness, advanced age, delayed lavage, and elevated NLR. Integrating the ANN model into clinical decision-making may enhance early risk identification and improve patient outcomes. Future studies should focus on prospective validation and real-time implementation.\u003c/p\u003e","manuscriptTitle":"ANN Model Excels in Predicting Aspiration During Gastric Lavage for Acute Poisoning: A Comparison with RF, XGBoost, and GBDT","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-11 10:07:09","doi":"10.21203/rs.3.rs-9227317/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-05T02:04:20+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"259731102254443918444488338025568505099","date":"2026-05-02T11:23:10+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-30T11:17:07+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-04-03T06:47:32+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-31T11:19:16+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-31T11:18:41+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Informatics and Decision Making","date":"2026-03-25T22:07:18+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-informatics-and-decision-making","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"midm","sideBox":"Learn more about [BMC Medical Informatics and Decision Making](http://bmcmedinformdecismak.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/midm/default.aspx","title":"BMC Medical Informatics and Decision Making","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"5dea9573-34f3-47d5-9e3a-5bd84203a0a8","owner":[],"postedDate":"May 11th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-05T02:04:20+00:00","index":40,"fulltext":""},{"type":"reviewerAgreed","content":"259731102254443918444488338025568505099","date":"2026-05-02T11:23:10+00:00","index":38,"fulltext":""},{"type":"reviewersInvited","content":"30","date":"2026-04-30T11:17:07+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-11T10:07:09+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-11 10:07:09","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9227317","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9227317","identity":"rs-9227317","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}