Machine learning predictive model for aspiration risk in early enteral nutrition patients with severe acute pancreatitis

doi:10.21203/rs.3.rs-3969940/v1

Background: The aim of this study was to build and validate a risk prediction model for aspiration in severe acute pancreatitis patients receiving enteral nutrition by identifying risk factors for aspiration in these patients.

Methods: The risk factors for aspiration were analyzed to build a prediction model based on the data collected from 339 patients receiving enteral nutrition. Subsequently, we used six machine learning algorithms and the model was validated by the area under the curve.

Results: In this study, the collected data were divided into two groups: a training cohort and a validation cohort. The results showed that 28.31% (77) of patients had aspiration and 71.69% (195) of patients had non-aspiration in training cohort. Moreover, age, consciousness, mechanical ventilation, aspiration history, nutritional risk and number of comorbidities were included as predictive factors for aspiration in patients receiving EN. The XGBoost model is the best of all machine learning models, with an AUROC of 0.992 and an F1 value of 0.902. The specificity and accuracy of XGBoost are higher than those of traditional logistic regression.

Conclusion: 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.

Health sciences/Diseases

Health sciences/Gastroenterology

Health sciences/Medical research

Health sciences/Risk factors

Aspiration

enteral nutrition

severe acute pancreatitis

predictive model

risk factors

Severe acute pancreatitis (SAP) has a rapid onset and progression with complex clinicopathologic changes. Systemic inflammatory response syndrome and multiple organ dysfunction syndrome (MODS) can occur at an early stage, with a mortality rate of 20%-30%[1]. SAP patients are in a state of hyper decomposition and hypermetabolism, often accompanied by severe metabolic disorders, and prone to malnutrition[2]. Early enteral nutrition (EEN) (≤48h) can reduce the incidence of complications in SAP patients[3]. Enteral nutrition (EN) could stimulate intestinal metabolic activity, which helps to maintain intestinal function and protect the intestinal mucosal barrier. Intestinal integrity plays a crucial role in regulating the imbalance of intestinal flora[4]. A study had shown that EN had fewer complications than parenteral nutrition, shorter hospital stays and a positive impact on patient outcomes[5]. However, due to complications of EN, there is still a proportion of patients who are unable to reach their energy goals[6]. Aspiration is one of the serious complications of EN, which increases the risk of aspiration pneumonia by 12 times, which results in acute respiratory distress syndrome (ARDS), with a case fatality rate of up to 70%[7, 8], and may also prolong the patient's hospital stay and increase hospitalization costs[9, 10].

Aspiration can be categorized into overt and micromisaspiration according to the symptoms[11]. Clinical symptoms such as choking, coughing and shortness of breath immediately after aspiration are characterized as overt aspiration. Microaspiration without any external signs. The incidence of overt aspiration in ICU patients ranges from 8.80% to 22.45%[11, 12], and the incidence of microaspiration reaches 88.00%[13, 14]. Microaspiration is an independent risk factor for ventilator-associated pneumonia and a major source of bacteria[15]. There are several methods for detecting aspiration, but all have their drawbacks, such as the radioactivity of technetium 99[16], the complexity of methylene blue[17], and the inaccuracy of alpha amylase[18]. Videofluoroscopic swallowing study (VFSS) can comprehensively and visually assess the swallowing function of patients, which are the gold standard for diagnosing swallowing dysfunction[19]. However, there are high requirements for operators, operating environment and patient cooperation, and the cost is expensive, so they are less used in clinical practice. In contrast, pepsin detection is easy to perform. Therefore, it is recommended to use pepsin detection as a marker for microaspiration[17, 18]. Currently, clinical nurses are not fully aware of aspiration, especially microaspiration[20]. It is recommended that medical staff first screen patients with EAT-10 questionnaire, and further test pepsin for positive patients to determine whether there is aspiration.

Machine learning (ML) algorithms are good at defining data attributes, using computers to mine the laws present in the data, and applying different classifiers to predict the occurrence of diseases[21]. ML has now become a new data analysis paradigm competing with multivariate linear regression in medical research and is mainly applied to clinically assisted medical decision making[22]. In particular, ML models for early detection of diseases had appeared in various fields such as coronary heart disease[23], diabetes[24], kidney disease[25] and lung diseases[26], and the results showed that ML predictive models were comparable to existing screening tools. In addition, 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[27]. Despite the rapidly expanding field of research on ML algorithms, there are no ML-based models for predicting EEN aspiration in patients with SAP.

The aim of this study was to establish an ML predictive model to screen for aspiration. Extreme gradient boosting (XGBoost), Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Stochastic Gradient Descent (SGD) were selected to mine target data sets and establish risk prediction models for aspiration. Then the prediction efficiency of the models under different algorithms is evaluated in order to construct the optimal prediction model. The aim is to help medical staff identify high-risk patients with aspiration as early as possible, enhance risk perception ability, and provide new ideas for reducing the incidence of aspiration and improving the prognosis of patients.

2.1 Study design and data collection

A total of 339 patients receiving EN in two hospitals from October 2019 to October 2023 were enrolled and their clinical data collected for this retrospective cohort study. Inclusion criteria: ①Age≥18 years old; ②Meet the Atlanta diagnostic criteria for SAP[28]; ③EN was given for 24-48 hours after admission, and the duration was 1 week; ④Patients and their families gave informed consent to the treatment. Exclusion criteria: ①malignant tumor patients; ②pregnancy; ③terminal state; ④incomplete data. For patients with multiple admissions take their first data.

2.2 Data management

The formulas of the unmatched case–control study were used to calculate the sample size[29]. The patients were randomized into the training and validation cohorts with a ratio of 8:2. Considering that a larger sample size would yield better performances of the prediction models, 339 patients were randomly selected for inclusion in the study, with 272 patients in the training group and 67 patients in the validation group.

2.3 Independent variables

The outcome of this study was whether aspiration occurred during EN. The following data were collected for modeling. (1) Demographic characteristics: gender, age (years), length of stay, aspiration history, number of comorbidities, consciousness, nutritional risk; (2) EN related index: intubation days, intubation depth, food type; (3) Nursing measures: mechanical ventilation; (4) medication use: gastric stimulants, antacids; (5) Laboratory test indicators: red blood cells (RBC) counts, haemoglobin (Hb) level, platelet (PLT) counts, white blood cell (WBC) counts, lymphocyte (LY) counts, neutrophil (NEUT) counts. All variables were screened, missing values greater than 25% were excluded, and patients with more than 5% missing personal data were excluded, and the remaining missing values were filled in according to multiple interpolation. Finally, a total of 20 variables were included, as shown in Table 1.

2.4 Predictive model development and validation

Univariate analysis was performed on the final complete dataset to screen out the factors associated with aspiration in patients with EEN. Subsequently, multivariate logistic regression analysis was performed to obtain the patient's independent risk factors. Based on ML, six prediction models were established using the data of the training and the validation cohort, which were as follows: (1) extreme gradient boosting (XGBoost); (2) Logistic Regression (LR); (3) Random Forest (RF); (4) Support Vector Machine (SVM); (5) K-Nearest Neighbor (KNN); (6) Stochastic Gradient Descent (SGD). Select 10-fold cross-validation for the selection and adjustment of predictive model hyperparameters. The discriminant discrimination ability of the model is measured by the area under the receiver operating characteristic curve (AUROC).

The medical records of SAP patients who received EN from a hospital in Shandong Province were used as an external validation dataset. After converting the optimal predictive model into a web-based calculator, the dataset was used as the basis for external validation to evaluate the accuracy and extrapolation power of the model.

2.5 Ethics statement

The study was approved by the Ethics Committee of the First Affiliated Hospital of Xi’an Jiaotong Universityy (No. XJTU1AF2023LSK-507). The requirement for individual informed consent was waived by the ethics committee of the First Affiliated Hospital of Xi’an Jiaotong University because of the retrospective nature of the study. This study was conducted in accordance with the principles of the Declaration of Helsinki.

2.6 Statistical analysis

R statistical software (version 4.0.1) and IBM SPSS26.0 statistical software were used for statistical analysis. The Shapiro-Wilk test was used to detect whether the continuous variables conform to a normal distribution. Categorical variables are expressed as frequency (proportion). Te chi-squared (trend) test was used for comparative analysis between the two groups. Te area under the receiver operating characteristic curve (AUROC) was analyzed using the “Epi” package in the R software. Factors with significant differences are used in regression analysis. p < 0.05 is considered to have a significant difference.

3.1 Characteristics of patients

A total of 339 patients receiving EEN who were randomly selected for inclusion in the study. They were randomly divided into training and validation cohorts by a ratio of 8:2. The comparison of demographic and clinical characteristics between the training cohort and validation cohort is summarized in Supplementary Information Table S1. The detailed baseline characteristics in the training cohort are displayed in Table 1. There were 77 aspiration patients and 195 non-aspiration patients in the training cohort. A total of 20 variables were adopted in this study, of which 7 variables were statistically significant (p < .05). Compared with non-aspiration patients, the age of patients with aspiration is significantly older (average age, 56 vs. 65, p < 0.001). Other basic characteristics such as number of comorbidities, nutritional risk, aspiration history and consciousness were also found to be statistically significant (p < .05). There were no significant differences in EN related indicators and medication use. Significant differences were found in mechanical ventilation among the nursing measures (p < .05). RBC counts showed better effects on the latest blood indexes of EN compared with those of PLT counts, Hb levels, WBC counts, lymphocyte count, neutrophil counts and K⁺.

3.2 Risk factor screening and model construction

Seven risk factors associated with EEN aspiration in SAP patients were taken as independent variables in univariate analysis, and the dependent variable was whether aspiration occurred. Logistic regression analysis was performed. The final results showed that age, consciousness, aspiration history, mechanical ventilation, nutritional risk and number of comorbidities were independent risk factors. The specific analysis results are shown in Figure 1 and Table 2.

A nomogram was developed and visualized with the six significant factors based on the final regression analysis (Figure 2). A score was assigned to the value of each variable on a point-scale axis. Predicted probabilities were converted into points on the coordinate axis from 0 to 100. Each evaluation criterion corresponded to a variable. Among the six factors, age had the highest effect (points:100). Patients with number of comorbidities and mechanical ventilation were respectively allocated 85 points and 45 points. Patients with aspiration history were assigned higher points than those without aspiration history (points:38 vs.0). Patients with nutritional risk were only slightly higher than patients without nutritional risk. The risk of aspiration signified the aspiration probability ranging from 0.01 to 0.99. The constructed risk prediction nomogram of aspiration is shown in Figure 2.

3.3 Model verification

Based on ML, six aspiration prediction models were established using the above four variables. The predictive power of the algorithm for EEN aspiration was validated and evaluated in the validation cohort. ROC curves were plotted with sensitivity as the ordinate and 1-specificity as the abscissa, and AUROC was used to evaluate the discrimination of each model. After comparing the six models, the AUC, specificity, accuracy, and F1 value of the XGBoost model ranked first, and its AUC was 0.992. The ROC curves for the six predictive models are shown in Figure 3.

The AUC of the SVM, SGD, KNN, RF, XGBoost and LR models is 0.904, 0.843, 0.964, 0.984, 0.992 and 0.903, respectively. XGBoost has the highest AUROC and specificity (Table 3), and has better predictive performance than traditional LR models. The most sensitive ones are SVM and KNN. The specificity, F1 and accuracy of XGBoost are higher than those of traditional LR, and XGBoost has a higher proportion of correct predictions for all samples, and more sensitivity for the diagnosis of aspiration.

DCA is a method of assessing whether clinical decisions are feasible by considering the likely range of risks and benefits for patients. The DCA curve was used to evaluate the actual clinical utility of the prediction model. The DCA curve of the aspiration risk prediction model in SAP patients with EEN was shown in Figure 4. The results showed that the DCA curve of XGBoost performed best.

The incidence of aspiration is extremely high in clinical practice. However, for critically ill (CCI) patients undergoing EN, especially those with impaired consciousness, there is still a lack of comprehensive and systematic risk assessment tools for aspiration[30]. The results of this study showed that impaired consciousness was one of the independent risk factors. That is, patients with SAP who have impaired consciousness are at increased risk. The state of consciousness was also one of the indicators reflecting the severity of the patient's disease, and patients who were critically ill or in impaired consciousness had a decreased cough reflex and reduced defensive airway protection, leading to swallowing disorders. At the same time, persistent impaired consciousness or coma could lead to relaxation of the muscles of the throat and increase the risk of accidental respiratory secretions from the pharynx, thus increasing the risk of aspiration[31, 32]. Therefore, clinical medical staff should pay attention to the prevention of the risk of aspiration in CCI patients and patients with impaired consciousness in the implementation of treatment and nursing intervention. Nurses can judge the severity of the patient's condition by dynamically assessing the state of consciousness, and then prejudge the risk of aspiration in patients according to the condition, and take preventive measures as soon as possible to prevent the occurrence of aspiration.

Age was one of the risk factors for aspiration. With the increase of age, the functions of each organ of the body shows different degrees of physiological decline, especially the reflex function and swallowing function[33]. Moreover, elderly patients are often complicated with a variety of internal medical diseases such as cardiovascular and cerebrovascular diseases and pulmonary diseases, so age becomes an important cause of aspiration[34].

North American expert consensus on aspiration in CCI patients[31] lists aspiration history as the main risk factor for aspiration. This study showed that patients with aspiration history had a much higher risk of aspiration than patients without aspiration history. A study[35] used ROC curve and multivariate analysis and found that aspiration history was a risk factor for complicated aspiration in patients with dysphagia. The pathogenesis of recurrent aspiration with aspiration history is complex, which may be related to age-related neurophysiological changes and changes in biological mechanisms controlling swallowing[36]. Therefore, patients with aspiration history should be screened and evaluated early in clinical practice and risk factors should be further evaluated through interdisciplinary team management. Timely preventive measures should be taken, such as posting warning signs for "aspiration prevention," notifying patients and family members to be vigilant, to prevent the occurrence of secondary aspiration.

Patients with mechanical ventilation are at high risk of aspiration, with a risk of 11.4 %[37]. A study had shown that the longer the implementation of mechanical ventilation, the greater the likelihood of aspiration[38], which was consistent with the results of this study. Mechanical ventilation destroys the swallowing and coughing reflex function of patients, making patients unable to defend themselves through swallowing and coughing mechanisms, and thus prone to aspiration. Sedative medications have been shown to be an independent factor in the risk of aspiration[39]. Therefore, it is clinically important to minimize unnecessary deep sedation in patients on mechanical ventilation, and health care providers should assess the need for sedative use on a daily basis. Airway management is an important part of aspiration prevention management in patients with mechanical ventilation. Subglottic suctioning can effectively drain the secretions from the tracheal cannula sac, which is an important measure to reduce aspiration. Suction of oral secretions every 4 hours can reduce microaspiration. It is recommended that oral and pharyngeal and underdoor aspiration should be performed first when changing body position, which is conducive to reducing aspiration[40].

Nutritional risk is one of the independent risk factors for predicting the occurrence of aspiration in SAP patients. High nutritional risk is positively associated with an increased risk of clinical outcomes such as in-hospital death[41]. Patients with high nutritional risk are subjected to additional therapeutic measures such as endotracheal intubation, tracheotomy and placement of a nasogastric tube, which can increase the risk of aspiration. However, these patients are most likely to benefit from aggressive nutritional therapy[42]. Nutritional risk is an important prognostic factor for critically ill patients, but it is often underappreciated[43]. Therefore, timely assessment of patients' nutritional risk and intervention for patients at high nutritional risk can improve the outcome of aspiration.

Disease factors were independent risk factors for EN complicated by aspiration (OR=15.44). Concurrent aspiration was associated with neurological comorbidities, hypertension, respiratory comorbidities and other diseases. Patients with lung disease were at increased risk of aspiration due to impaired respiratory defense mechanisms, such as a weakened cough reflex and a weakened mucociliary barrier[44-46]. A study identified cardiovascular comorbidities and neurological comorbidities in stroke patients with indwelling gastric tubes as important causes of aspiration[47]. Aspiration had been found to be associated with poorer pre-stroke functional status, hemorrhagic stroke, and more severe stroke[47]. A prospective study showed that stroke patients with dysphagia were 4.69 times more likely to develop aspiration than non-dysphagia stroke patients within one year, and the risk of aspiration peaked in patients with hemorrhagic stroke at the third year[48]. Patients with neurological complications were highly susceptible to acute respiratory distress syndrome due to impaired neurocognitive function and difficulty in actively removing respiratory foreign bodies during aspiration, resulting in acute lung injury and aspiration pneumonia[49]. The American Geriatric Society does not recommend nasal feeding in patients with dementia because it increases the risk of irritability and unplanned extubation[50].Therefore, medical staff should treat the primary disease of patients early, pay attention to the mechanism of the disease on aspiration, and reduce the incidence of aspiration in multiple dimensions, measures and channels.

There are some limitations should be addressed. First, the datasets were retrospective information although they were from two large domestic general hospitals. Second, the sample size was small and patient enrollment bias was difficult to avoid. Therefore, prospective, large-sample studies should be conducted in subsequent studies to explore and validate the clinical value of the prediction model.

In summary, our results showed that age, consciousness, aspiration history, mechanical ventilation, nutritional risk and number of comorbidities were risk factors for patients with aspiration. Overall, aspiration prediction models based on six ML algorithms is feasible, and the XGBoost model had the most satisfactory prediction performance in terms of several aspects after internal validation. The findings may provide an important reference for aspiration risk prediction and promote prevention and treatment of aspiration.

Hines, O.J. and S.J. Pandol, Management of severe acute pancreatitis. BMJ, 2019. 367: p. l6227.
Jablonska, B. and S. Mrowiec, Nutritional Support in Patients with Severe Acute Pancreatitis-Current Standards. Nutrients, 2021. 13(5).
Vaughn, V.M., et al., Early Versus Delayed Feeding in Patients With Acute Pancreatitis: A Systematic Review. Ann Intern Med, 2017. 166(12): p. 883-892.
Allen, K. and L. Hoffman, Enteral Nutrition in the Mechanically Ventilated Patient. Nutr Clin Pract, 2019. 34(4): p. 540-557.
Elke, G., et al., Enteral versus parenteral nutrition in critically ill patients: an updated systematic review and meta-analysis of randomized controlled trials. Crit Care, 2016. 20(1): p. 117.
Marik, P.E., Enteral nutrition in the critically ill: myths and misconceptions. Crit Care Med, 2014. 42(4): p. 962-9.
DiBardino, D.M. and R.G. Wunderink, Aspiration pneumonia: a review of modern trends. J Crit Care, 2015. 30(1): p. 40-8.
Cohen, D.L., et al., Post-stroke dysphagia: A review and design considerations for future trials. Int J Stroke, 2016. 11(4): p. 399-411.
Schwarz, M., et al., The impact of aspiration pneumonia and nasogastric feeding on clinical outcomes in stroke patients: A retrospective cohort study. J Clin Nurs, 2018. 27(1-2): p. e235-e241.
Huang, S.T., C.C. Chiou and H.Y. Liu, Risk factors of aspiration pneumonia related to improper oral hygiene behavior in community dysphagia persons with nasogastric tube feeding. J Dent Sci, 2017. 12(4): p. 375-381.
Byun, S.E., et al., Risk factors and prognostic implications of aspiration pneumonia in older hip fracture patients: A multicenter retrospective analysis. Geriatr Gerontol Int, 2019. 19(2): p. 119-123.
Benjamin, E., et al., Witnessed aspiration in trauma: Frequent occurrence, rare morbidity--A prospective analysis. J Trauma Acute Care Surg, 2015. 79(6): p. 1030-6; discussion 1036-7.
Jaillette, E., et al., Optimal care and design of the tracheal cuff in the critically ill patient. Ann Intensive Care, 2014. 4(1): p. 7.
Metheny, N.A., et al., Tracheobronchial aspiration of gastric contents in critically ill tube-fed patients: frequency, outcomes, and risk factors. Crit Care Med, 2006. 34(4): p. 1007-15.
Doudakmanis, C., et al., Relationship Between Intra-Abdominal pressure and microaspiration of gastric contents in critically ill mechanically ventilated patients. J Crit Care, 2023. 74: p. 154220.
Heyland, D.K., et al., Effect of postpyloric feeding on gastroesophageal regurgitation and pulmonary microaspiration: results of a randomized controlled trial. Crit Care Med, 2001. 29(8): p. 1495-501.
Talbert, S., et al., Pepsin A in Tracheal Secretions From Patients Receiving Mechanical Ventilation. Am J Crit Care, 2021. 30(6): p. 443-450.
Dewavrin, F., et al., Accuracy of alpha amylase in diagnosing microaspiration in intubated critically-ill patients. PLoS One, 2014. 9(6): p. e90851.
Schindler, A., et al., Fiberoptic endoscopic evaluation of swallowing in early-to-advanced stage Huntington's disease. Sci Rep, 2020. 10(1): p. 15242.
Klompas, M., Prevention of Intensive Care Unit-Acquired Pneumonia. Semin Respir Crit Care Med, 2019. 40(4): p. 548-557.
Nwanosike, E.M., et al., Potential applications and performance of machine learning techniques and algorithms in clinical practice: A systematic review. Int J Med Inform, 2022. 159: p. 104679.
Huang, L., et al., Comparing Multiple Linear Regression and Machine Learning in Predicting Diabetic Urine Albumin-Creatinine Ratio in a 4-Year Follow-Up Study. JOURNAL OF CLINICAL MEDICINE, 2022. 11(13).
Krittanawong, C., et al., Machine learning and deep learning to predict mortality in patients with spontaneous coronary artery dissection. Sci Rep, 2021. 11(1): p. 8992.
Segar, M.W., et al., Machine Learning to Predict the Risk of Incident Heart Failure Hospitalization Among Patients With Diabetes: The WATCH-DM Risk Score. Diabetes Care, 2019. 42(12): p. 2298-2306.
Makino, M., et al., Artificial intelligence predicts the progression of diabetic kidney disease using big data machine learning. Sci Rep, 2019. 9(1): p. 11862.
He, W., et al., Risk factors and machine learning prediction models for bronchopulmonary dysplasia severity in the Chinese population. World J Pediatr, 2023. 19(6): p. 568-576.
Bacchi, S., et al., Machine learning in the prediction of medical inpatient length of stay. Intern Med J, 2022. 52(2): p. 176-185.
Banks, P.A., et al., Classification of acute pancreatitis - 2012 : revision of the Atlanta classification and definitions by international consensus. GUT, 2013. 62(1): p. 102-111.
Dupepe, E.B., et al., What is a Case-Control Study? Neurosurgery, 2019. 84(4): p. 819-826.
Martino, R., et al., The Toronto Bedside Swallowing Screening Test (TOR-BSST): development and validation of a dysphagia screening tool for patients with stroke. Stroke, 2009. 40(2): p. 555-61.
McClave, S.A., et al., Guidelines for the Provision and Assessment of Nutrition Support Therapy in the Adult Critically Ill Patient: Society of Critical Care Medicine (SCCM) and American Society for Parenteral and Enteral Nutrition (A.S.P.E.N.). JPEN J Parenter Enteral Nutr, 2016. 40(2): p. 159-211.
McClave, S.A., et al., ACG Clinical Guideline: Nutrition Therapy in the Adult Hospitalized Patient. Am J Gastroenterol, 2016. 111(3): p. 315-34; quiz 335.
Zhao, T., et al., Identifying Risk Factors for Aspiration in Patients Hospitalized with Community-Acquired Pneumonia. Int J Clin Pract, 2023. 2023: p. 2198259.
Almirall, J., et al., Aspiration pneumonia: A renewed perspective and practical approach. Respir Med, 2021. 185: p. 106485.
Chiba, Y., et al., Predictive value of the Hyodo score in endoscopic evaluation of aspiration during swallowing. Auris Nasus Larynx, 2018. 45(6): p. 1214-1220.
Gomes, G.F., et al., The nasogastric feeding tube as a risk factor for aspiration and aspiration pneumonia. CURRENT OPINION IN CLINICAL NUTRITION AND METABOLIC CARE, 2003. 6(3): p. 327-333.
Metheny, N.A., et al., Tracheobronchial aspiration of gastric contents in critically ill tube-fed patients: frequency, outcomes, and risk factors. Crit Care Med, 2006. 34(4): p. 1007-15.
Deem, S., et al., Randomized Pilot Trial of Two Modified Endotracheal Tubes To Prevent Ventilator-associated Pneumonia. Ann Am Thorac Soc, 2016. 13(1): p. 72-80.
Noguchi, S., et al., Impact of the number of aspiration risk factors on mortality and recurrence in community-onset pneumonia. Clin Interv Aging, 2017. 12: p. 2087-2094.
Sole, M.L., et al., Nursing oral suction intervention to reduce aspiration and ventilator events (NO-ASPIRATE): A randomized clinical trial. J Adv Nurs, 2019. 75(5): p. 1108-1118.
Occhiali, E., et al., Risk factors for poor outcome in older patients admitted in a surgical intensive care unit. NURSING IN CRITICAL CARE, 2023. 28(1): p. 40-46.
Rahman, A., et al., Identifying critically-ill patients who will benefit most from nutritional therapy: Further validation of the "modified NUTRIC" nutritional risk assessment tool. Clin Nutr, 2016. 35(1): p. 158-162.
Campos, L.S., et al., Impact of nutritional risk on 28-day mortality and the prevalence of underfeeding in critically ill patients: A prospective cohort study. Nutr Hosp, 2020. 34(3): p. 414-421.
Lee, A.S., et al., Reflux-Aspiration in Chronic Lung Disease. Ann Am Thorac Soc, 2020. 17(2): p. 155-164.
Houghton, L.A., et al., Respiratory disease and the oesophagus: reflux, reflexes and microaspiration. Nat Rev Gastroenterol Hepatol, 2016. 13(8): p. 445-60.
Zheng, Z., et al., Silent aspiration in patients with exacerbation of COPD. Eur Respir J, 2016. 48(2): p. 570-3.
Teuschl, Y., et al., Systematic dysphagia screening and dietary modifications to reduce stroke-associated pneumonia rates in a stroke-unit. PLoS One, 2018. 13(2): p. e0192142.
Feng, M.C., et al., The Mortality and the Risk of Aspiration Pneumonia Related with Dysphagia in Stroke Patients. J Stroke Cerebrovasc Dis, 2019. 28(5): p. 1381-1387.
Schwartz, D.B., Enteral Nutrition and Dementia Integrating Ethics. Nutr Clin Pract, 2018. 33(3): p. 377-387.
Flaschner, E. and D. Katz, American Geriatrics Society Position Statement on Feeding Tubes in Advanced Dementia. J Am Geriatr Soc, 2015. 63(7): p. 1490-1.

Table 1. Training cohort baseline

Characteristic	Non-Aspiration [cases (%)]	Aspiration [cases (%)]	P value
Total	195	77
Age	56.00 [51.00, 66.00]	65.00 [59.00, 69.00]	<0.001
Gender			0.456
Male	135 (69.2)	49 (63.6)
Female	60 (30.8)	28 (36.4)
Food type			0.341
Commercial nutrition	161 (82.6)	59 (76.6)
Home-made meals	34 (17.4)	18 (23.4)
Consciousness			<0.001
Conscious	120 (61.5)	12 (15.6)
Sleepiness	38 (19.5)	31 (40.3)
Lethargy	17 ( 8.7)	12 (15.6)
Coma	20 (10.3)	22 (28.6)
Mechanical ventilation			<0.001
NO	163 (83.6)	19 (24.7)
YES	32 (16.4)	58 (75.3)
Aspiration history			<0.001
NO	154 (79.0)	33 (42.9)
YES	41 (21.0)	44 (57.1)
Nutritional risk			<0.001
NO	156 (80.0)	37 (48.1)
YES	39 (20.0)	40 (51.9)
Length of Stay (d),median (IQR)	15.00 [11.00, 22.50]	15.00 [10.00, 26.00]	0.735
Intubation depth (cm),median (IQR)	100.00 [95.00, 105.00]	100.00 [95.00, 105.00]	0.073
Intubation days (d),median (IQR)	13.00 [10.00, 20.00]	14.00 [9.00, 20.00]	0.356
Number of comorbidities			<0.001
0	125 (64.1)	4 ( 5.2)
1	40 (20.5)	15 (19.5)
2	19 ( 9.7)	31 (40.3)
3	8 ( 4.1)	18 (23.4)
4	3 ( 1.5)	9 (11.7)
Antacids			0.435
NO	31 (15.9)	16 (20.8)
YES	164 (84.1)	61 (79.2)
Gastricstimulants			0.935
NO	61 (31.3)	23 (29.9)
YES	134 (68.7)	54 (70.1)
RBC (10¹²/L),median (IQR)	3.75 [3.20, 4.50]	3.55 [3.12, 4.15]	0.047
Hb (g/L), mean ± SD	116.00 [111.0, 130.0]	115.00 [110.0, 129.0]	0.356
PLT (10⁹/L) ,median (IQR)	208.00 [148.5, 307.0]	190.00 [122.0, 284.0]	0.056
WBC(10⁹/L),median (IQR)	11.54 [8.30, 15.47]	10.94 [7.90, 14.25]	0.345
Lym(%),median (IQR)	0.92 [0.66, 1.31]	0.80 [0.62, 1.14]	0.176
NEUT(%),median (IQR)	9.70 [6.32, 13.53]	9.18 [6.41, 12.53]	0.572
K+ (mEq/L)	4.00 [3.55, 4.43]	3.96 [3.66, 4.32]	0.469

Note: Number of comorbidities included the total diseases of neurological complications, hypertension, respiratory complications and heart disease.

Age is presented by mean and interquartile range (IQR).

Hb: Hemoglobin; WBC: white blood cell; RBC: red blood cell; PLT: platelet; Lym: lymphocyte; NEUT: neutrophil.

Abbreviations: IQR, interquartile range

Table 2 Logistic regression of risk factors for aspiration

Variables	Beta coeff.	SE	Z	p	Odds ratio	95% CI
Intercept	-2.849	3.376	-0.84
Age	-0.068	0.026	-2.66	0.008	0.93	0.89-0.98
Consciousness	0.466	0.222	2.10	0.036	1.59	1.03-2.46
Mechanical ventilation	2.309	0.525	4.39	＜0.001	10.07	3.59-28.20
Aspiration history	2.192	0.545	4.02	＜0.001	8.96	3.08-26.09
Nutritional risk	1.870	0.672	2.78	0.005	6.49	1.74-24.23
Number of comorbidities	1.177	0.261	4.52	＜0.001	3.25	1.95-5.41

Abbreviations: Beta coeff., beta coefficient; SE, standard error.

No competing interests reported.

supplementTableS1.docx

Machine learning predictive model for aspiration risk in early enteral nutrition patients with severe acute pancreatitis

Status:

Version 1

Abstract

Figures

1. Introduction

2. Methods

3. Results

4. Discussion

5. Conclusion

References

Tables

Additional Declarations

Supplementary Files

Status:

Version 1