Development of a prediction model for acute respiratory distress syndrome in ICU patients with acute pancreatitis based on machine learning algorithms
"Objective To develop and validate a predictive model based on machine learning algorithms to assess the risk of acute respiratory distress syndrome（ARDS）in patients with acute pancreatitis（AP）admitted to the intensive care unit（ICU）. Methods The relevant data of 857 AP patients from the Medica...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
The Editorial Department of Chinese Journal of Clinical Research
2025-08-01
|
| Series: | Zhongguo linchuang yanjiu |
| Subjects: | |
| Online Access: | http://zglcyj.ijournals.cn/zglcyj/ch/reader/create_pdf.aspx?file_no=20250808 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | "Objective To develop and validate a predictive model based on machine learning algorithms to assess the risk of acute respiratory distress syndrome（ARDS）in patients with acute pancreatitis（AP）admitted to the intensive care unit（ICU）. Methods The relevant data of 857 AP patients from the Medical Information Mart for Intensive CareⅣ v2.2（MIMIC⁃Ⅳ v2.2）database were retrospectively analyzed and were randomly divided into a training set（n=601）and an internal validation set（n=256）in a 7∶3 ratio. Additionally，the relavent data of 126 AP patients from the ICU of Changshu Hospital Affiliated to Soochow University from January 2019 to March 2024 were collected as an external test set. Patients were categorized into ARDS and non ⁃ ARDS groups based on the occurrence of ARDS. Demographic characteristics，initial vital signs，laboratory data，functional scores，and complications within the initial 24⁃hour of ICU admission were collected. Feature selection was performed using least absolute shrinkage and selection operator（LASSO）regression. Predictive models were constructed using seven machine learning algorithms：random forest（RF），extreme gradient boosting（XGBoost），light gradient boosting machine（LightGBM），decision tree（DT），logistic regression（LR），support vector machine（SVM），and K⁃nearest neighbors（KNN）. Model performance was evaluated using receiver operating characteristic （ROC） curves，calibration curves，and decision curve analysis（DCA）. Finally，model interpretability was enhanced through Shapley additive explanations（SHAP）analysis. Results In the MIMIC⁃Ⅳ database，202 patients（23.57%）developed ARDS，while 26 patients（20.63%）developed ARDS in the external test set. Seven key variables were selected by LASSO regression from 43 variables in the training set to construct the models. Among various machine learning models，the RF model demonstrated the best performance with an
area under the curve（AUC）of 0.780（95%CI：0.721-0.846）in the internal validation set and 0.842（95%CI：0.751-0.917）in the external test set，outperforming the other six models. The calibration curve indicated that the predicted probabilities from the RF model had the smaller deviation from the actual probabilities compared to other models，showing the best overall predictive performance. SHAP analysis based on the RF model revealed that mechanical ventilation，sequential organ failure assessment（SOFA）score，body mass index（BMI），peripheral oxygen saturation（SpO2）and simplified acute physiology score（SAPS Ⅱ）were the main factors influencing ARDS risk. Mechanical ventilation increased the risk of ARDS from 16% to 37% . When the SOFA score exceeded 8, the ARDS risk rose significantly. The risk of ARDS elevated with increased BMI. While SpO2 remained below 90%, ARDS risk stabilized at 30%， once SpO2 surpassed 90%, the risk demonstrated a declining trend with further increases in SpO2. For SAPS⁃Ⅱ scores between 46 and 60, ARDS risk showed a pronounced upward trend. Conclusion The RF predictive model provides a reliable tool for assessing the risk of ARDS in AP patients and enhances model interpretability through the SHAP method，aiding in clinical decision⁃making."
|
|---|---|
| ISSN: | 1674-8182 |