Chrome Extension
WeChat Mini Program
Use on ChatGLM

Mortality risk model to predict bacterial infection in Pediatric Intensive Care Unit

Research Square (Research Square)(2022)

Cited 0|Views4
No score
Abstract
Abstract BackgroundBacterial infections have long been a significant cause of child mortality worldwide; however, models for predicting the risk of death in children with bacterial infections that combine predictive ability and interpretability are scarce. Here we try to explore a new method to complete the task. MethodsWe use Paediatric Intensive Care Database (PIC) to carry out the research. Data from hospitalized children with positive bacterial culture results was extracted and categorized into three groups: the positive culture group, the gram-positive group, and the gram-negative group, which were divided into 80% training and 20% test sets. Then we extracted the demographic information, vital signs, and laboratory data within 24 hours of admission of the datasets. We use the XGBoost algorithm to select the features and rank their importance,and the Logistic Regression (LR) algorithm for model development based on various numbers of feature. All the models were evaluated by Receiver Operating Characteristic curve (ROC), area under the ROC curve (ROC-AUC), Precision-Recall curve (PR), the area under the PR curve (PR-AUC), and compared with Paediatric Mortality Risk Score III (PRISM III), the paediatric logistic organ dysfunction score-2 (PELOD-2), and the paediatric multiple organ dysfunction scores (P-MODS).ResultIn total, 3695 children with bacterial infection were included, with an average age of 20.17 ± 36.74 months, average paediatric intensive care unit stay of 18.51 ± 28.84 days, and overall mortality rate of 8.39%. The following predictors appeared in the 64 most important predictors of the three datasets: average white blood cell count, maximum value of white blood cell technology, average value of anion gap, minimum value of anion gap, maximum value of type B natriuretic peptide, and maximum value of thrombocytocrit. Finally, we established a LR model for bacterial infected children with 4 features and a LR model for gram-negative bacterial infected children with 10 features, which achieve a 0.7244 and a 0.7848 ROC-AUC score respectively. The ROC-AUC scores of the two models were better than PRISM III、PELOD-2 and P-MODS. ConclusionThis study developed models to predict the risk of death in children with bacterial infections. The final models use fewer features and achieve better mortality prediction performance than traditional scoring models, and the models are easier for paediatricians to understand.
More
Translated text
Key words
intensive care,bacterial infection,mortality risk
AI Read Science
Must-Reading Tree
Example
Generate MRT to find the research sequence of this paper
Chat Paper
Summary is being generated by the instructions you defined