D2.131 - Application of a Neural Network Model for Predicting Clinical Outcomes of Acute Bronchitis with Bronchospasm Based on Comorbidities

To evaluate the use of a neural network model in predicting clinical outcomes of acute bronchitis with bronchospasm in children, based on comorbidities.

A multilayer perceptron (MLP) model was developed to assess the probability of unfavorable clinical outcomes in children with acute bronchitis and bronchospasm. The training dataset included 223 cases (70.8%), while the test dataset comprised 92 cases (29.2%). The model’s input layer incorporated the following predictors: infectious diseases, allergic diseases, parasitic diseases, anemia.

Analysis of the area under the ROC curve (AUC) showed identical values of 0.598 for both “Recovery” and “Exacerbation,” indicating low discriminative performance. Assessment of variable importance revealed that the most influential predictors were: anemia — normalized importance 100.0%; allergic diseases — 78.0%; parasitic diseases — 66.6%; and vaccination — 61.0%. Despite the model’s low sensitivity for predicting complicated outcomes, it underscores the significance of certain risk factors, particularly anemia and allergic conditions.

The model can serve as an auxiliary screening tool for the comprehensive evaluation of children with a burdened premorbid background. Classification results demonstrated high sensitivity to favorable outcomes but very low sensitivity to complications, which limits the clinical applicability of the model in risk assessment, despite its relatively high overall accuracy.