Prediction on influenza-like virus pathogen and its effects on prognosis of patients with community acquired pneumonia under long and short term memory neural network model

Abstract

This study was to explore the application value of long and short term memory neural network model (LSTM) in prediction of influenza-like virus pathogen, and analyze the prognostic factors of patients with community acquired pneumonia (CAP). Based on the LSTM, the prediction model of influenza-like virus pathogen (PALSTM) was constructed, and compared with the mathematical model in prediction performance. 716 CAP patients caused by influenza pathogens were selected as the research objects, and they were divided into a hormone group (321 cases) and a non-hormonal group (395 cases) according to whether they were treated with glucocorticoids (GC). The gender, age, concomitant diseases, laboratory indicators, and imaging examination results on the day of admission were compared for patients between the two groups. The accuracy, sensitivity, and specificity of the PRLSTM model were higher than those of autoregressive moving average model (ARMA) and moving average model (MA). The proportions of patients with cardiovascular disease (CVD), diabetes, and chronic kidney disease (CKD) in the non-hormonal group were much higher in contrast to those in the hormone group (P < 0.01); the proportion of patients with chronic obstructive pulmonary disease (COPD) in the hormone group was greatly higher than that in the non-hormonal group (P < 0.01). The age, gender proportion, the proportion of patients with procalcitonin (PCT) ≥ 0.1 ug/L, the proportion of patients with arterial blood oxygenation index (ABOI) < 300 mmHg, and the proportion of patients treated with neuraminidase inhibitor within 48 h in the hormone group were compared with the non-hormonal group, showing remarkable differences (P < 0.01). The proportion of patients with nursing in intensive care unit (ICU) and the 30-day mortality rate in the hormone group were higher greatly in contrast to those in the non-hormonal group (P < 0.05). Asthma, COPD, and ABOI < 300 mmHg were independent risk factors for patients with FLU-A and CAP receiving glucocorticoid therapy (P < 0.05). It showed that the PRLSTM model based on the LSTM was better than the mathematical models in predicting influenza-like virus pathogens. GC therapy was related to many adverse clinical outcomes of patients with FLU-A and CAP, so it should not be used routinely.