Respiratory Syncytial Virus Infection in Infants: A Comparative Study Using Discriminant, Probit and Logistic Regression Analysis

Abstract

In babies, respiratory syncytial virus (RSV) is the most common cause of lung inflammation (pneumonia) or bronchiolitis (inflammation of the lungs' airways). This virus comes with several symptoms such as congested or runny nose, dry cough, low-grade fever, sore throat, sneezing, headache, difficulty in breathing etc. The virus can cause death in babies if not properly managed and therefore calls for immediate investigations to reveal the significant causes. Several research works have been conducted but the idea of investigating more potential predictor variables and the application of both regression and classification models have been grossly understudied. Therefore, unpublished secondary data collected from three different hospitals in Port Harcourt, Rivers State, Nigeria on fifteen predictor variables which are potential causes of RSV are modeled using two categorical regression approaches – logistic and probit regression models and one classification model – discriminant function analysis. The models were compared using misclassification errors, Receiver Operating Characteristic (ROC) plot, concordance, sensitivity, specificity and pseudo R-square values. The linear discriminant function model outperformed both the logit and probit models. The results showed that paternal history of asthma, maternal history of asthma, mother’s occupation, mother’s smoking habit and mother’s education level were the most important variables to linearly classify seropositive and seronegative RSV patients.