Prediction of disease severity in Jordanian patients with respiratory syncytial virus (RSV) in the presence of missingness

Abstract

The global burden of respiratory syncytial virus (RSV) is immense. Of the estimated 33.8 million annual RSV-associated cases of acute respiratory infection (ARI), 3.4 million (roughly 10.1%) require hospitalization. The majority of these hospitalizations involve previously healthy children, thus begging the question of whether there exist particular subgroups of patients in less need of intervention relative to others. Also of interest is how patient-level risk of experiencing a major medical intervention and/or event (MMIE) can be quantified when dealing with incomplete covariate information. To investigate, we invoke a dataset of 1,397 RSV-positive children hospitalized at Al-Bashir Government Hospital between March 2010 and March 2013. We propose a risk index that estimates a patient’s proba- bility of experiencing an MMIE contingent on his/her covariate information, which includes medical history, clinical, and medical examination data. We generate these probabilities (or risk scores) by ridge regression coupled with M-fold cross-validation to identify an optimal shrinkage penalty term λ, and we handle missing data by implementing multiple imputation by chained equations. We then explore how to properly characterize patient-level risk where partial covariate information is available by studying how factors such as the extent of overall missingness and missingness in variables deemed by scientific literature to be associated with disease severity influence subject risk score estimation. The described procedure can provide unique insights regarding risk of MMIE experience among hospitalized RSV patients.