Parainfluenza Virus Infection Among Human Immunodeficiency Virus (HIV)-Infected and HIV-Uninfected Children and Adults Hospitalized for Severe Acute Respiratory Illness in South Africa, 2009–2014

Cohen, Adam L.; Sahr, Philip K.; Treurnicht, Florette; Walaza, Sibongile; Groome, Michelle J.; Kahn, Kathleen; Dawood, Halima; Variava, Ebrahim; Tempia, Stefano; Pretorius, Marthi; Moyes, Jocelyn; Olorunju, Steven A. S.; Malope-Kgokong, Babatyi; Kuonza, Lazarus; Wolter, Nicole; von Gottberg, Anne; Madhi, Shabir A.; Venter, Marietjie; Cohen, Cheryl

The second and third analyses models were built using manual backward elimination in which nonsignificant variables were removed from the model one at a time starting with the variables with smallest magnitude of effect until all remaining variables had a P value of <.05. Covariates with a P value of <.2 at univariate analysis were assessed for significance with multivariable analysis; statistical significance was assessed at P < .05 for all multivariable models. Two-way interactions were assessed by inclusion of product terms for all variables remaining in the final additive models. For each univariate analysis, we used all available case information. For important variables in the HIV association and AF analyses that had substantial missing data, namely HIV infection status, we multiply imputed that variable as well as any variables that were incomplete and associated with HIV using chained equation multiple imputation over 10 iterations. When adjusting for respiratory viral coinfection in our models, we evaluated coinfection with each virus separately and also as a combined variable of coinfection with any of the tested viruses.