Handling multiple time-varying exposures in survival analysis using real-world pediatric data from the pedianet database

In survival analysis, models typically assess the impact of a single Time-Varying Exposure (TVE), where the exposure status can change over time. However, situations with multiple TVEs frequently arise, and adequate statistical handling remains an area of active research. To apply multiple time-varying approaches and to compare estimates derived from different models using an application with real-world data in the paediatric field. The Italian national paediatric database Pedianet was used to identify children aged between six months and 14 years at the beginning of the epidemiological season (October 1, 2017, to May 31, 2018). Influenza vaccine administrations and antibiotic prescriptions were modeled using both time-fixed and time-varying approaches. Cox proportional-hazard models with random intercept for the region of residence to address the association between antibiotic use, influenza vaccination, and the onset of influenza/influenza-like illness (ILI). Estimates for influenza vaccination remained relatively stable across the different modeling approaches, likely due to the relatively short length of the potential misspecification window. In contrast, estimates for antibiotic use varied significantly between the different scenarios, highlighting the need for careful evaluation and selection of the most appropriate statistical handling approach. It is of utmost importance to carefully evaluate the characteristics of each exposure included in the analysis. Statistical tools and techniques tailored for multiple TVEs need to be acknowledged.