Assessing spillover effects: Handling missing outcomes in network-based studies

Area of Science:

• Epidemiology
• Social Network Analysis
• Causal Inference

Background:

• Estimating causal effects in social networks is complex due to spillover effects and missing outcome data (censoring).
• Spillover effects occur when an intervention impacts individuals not directly exposed but connected within a network.

Purpose of the Study:

• To develop and validate a novel statistical method for estimating causal effects in network-based studies with censored outcome data.
• To assess the impact of community alerts on human immunodeficiency virus (HIV) risk behavior, considering both direct and spillover effects.

Main Methods:

• Introduction of an inverse probability censoring weighted (IPCW) estimator, extending existing methods for network data with censoring.
• Theoretical proofs of the estimator's consistency and asymptotic normality, with derivation of its asymptotic variance.
• Application of the IPCW estimator to real-world data from the Transmission Reduction Intervention Project (TRIP) and simulation studies for performance evaluation.

Main Results:

• The proposed IPCW estimator demonstrates consistency and asymptotic normality.
• Simulation studies confirm the estimator's effectiveness with adequate sample sizes and network structures.
• Analysis of the TRIP data revealed that community alerts significantly reduced HIV risk behavior among individuals and their network contacts.

Conclusions:

• The developed IPCW method is a valuable tool for analyzing network data with censored outcomes.
• Community alerts can effectively reduce HIV risk behavior through direct and indirect (spillover) pathways within social networks.
• Interventions targeting social networks show promise for public health initiatives, like reducing HIV transmission risk.