Causal Inference and Survey Data in Paediatric Epidemiology: Generalising Treatment Effects From Observational Data

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Survey data are vital for understanding child health outcomes in epidemiology.
• Causal inference from observational data is advanced by the potential outcomes framework.
• Traditional methods often neglect sample weights, limiting generalizability.

Purpose of the Study:

• To demonstrate methods for estimating population average treatment effect (PATE).
• To examine the impact of household second-hand smoke (SHS) exposure on children's blood pressure.
• To compare different causal inference estimators.

Main Methods:

• Utilized National Health and Nutrition Examination Survey (NHANES) 2017-2020 data.
• Assessed SHS exposure's effect on blood pressure in school-aged children.
• Applied Inverse Probability of Treatment Weighting (IPTW), G-computation, and Targeted Maximum Likelihood Estimation (TMLE).

Main Results:

• Fully adjusted methods (IPTW, G-computation, TMLE) differed significantly from unadjusted models.
• Including sample weights widened confidence intervals but improved generalizability.
• G-computation and TMLE yielded narrower confidence intervals compared to other adjusted methods.

Conclusions:

• Sample weights are essential for causal inference and generalizability in survey data.
• Estimators incorporating sample weights provide a robust framework for health research.
• Accurate estimation of average treatment effects requires accounting for survey design.