Prenatal paracetamol exposure and wheezing in infancy: a targeted maximum likelihood estimation application

Area of Science:

• Epidemiology
• Biostatistics
• Perinatal Health

Background:

• Targeted Maximum Likelihood Estimation (TMLE) is a robust statistical method for causal inference.
• TMLE utilizes SuperLearner for flexible modeling of complex relationships.
• Understanding early life exposures and their impact on child health is crucial.

Purpose of the Study:

• To evaluate the association between first-trimester maternal paracetamol use and child wheezing up to 18 months.
• To apply TMLE for robust estimation of this effect.
• To compare TMLE with traditional regression and propensity score methods.

Main Methods:

• Employed TMLE with three implementations: parametric, SuperLearner (parametric/semiparametric), and SuperLearner (parametric/semiparametric/non-parametric).
• Adjusted for progressively larger sets of covariates, including socioeconomic, maternal, and health factors.
• Compared TMLE results with multivariable regression, propensity score regression, and inverse probability weighting.

Main Results:

• All methods indicated a weak positive association between maternal paracetamol use and child wheezing.
• The association attenuated towards the null as more covariates were included in the analysis.
• TMLE with extensive covariate adjustment (Set 3) yielded a risk ratio of 1.10 (95% CI 0.97 to 1.26).

Conclusions:

• The observed association may represent a small positive effect or residual confounding.
• Consistency across different TMLE implementations suggests robustness against model misspecification.
• Further research may be needed to fully elucidate the role of unmeasured confounding.