Causal inference in studies of preterm babies: a simulation study.

Area of Science:

• Perinatal epidemiology
• Biostatistics
• Clinical research methodology

Background:

• Studies focusing on preterm births often face challenges in causal interpretation.
• Pre-existing pathologies in fetuses can influence gestational length and neonatal outcomes.
• Understanding these biases is crucial for accurate risk assessment in neonatal care.

Purpose of the Study:

• To illustrate how associations in preterm birth studies can be causally misinterpreted.
• To demonstrate the impact of selection bias on estimating risks of neonatal death.
• To highlight the need for careful study design in perinatal research.

Main Methods:

• A simple data simulation of a hypothetical fetal cohort with varying pathological factors (A-D) was employed.
• The study focused on births at or before 32 weeks of gestation.
• Associations between specific pathologies (pre-eclampsia, chorioamnionitis) and neonatal death were analyzed using odds ratios.

Main Results:

• Simulated odds ratios for neonatal death were substantially biased, underestimating true risks.
• Factor A (pre-eclampsia) showed a protective effect (OR 0.39) in preterm births, despite a true causal OR of 1.50.
• Factor D (chorioamnionitis) also exhibited biased associations, complicating risk interpretation.

Conclusions:

• Selection bias is inherent in studies of very preterm births.
• Babies with one pathology are often compared to those with multiple, leading to biased associations.
• Findings underscore that associations in preterm birth studies may not reflect true causal risks.