On the use of the E-value for sensitivity analysis in epidemiologic studies.

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Observational studies are prone to bias from unmeasured confounding variables.
• Strengthening causal inference from observational data is a critical challenge in epidemiology.

Purpose of the Study:

• To illustrate the application and interpretation of the E-value, a novel sensitivity analysis metric.
• To assess the robustness of the association between prenatal care indicators and low birthweight using the E-value.

Main Methods:

• The E-value formula (E-value = RR + sqrt(RR * (RR - 1))) was applied.
• Observational data on prenatal care adequacy (GINDEX, APNCU) and low birthweight were used.
• Sensitivity analyses were conducted to quantify the minimum confounder-outcome association required to negate observed exposure-outcome associations.

Main Results:

• E-values ranged from 1.45 to 5.63, varying by prenatal care index and category.
• The highest E-value was observed for "no prenatal care" (GINDEX), and the lowest for "intermediate prenatal care" (APNCU).
• For "inappropriate prenatal care," E-values ranged from 2.76 (GINDEX) to 4.99 (APNCU), indicating substantial required confounder strength.

Conclusions:

• The E-value demonstrates that strong associations between unmeasured confounders and low birthweight (over 400% risk increase) are necessary to explain away the observed prenatal care-low birthweight link.
• The E-value serves as a valuable and intuitive tool for enhancing causal inference in epidemiological research.
• This analysis supports the causal interpretation of the association between inadequate prenatal care and increased risk of low birthweight.