How can sensitivity analyses assess the potential impact of unmeasured confounding in a cohort study?

• A. Using stratification by measured confounders only
• B. Use the E-value to quantify how strong an unmeasured confounder would need to explain away the observed association
• C. Increasing sample size
• D. Collecting more outcome events

Answer: (B)

The main idea here is using sensitivity analyses to gauge how unmeasured confounding might affect the observed association. The E-value provides a concrete, quantitative way to do this: it tells you the minimum strength that an unmeasured confounder would need to have with both the exposure and the outcome to explain away the observed association after accounting for measured confounders. If you find a large E-value, it means you’d need a fairly strong, perhaps unlikely confounder to negate your result, suggesting the finding is fairly robust to unmeasured confounding. This makes the sensitivity analysis directly focused on potential hidden bias rather than just increasing precision or adjusting for known factors.

Stratifying by measured confounders only addresses bias from known factors and doesn’t quantify the impact of unmeasured ones. Increasing sample size or collecting more outcome events improves precision and statistical power but does not directly assess or quantify unmeasured confounding.