Which stratification and multivariable adjustment methods help control confounding in cohort data?

• A. Stratification by confounders to estimate stratum-specific effects.
• B. Multivariable regression with single confounder adjustment.
• C. Propensity score matching only.
• D. Stratification by confounders and multivariable regression (logistic, Poisson/ NB, Cox) to adjust for multiple confounders.

Answer: (D)

The main idea is how to separate the effect of the exposure from other factors that could distort it. Stratifying by confounders means dividing the data into groups where those confounding factors are similar and then comparing the exposure’s effect within each group. This approach reduces confounding within each stratum and can also reveal if the exposure effect changes across different levels of a confounder (effect modification). Yet, when there are many confounders, the number of strata grows quickly and you run into sparse data problems, making stratification alone impractical.

Multivariable regression tackles confounding by adjusting for several confounders simultaneously in a single model. The choice of model depends on the outcome: logistic regression for binary outcomes, Poisson or negative binomial regression for count or incidence data, and Cox regression for time-to-event data. The exposure coefficient from these models represents the association after holding the included confounders constant.

Combining these approaches leverages their strengths: stratification can help identify and explore potential effect modification and reduce confounding for key factors, while multivariable regression provides a scalable way to adjust for multiple confounders efficiently. Propensity score methods are useful for balancing measured confounders, but they are most effective when used alongside regression or stratification rather than as the sole method.