Which method adequately handles time-varying exposures in a cohort study?

• A. Use time-dependent covariates in Cox models
• B. Use baseline exposure only
• C. Randomize exposure status over time
• D. Treat exposure as fixed and ignore changes

Answer: (A)

The key idea is modeling exposures that can change as people are followed. In cohort studies, an exposure like smoking, medication use, or a risk factor can start or stop during the observation period. To correctly estimate how that exposure affects the risk of the outcome, you need a method that updates the exposure status as it actually changes.

Using time-dependent covariates in a Cox model does exactly that. The hazard at any moment is allowed to depend on the exposure status at that moment, and the data are often split into intervals where the exposure is constant. This counting-process or interval-based approach keeps the risk sets properly aligned with who is exposed at each time, preventing misattribution of person-time to the wrong exposure category. It also avoids immortal time bias, where time during which the event could not occur is incorrectly allotted to a particular exposure group.

If you use only baseline exposure, you’re assuming the exposure never changes, which is rarely true. Any later changes in exposure would be ignored, leading to misclassification and biased estimates of the exposure effect. Treating exposure as fixed and ignoring changes has the same problem, and randomly changing exposure over time isn’t a standard, feasible method in observational cohorts.

So, the best choice is using time-dependent covariates in the Cox model because it faithfully represents how exposure evolves over follow-up and yields more valid estimates.