Which models are commonly used to analyze time-to-event data in cohort studies?

• A. Logistic regression for time-to-event
• B. Cox proportional hazards models and accelerated failure time models
• C. Linear regression for time-to-event
• D. Poisson regression for rate data

Answer: (C)

When analyzing time-to-event data in cohort studies, you use survival analysis techniques that handle not only whether an event occurs but also when it happens, including the fact that some individuals may not have the event by the end of follow-up (censoring). The two commonly used modeling approaches are Cox proportional hazards models and accelerated failure time models.

Cox models are semi-parametric and focus on how covariates influence the hazard, or instantaneous risk, of the event over time. They yield hazard ratios that tell you whether a factor increases or decreases the risk at any given moment, without needing to specify the underlying shape of the baseline hazard function. This makes them very flexible for many cohort studies.

Accelerated failure time models, on the other hand, are parametric and model the actual survival time directly. They produce time ratios that describe how covariates stretch or shrink the expected time to event, assuming a particular distribution for event times (such as Weibull or log-normal). These are useful when you want a direct interpretation in terms of time.

Linear regression is not appropriate as the main tool for time-to-event outcomes because it assumes normal, uncensored data and constant variance, which is violated when many participants are censored or when event times are skewed. Poisson regression is designed for counts or incidence rates over person-time rather than the timing of a first event, so it doesn’t directly model time to event.

In summary, for time-to-event analysis in cohort studies, the standard choices are Cox proportional hazards models and accelerated failure time models.