What is the most common form of multivariable analysis for survival data?

• A. Linear regression.
• B. Logistic regression.
• C. Cox proportional hazards model.
• D. Poisson regression.

Answer: (C)

Survival data involve time to an event and censoring, so the analysis needs to relate covariates to the hazard of the event rather than to a mean time or a simple probability. The Cox proportional hazards model does this directly by linking covariates to the hazard function—the instantaneous risk of the event at any given time—through hazard ratios. It handles censoring naturally, uses all available follow-up information, and does not require specifying the exact distribution of survival times. The baseline hazard is left unspecified, making the method flexible across diverse datasets, while still allowing multiple covariates to be included simultaneously.

Linear regression isn’t appropriate because survival times are often skewed and many observations are censored, violating normality and constant variance assumptions. Logistic regression models the probability of an event by a fixed time point and ignores when the event occurs or how long someone is followed. Poisson regression models counts or rates and assumes a specific time-constant rate, which doesn’t naturally handle time-to-event with censoring.