OBJECTIVES :

To illustrate the dynamic nature of time, it’s potential impact on risk estimates, and approaches to account for time dependencies in time to event analyses.

METHODS :

Survival analysis techniques will be illustrated including time to event concepts, the relation between survival and hazard functions, censoring, and truncation. Further, exploration of time dependencies will be demonstrated and methodological approaches to account for and visualize variations in hazard estimates over time will be shown.

RESULTS :

Descriptive statistics and unadjusted product-limit estimators will be presented including frequencies, chi-square statistics, and Kaplan-Meier estimates to illustrate unadjusted associations. Graphical exploration of time-varying functions will be conducted to display time interactions and violation of proportionality. Cox proportional hazards modeling will be demonstrated and compared to the extended Cox model to account for time dependencies and covariate adjustment.

CONCLUSIONS :

Temporal deviations including loss to follow-up and time-dependencies can occur in even the most controlled research designs and accounting for time offers a unique and more granular view of the outcome in the context of treatment and other covariates. It is incumbent upon researchers to explore study outcomes as a function of time when possible in survival analytic approaches since time dependencies, when unaccounted for, can drastically alter the interpretation of results. Techniques exist to address time-varying hazards when identified and should be employed.