Introduction to Survival Analysis Flashcards
Does survival analysis necessarily mean death?
The methods for analysing time-to-event data are usually called survival analysis, even though the event of interest does not have to be death.
Why do we use rates rather than proportions with survival analysis?
To allow for differing length of follow-up between groups of patients.
What does it mean to have censored data?
In survival analysis, observations are censored if their follow-up finishes before they have had an event.
In general we assume that censored data happens randomly. How might censored data undermine the validity of the analysis if it is not random?
In practice, people who are censored may be more likely (or less likely) to experience the outcome. This would limit the validity of the analysis, especially if many people are censored.
What is a suitable method of survival analysis when we do not have access to exact dates of events/censoring?
Life table method (requires probabilities to be adjusted for censoring)
How is survival data usually presented in clinical trials?
Kaplan-Meier Method
What two types of Kaplan-Meier curve can you construct?
Survival rate Failure rate (1 - S-rate)
Why is the failure rate often more useful visually?
The upward graph is more informative if the cumulative probability of the event is low until the end of the trial (as in the RITA-2 dataset). In such cases, the downward graph uses only a small part of the data range, e.g. the RITA-2 survival curve goes from 1 to 0.81, but the axis range is from 1 to 0. By contrast, the upward graph can cover all its data range (0 to 0.19 for RITA-2) with no empty space.
How can we assess if there is a significant difference in survival outcomes using Kaplan-Meier Curves?
Logrank tests (for of Chi-squared tests)
What statistical components go into survival analysis?
Plots of survival probability
Hypothesis test (Logrank test)
Estimand for magnitude of difference
How do we estimate the magnitude of the difference in survival analysis in clinical trials?
Poisson or Cox regressions (most commonly Cox)
What does a COx regression model?
Hazard ratios (the proportional effect of treatment on event rates in those still at risk at each timepoint)
Why is Cox regression most popular for calculation of proportional hazards?
It does not assume any particular for for the hazard rate in the control arm. The baseline hazard rate can vary over time.
What assumption does the Cox regression method make?
It assume that the ratio of hazard rates between arms is constant. Called ‘proportional hazards’.
What is the best way to test Cox regression validity for proportional hazard ratio?
Look at the KM curve (there are also formal tests)
