Survival analysis Flashcards
what is the dependent variable in survival analysis?
The time until a well defined event occurs
what kind of analysis is survival?
time-to-event analysis
What is time in survival analysis?:
the time interval between the beginning of a study and the occurrence of the event.
what is the duration of being at risk of the event?
time between the start of measuring and the event
what is the main goal of survival analysis?
to estimate the survival probability from survival time and assess the effect of predictors (wealth, local conditions, etc.) on survival
what is censoring?
when you don’t know what happened, if the event happened to the subject or not.
what is right censoring?
the event happening after the measuring ended, or if it’s unknown excactly when it happened
what is left censoring?
when the event happened before you started measuring
what is not censored data called?
complete
how is censored data represented in a graph?
with a plus.
how do you account for censoring?
a survival analysis
S(t)=
survival function. The probability that a subject survives from the starting time to beyond a specific time t.
H(t)=
hazard function. The instantaneous probability for the event to occur, if the subject survived to time t.
what does a constant hazard function mean for survival function?
exponential decrease
how can you know if hazard function is constant over time?
Ln(survival function). if the line is straight the hazard function is constant over time.
whats the probability of the event happening?
1- survival probability
what is on the y-axis in survival probability graph?
probability of survival
how to calculate the probabily of survival and event happening for a time that wasn’t monitored?
you convert P to a rate at which events take place over time (r). then if occurs at constant rate you can calculate the P for a different time, and 1 - P is chance of event.
in survival graph how to know if event rate increases or decreases?
upward curve (sad smile) = increase. flat line is constant. downward curve (happy smile) = decrease
how to account for local conditions for survival rate?
you use Kaplan Meier estimation or cox proportional hazards model
whats an assumption for cox?
the groups have to be proportional, the lines can’t cross. you can test this using Scaled Schoenveld residuals.
what does cox estimate?
relative risk
can you remove censored data in Kaplan Meier?
no
if you have two groups in a Kaplan Meier graph, when is the difference likely significant?
if they don’t overlap.
what’s the null hypothesis for cox?
Beta (regression coefficients for independent variables) are 0.
Why don’t you compare the mean time- event between groups using t-test or regression?
Because this ignores censoring
Why can’t you use proportion of events and logistic regression to compare mean - time events?
this ignores time
what is ‘median’ in kaplan output?
time to event
in a cox proportional hazard output, what is the hazard of the individual that got the treatment relative to the baseline hazard?
100* Exp(coef)