Survival analysis

Published on October 31, 2016   26 min

Other Talks in the Series: The Risk of Bias in Randomized Clinical Trials

Please wait while the transcript is being prepared...
Hi, my name is Isoken Odia and I'm a biostatistician for the Jaeb Center for Health Research. I'm going to be giving a talk on the key concepts of survival analysis.
The outline for this presentation includes, first I'm going to talk on what survival analysis is. I'm going to discuss probability functions, censoring and truncation, estimators of survival function, the Cox regression model, and the proportional hazards assumption, competing risk, selection bias in randomized clinical trials, frailty models and statistical software.
First, I want to use an illustration to describe what survival analysis means. During a soccer game there's a draw, and we need a winner. Usually you have to go for a penalty kick. There are two teams, and each person has to play the ball into the net. They keep on playing the ball until someone loses. In survival analysis it's a bit similar. We keep on following the participants till they have a failure, usually it's called 'the event'. So if Beckham shoots the first goal, we keep on following everybody until someone loses the ball. That's kind of what survival analysis looks at. We are looking at the timing to failure, timing to losing the game, timing to lose the ball.
So what is survival analysis? Survival analysis is a statistical method for analyzing survival data, which is longitudinal time to event data. In this situation, our outcome of interest is the time to an event. An event is usually the transition from one state to the other. It may be disease relapse, it could be death, it could be disease remission- just moving from one state to the other. Initially survival analysis was termed survival because they were always looking at the time to death. So how many people survived past the particular time? Objectives of survival analysis: What do we want to look at? We want to look at the time to event for certain groups. We want to estimate the timing to the event. We want to compare the time to event for different groups. We want to assess how setting variables affects the time to event.