I am working on finding the optimum follow-up interval for a failure outcome. My data is survival data including a time variable and an event variable and a number of exposure variables. I was wondering if I can set up the ROC curve by placing time-to-follow-up as the exposure variable and the event as the classifying variable, and then take the optimum cutoff point of this ROC curve as the optimum follow-up interval to best screen my future subjects?
The optimum has to be optimal in relation to some criterion. You are best looking at the cumulative distribution of events and relating that to the costs of increasing follow up. For example, the first 50% of events may occur in 1 year, but in the next year only an additional 20% occur, so you are paying a lot extra for those events.
However, this brings up the question of whether all events are equal. Are late-occurring events qualitatively different? For example, event rates in the first year after MI are higher than in subsequent years, but they are partly driven by the severity of the initial episode. It is only by following after the first year that relationships between event rates and risk factors becomes clear.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis