I am conducting a study on neurological diseased patients. I am looking into possible factors related to time-to-dementia. Costs associated with neurological testing allow us to assess all patients in the prospective cohort on an annual basis. We have (can afford) 4 fixed follow up visits for all patients - baseline, 1-year, 2-year, 3-year. We assess dementia at each visit. My data look roughly as follows:
- Baseline: 0/140 patients with dementia.
- 1-year: 10/120 patients with dementia.
- 2-year: 20/100 patients with dementia.
- 3-year: 30/80 patients with dementia.
A lot of censoring is occurring year-to-year. Further, I don't know when patients actually develop/convert to dementia. Just know it occurred between:
- (0,1]
- (1,2]
- (2,3]
A sensible looking reference on "fixed interval" survival data seemed to suggest that I could analyze these data (after manipulating the dataset) using GLM (binomial distribution, and complementary log-log link). Also compared to Cox PH using Efron method to account for ties. Thoughts on this approach?
http://www.ics.uci.edu/~vqnguyen/stat255/Lecture13.pdf
If I want a "regression" approach to investigate time to dementia what might be possible alternatives to the above model(s)?
