Repeated measures with missing data.

[Bob Galbreath]

I have PSA values for 316 prostate cancer patients.These PSAs were obtained at intervals of 6 months, for 5 years. The problem is that there is missing data. Some of the missing data is due to unknown reasons, and some is due to death of a patient. Is there a best way to analyze this data to show how PSAs change after the radaition treatment that they all had.

[Felix Bittmann]

If a person died this is a censoring event and this is fine, you can use this information in a cox regression (for example). If data are missing and you know for sure the patient is still alive (because you have another, later measurement available) you can use multiple imputation to combat this problem, see https://stats.oarc.ucla.edu/stata/se...stata_pt1_new/

[Carlo Lazzaro]

Bob:
deaths apart, attrition is a common nuisance in panel datasets.
How many missing values you have for patients who did not pass away?

[Bob Galbreath]

I saw a typo in my original question. These patients have been followed for up to 15 years, not 5 years. So, if all patients had 2 PSAs per year and if all patients had all 29 PSAs (the first reading was at 9 months post-treatment) there would be 9,164 data points. However, there are only 5,246. Another problem with the data is that not only are there missing data due to deaths, the patients did not all start at the same time and some patients may have fewer PSAs because they have not been followed as long as others. My belief is that I might be only able to present this data descriptively.