# Recurrent time-to-event?

I am working on a research paper looking at hospital readmissions after surgery and I have data for about 200 patients. I have the readmission dates for the first year after initial hospital discharge and I can accordingly calculate time between any two events (readmissions). In my sample, it turned out that the number of readmission events per patient are between 0 and 8; i.e. some patients have no readmissions and some have up to 8. I am aware that this is a recurrent time-to-event; but our focus is not on creating survival curves. Our focus is on those admitted (165 patients) and we are wondering if patients with multiple readmissions have their events close together or scattered throughout the year. Any suggestions of what analysis to use in order to make a conclusion or not whether that repeated readmissions are close together or spread over time? If not a specific analysis, is there any graph that can visualize this (I was thinking at the single patient level)? I would appreciate if anybody have the syntax or link to that. Thanks!

• I'm not sure I follow your thinking. If there are more readmissions within a given interval, don't the readmissions have to be closer together? – gung - Reinstate Monica May 4 '16 at 2:10
• Thanks guns for your prompt reply! that is true that there would be more readmissions in a certain interval that there are close together if I have a case for one patient, however, I have 200 patients. For example, if a patient had 3 readmissions at 60, 70, and 80 days after surgery they would be considered close together compared to another who had 3 readmission but at 60 150 and 300 days post surgery. I am trying to quantify the aggregation of readmission for each patient and then draw conclusion based on the 200 patients. – Moe May 4 '16 at 6:40
• This would clinically help by stating that many of readmission are/ are not (depends on what we find) close together indicating that the following close readmissions could have been avoidable if the patient had appropriate care. I hope this answers. Thanks again for your help! – Moe May 4 '16 at 6:40
• Could you ignore the time to the first event and then treat each interval as an observation? You would then have between 1 and 7 observations per patient and could potentially fit a model with a random effect to account for the clustering. I am not absolutely sure this fits your clinical question though. – mdewey May 4 '16 at 12:11
• Hi mdewey! I like your idea and actually I was thinking before about ignoring the time to the first even, but I did not think about a model with random effects. Could you please give more directions? What would be my random effect variable in this case? In other words, how would I cluster the visits? Thanks very much for your help! – Moe May 4 '16 at 23:23