# Calculation of person years in a cohort study for survival analyses

I am analysing data from a cohort study using primary care data (but I linked them up to secondary care data only for the outcome). Specifically, I am looking at the adoption of a specific care pathway and risk of hospital admission for respiratory conditions. Every time a patient visits his general practitioner I have data recorded in my database. I defined my cohort based on age range in my baseline year and I followed my cohort for ten years (120 months). If a patient moves out of that practice, or dies without being previously admitted to the hospital for respiratory conditions (my failure) stops contributing in the calculation of the person years the month he moves out/dies.

My doubt is about those who are not admitted to the hospital and who are alive and registered with the practices included in my cohort for the whole study period.

Shall I consider them contributing to the calculation of the persons year for the whole duration of the study period or shall I consider as last 'contribution' the month of the last primary care visit?

Some might argue that if the patient is not visiting his general practitioner it is because he/she doesn't need to but he/she is still contributing to the calculation of the person years because I would know whether he has been admitted to the hospital or not. On the other hand we don't really know with censored data what might happen to that patient if we also consider him/her beyond the last visit recorded in primary care.

My doubt stands because I am plotting a KM curve and if I consider one or the other option the shape of its tale changes (not the main part of the curve, but the ending), as well as the calculation of the person years.

Below an example of the data for clarity

• ID MOVE-OUT FAILURE MONTH DEATH
• 1 NO YES 25 NO
• 2 YES NO 45 NO
• 3 NO NO 66 YES
• 4 NO NO 102 NO
• 4bis NO NO 120 NO

My doubt is about the 4th patient. Last visit was recorded in the 102th month, but the study duration is 120 months. Is the first specification correct or is the second more appropriate? Can you suggest a nice reference for one case or the other?

Thanks

• Are they at risk of the event? – mdewey Oct 26 '17 at 11:04
• Yes, they all are – Vincent Oct 26 '17 at 12:26
• My comment was obviously too cryptic. During what time were they at risk of the event? That gives you your answer. – mdewey Oct 26 '17 at 13:45