# How to model survival analysis for "lost disease-free years"

My research team suggested we use in a Cox regression the "lost disease-free years" as outcome measures. It works like this: we have the median age at which the general population gets cardiovascular disease, say 75 years. In my sample, I subtract the age that every subject got cardiovascular disease from that age; e.g. if somebody got cardiovascular disease at age 60, their lost disease-free years is 15. IF somebody got cardiovascular disease at age 81, their lost disease-free years is -6 (they "gained" disease-free years compared to the general population).

However, this is not a typical time-to-event variable to be modelled using Cox regression. We want a model in which our risk estimate is positive for positive lost disease-free years and increases with their absolute value; the earlier you get the disease before the population median, the higher the risk. Conversely, it should be negative for negative lost disease-years (protection effect) proportionally to their absolute values (-10 is more protective than -5).

How would you model this?

• Did everyone in your study have cardiovascular disease? That is, is there any censoring at all? How were participants enrolled; were they a cohort or were they enrolled after a cardiovascular event? Jun 9, 2017 at 13:07
• Hi @DavidSmith, it is a traditional cohort study; they were enrolled and then followed. Only 25% developed a cardiovascular disease; the rest would be, in a standard survival analysis using Cox regression, censored. Jun 10, 2017 at 13:28
• By the way, I am now looking into ways to approach the problem by doing a traditional Cox regression and then estimating the "years of healthy life lost" because of each covariate, using some kind of postestimation (I use R but am not averse to Stata). Still unclear how, though. Jun 10, 2017 at 13:29
• I suggest that parametric survival analysis would make post-estimation easier. Jun 10, 2017 at 14:00