From Hospital Trajectories to Years Without Severe Disease

The subject here is the survival-analysis machinery behind the French study’s Disease-Free Life Expectancy, or Dis-FLE. The specific problem is how to turn patient records that begin late, end for different reasons, and do not all reveal an event into an estimate of remaining disease-free time. Following that process lets the reader interpret Dis-FLE as a conditional survival quantity rather than as an unqualified subtraction from life expectancy.

A trajectory with a late start and several endings

The study treats health as a duration without a new event from its selected set of severe clinical conditions. For each person, the relevant time begins at the later of 1 January 2010 or the person’s 50th birthday. That rule gives the analysis a common observation boundary while respecting the fact that people enter the age range at different times. A person already 55 on 1 January 2010 starts then; a person who turns 50 in 2012 enters on that birthday.

This is not a complete life history. The study has information from 2008 and 2009 for exclusion and baseline construction, but the main duration is observed from the defined start through the 2010–2013 period. Starting observation at 50 or at the calendar boundary creates left truncation: earlier time is not part of the observed risk period, and only people who are alive and eligible at entry can contribute to the analysis. The estimate is consequently about disease-free duration after entry, not about all the healthy time a person may have lived before entering the study.

After entry, several kinds of endpoint are possible. A new severe event ends the disease-free state. Death also ends the possibility of remaining disease-free. The observation can instead stop at 31 December 2013 or when the person is lost to follow-up, without revealing what would have happened afterward. Those latter endings are forms of right censoring: the study knows that the person remained under observation without a recorded event up to a point, but it does not know the person’s later event time from the available data.

The distinction between an event and censoring is essential. If someone is censored at the end of 2013, the record does not say that they became ill on that date. It says that the study’s observation window ended while they were still in the risk set. Treating every last observed date as an event would make follow-up look shorter and disease-free survival look worse. Treating a documented severe event as mere loss to follow-up would have the opposite problem. The analysis needs to preserve what each ending means.

The timing rule also affects the meaning of “remaining disease-free.” Someone who enters at 50 and is observed for three years without a qualifying event contributes three years of event-free follow-up, but the study does not assume that those three years predict the same future as three years observed in a different age range. Age is the analysis time scale, so the risk set is continually changing as people move through later ages. This lets the estimate reflect age-specific event rates rather than treating all elapsed follow-up as equivalent.

A censored trajectory is therefore not useless data. It confirms that the person was event-free for the period in which they were observed and helps define how many people were still at risk at later event times. Its contribution becomes uncertain only after observation stops. Survival methods use that partial information under the independence assumption instead of discarding every person who lacks a complete endpoint.