Whose Health Does the Number Belong To?

The subject of this chapter is the population behind the French study’s Dis-FLE estimate, and the specific problem is selection: who appears in the hospital record, who is removed before analysis, and who is absent altogether. The practical judgment is to distinguish an estimate adjusted toward a general-population target from a model that remains about hospitalized people, because the same clinical data can support both only when their scopes are kept separate.

The target population and the sample are not the same

A large database can still be a selected sample. The PMSI data include people who were hospitalized at some point in Metropolitan France, not everyone who was alive in the relevant age range. The study reports that more than 75% of the general population aged 50 and older appeared in the database during the covered period. That is substantial visibility, but the remaining population is not necessarily a random remainder. People who never require hospital care may differ from hospitalized people in health, access, risk, and the likelihood that a severe condition is recorded.

The numbers make the denominator problem tangible. In 2010, France had about 22.5 million people aged 50 and over, while about 10.5 million people observed in hospital remained in the study after the relevant exclusions. The post-exclusion analysis is large enough to estimate trajectories with small statistical errors, but it does not directly observe the other half of the age group. Precision within the observed cohort cannot by itself establish representativeness of the unseen population.

There are two separate questions hidden in the word “coverage.” The first is whether people appear in the database at all. The second is whether the recorded event and health history represent them comparably once they appear. A hospital system can cover many residents while still recording different kinds of encounters for people with different access, referral patterns, or illness severity. The study’s clinical trajectory is strongest for the portion of health that leads to the relevant hospital records. It is not evidence that the people outside those records have no severe illness or no functional limitation.

This distinction matters when the clinical estimate is placed beside HALE or HLY. HALE’s population summary draws on mortality and health information; HLY’s survey framework asks about long-term activity limitation. Their denominators and observation mechanisms are different from PMSI even when the country and age range are similar. A hospital-based estimate may be internally consistent and still require a population bridge before it can answer the same broad national question as a general-population indicator.

The cohort is selected twice. First, the hospital system determines who can appear at all. Second, the study applies exclusions to create a population considered healthy with respect to the chosen severe conditions at 1 January 2010. People observed with any of the severe conditions during 2008–2009 are excluded, as are people with additional chronic, neurological, infectious, genetic, or mental conditions specified by the study. The goal is to avoid starting a disease-free trajectory with a known history that already violates the study’s definition.

That goal produces a coherent starting state, but it also changes the denominator. A person removed because of an earlier qualifying condition is not returned as an unhealthy observation in the final cohort. The remaining participants may be healthier than the original hospitalized population even though they are still drawn from people who entered the hospital system. An estimate based on this group therefore cannot be interpreted by looking only at the final sample size.