Why the Average Splits Apart

The subject here is the variation in Dis-FLE that the French clinical model finds across behavioral risk profiles, sex, and place. The specific question is what a subgroup pattern can legitimately tell us about disease-free trajectories—and what it cannot tell us about causes, personal futures, or policy effects. The reader’s practical gain is a disciplined way to read hazard ratios and model predictions as conditional evidence rather than as rankings of people or interventions.

What the model reveals about unequal trajectories

An overall Dis-FLE average compresses people with different observed risk profiles into one curve. The French study uses the Cox model to separate some of that variation. For its illustrative profiles, the lowest-risk group has none of the selected risk-increasing behaviors, the intermediate group has one, and the highest group has two. These are model groupings built from clinically coded tobacco use, alcohol use disorders, and obesity; they are not a complete survey of everyone’s exposure or a moral classification of individuals.

The contrast is large in the model’s terms. At age 50, the study reports roughly a ten-year range in conditional Dis-FLE between the lowest and highest selected profiles. Having at least one risk-increasing behavior is associated with an approximately five-year reduction in the age-50 Dis-FLE comparison. The exact value depends on sex, the particular risk factor, and the model’s other covariates, but the broad conclusion is clear: a population average can hide a wide spread in predicted disease-free trajectories.

That finding is important without being a personal forecast. The profile values are conditional predictions for the hospitalized population and the study’s severe-event definition. They do not mean that every person classified in a category will lose a fixed number of years, or that changing one coded characteristic would mechanically restore those years. The model identifies a pattern in observed records. It does not observe the counterfactual future in which one person’s exposure is changed while everything else remains the same.

The behavioral factors also differ in their modeled strength. Category-2 alcohol exposure has the largest reported effect, followed by smoking and obesity. In the study’s conclusion, each behavioral factor has a hazard ratio above 2 before age 80, and category-2 alcohol has a hazard ratio above 3 before age 60. These are relative hazards within the fitted model, not absolute probabilities. A ratio above 1 means a higher event rate among people still at risk relative to the selected reference; the corresponding absolute chance over a period depends on the baseline hazard and age.

The categories themselves need to be read carefully. Category 0 means that no disorder was recorded under the study’s coding rules. Categories 1 and 2 represent increasing clinical severity for the relevant factor.