Background and objective:
Korian® group, specialized in care and support for fragile people, manages 280 French nursing homes (NH). Its datawarehouse clinical narratives (CN), daily-fed by caregivers in the transmissions’ table, contain key textual data about residents’ care and health. Denutrition and malnutrition are highly prevalent among elderly inside and outside NH and impact health and autonomy among this fragile population. By building a denutrition cohort, computing successive body-mass indexes (BMI), following all the residents’ health events, including falls, hospitalizations and deaths, adding pathologies, risk factors and CN, we can predict residents’ denutrition.

Methods:
First, we selected all residents with at least one transmission, from August 29, 2022 to October 02, 2022 from all 280 NH. We will follow them for twenty weeks, starting August 29. We extracted residents’ age, sex, autonomy level, comorbidities, risks, hospitalizations, deaths, exits events as well as their BMI over time. Second, we labelled four BMI classes according to HAS (High Health Authority) thresholds: emaciated = below 20, undernourished = [20 – 22[, well fed = [22 – 30[, heavy = above 30. We added also 30 main pathologies classes according to four medical experts using the Delphi method. Third, by assessing correlations between our features and then, using tree classifiers we will be able to predict denutrition. Finally, combining clustering and survival analysis will help us grasp the whole denutrition process.

Results:
We built a 21,856 residents cohort with 15,886 (73.4%) women and 5,771 men of mean age 87.6 +/- 8.2. We found 5,418 (25%) emaciated, 4.9 times more women of mean age 90, men being five years younger. Both sexes of this group showed a 0.4 autonomy level loss.

Conclusion:
The ’emaciated’, being slightly older, are particularly at risk. Adding CN labelled health information will deepen our residents’ profile understanding and improve survival.