Introduction. - Job retention has become a priority in terms of occupational health. The risk of unfitness for work due to medical reasons is the main risk of professional disinsertion with health consequences. The objective is to create a "sector-job retention/disability matrix" as indicators of sectors at risk, using data from the local observatory on the island of Reunion. Method. - A retrospective study of data collected between 2017 and 2021 by the two occupa-tional health and prevention services on Reunion Island (Intermetra and SISTBI, excluding the civil service), was conducted. According to the activity sector classified by the French Nomen-clature of Activities 2008 (NAF), the number of incapacities per year and their proportion was described, as well as the number of employees benefiting from the obligation of employment of disabled worker "OETH" among these incapacities. Descriptive and predictive analyses using Machine Learning were performed. Results. - Between 2017 and 2021, an increase in incapacities and disabled worker employment obligations was observed on the observatory data. The main sectors, which are accommoda-tion and catering, human health and social action (including the home help sector), certain businesses such as food sales, the construction sector with masonry work, and the cleaning sector are at risk of professional displacement on these data in classic analysis. The creation of Machine Learning (in particular Elastic Net, Random Forest, Decision Tree) makes it possible to make predictions that must be verified over the following years. Conclusion. - In conclusion, this original approach by "sector-job retention/disability matrix"highlights the importance of the use of remarkable data collected daily by the occupational physicians of the Reunion Island in the framework of the prevention of job loss.

Le maintien en emploi est devenu une priorité en termes de santé travail. Le risque d’inaptitude pour cause médicale est le principal risque de désinsertion professionnelle avec des conséquences sanitaires. L’objectif est de créer des matrices emplois-inaptitudes/handicap comme indicateurs de secteurs à risque en utilisant les données de l’observatoire local de l’île de la Réunion. Une étude rétrospective sur les données recueillies entre 2017 et 2021 par les deux services de prévention et de santé au travail de l’île de la Réunion (Intermetra et SISTBI, hors fonction publique) a été effectuée. En fonction du secteur d’activité classé par la Nomenclature française des activités 2008 (NAF), le nombre d’inaptitudes par année et leur proportion a été décrit, de même que le nombre de salariés bénéficiant de l’obligation d’emploi de travailleur handicapé (OETH) parmi ces inaptitudes. Des analyses descriptives et de prédiction par Machine Learning ont été réalisées. Entre 2017 et 2021, il a été observé une augmentation des inaptitudes et des obligations d’emploi de travailleurs handicapés sur les données de l’observatoire. Les principaux secteurs que sont l’hébergement et la restauration, la santé humaine et l’action sociale (dont le secteur d’aide à domicile), certains commerces comme la vente alimentaire, le secteur de la construction avec les travaux de maçonnerie, le secteur du nettoyage sont à risque de désinsertion professionnelle en analyse classique. La création par Machine Learning (notamment Elastic Net, Random Forest, arbre de décision) permet d’envisager des prédictions qui devront être vérifiées les années suivantes. En conclusion, cette approche originale par matrices secteurs inaptitudes/handicap met en lumière l’importance de l’utilisation des données remarquables recueillies au quotidien par les médecins du travail de l’île de la Réunion dans le cadre de la prévention de la désinsertion professionnelle. Job retention has become a priority in terms of occupational health. The risk of unfitness for work due to medical reasons is the main risk of professional disinsertion with health consequences. The objective is to create a “sector-job retention/disability matrix” as indicators of sectors at risk, using data from the local observatory on the island of Reunion. A retrospective study of data collected between 2017 and 2021 by the two occupational health and prevention services on Reunion Island (Intermetra and SISTBI, excluding the civil service), was conducted. According to the activity sector classified by the French Nomenclature of Activities 2008 (NAF), the number of incapacities per year and their proportion was described, as well as the number of employees benefiting from the obligation of employment of disabled worker “OETH” among these incapacities. Descriptive and predictive analyses using Machine Learning were performed. Between 2017 and 2021, an increase in incapacities and disabled worker employment obligations was observed on the observatory data. The main sectors, which are accommodation and catering, human health and social action (including the home help sector), certain businesses such as food sales, the construction sector with masonry work, and the cleaning sector are at risk of professional displacement on these data in classic analysis. The creation of Machine Learning (in particular Elastic Net, Random Forest, Decision Tree) makes it possible to make predictions that must be verified over the following years. In conclusion, this original approach by “sector-job retention/disability matrix” highlights the importance of the use of remarkable data collected daily by the occupational physicians of the Reunion Island in the framework of the prevention of job loss.