A machine learning approach to determine the influence of specific health conditions on self-rated health across education groups

• Main Authors: Jordi Gumà-Lao (Author), Bruno Arpino (Author)
• Format: Book
• Published: BMC, 2023-01-01T00:00:00Z.
• Subjects: article
• Online Access: Connect to this object online.
• Notes: 10.1186/s12889-023-15053-8; 1471-2458
• Keywords: Self-rated health; Health conditions; Education; Machine learning; SHARE survey; Public aspects of medicine; RA1-1270
• Source: BMC Public Health, Vol 23, Iss 1, Pp 1-11 (2023)

Abstract Background Self-rated health, a subjective health outcome that summarizes an individual's health conditions in one indicator, is widely used in population health studies. However, despite its demonstrated ability as a predictor of mortality, we still do not full understand the relative importance of the specific health conditions that lead respondents to answer the way they do when asked to rate their overall health. Here, education, because of its ability to identify different social strata, can be an important factor in this self-rating process. The aim of this article is to explore possible differences in association pattern between self-rated health and functional health conditions (IADLs, ADLs), chronic diseases, and mental health (depression) among European women and men between the ages of 65 and 79 according to educational attainment (low, medium, and high). Methods Classification trees (J48 algorithm), an established machine learning technique that has only recently started to be used in social sciences, are used to predict self-rated health outcomes. The data about the aforementioned health conditions among European women and men aged between 65 and 79 comes from the sixth wave of the Survey of Health, Ageing and Retirement in Europe (SHARE) (n = 27,230). Results It is confirmed the high ability to predict respondents' self-rated health by their reports related to their chronic diseases, IADLs, ADLs, and depression. However, in the case of women, these patterns are much more heterogeneous when the level of educational attainment is considered, whereas among men the pattern remains largely the same. Conclusions The same response to the self-rated health question may, in the case of women, represent different health profiles in terms of the health conditions that define it. As such, gendered health inequalities defined by education appear to be evident even in the process of evaluating one's own health status.