What influences individual perception of health? Using machine learning to disentangle self-perceived health

Self-perceived health is a subjective health outcome that summarizes all the health conditions and is widely used in population health studies. Yet, despite its well-known relationship with survival, it is still unclear as to which health conditions are actually taken into account when making an individual assessment of one's own health. The aim of this paper is to assess the influence of four objective health conditions – IADLs, ADLs, chronic diseases, and depression – in predicting self-perceived health among Europeans by age group (50–64 and 65–79) and by sex. Classification trees (J48 algorithm), which pertains to the emerging Machine Learning techniques, were applied to predict self-perceived health according to the four abovementioned objective health conditions of European individuals in the sixth wave of the Survey of Health, Ageing and Retirement in Europe (SHARE) (n = 55,611). The four variables present different degrees of relevance in establishing predictions of self-perceived health values by age and by sex. Before the age of 65, chronic diseases have the greatest importance, while IADL limitations are more important in the 65–79 age group. Likewise, ADL limitations are more important for women free of chronic diseases in the 50–64 age group; however, these differences disappear among women in the older group. There is an evident degree of interplay between the objective health indicators of chronic diseases, ADLs, IADLs, and depression when predicting self-perceived health with a high level of accuracy. This interplay implies that self-perceived health summarizes different health conditions depending on age. Gender differences are only evident for the younger age group, whereas construction of self-perceived is the same for women and men among the older group. Therefore, none of these four indicators on its own is able to totally substitute self-perceived health.