Predicting self-rated health (SRH) in elderly hypertensive demographic: The role of Social Determinants of Health (SDOH) explored through Ensemble machine-learning methods

Abstract

Introduction

Both the World Health Organization (WHO) and the Centers for Disease Control and Prevention (CDC) recognize self-rated health (SRH) as a valuable measure for understanding individuals' perceptions of their own health, which can provide insights into their overall well-being and predict future health outcomes [1]. SRH is distinct from most health indicators because it stems from an individual's active cognitive process, not governed by formal, standardized rules or definitions[2, 3]. It can be viewed as a comprehensive assessment, where both subjective experiences and objective health factors are integrated within the individual's perceptual framework [1, 4]. Despite its subjective nature, SRH has been recognized as a reliable predictor of mortality, healthcare utilization, and physical functioning in the adult population [4-6]. The standard approach involves asking individuals to rate their current health with options such as “very good”, “good,” “fair,” “poor,” and “very poor”. This subjective measure helps in understanding how people perceive their own health status, which can be influenced by various factors, including physical health, mental health, and socio-economic conditions as well as the non-medical factors that determine health outcomes also known as SDH [7, 8]. The current study derived its curiosity on how these SDH play a role in SRH in a specific demographic of elderly hypertensive residing in a rural locality. The primary SDH influencing SRH encompass various factors, including individuals' socioeconomic status (SES), educational attainment, occupation, as well as elements of the physical environment and neighborhood.   Various reports highlight the growing recognition that medical care alone is insufficient to significantly improve overall health or reduce health disparities without also addressing the living conditions and environments of individuals. A substantial body of relevant knowledge has accumulated, documenting associations, exploring pathways and biological mechanisms, and providing a previously unavailable scientific foundation for understanding the role of SDOH in health outcomes [9, 10]. The WHO defines SDOH as the conditions in which people are born, grow, live, work, and age. These conditions are shaped by the distribution of socioeconomic status, and resources at global, national, and local levels and further emphasizes that these SDOH are primarily responsible for health inequities the unfair and avoidable differences in health status seen within and between countries [11]. SDOF is also used to refer broadly to any nonmedical factors influencing health, including health-related knowledge, attitudes, beliefs, or behaviors including such as smoking, alcohol consumption and physical activity.  These factors, however, represent only the most immediate determinants in the causal pathways influencing health; they are shaped by more fundamental upstream determinants. While these concepts may intuitively make sense, the causal pathways linking upstream determinants with downstream determinants, and ultimately with health, are typically lengthy and complex, often involving multiple intervening and potentially interacting factors along the way. A study conducted by Almevall et al on SRH in old age indicate that while most participants initially rated their health as "Quite good," changes in SRH over the study period were significantly associated with factors such as age, pain, nutrition, cognition, physical activity, and depressive symptoms. Additionally, SRH at baseline was significantly linked to survival, highlighting its predictive value for longevity in the elderly population [12]. The variation in how individuals rate their health status may indicate the presence of healthcare disparities related to the social determinants of health in their respective localities [13, 14]. Our ability to mitigate these disparities remains constrained due to a limited understanding of the factors that contribute to them. The SDOH have garnered increasing attention from public health and nonprofit agencies. This growing momentum is driven by several factors. Primarily, there is an expanding body of knowledge in the social and biomedical sciences, both from China and internationally, which has deepened the understanding of how social factors impact health [15-17]. This knowledge has subsequently enhanced the scientific credibility of efforts aimed at addressing these determinants. Although research into the relationship between SDOH and SRH status is prevalent, limited studies investigated this relationship in the hypertensive elderly population in rural localities of China. The current study investigated the dynamics of social determinants of health (SDOH) on self-rated health (SRH) status using logistic regression models and compared them with ensemble machine learning (eML) models, including LightGBM, XGBoost, CatBoost, AdaBoost, and Gradient Boosting. Ultimately, we assessed and compared their predictive performances to determine which approach offers greater accuracy in predicting poor SRH.