In this analysis of nearly 15,000 middle-aged adults from four communities in the United States, we found that participants who self-rated their health as “poor” compared to others their age experienced higher rates of hospitalization and death, even after adjusting for demographic factors and medical history. The risk increased with each level of self-rated health below “excellent,” including even those who reported “good” health at baseline. This association was strengthened after accounting for the correlation between hospitalization and death, suggesting that modeling hospitalization and death separately underestimates the association between poor self-rated health and increased hospitalization.

Several theories exist for why poor self-rated health is independently associated with poor outcomes.22,23 One explanation is that self-rated health captures health information that is not measured or fully accounted for by traditional risk factors.2 For example, behavioral factors such as higher quality diet and greater physical activity improve both objective and subjective health, but measurement is often incomplete or imprecise. There may also be specific deleterious conditions or behaviors undisclosed by the participant, which may be captured in more non-specific measures of health. An individual’s dynamic projection of their future health, based on their own health trajectory or in comparison to others, can also influence the response.22,24 Another explanation is that self-rated health reflects subconscious bodily sensations that provide a direct sense of health unavailable to external observation.2 Lastly, self-rated health may potentially be a self-fulfilling prophesy, reinforcing beneficial behaviors in those with better self-rated health and harmful behaviors in those without.25

Our study provides evidence for the first explanation, but as designed, could not address the others. The unadjusted model estimated a substantial difference in event rates between the self-rated health categories which was attenuated after adjusting for medical conditions. This recapitulates findings from prior studies that suggest self-rated health is an independent risk factor for health outcomes.8,10,11,23,26,27,28,29,30 However, as demonstrated in a study that incorporated number of diseases, physical functioning, and 150 biomarkers, simply accounting for more measures of objective health does not appear to eliminate the prognostic power of self-rated health.31 If self-rated health and objective biomarkers indeed capture wholly different domains of health, using both in conjunction can provide more information than each individually.32

Previous studies of self-rated health in the ARIC study demonstrated that self-rated health declines after a diagnosis of a major disease such as myocardial infarction, stroke, heart failure, or cancer.33,34 Our analysis here focused on self-rated health in midlife as an exposure and tracked the accumulating morbidity and mortality over time as measured by rates of hospitalization and death. Together, these analyses provide support for the bi-directional nature of self-rated health and illness. Individuals with poor self-rated health tend to accumulate more disease, and individuals diagnosed with major diseases tend to have worsening trajectories of self-rated health.34

Our study is novel in that it demonstrates the longitudinal association of self-rated health with hospitalizations over three decades of follow-up time, in a large and diverse cohort of middle-aged adults. To our knowledge, no other study of recurrent hospitalizations and self-rated health has had nearly this length of outcome ascertainment. Additionally, the joint survival model provided flexibility over traditional single event or recurrent event analysis and addressed the conservative bias incurred by non-independent censoring. Because our analysis is computationally reproducible and was conducted with open-source software, future investigators can readily apply these methods to study cause-specific hospitalizations in ARIC and other populations.

Several limitations are present, which involve the measurement of self-rated health and construct validity of hospitalization as a proxy for morbidity. The particular wording of the self-rated health question in ARIC included a phrase that prompted individuals to compare their health to other people their age. This qualifier could have diminished the age-related changes in self-rated health and would complicate the direct comparison of these findings to other studies using a different wording. This study did not use information from repeated measurements of self-rated health, but more complex models using time-dependent covariates or marginal structural models could incorporate the repeated measurements to better derive causal effect estimates.35 Likewise, interventions that directly or indirectly improve self-rated health should be studied to examine whether improvements could decrease risk of future hospitalization or death.

In this study, each hospitalization was considered a single event with equal weight. However, hospital admissions are heterogeneous events with potentially widely differing determinants and effects with respect to self-rated health. Hospitalizations following elective or cosmetic procedures, for example, would be indicators of well-being, rather than illness. Further studies could focus on disease-specific hospitalizations and identify self-rated health trajectories or participant-level characteristics specific to a single disease process.

Our study has clinical implications because it reinforces the prognostic power of self-rated health beyond traditional risk factors, and extends our knowledge to the outcome of all-cause hospitalization. The major finding is that self-rated health status informs an individual’s future risk of hospitalization beyond factors that are conventionally observed. Clinicians can use this simple and convenient measure for individual patients to provide more accurate and personalized risk assessments. Additionally, self-rated health assessments using standardized workflows may be incorporated into routine care to predict future hospitalization rates for clinics, hospitals, and communities.36 Improvements in factors that influence how people self-rate their health could also improve clinical outcomes such as hospitalization risk. Thus, our study provides a rationale for future inquiry into routine assessment of self-rated health or even targeted interventions that may improve self-rated health and its determinants.