The team, led by cardiologists at Shanghai East Hospital of Tongji University and Suzhou Municipal Hospital affiliated with Nanjing Medical University, analyzed data from the 2018 wave of the Mexican Health and Aging Study, a nationally representative longitudinal survey. Their analytical cohort included 12,709 adults aged fifty or older, of whom 1,585 reported having physician-diagnosed heart disease and 11,124 served as controls. Rather than treating sleep as a single vague complaint, the researchers decomposed it into four distinct indicators: difficulty falling asleep, waking up during the night, waking up too early, and feeling refreshed upon waking. Each was measured on a frequency scale, allowing the team to ask not just whether a symptom existed but how often it occurred.

The statistical architecture of the study was deliberately multimodal. In the primary analysis, multivariable logistic regression models estimated the total association between each sleep indicator and heart disease, adjusting for a standard panel of demographic and clinical covariates without explicitly accounting for depressive symptoms or fatigue. The results were striking in their specificity. Participants who reported waking up at night most of the time had 70 percent higher odds of heart disease compared with those who never experienced the symptom, with an adjusted odds ratio of 1.70 and a 95 percent confidence interval of 1.41 to 2.03. Even occasional nighttime awakenings carried a measurable signal, with an adjusted odds ratio of 1.28. Difficulty falling asleep most of the time was associated with a 27 percent increase in odds, while the protective side of the sleep spectrum was equally informative: people who reported feeling refreshed upon waking most of the time had 19 percent lower odds of heart disease, an adjusted odds ratio of 0.81. Notably, waking up too early, often assumed to be a cardinal marker of poor sleep in aging, showed no statistically significant association at all.

This dissociation between different sleep symptoms is one of the study’s most interesting technical contributions. It suggests that the cardiovascular relevance of disturbed sleep is not a diffuse property of sleeping badly in general, but is concentrated in specific phenomena, particularly nocturnal awakenings. From a pathophysiological standpoint, repeated nighttime waking is plausibly linked to cardiovascular stress through several mechanisms, including intermittent activation of the sympathetic nervous system, surges in blood pressure and heart rate during arousals, and the well-documented association between sleep fragmentation and conditions such as sleep apnea, in which each awakening may reflect an underlying episode of oxygen desaturation. The inverse association with feeling refreshed upon waking fits the same picture, since restorative sleep is the functional opposite of fragmented sleep.

The researchers were careful to stress-test their findings. In a sensitivity analysis that incorporated survey weights to better reflect the national population, the associations became less precise and were no longer statistically significant, a caution the authors acknowledge openly. However, the nighttime-awakening association proved robust in a different and arguably more important test: when the team removed the three items of the CESD-9 depression scale that overlap conceptually with sleep and fatigue, the association persisted. This matters because depression scales often contain sleep-related questions, and failing to account for that overlap can artificially inflate the apparent link between sleep symptoms and disease. By demonstrating that the signal survives the removal of these overlapping items, the authors strengthened the case that nighttime awakening is not merely a proxy for depressed mood.

The machine learning component of the study took a deliberately exploratory form. The team trained a random forest classifier, an ensemble method that builds many decision trees on random subsets of the data and aggregates their votes, to see which variables carried the most information for distinguishing participants with heart disease from those without. Performance was assessed with ten-fold cross-validation, in which the data are split into ten parts and the model is trained and tested ten times to avoid overoptimistic estimates. The resulting area under the receiver operating characteristic curve, or AUC, was 0.72, with a precision-recall AUC of 0.24. At the Youden-optimal threshold, the point that maximizes the sum of sensitivity and specificity, both metrics reached 0.67. The authors are explicit that these numbers reflect internal classification performance in an exploratory exercise, not a clinically validated prediction model, and the modest precision-recall value reflects the inherent difficulty of the task given the low prevalence of heart disease in the cohort.

What the random forest lacked in predictive power, it compensated for in interpretability. The most informative variables for classification were hypertension, age, fatigue, the activities of daily living score, the CESD-9 depression score, and history of stroke. This hierarchy is telling: it places fatigue, a subjective and often dismissed symptom, alongside established cardiovascular risk factors. It also hints that the relationship between sleep and heart disease in older adults may run partly through a cluster of psychological and functional states, depression and exhaustion, rather than through sleep mechanics alone.

That hypothesis was tested more directly using exploratory path analysis, a statistical technique that maps out hypothesized directional pathways among variables rather than estimating a single aggregate effect. The path analysis indicated that depressive symptoms and fatigue were associated both with the sleep disturbance indicators and with heart disease status, suggesting a web of interconnections in which poor sleep, low mood, and exhaustion travel together and jointly relate to cardiac outcomes. The authors are careful to frame these indirect associations as descriptive rather than as evidence of temporal mediation, an appropriate caution given that the cross-sectional design captures a single moment in time and cannot establish which came first.

This limitation is fundamental and the authors do not shy away from it. Because the study is cross-sectional, all findings should be interpreted as associative rather than causal. It is entirely possible that subclinical heart disease, through mechanisms such as nocturnal dyspnea, orthopnea, or medication effects, causes the sleep disturbances rather than the reverse, or that shared upstream factors such as chronic inflammation, obesity, or sleep apnea drive both. Longitudinal follow-up within aging cohorts, and ultimately interventional studies targeting sleep quality, would be needed to establish directionality. The MHAS cohort, with its repeated waves, offers a natural platform for such future work.

Even with these caveats, the study carries practical weight. Sleep complaints are among the most common reasons older adults consult physicians, yet they are rarely treated as cardiovascular information. The finding that the frequency of nighttime awakenings, a question that takes seconds to ask, discriminates heart disease cases with an effect size comparable to established risk factors suggests that sleep history deserves a more systematic place in cardiovascular risk assessment for aging populations. The authors’ methodological transparency, from publishing analysis scripts that reproduce every estimate from the public-use MHAS data to explicitly labeling the machine learning component as exploratory, sets a standard for how observational sleep-cardiology research should be reported. For now, the message for clinicians and patients alike is measured but concrete: when an older patient says they keep waking up at night, that fragment of the medical history may be worth more than it seems.

Subject of Research: Associations between sleep disturbance indicators and heart disease in older adults analyzed with machine learning and exploratory path analysis

Article Title: Multimodal analysis of sleep disturbances and heart disease in older adults: a machine learning and exploratory path analysis approach

Article References: Wei, T., Wan, T., Huang, S., Xu, H., Sun, Y., Zhang, Q., & Li, Y. (2026). Multimodal analysis of sleep disturbances and heart disease in older adults: a machine learning and exploratory path analysis approach. BMC Public Health. https://doi.org/10.1186/s12889-026-29774-z

Image Credits: AI Generated

DOI: 10.1186/s12889-026-29774-z

Keywords: sleep disturbance, heart disease, older adults, nighttime awakenings, machine learning, random forest, path analysis, depressive symptoms, fatigue, Mexican Health and Aging Study, cardiovascular risk, CESD-9