A machine learning-based approach to predict depression in Chinese older adults with subjective cognitive decline: a longitudinal study

Using the fourth and fifth wave data from the CHARLS survey, this study analyzed the demographic variables, health status, and chronic medical history of 1,921 middle-aged and elderly individuals. A predictive model for depression risk in the elderly with SCD was constructed. Figure 8 illustrates the overall conceptual framework of this study, summarizing the accuracy of the predictive models, presenting the research findings, and highlighting the potential clinical applications of the models.

Conceptual framework of discussion in this study.

Among the three machine learning models evaluated, both Boosted XGBoost and RF demonstrated distinct yet complementary advantages in predicting depression risk among older adults with SCD. Notably, education level emerged as the top-ranked predictor in SHAP analysis for both models.

These findings suggest that patients’ education level may also play a crucial role in influencing depression rates among individuals with SCD.This finding is similar with results from a study conducted in Japan¹⁹.Patients with lower educational levels may lack sufficient cognitive reserve to cope with the challenges of cognitive decline, thus increasing the risk of depression. Moreover, individuals with lower education levels may experience exacerbated depressive symptoms due to limited social networks and a lack of health-related knowledge.

Furthermore, other studies have shown that educational level can significantly predict cognitive ability levels²⁰.Theoretically, enhancing the educational attainment of patients from less educated backgrounds could aid in averting cognitive deterioration. Therefore, interventions that promote education for the elderly could contribute to enhancing cognitive function in these patients. In conclusion, relevant institutions should emphasize late-life education for the elderly population. Providing specialized psychological care and guidance through social channels for individuals with low education and income levels is crucial.Additionally, social support through media campaigns, financial assistance, and other means can help reduce the risk of depression for those in need.

SHAP analysis revealed that arthritis ranked as the second most important predictor in the Boosted XGBoost model and third in the RF model, suggesting that arthritis is also a risk factor for DS in individuals with SCD. Chronic pain may contribute to depression risk through inflammatory mechanisms.

As a chronic condition, arthritis causes persistent pain and activity limitations, which over time restrict daily social interactions and engagement. This, in turn, can lead to negative emotions and feelings of loneliness, eventually manifesting as depressive symptoms. A study by Su et al.which utilized machine learning to predict the risk of depression in elderly patients over a two-year period, found that arthritis, along with other factors, plays a significant role in the onset of depression. Chronic disease patients are more prone to depression^21,22, which implies that the presence of arthritis and other such conditions in patients with pre-existing depressive symptoms may exacerbate their mental health issues. Therefore, greater attention should be given to depression screening and treatment among individuals at risk of chronic diseases. Collaboration between government agencies, healthcare organizations, and local communities is crucial for patient rehabilitation, including providing effective disease management strategies, financial support, and psychological assistance. In conclusion, the close association between arthritis, depression, and other chronic diseases necessitates comprehensive disease management strategies. Our goal should be to reduce the depression risk in arthritis patients and improve their overall quality of life through a broad approach that integrates physiological, psychological, and social strategies.

The other three significant predictors were digestive health status, residence location, and sleep duration.There is a significant correlation between gastrointestinal abnormalities and the risk of DS in patients with SCD. One potential explanation is that individuals with digestive issues often experience prolonged discomfort and require long-term medication, leading to a decline in quality of life and negative impacts on mental health. A study conducted in the United States found that individuals reporting gastrointestinal, respiratory, and cardiovascular problems were more likely to exhibit depressive symptoms. This suggests that mental health issues resulting from these three chronic conditions may outweigh those associated with other chronic diseases²³. Furthermore, changes in gastrointestinal metabolites under pathological conditions may influence brain activity through the gut-brain axis²⁴. Maintaining a healthy digestive system helps prevent neuroinflammation, thereby protecting against cognitive decline, and may also be a key factor in slowing the onset and progression of depression.

Therefore, to reduce DS in SCD patients, the following strategies are essential: timely treatment and management of gastrointestinal diseases. Additionally, healthcare professionals should provide comprehensive care plans, including psychological support, lifestyle guidance, and other measures to help patients manage the mental stress associated with chronic diseases. Emphasizing the link between gut and brain health is crucial in reducing the likelihood of depression and neuroinflammation in SCD patients.

The results of this study show that the incidence of DS is higher among rural patients, which is consistent with findings from similar studies²⁵. This may be due to the limited healthcare resources, inadequate social support, and lower living standards in rural areas. Furthermore, the social and cultural environment in rural regions can influence patients’ understanding and perceptions of their illness, potentially worsening mental health issues. Therefore, while evaluating individual patient factors, relevant government agencies should focus on strengthening public services and healthcare systems in rural areas to reduce the likelihood of depression in these populations.

Sleep duration plays a critical role in the development of DS in patients with SCD. Studies have shown that both reduced sleep time and prolonged rest periods can trigger the onset of depressive symptoms^26,27. A longitudinal study lasting more than two years on elderly individuals in community settings found that sleep disturbances are a significant factor contributing to persistent depression²⁸. This may be due to the close relationship between sleep and mood regulation, where poor sleep quality directly affects brain activity, impairs stress adaptation, and increases the risk of depression. Indeed, related studies suggest that prolonged instability in sleep duration can impair cognitive function in older adults, potentially leading to cognitive decline²⁹.

This highlights the strong association between poor sleep habits and cognitive deficits in the elderly³⁰, where sleep disturbances can unknowingly contribute to cognitive decline, often resulting in DS. Neurological and clinical studies have shown that insufficient sleep disrupts the continuous flow of cerebrospinal fluid and interstitial fluid, which may lead to brain function deterioration and permanent cognitive impairment^31,32. Sleep deprivation may also increase tau protein synthesis, reduce brain-derived neurotrophic factors, and stimulate the formation of new neurons and blood vessels³³. Chronic sleep deprivation may exacerbate synaptic plasticity in the hippocampus, potentially leading to cognitive decline³⁴. Furthermore, disrupted sleep patterns may be indicative of daytime brain fatigue, adversely affecting multiple cognitive regions. Therefore, sleep problems in SCD patients may exacerbate their cognitive impairment and emotional distress, leading to the onset of depressive symptoms.

There’s been a notable increase in the interest in adopting virtual reality (VR) in managing senior patients’ conditions.people with physical disabilities can discover beneficial physical endeavors, supported by virtual reality (VR).This action enhances life quality and additionally helps in lowering adverse emotional responses.Hence, investigating the effective incorporation of VR technology into patients with physical impairments might be a crucial research path, targeting the alleviation of patient solitude, enhancement of mental health, and the encouragement of both physical and mental recuperation.

In summary, the Random Forest (RF) model may be preferable for clinical screening applications requiring high sensitivity, whereas the Boosted XGBoost model offers superior stability when considering comprehensive performance metrics. Our findings underscore the importance of implementing early screening for high-risk populations in clinical and community health settings, particularly targeting individuals with lower educational attainment, digestive disorders, or arthritis comorbidities, through targeted health education programs to mitigate depression progression.Future research directions should focus on translating our predictive model into practical applications. Developing a web-based risk assessment tool by deploying our optimal model within a theoretical framework could facilitate widespread implementation. Such digital solutions would enable healthcare providers to conduct efficient depression risk evaluations and design personalized intervention strategies based on individualized risk profiles, ultimately advancing precision medicine approaches for depression prevention and management.

Limitations

This study has several limitations. First, the CHARLS data used in this research is specific to the elderly population in China, meaning that the findings may not be directly applicable to older adults in other countries or regions. Additionally, the effectiveness of machine learning models is intrinsically linked to data quality and its associated factors. Although machine learning techniques were employed for feature selection in this study, some potentially important variables may not have been fully explored. These factors could affect the reliability of the model’s predictions. Future research could improve the model’s predictability by incorporating more potential influencing factors, utilizing advanced machine learning techniques, and combining data from different countries.