This study collected data from the China Health and Retirement Longitudinal Study (CHARLS) baseline dataset, which is operated by the National School of Development at Peking University. ¹ CHARLS has established a high-quality open-access database encompassing various details regarding individuals, families, health status, and socio-economic aspects, including “Health Status and Functioning,” “Cognition,” “Work Retirement,” and “Family Information.” The study cohort comprised individuals aged 45 and above randomly selected from 150 counties or districts and 450 villages or urban areas across 28 provinces, representing the middle-aged and older adult population in China (24). Utilizing the 2018 CHARLS dataset, this study encompassed a total of 20,813 samples. Participants under 45 years old were excluded from the analysis, along with those with insufficient demographic or health data, resulting in a final sample size of 9,766 individuals.

The assessment was conducted based on the records of emotional problems data “Diagnosed with Emotional Problems by a Doctor” in CHARLS. The samples were divided into two cohorts: those affirmed as “Diagnosed with Emotional Problems by a Doctor” and those negated as “Not diagnosed with Emotional Problems by a Doctor.”

The Barthel Index (BI) is a recognized method used to assess the level of ADL. In this study, the BI was employed to evaluate the ADL of all samples. The BI categorizes daily activities into 10 independent elements, including “Feeding,” “Bathing,” “Grooming,” “Dressing,” “Bowel Control,” “Bladder Control,” “Toilet Use,” “Transfers,” “Mobility,” and “Stairs,” each element being assessed with a score (Table 1), reflecting the individual’s ability and proficiency in completing the respective tasks. By assigning scores to each element and summarizing these scores, the BI is derived, ranging from 0 to 100. A higher BI score indicates greater autonomy and proficiency in performing daily life activities (6, 25–27).

The covariates in this study were derived from the CHARLS dataset, encompassing variables such as age, gender (male, female), residence (Central of City/Town, Urban–Rural Integration Zone, Rural, Special Zone), education (ranging from No Formal Education (Illiterate) to Doctoral Degree/Ph.D.), smoking status (Still Smoke, Quit or No, Never Smoked), drinking status (Drink more than once a month, Drink but less than once a month, None), as well as hypertension, diabetes, and dyslipidemia. For age, middle-aged individuals were defined as those aged 45 to below 60 years, while older adult individuals were defined as those aged 60 years and above. Regarding hypertension, as per the 2010 Chinese Hypertension Guidelines, hypertension was diagnosed based on an average systolic blood pressure ≥ 140mmHg, and/or an average diastolic blood pressure ≥ 90mmHg, and/or self-reported use of antihypertensive medications within the past 2 weeks (28). Diabetes diagnosis criteria included a fasting blood glucose level ≥ 7.0 mmol/L or current treatment with antidiabetic medication. Dyslipidemia diagnosis criteria, following the 2016 Chinese Adult Dyslipidemia Guidelines, included total cholesterol level ≥ 240 mg/dL, high-density lipoprotein cholesterol level < 40 mg/dL, low-density lipoprotein cholesterol level > 160 mg/dL, or triglyceride level ≥ 200 mg/dL (29).

In the demographic characteristics analysis, continuous variables such as age and BI were summarized using mean and standard deviation (SD), while categorical variables like gender, residence, and education were represented as counts and percentages. To examine demographic characteristics, independent sample t-tests were employed for continuous variables to discern differences, while chi-square tests were utilized for categorical variables. Correlation analysis involved Pearson correlation tests to assess the relationship between independent variables, covariates, and dependent variables. Subsequently, a comprehensive evaluation of the relationship between ADL and EPs was conducted through hierarchical multiple linear regression analysis. This regression underwent three iterations with distinct models, incorporating dummy variables as references in each variable. Model 1 included three variables: age, gender, and ADL. Model 2 extended Model 1 by adding residence, education, smoking status, and drinking status, resulting in a total of seven variables. Model 3 further expanded upon Model 2 with the inclusion of hypertension, diabetes, and dyslipidemia, encompassing a total of 10 variables.

The study rigorously excluded ineligible subjects, namely those aged 45 and below, as well as individuals lacking adequate demographic or health data. Ultimately, the cohort comprised 9,766 participants, among whom 3,904 were male (39.98%) and 5862 were female (60.02%). Table 2 encapsulates the demographic characteristics of the study cohort, wherein 9539 individuals (97.68%) remained undiagnosed with EPs by medical professionals, while 227 individuals (2.32%) received EP diagnoses. Remarkably, those diagnosed with EPs demonstrated substantially lower average BI scores (65.044, standard deviation = 30.6421) compared to their undiagnosed counterparts (85.128, standard deviation = 17.6386), indicating a significant disparity (p < 0.001, t = 16.573) in ADL between the diagnosed and undiagnosed EP cohorts. Furthermore, significant differences (p < 0.001) were discerned between individuals diagnosed with EPs and those without in terms of gender (p = 0.015), smoking status (p = 0.005), alcohol consumption (p < 0.001), hypertension (p = 0.001), and dyslipidemia (p = 0.011). Analysis of the demographic features of the sample underscores the influence of ADL, gender, education, smoking and alcohol habits, as well as the presence of hypertension or dyslipidemia on the prevalence of EPs.

We employed Pearson correlation analysis to explore the relationship between EPs and both independent and covariate variables (Table 3). The results reveal a significant negative correlation between ADL and EPs (r = −0.165, p < 0.001), indicating an increased susceptibility to EPs among individuals with impaired ADL. Additionally, a slight yet significant positive correlation was observed between gender and EPs (r = 0.025, p = 0.015), suggesting a higher likelihood of females being diagnosed with EPs. Regarding smoking status, a significant positive correlation with EPs was found (r = 0.032, p = 0.002), implying a higher susceptibility to EPs among non-smokers. Similarly, a significant positive correlation was observed between alcohol consumption and EPs (r = 0.044, p < 0.001), indicating that individuals with lower alcohol consumption frequency are more likely to be diagnosed with EPs. Furthermore, hypertension (r = 0.032, p = 0.001) and dyslipidemia (r = 0.026, p = 0.011) exhibited significant positive correlations with EPs, suggesting a greater likelihood of individuals with hypertension or dyslipidemia being diagnosed with EPs.

To delve deeper into the correlation between EPs and ADL, we employed hierarchical multiple linear regression to systematically investigate the association between the independent variables (ADL, age, gender, residence, smoking status, etc.) and the dependent variable EPs across three distinct models (Model 1, Model 2, Model 3; Table 4).

In Model 1, we included ADL, age, and gender (R = 0.170, R² = 0.029, F = 96.764, p < 0.001). We observed a significant negative correlation between ADL and EPs (B = −0.001, β = −0.175, t = −16.777, 95% CI = −0.002, −0.001, p = 0.000), underscoring the heightened susceptibility to EPs among individuals with impaired daily living abilities. In Model 2, additional covariates were incorporated (residence, education, smoking, and drinking status), where ADL continued to exhibit a pronounced negative correlation with EPs (B = −0.001, β = −0.174, t = −16.528, 95% CI = −0.002, −0.001, p = 0.000), maintaining statistical significance. In the final iteration, Model 3 introduced hypertension, dyslipidemia, and diabetes, reaffirming the negative correlation between ADL and EPs (B = −0.002, β = −0.186, t = −16.476, 95% CI = −0.002, −0.001, p = 0.000), with an overall significant fit (R = 0.190, R² = 0.036, F = 15.870, p < 0.001).

All three models demonstrated statistically significant associations between the independent and dependent variables (p < 0.001). The R value increased gradually from 0.170 in Model 1 to 0.190 in Model 3, indicating an enhanced ability to explain the variability of outcomes. Across all models, despite adjustments for covariates, our study consistently emphasized the significant association between declining ADL and increased susceptibility to EPs, while underscoring the robustness of this relationship.

In our investigation, hypertension emerged as a consistent factor exhibiting correlation with EPs across all analyses, suggesting its potential role as a significant confounding factor in this study. To mitigate this confounding effect, we partitioned the entire sample into two cohorts: a hypertension group (Table 5) and a non-hypertension group (Table 6). Through this stratification, we aimed to delve into the relationship between ADL and EPs in detail.

Within the hypertension group, iterative modeling across three iterations continued to reveal a statistically significant negative correlation between ADL and EPs (B = −0.001, β = −0.123, t = −4.784, 95% CI = −0.001, −0.001, p = 0.000). Similarly, within the non-hypertension group, this negative correlation persisted (B = −0.002, β = −0.202, t = −17.450, 95% CI = −0.002, −0.002, p = 0.000). These findings underscore that irrespective of hypertension status and adjustments for covariates, ADL continues to influence susceptibility to EPs (p < 0.001).

Through statistical analysis, we have identified a significant negative correlation between ADL and EPs, indicating that individuals with impaired daily living abilities are more susceptible to EPs. Moreover, employing hierarchical multiple linear regression, adjusting for and iterating covariates, we have established the stability of the negative correlation between ADL and EPs. Consequently, we employed three machine learning algorithms—SVM, DT, and LR—to validate the association between the two.

The results (see Figure 1A) demonstrate that the Area Under the Curve (AUC) values of SVM, DT, and LR exceeded the critical threshold of 0.7, reaching 0.700, 0.742, and 0.711, respectively, confirming the diagnostic capability of these models. Comprehensive evaluation of the three machine learning algorithms using Accuracy, Precision, Recall, and F1 Score metrics (see Figure 1B; Table 7) reveals that DT exhibited outstanding Accuracy (0.677) and Precision (0.735), while SVM demonstrated superiority in Recall (0.600) and F1 Score (0.637). The combined statistical analysis and machine learning algorithm validation affirm a substantial association between ADL and EPs, indicating that impaired daily living abilities increase susceptibility to emotional problems.

To enhance the credibility of the association between ADL and EPs and underscore the predictive role of ADL in EPs, we partitioned the dataset into two groups with distinct features. The first group included all covariates except for the Bath Index (BI), while the second group incorporated BI while retaining the same set of covariates. Machine learning algorithms rigorously validated these two data subsets. The results demonstrate that upon inclusion of BI, the AUC values of SVM and LR models increased from 0.825 and 0.627 to 0.903 and 0.743, respectively, whereas in the DT model, the AUC value decreased from 0.966 to 0.956. Overall, the addition of BI enhanced the predictive AUC values and improved model performance (see Figures 2A,B). Further analysis using Accuracy, Precision, Recall, and F1 Score metrics (see Figures 2C–E; Table 8) revealed that the SVM model exhibited the best performance, with Accuracy increasing from 0.749 to 0.823, Precision from 0.698 to 0.794, and F1 Score from 0.776 to 0.830, demonstrating robust overall performance.

All three machine learning algorithms verified the association between ADL and EPs and affirmed the positive role of BI in overall prediction, enhancing model performance. The validation of machine learning algorithms confirms our finding that impaired daily living abilities increase susceptibility to EPs.

Emotional Problems (EPs) have become a significant challenge affecting the quality of life in middle-aged and older adult populations, garnering increasing attention in public health. Early detection of potential EPs among middle-aged and older adults is crucial. This study explores the potential of Activities of Daily Living (ADL) as predictive indicators for EPs. Using data from the 2018 China Health and Retirement Longitudinal Study (CHARLS) national baseline survey, which includes 9,766 individuals aged 45 and above, we assessed ADL using the Barthel Index (BI). Statistical analyses were conducted to investigate the correlation between ADL and EPs, followed by validation using machine learning algorithms (Support Vector Machine, Decision Tree, and Logistic Regression) to elucidate the underlying relationship between ADL and EPs.