Construction and validation of a risk prediction model for depressive symptoms in a middle-aged and elderly arthritis population

Abstract

BACKGROUND

Arthritis is a prevalent and debilitating condition that affects a significant proportion of middle-aged and older adults worldwide. Characterized by chronic pain, inflammation, and joint dysfunction, arthritis can severely impact physical function, quality of life, and mental health. The overall burden of arthritis is further compounded in this population due to its frequent association with depression. As the global population both the prevalence and severity of arthritis are anticipated to increase.

AIM

To investigate depressive symptoms in the middle-aged and elderly arthritic population in China, a risk prediction model was constructed, and its effectiveness was validated.

METHODS

Using the China Health and Retirement Longitudinal Study 2018 data on middle-aged and elderly arthritic individuals, the population was randomly divided into a training set (n = 4349) and a validation set (n = 1862) at a 7:3 ratio. Based on 10-fold cross-validation, least absolute shrinkage and selection regression was used to screen the model for the best predictor variables. Logistic regression was used to construct the nomogram model. Subject receiver operating characteristic and calibration curves were used to determine model differentiation and accuracy. Decision curve analysis was used to assess the net clinical benefit.

RESULTS

The prevalence of depressive symptoms in the middle-aged and elderly arthritis population in China was 47.1%, multifactorial logistic regression analyses revealed that gender, age, number of chronic diseases, number of pain sites, nighttime sleep time, education, audiological status, health status, and place of residence were all predictors of depressive symptoms. The area under the curve values for the training and validation sets were 0.740 (95% confidence interval: 0.726-0.755) and 0.731 (95% confidence interval: 0.709-0.754), respectively, indicating good model differentiation. The calibration curves demonstrated good prediction accuracy, and the decision curve analysis curves demonstrated good clinical utility.

CONCLUSION

The risk prediction model developed in this study has strong predictive performance and is useful for screening and assessing depression symptoms in middle-aged and elderly arthritis patients.

Key Words: Middle-aged and elderly individuals; Arthritis; Depression symptoms; Current status; Influencing factors; Risk prediction models

Core Tip: This study focuses on developing and validating a risk prediction model specifically for depressive symptoms in middle-aged and elderly individuals with arthritis utilizing data from the China Health and Retirement Longitudinal Study national survey. By addressing the unique challenges of this population, the model aims to improve early detection and management of depression, ultimately enhancing patient outcomes and guiding more effective interventions.

INTRODUCTION

Arthritis, a degenerative autoimmune disorder linked to ageing and classified as osteoarthritis or rheumatoid arthritis, is recognized as a prevalent and significant cause of disability. It greatly impacts a patient’s quality of life and can result in a loss of independence[1-4]. According to the China Health and Retirement Longitudinal Study (CHARLS) national survey, arthritis in middle-aged and older individuals in China is 31.4%, and the occurrence increases with age[1]. According to the Chronic Disease Center of China, life expectancy is projected to rise consistently and is anticipated to reach 81.3 years by 2035[5]. Additionally, the prevalence of arthritis is predicted to increase. Depression symptoms frequently coexist with chronic diseases, making it a prevalent comorbidity[6]. Research has indicated a global increase in the prevalence of depression, with a notable 16.4% increase in cases between 2010 and 2021. By 2021, the number of individuals affected by depression exceeded 332 million, highlighting its substantial impact on global health and disease burden[7-10]. Depression is associated with several factors, including the presence of multiple chronic conditions, duration of nighttime sleep, level of literacy, and chronic pain[11,12]. The objective of this study was to evaluate the present condition of depression in the Chinese middle-aged and older population with arthritis by utilizing data from the CHARLS survey. Additionally, this study attempted to determine the components linked to depression symptoms in this population and develop a risk prediction model.

MATERIALS AND METHODS

Data sources

The data used in this study were obtained from the 4^th round of national survey data conducted by CHARLS in 2018. These data can be accessed by the public at https://charls.pku.edu.cn/.

Crowd selection

The 2018 CHARLS nationwide survey data recruited a total of 19816 individuals. This study excluded people under the age of 45, years and participants without arthritis (those who answered no to the questionnaire about whether a doctor has ever told them that they have arthritis or rheumatism), participants without depression and demographic data, and participants with missing values for critical variables. We successfully included a total of 6211 individuals in our study. The CHARLS research project obtained approval from the Biomedical Ethics Committee of Peking University (IRB00001052-11015), and all participants were required to provide their signatures on an informed consent form.

Data extraction

Assessment of depression: Depressive symptoms were assessed using the CESD10, a widely used tool for measuring mental health. The CESD10 consists of 10 items, each scored on a scale from 0 to 3. A score of 0 indicates “rarely or not at all”, 1 indicates “not very much”, 2 indicates “sometimes or half the time”, and 3 indicates “most of the time”. The total score for the 10 items ranges from 0 to 30. A response indicating occasional or moderate occurrence was assigned a score of 2, whereas a response indicating frequent occurrence was assigned a score of 3. The cumulative score for the 10 questions was 30. Scores equal to or greater than 10 were classified as indicative of depression symptoms, whereas scores less than 10 were considered within the normal range[13].

Independent variables: The social factors of the population under consideration encompassed age and gender (female/male). The socioeconomic factors included literacy level (below primary/elementary/middle/high school and above), marital status categorized as married and unmarried (including divorced/widowed/single), place of residence (rural/urban), retirement status (yes/no), health insurance coverage (yes/no), and old age insurance coverage (yes/no). Health-related factors included health status (rated as very good, good, fair, poor, or very poor), activities of daily living (ADL) categorized as either obstacle-free or impaired, medical history of chronic diseases, hearing ability (rated as excellent, very good, good, fair, or bad), exercise habits (yes or no), and the number of pain sites (counted based on the number of reported sites). The lifestyle characteristics were smoking status (yes or no), current alcohol use (yes or no), and overnight sleep length.

Statistical analyses

R software (version 4.4.1) was used to perform the statistical analyses in this investigation. Numerical variables were characterized by utilizing the mean ± SD and the maximum and minimum values. The analysis of categorical variables involved using either the χ² test or Fisher’s exact test to compare different groups. The results are presented as the number of cases and the corresponding percentage. The characteristic variables were found using the least absolute shrinkage and selection (LASSO) regression technique, and the variables were chosen using 10-fold cross-validation. The investigators utilized multivariate logistic regression analysis to construct a prognostic model and generate nomograms. The model’s discriminative power was evaluated via the receiver operating characteristic (ROC) curve and the area under the curve. Calibration curves were employed to evaluate the agreement level between the predicted probabilities and observed probabilities. Decision curve analysis was utilized to evaluate the practical usefulness in a clinical setting, with P < 0.05 considered as a statistically significant difference.

RESULTS

Demographics and clinical characteristics

A total of 6211 middle-aged and elderly individuals diagnosed with arthritis were included in this study. The sample comprised 2628 (42.3%) males and 3583 (57.7%) females. Among them, 5219 (84%) were married and 992 (16%) were unmarried. Additionally, 2116 (34.1%) resided in urban areas, whereas 4095 (65.9%) lived in rural areas. In terms of education, 2943 (47.4%) had finished elementary school, 1683 (27.1%) had gone to middle school, 1067 (17.2%) had finished high school, and 518 (8%) had furthered their education beyond high school. A total of 27.1% of the students were enrolled in primary schools, whereas 17.2% were enrolled in secondary schools. Additionally, 8.3% of the students were enrolled in high schools and above. Table 1 displays the fundamental demographic and clinical characteristics.

Table 1 Fundamental demographic and clinical characteristics, n (%).

	Non-depression (n = 3288)	Depression (n = 2923)	Total (N = 6211)	P value
Gender				< 0.001
Male	1595 (48.5)	1033 (35.3)	2628 (42.3)
Female	1693 (51.5)	1890 (64.7)	3583 (57.7)
Marital status				< 0.001
Married	2847 (86.6)	2372 (81.1)	5219 (84.0)
Unmarried	441 (13.4)	551 (18.9)	992 (16.0)
Residence				< 0.001
Urban	1287 (39.1)	829 (28.4)	2116 (34.1)
Rural	2001 (60.9)	2094 (71.6)	4095 (65.9)
Age (year), mean ± SD	63.3 ± 9.38	63.5 ± 9.03	63.4 ± 9.22
Median (min, max)	63.0 (45.0, 95.0)	64.0 (45.0, 94.0)	64.0 (45.0, 95.0)
Education				< 0.001
Below primary	1349 (41.0)	1594 (54.5)	2943 (47.4)
Secondary school	943 (28.7)	740 (25.3)	1683 (27.1)
Middle school	642 (19.5)	425 (14.5)	1067 (17.2)
High school and above	354 (10.8)	164 (5.6)	518 (8.3)
Nighttime sleep duration (hour)				< 0.001
mean ± SD	6.28 ± 1.90	5.42 ± 2.17	5.88 ± 2.07
Median (min, max)	6.00 (0, 15.0)	5.00 (0, 20.0)	6.00 (0, 20.0)
Health status				< 0.001
Excellent	308 (9.4)	94 (3.2)	402 (6.5)
Rather or relatively good	382 (11.6)	136 (4.7)	518 (8.3)
General	1786 (54.3)	1238 (42.4)	3024 (48.7)
Mediocre	668 (20.3)	1065 (36.4)	1733 (27.9)
Bad	144 (4.4)	390 (13.3)	534 (8.6)
ADL-6				< 0.001
Not-difficult	2708 (82.4)	1809 (61.9)	4517 (72.7)
Difficult	580 (17.6)	1114 (38.1)	1694 (27.3)
No. of chronic diseases	2.96 (1.72)	3.61 (1.98)	3.26 (1.88)	< 0.001
Median (min, max)	3.00 (1.00, 12.0)	3.00 (1.00, 13.0)	3.00 (1.00, 13.0)
Hypertensive				< 0.001
Yes	1338 (40.7)	1338 (45.8)	2676 (43.1)
No	1950 (59.3)	1585 (54.2)	3635 (56.9)
Diabetes				< 0.001
Yes	434 (13.2)	496 (17.0)	930 (15.0)
No	2854 (86.8)	2427 (83.0)	5281 (85.0)
Tobacco use				< 0.001
Yes	886 (26.9)	647 (22.1)	1533 (24.7)
No	2402 (73.1)	2276 (77.9)	4678 (75.3)
Alcohol abuse				< 0.001
Yes	1178 (35.8)	804 (27.5)	1982 (31.9)
No	2110 (64.2)	2119 (72.5)	4229 (68.1)
Retired				< 0.001
Yes	548 (16.7)	259 (8.9)	807 (13.0)
No	2740 (83.3)	2664 (91.1)	5404 (87.0)
Audiological				< 0.001
Excellent	55 (1.7)	30 (1.0)	85 (1.4)
Very good	413 (12.6)	213 (7.3)	626 (10.1)
Good	586 (17.8)	371 (12.7)	957 (15.4)
General	1783 (54.2)	1709 (58.5)	3492 (56.2)
Bad	451 (13.7)	600 (20.5)	1051 (16.9)
Exercise				0.084
Yes	3040 (92.5)	2657 (90.9)	5697 (91.7)
No	248 (7.5)	266 (9.1)	514 (8.3)
Medical insurance				0.201
Yes	3211 (97.7)	2833 (96.9)	6044 (97.3)
No	77 (2.3)	90 (3.1)	167 (2.7)
Old-age insurance				0.961
Yes	2135 (64.9)	1888 (64.6)	4023 (64.8)
No	1153 (35.1)	1035 (35.4)	2188 (35.2)
No. of painful sites	3.42 (3.70)	5.89 (4.42)	4.58 (4.24)	< 0.001
Median (min, max)	2.00 (0, 15.0)	5.00 (0, 15.0)	4.00 (0, 15.0)

P values < 0.05 were considered statistically significant. ADL: Activities of daily living.

Prevalence of depressive symptoms in the middle-aged and elderly arthritic population

In this study, the incidence of depressive symptoms among arthritis patients was 47.1% (2923 out of 6211). Of the 2923 patients diagnosed with depressive symptoms, 64.7% (1890/2923) were female and 35.3% (1033/2923) were male.

Predictive model construction

The 6211 middle-aged and elderly arthritis patients were divided at a 7:3 ratio, with 70% (n = 4349) randomly assigned to the training set and the remaining 30% (n = 1862) randomly assigned to the validation set, we used a random seed to ensure that the sampling was random and repeatable. This study included a total of 28 independent variables, with depressive symptoms being the dependent variable. The feature variables were screened using the LASSO regression method, and the variables were further screened via 10-fold cross-validation (Figure 1A and B). The independent variables screened by LASSO regression were subjected to multifactorial logistic regression, and the results revealed that gender, number of pain sites, number of comorbidities, and place of residence were risk factors for depressive symptoms in middle-aged and elderly arthritis populations (Table 2). A nomogram-predicting model was constructed (Figure 2), and the instance of variable assignment are shown in Table 3.

Figure 1 Least absolute shrinkage and selection regression model. A: The least absolute shrinkage and selection regression model uses log(lambda) sequences with nonzero coefficients produced by the optimal lambda to create coefficient distributions for the selection of clinical and demographic characteristics; B: The log(lambda) sequence was used to generate the coefficient distribution, and vertical lines were created at the values determined by 10-fold cross-validation. Lambda.1-se was chosen as the ideal lambda in this study, resulting in 16 features with non-zero coefficients.

Figure 2 A risk prediction model for depression in a middle-aged and elderly arthritic population. ADL: Activities of daily living.

Table 2 Results of multifactor logistic regression analysis.

Variant	β	SE	Wald χ2	P value	OR	95%CI
Constant	1.130	0.350	3.23	0.001	3.094	1.559-6.142
Gender (female)	0.292	0.072	4.07	< 0.001	1.339	1.163-1.542
Residence (rural)	0.330	0.074	4.46	< 0.001	1.391	1.203-1.608
Health status (excellent)	-1.182	0.202	-5.84	< 0.001	0.307	0.206-0.456
Health status (bad)	-0.392	0.140	-2.81	0.005	0.675	0.514-0.888
Health status (rather or relatively good)	-1.167	0.183	-6.36	< 0.001	0.311	0.217-0.446
Health status (mediocre)	-0.819	0.138	-5.92	< 0.001	0.441	0.336-0.578
ADL-6 (difficult)	0.511	0.082	6.26	< 0.001	1.667	1.420-1.955
No. of chronic diseases	0.041	0.020	2.04	0.041	1.042	1.002-1.083
No. of painful sites	0.073	0.010	8.04	< 0.001	1.076	1.057-1.095
Nighttime sleep duration	-0.150	0.017	-8.75	< 0.001	0.082	0.833-0.891
Audiological status (very good)	-0.398	0.142	-2.80	0.005	0.672	0.508-0.888
Audiological status (good)	-0.286	0.122	-2.35	0.020	0.751	0.591-0.954
Audiological status (excellent)	-0.156	0.315	-0.49	0.622	0.856	0.461-1.589
Audiological status (general)	0.005	0.095	0.05	0.962	1.001	0.834-1.210
Education (below the primary)	0.758	0.141	5.39	< 0.001	2.133	1.620-2.810
Education (secondary school)	0.502	0.142	3.53	0.004	1.653	1.250-2.184
Education (middle school)	0.364	0.151	2.42	0.016	1.440	1.071-1.933
Age	-0.018	0.004	-4.66	< 0.001	0.982	0.974-0.989

P values < 0.05 were considered statistically significant. SE: Standard Error; OR: Odds ratio; CI: Confidence interval; ADL: Activities of daily living.

Table 3 Instance of variable assignment.

Variable	Assignment mode
Health status	Bad = 0. Mediocre = 1. General = 2. Rather or relatively good = 3. Excellent = 4
ADL-6	Not-difficult = 0. Difficult = 1
Education	Below primary = 0. Secondary school = 1. Middle school = 2. High school and above = 3
Audiological status	General = 0. Bad = 1. Excellent = 2. Good = 3. Very good = 4
No. of chronic diseases	Measured value
No. of painful sites	Measured value
Nighttime sleep duration	Measured value
Age	Measured value

ADL: Activities of daily living.

Predictive model validation

The area under the curve is a statistical metric to assess a classifier’s performance. This figure shows the likelihood that a randomly selected positive sample will outperform a randomly chosen negative sample. Bootstrap 1000 iterations were used for internal validation and ROC curves were plotted. The training model’s area under the ROC curve was 0.740 (95% confidence interval: 0.726-0.755), while the validation set’s area under the ROC curve was 0.731 (95% confidence interval: 0.709-0.754), indicating strong model differentiation (Figure 3A and B).

Figure 3 Training set, and validation set receiver operating characteristic curves. A: Receiver operating characteristic curves for the training set; B: Receiver operating characteristic curves for the validation set. ROC: Receiver operating characteristic; CI: Confidence interval; AUC: Area under the curve.

Predictive model calibration

The calibration curves (Figure 4A and B), revealed that the model had a good fit on both the training and validation sets, and that the model had good prediction accuracy.

Figure 4 Calibration curves for the training set, and the validation set depression risk model. A: Calibration curves for the training set; B: Calibration curves for the validation set.

Evaluation of clinical validity

The clinical usefulness of the model was evaluated via decision curve analysis. The findings indicate that the predictive model provides a greater net benefit than the net benefit of the two extreme strategies, suggesting that the line-plot model has a better net benefit and predictive accuracy (Figure 5A and B).

Figure 5 Decision curve analysis curves for the training set, and the validation set depression risk model. A: Decision curve analysis curves for the training set; B: Decision curve analysis curves for the validation set.

DISCUSSION

This study exploited data from the 2018 CHARLS national survey to investigate the frequency of depressive symptoms among middle-aged and elderly adults in China who have arthritis. The findings revealed that 47.1% (2923 out of 6211) of this population experienced depressive symptoms. Additionally, Duan and Wang[14] reported that the prevalence of depressive symptoms among the general middle-aged and elderly population in China was 35.0% based on the CHARLS survey data. Additionally, the prevalence of depressive symptoms among middle-aged and elderly individuals with arthritis was greater than that observed in the general middle-aged and elderly population in China. The mental well-being of individuals with arthritis in China requires immediate and thorough attention.

The current study found that both gender and age were important predictors of depressive symptoms in middle-aged and elderly individuals with arthritis. The findings revealed that women had a greater likelihood of experiencing depression than males. Among the total population of middle-aged and older individuals suffering from arthritis and experiencing depression, 64.7% (1890 out of 2923) were women, whereas 35.3% (1033 out of 2923) were men. Previous studies have indicated that women experience depression at a rate that is twice as that of men[15,16]. This current study aligns with the findings of those previous investigations. This phenomenon can be attributed to various factors. First, from a medical standpoint, arthritis is two to three times more prevalent in women than in men[17]. This suggests that a significant number of women in the middle-aged and elderly population have arthritis, leading to a relatively high proportion of depression among them. Furthermore, this study included a sample of individuals in the middle-aged and elderly age range. Research has indicated that women tend to experience a greater number of depressive symptoms both before and following changes in sex hormones[18,19]. Additionally, women are more susceptible to depression than males. This study also revealed that the likelihood of experiencing depression decreases as individuals grow older and is more prevalent during the menopausal period, which aligns with other research[20,21]. From a physiological standpoint, brain chemicals and neurotransmitters can change as a person ages. These changes may play a role in regulating emotional reactions and potentially decreasing the likelihood of experiencing depression. Furthermore, prior research has revealed that the typical age at which arthritis first appears is 45 years, with the highest occurrence at 50 years[22]. Additionally, individuals in the middle stage of their lives, known as the middle-life transition, may have increased susceptibility to developing depression[16].

This study revealed that among middle-aged and senior individuals with arthritis, the number of locations where they experienced pain was a separate factor that increased the likelihood of depression. Furthermore, individuals with more painful joints were more prone to experiencing depression, which aligns with prior research findings[23-27]. There is a complex interactive relationship between pain and depression that involves many aspects, such as physiological, psychological, and social factors. Pain can disrupt the balance of neurotransmitters like serotonin and dopamine, which are essential for mood regulation, and both factors may share overlapping pathological pathways that could influence each other[11,28,29]. Repeated episodes of arthritis pain can lead to emotional burdens and depressive symptoms, whereas pain may limit an individual’s physical activity and social communication. Furthermore, this study revealed a link between sleep disorders and depression risk[30,31], and past reports suggest a reciprocal relationship between pain and sleep[32]. Future research could delve deeper into the interplay between pain, sleep, and depression. ADL dysfunction was also a risk factor for depression in this study, which is consistent with the findings of previous studies[33]. Pain and ADL dysfunction increases the risk of depression and social isolation. Future studies are needed to explore the complex biological and psychological mechanisms involved in the relationship between pain and depression and develop new research strategies. Simultaneously, we should prioritize symptom management and health education to improve the cognitive and emotional pain management of middle-aged and elderly arthritis patients, aid in self-management, and enhance their self-care abilities.

Our predictive modelling revealed that self-rated health and the number of comorbidities associated with chronic diseases were influential factors for depression. The poorer the self-rated health status and the greater the number of comorbidities, the greater the risk of depression. Studies have shown that individuals with multiple chronic conditions are more likely to develop depression, further emphasizing the prevalence and severity of depression among chronic conditions[6]. In addition, the degree of comorbidity is significantly associated with self-rated health, with several studies indicating poorer self-rated health in chronic disease comorbid populations[34,35]. The presence of multiple chronic diseases may lead to greater economic and psychological burdens for patients, and appropriate strategies for the prevention and control of chronic disease should be adopted to assist middle-aged and elderly individuals in managing comorbidities and improving their health status and quality of life. The complex relationships between different chronic diseases and depression should be further explored in the future.

As individuals age, their physical functioning declines. Research indicates that over 60% of individuals over the age of 60 years experience hearing impairment[36]. Previous studies have also reported a greater prevalence of depression among individuals with hearing impairments[36-39]. Our predictive modelling further supports the notion that individuals with hearing impairments are more prone to experiencing depressive symptoms. The location of residence and level of literacy were significant determinants of depression among middle-aged and elderly individuals with arthritis. Individuals residing in urban areas and possessing a high level of literacy had a reduced likelihood of experiencing depression. We can attribute this phenomenon to the greater availability of healthcare resources and higher economic levels in metropolitan areas. Individuals with a high level of education typically possess a more extensive network of social connections and support networks, which can potentially alleviate the psychological strain experienced by patients. Furthermore, individuals with elevated literacy levels typically possess enhanced cognitive capacities and demonstrate greater receptiveness towards standardized medical interventions for their ailments, and a commitment to preserving their mental well-being. The survey data in this study indicate a pressing requirement to enhance mental health assistance for rural and low-literacy communities. This study not only reveals the mental health challenges faced by middle-aged and older individuals with arthritis but also provides valuable data to inform future policy development, considering the cultural and societal adaptations needed for diverse groups.

CONCLUSION

This study employed the 2018 CHARLS data to investigate the frequency of depression among the Chinese middle-aged and senior populations with arthritis. The results indicated that the rate of depression in this group was 47.1%, surpassing the that of the general middle-aged and elderly population. Gender, age, number of chronic diseases, number of pain sites, nighttime sleep duration, educational attainment, audiological status, overall health status, and place of residence all significantly affect depression in middle-aged and elderly arthritis patients. The risk prediction model developed using these influencing factors demonstrates excellent predictive efficacy, making it valuable for identifying and evaluating depression symptoms in middle-aged and older individuals with arthritis. This study has several limitations. First, it is important to note that the study is based on the 2018 CHARLS data, meaning that the results reflect outcomes from a specific time duration. The characteristics and health status of the study population may have changed over time. Therefore, it is necessary to carefully evaluate the long-term applicability of the results. Furthermore, future follow-up should continuously enhance the study’s model. Although this study identified several influencing factors, additional research is needed to explore the impact of these various factors on depression. In this work, we constructed a prediction model using Chinese data. We might perform external validation to ensure the model’s reliability across other countries and locations.