The associations between patterns of Internet use and depressive symptoms among older adults in China: A latent class analysis

Abstract

Objective

Depression is increasingly becoming a global concern among older adults. Many existing studies have found an association between Internet use and mental health in later life. However, most of this research has relied on a variable-centered approach, which may overlook the heterogeneity in Internet use behaviors. We adopted a person-centered approach to explore distinct patterns of Internet use and their associations with depressive symptoms among older adults.

Methods

Using data from the 2022 China Family Panel Studies, 3975 older adults (aged ≥ 60 years) reported their Internet use and depressive symptoms after excluding samples with missing core values. Latent class analysis (LCA) was employed to analyze the potential classiﬁcation of Internet use.

Results

LCA identified three distinct Internet use profiles: low digital engagement (Class 1, 51.4%), active social engagement (Class 2, 36.1%), and high comprehensive digital engagement (Class 3, 12.5%). Compared to Class 1, both Classes 2 and 3 showed negative associations with depressive symptoms. Heterogeneity analyses revealed that adults under 70 years, males, and rural residents demonstrated stronger associations between Internet use and reduced depressive symptoms.

Conclusion

Internet use has a significant negative impact on depressive symptoms. The results provide an empirical reference for the prevention and intervention of mental disorders in older adults.

Keywords

Internet use depressive symptoms older adults latent class analysis

Introduction

With the rapid growth of global aging, depression has become a significant risk factor, associated with increased disability, higher mortality rates,¹ and a greater prevalence of comorbidities.² Due to the spread of the COVID-19 pandemic and prolonged economic pressures, the incidence of depression has increased rapidly around the world.³ It is particularly important to note that depression not only directly harms the physical health of older adults, but also complicates their care, as it is often associated with other health problems.⁴ In this context, it is extremely important to conduct exhaustive research into the causes of depression and to investigate effective interventions to reduce its prevalence.

With the improvement of Internet accessibility, Internet use rates among older adults continued to grow. Amichai-Hamburger and Ben-Artzi⁵ pointed out that the Internet has evolved from a uniform and single platform into a complex network system offering a variety of services. Schehl et al.⁶ identified three main reasons why older adults participate in online activities, including informational purposes (such as watching videos, reading news, and accessing health information), social purposes (connecting with friends and family), and instrumental purposes (such as banking, shopping, and taking online courses). Several studies confirmed that the mental health of older adults can be improved through digital technology. According to the Amplification Theory of Technology, Internet use enabled older adults to maintain their existing relationships and expand their social networks, which can have a positive impact on their mental health.⁷ Previous studies have shown a correlation between the use of digital technology and reduced depressive symptoms among older adults.^8,9

Current research focuses on how specific digital platforms and applications affect mental health. Zhang and Liang¹⁰ noted that the use of WeChat helps middle-aged and older adults improve their mental health by maintaining close social relationships and building trust between people. Similarly, studies have shown that short video platforms strengthen social bonds, provide social support, and can slow cognitive decline in older adults.^11,12 In addition to social media, Liu and Guo¹³ have found that mobile payment applications significantly improve the physical and mental health of older adults.

Existing studies mainly use factor analysis or classification by functional dimensions to classify Internet usage patterns. Although these methods provide valuable insights, their internal limitations should not be underestimated. They fail to accurately reflect dynamic individual changes in Internet use patterns and may not consider important specific information about individuals. Consequently, the association between Internet use patterns and depression has not yet been systematically explained.

To overcome the above limitations, the study employed the latent class analysis (LCA) method to address the limitations of these issues. As a person-centered approach, LCA can identify subgroups of individuals who possess similar characteristics and ensure that individuals in the same group exhibit comparable patterns in the observed variables.¹⁴ Although this method cannot overcome the limitations of cross-sectional data in terms of drawing causal conclusions, it accurately identified heterogeneous patterns of Internet user behavior in a single sample.¹⁵ The classification of LCA is model-based, and the quality of these classifications can be evaluated using statistical diagnostic tools.¹⁴ Based on the data of the China Family Panel Study (CFPS) in 2022, we explored the relationship between Internet use patterns and depressive symptoms of the older adults.

Methods

Study population and sample

This study extracted data from CFPS. CFPS consists of seven waves with baseline surveys conducted in 2010, and subsequent evaluations conducted every 2 years until 2022. CFPS conducted interviews with family members at different developmental stages (from childhood to old age) in selected households. Data collection involved conducting face-to-face interviews using standardized questionnaires to gather information on sociodemographic factors, lifestyle behaviors, and mental health status. CFPS had been approved by the Biomedical Ethics Review Committee of Peking University (ID: IRB00001052-14010).

The current research is a secondary analysis of prospective data from CFPS. If participants met the following criteria, they were included (1) age 60 years or older; (2) having complete information on the Center for Epidemiological Studies Depression Scale (CES-D) scores in 2022; and (3) having complete information on Internet use. As a result, the final sample size was 3975.

Dependent variable

This study used the CES-D to evaluate depressive symptoms. The CFPS 2022 survey adopted an 8-item condensed version. Participants rated the frequency of each item in the past week on a 4-point scale: 0 = not at all, 1 = several days, 2 = more than half of the days, and 3 = almost every day. The Chinese version of CES-D had been used in previous research, and its reliability and validity have been tested among Chinese populations.^16,17 The total score was from 0 to 24, with higher scores indicating more severe depressive symptoms. The Cronbach's alpha coefficient of the CES-D-8 in this study was 0.781.

Independent variable

We consider six variables to capture respondents’ Internet use patterns: playing online games, online shopping, watching short videos, online studying, using WeChat, and sharing content on WeChat moments. They are all measured as binary variables, with “no” coded as 0 and “yes” coded as 1.

Covariates

Building upon previous studies,^18,19 the following variables were included as covariates in this study: age, sex, education level, marital status, residence, employment status, exercise frequency, self-reported health status, the number of chronic diseases, drinking status, smoking status, life satisfaction, and region.

Statistical analysis

First, baseline characteristics were described as means and standard deviations for normally distributed continuous variables, and frequencies with percentages for categorical variables. Data are summarized according to Internet use categories and compared using the chi-square test and ANOVA, as appropriate.

Second, we employed an LCA approach to identify distinct patterns of Internet use. As a typical unsupervised machine learning method,²⁰ LCA was performed to classify participants into homogenous classes or groups based on a set of observed variables.²¹ The LCA was conducted using R software to fit models with a varying number of latent classes, ranging from one to six. The fit of each model was assessed using three criteria: the Akaike information criterion (AIC), the Bayesian information criterion (BIC), and the entropy. Speciﬁcally, smaller AIC and BIC indicate better model ﬁt. Entropy is used to evaluate the classiﬁcation accuracy of the model, ≥ 0.8 means the classification accuracy is > 90%.²² Additionally, all models should exhibit a minimum profile proportion above 5%, a criterion recommended by Tein et al.²³ to ensure representativeness. Since LCA is an unsupervised method, we selected random forest—a supervised machine learning algorithm with demonstrated effectiveness in classification tasks—to assess whether the data-driven clusters from LCA could be reliably reproduced. The dataset was randomly split into 70% training and 30% testing sets. Five-fold cross-validation was conducted within the training set, and the optimal hyperparameter values were selected to improve model performance.

Third, the variance of depressive symptoms exceeded the mean, making Poisson regression inappropriate. Therefore, we used quasi-Poisson regression, which allows the variance to differ from the mean and provides more reliable estimates.

R version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria) was used to analyze the data and perform descriptive statistics for each variable. LCAs were performed using Mplus version 8.3.

Results

Evaluation and selection of the LCA model

Table 1 presents the LPA fit indices from 1-class to 6-class models. We selected the three-class model for the following reasons: Firstly, it demonstrated optimal performance in terms of AIC (14993.63) and BIC (15119.38), showing relatively low values among all models. Secondly, the three-class model has a robust entropy value (0.816), indicating good classification quality. Thirdly, the three-class model showed a clear distinction with meaningful class proportions (51.42%/36.07%/12.50%), where all classes exceeded 5% of the sample size. In contrast, both the four-class and five-class models included at least one class comprising < 5% of the sample.

Table 1.

Fit statistics results of the LCA.

Model	AIC	BIC	aBIC	Entropy	LMR (p)	BLRT (p)	Class proportion (%)
1	19,320.43	19,358.15	19,339.09	–	–	–	–
2	15,282.35	15,364.09	15,322.78	0.859	<0.001	<0.001	66.36/33.63
3	14,993.63	15,119.38	15,055.83	0.816	<0.001	<0.001	51.42/36.07/12.50
4	14,978.77	15,148.54	15,062.74	0.812	<0.001	0.012	49.15/2.40/36.83/11.59
5	14,987.56	15,201.34	15,093.30	0.597	0.371	0.561	49.10/11.59/2.41/0.05/36.83
6	14,996.98	15,254.77	15,124.50	0.603	0.365	0.601	0.05/3.89/9.13/20.42/2.41/64.07

Note. LCA: latent class analysis; AIC: Akaike information criterion; BIC: Bayesian information criterion; aBIC: adjusted BIC; LMR: Lo–Mendell–Rubin adjusted likelihood ratio test; BLRT: bootstrapped likelihood ratio test.

The number and structure of latent classes may vary with different sample sizes or observed variables. Therefore, we performed cross-validation to validate the latent classes.²⁴ We assessed the performance of the random forest model through 5-fold cross-validation, and the results demonstrated its strong predictive capability, with an accuracy of 0.901, recall of 0.911, precision of 0.921, and an F1 score of 0.912.

Figure 1 exhibits the cluster-specific estimated probabilities of Internet use patterns for the three-class model from LCA. Class 1 (51.4%), named “low digital engagement,” was characterized by low probabilities across all Internet activities. Class 2 (36.1%), termed “active social engagement,” exhibited high probabilities of WeChat use and video consumption. Class 3 (12.5%), labeled “high comprehensive digital engagement,” exhibited high probabilities across all dimensions, particularly in shopping, video watching, and WeChat use, demonstrating comprehensive digital engagement.

Figure 1.

Conditional probabilities of three-class Internet use by latent class. Note. The definition of the classes: Class 1, low digital engagement; Class 2, active social engagement; and Class 3, high comprehensive digital engagement.

Sample characteristics

Based on Table 2, this study analyzed 3975 participants across three Internet use patterns, with Class 1 (low digital engagement) comprising 2044 participants (51.4%), Class 2 (active social engagement) including 1434 participants (36.1%), and Class 3 (high comprehensive digital engagement) containing 497 participants (12.5%).

Table 2.

Baseline characteristics of the total sample and the sample by the different pattern groups.^a

Variables	Overall (n = 3975)	Internet use patterns			p ^b
Variables	Overall (n = 3975)	Class 1(n = 2044)	Class 2(n = 1434)	Class 3(n = 497)	p ^b
Depressive symptoms, mean (SD)^c	5.77 (4.46)	6.28 (4.66)	5.42 (4.33)	4.72 (3.66)	<0.001
Internet use time, mean (SD)^c	0.38 (0.80)	0.05 (0.27)	0.47 (0.77)	1.47 (1.21)	<0.001
Sex					0.511
Female	1873 (47.1)	976 (47.7)	674 (47.0)	223 (44.9)
Male	2102 (52.9)	1068 (52.3)	760 (53.0)	274 (55.1)
Age (years), mean (SD)^c	68.29 (6.01)	70.13 (6.16)	66.46 (5.25)	66.03 (4.91)	<0.001
Marital status					<0.001
Unmarried	657 (16.5)	404 (19.8)	200 (13.9)	53 (10.7)
Married	3318 (83.5)	1640 (80.2)	1234 (86.1)	444 (89.3)
Residence					<0.001
Rural	3239 (81.5)	1825 (89.3)	1119 (78.0)	295 (59.4)
Urban	736 (18.5)	219 (10.7)	315 (22.0)	202 (40.6)
Education, mean (SD)^c	6.50 (4.63)	4.73 (4.27)	7.67 (4.30)	10.39 (3.32)	<0.001
Family income (logarithmic scale) and mean (SD)^c	10.85 (1.35)	10.55 (1.40)	11.04 (1.27)	11.50 (0.99)	<0.001
Current work status					<0.001
Not working	1954 (49.2)	903 (44.2)	741 (51.7)	310 (62.4)
Working	2021 (50.8)	1141 (55.8)	693 (48.3)	187 (37.6)
Exercise frequency per week					<0.001
None	2352 (59.2)	1395 (68.2)	779 (54.3)	178 (35.8)
1–4	347 (8.7)	138 (6.8)	135 (9.4)	74 (14.9)
≥ 5	1276 (32.1)	511 (25.0)	520 (36.3)	245 (49.3)
Self-rated health, mean (SD)^c	2.68 (1.25)	2.60 (1.31)	2.77 (1.21)	2.73 (1.11)	<0.001
Chronic disease					0.058
None	2697 (67.8)	1386 (67.8)	995 (69.4)	316 (63.6)
Yes	1278 (32.2)	658 (32.2)	439 (30.6)	181 (36.4)
Smoking status					0.248
No	2913 (73.3)	1520 (74.4)	1030 (71.8)	363 (73.0)
Yes	1062 (26.7)	524 (25.6)	404 (28.2)	134 (27.0)
Drinking status					0.949
No	3392 (85.3)	1747 (85.5)	1223 (85.3)	422 (84.9)
Yes	583 (14.7)	297 (14.5)	211 (14.7)	75 (15.1)
Life satisfaction, mean (SD)^c	4.22 (0.87)	4.26 (0.89)	4.18 (0.87)	4.15 (0.76)	0.005
Region					0.005
Central	1286 (32.4)	642 (31.4)	475 (33.1)	169 (34.0)
East	1759 (44.3)	890 (43.5)	626 (43.7)	243 (48.9)
West	930 (23.4)	512 (25.0)	333 (23.2)	85 (17.1)

Data are presented as counts (percentages) unless otherwise indicated.

p value determined using χ² test or analysis of variance F-test.

For continuous variables, mean (SD) for each group and significance from analysis of variance F-test are reported.

The definition of the classes: Class 1, low digital engagement; Class 2, active social engagement; and Class 3, high comprehensive digital engagement.

Class 1 (low digital engagement) exhibited the highest depressive symptoms score (6.28), the lowest average Internet use time (0.05 hours), the oldest mean age (70.13 years), the lowest educational attainment (4.73 years), and the highest proportion of rural residence (89.3%). This group also demonstrated challenges in health behaviors, with 68.2% reporting no regular exercise (p < 0.001) and lower self-rated health scores (2.60). In contrast, Class 3 (high comprehensive digital engagement) is characterized by the lowest depressive symptoms score (4.72), the longest Internet usage duration (1.47 hours), the youngest mean age (66.03 years), the highest educational level (10.39 years), the highest proportion of urban residence (40.6%), and the highest proportion of individuals engaging in frequent exercise (≥5 times per week: 49.3%).

Notably, several variables showed no significant differences across groups, including sex (p = 0.511), chronic disease status (p = 0.058), smoking status (p = 0.248), and drinking status (p = 0.949).

Relationship between Internet use and depressive symptoms

Table 3 presents the quasi-Poisson regression results for the association between Internet use patterns and depressive symptoms, with Class 1 (low digital engagement) serves as the reference group. Models 1 and 3 included only independent variables, while Models 2 and 4 incorporated additional control variables. Based on Model 2, both Class 2 (−0.027, p < 0.01) and Class 3 (−0.043, p < 0.001) showed significant negative associations with depressive symptoms compared to Class 1. Model 4 revealed that compared to non-Internet users, 2–3 hours per day had the strongest protective effect against depression (−0.077, p < 0.001), while more than 3 hours showed no significant association.

Table 3.

Association between Internet use patterns and depressive symptoms by quasi-Poisson regression models (reference: Class 1, low digital engagement).

Variables	Model 1	Model 2	Model 3	Model 4
Internet use patterns (Ref. = Class 1)
Class 2	−0.062*** (0.011)	−0.027** (0.011)
Class 3	−0.116*** (0.017)	−0.043*** (0.017)
Internet use time (Ref. = 0)
0–1 hours			−0.073*** (0.012)	−0.024^** (0.012)
1–2 hours			−0.100*** (0.022)	−0.039* (0.020)
2–3 hours			−0.121*** (0.030)	−0.077^*** (0.027)
≥3 hours			−0.130*** (0.048)	−0.069 (0.042)
Sex (Ref. = female)
Male		−0.064^*** (0.011)		−0.063^*** (0.011)
Age		−0.002^*** (0.001)		−0.002^** (0.001)
Marital status (Ref. = unmarried)
Married		−0.088^*** (0.012)		−0.088^*** (0.012)
Residence (Ref. = rural)
Urban		−0.046^*** (0.013)		−0.044^*** (0.013)
Educational level		−0.004^*** (0.001)		−0.004^*** (0.001)
Family income (logarithmic scale)		−0.019^*** (0.003)		−0.019^*** (0.003)
Current work status (Ref.= not working)
Working		0.011 (0.010)		0.011 (0.010)
BMI (Ref.= obese)
Normal		−0.004 (0.016)		−0.004 (0.016)
Exercise frequency per week (Ref. = none)
1–4		−0.011 (0.017)		−0.011 (0.017)
≥ 5		−0.050^*** (0.011)		−0.050^*** (0.011)
Self-rated health		−0.055^*** (0.004)		−0.056^*** (0.004)
Chronic disease (Ref. = none)
Yes		0.93*** (0.14)		0.066^*** (0.010)
Smoking status (Ref. = no)
Yes		0.023* (0.012)		0.022* (0.012)
Drinking status (Ref. = no)
Yes		0.009 (0.014)		0.009 (0.015)
Life satisfaction		−0.075^*** (0.005)		−0.075^*** (0.005)
Region (Ref. = West)
Central		−0.061^*** (0.012)		−0.059^*** (0.012)
East		−0.076^*** (0.012)		−0.074^*** (0.012)

Note. * p < 0.05; ** p < 0.01; *** p < 0.001. AIC: Akaike information criterion. The definition of the classes: Class 2, active social engagement; and Class 3, high comprehensive digital engagement.

In Model 2, being male (−0.064, p < 0.001), older age (−0.002, p < 0.01), being married (−0.088, p < 0.001), urban residence (−0.046, p < 0.001), higher education (−0.004, p < 0.001), higher family income (−0.019, p < 0.001), exercising at least five times per week (−0.050, p < 0.001), better self-rated health (−0.055, p < 0.001), greater life satisfaction (−0.075, p < 0.001), and residence in the central (−0.061, p < 0.001) and eastern regions (−0.076, p < 0.001) were all negatively associated with depressive symptoms. Conversely, chronic disease (0.066, p < 0.001) and smoking (0.023, p < 0.05) showed a significant positive association with depressive symptoms.

The results of the restrictive cubic splines (RCSs) on the association between Internet use time and depressive symptoms are shown in Figure 2. A significant nonlinear association was observed between Internet use time and depressive symptoms (p-overall association = 0.028, p-nonlinearity = 0.034). The level of depressive symptoms declined sharply as Internet use increased from 0 to 2 hours per day, after which the curve gradually plateaued.

Figure 2.

Associations between Internet use time and depressive symptoms using restricted cubic splines. Note. The restricted cubic spine model was adjusted for age, sex, marital status, family income, education level, residential area, working status, chronic disease status, self-rated health, exercise frequency, smoking status, drinking status, life satisfaction, and region.

Heterogeneity analyses

Table 4 presents a heterogeneity analysis, using quasi-Poisson regression, to examine the associations between Internet use patterns and depressive symptoms across different demographic subgroups (age, sex, and residence), with Class 1 (low digital engagement) as the reference group.

Table 4.

Heterogeneity analysis (reference: Class 1, low digital engagement).

Variables	Age		Sex		Residence
Variables	< 70	≥ 70	Female	Male	Rural	Urban
Class 2	−0.03* (0.01)	−0.01 (0.02)	−0.01 (0.02)	−0.05*** (0.02)	−0.02* (0.01)	−0.01 (0.03)
Class 3	−0.05* (0.01)	−0.02 (0.03)	−0.01 (0.03)	−0.07** (0.02)	−0.04* (0.02)	−0.04 (0.03)
N	2727	1248	1873	2102	3239	736

Note. All control variables are controlled. * p < 0.05; ** p < 0.01; *** p < 0.001.

The definition of the classes: Class 2, active social engagement; and Class 3, high comprehensive digital engagement.

Regarding age differences, both Class 2 (−0.03, p < 0.05) and Class 3 (−0.05, p < 0.05) exhibited a significant negative relationship with depressive symptoms only in the < 70 group, while no signiﬁcant association was found in the ≥ 70 group. In terms of sex differences, Class 2 (−0.05, p < 0.001) and Class 3 (−0.07, p < 0.01) showed negative associations with depressive symptoms only among male participants, while no signiﬁcant association was observed in female participants. With respect to residence differences, Class 2 (−0.02, p < 0.05) and Class 3 (−0.04, p < 0.05) were signiﬁcantly negatively associated with depressive symptoms only among rural older adults, while no signiﬁcant association was observed in urban participants.

Discussion

This study investigated the relationship between Internet use patterns and depressive symptoms among older adults using the LCA. Further examination revealed distinct characteristics across three Internet use profiles. Class 1 (low digital engagement, 51.4%) exhibited the highest depressive symptoms scores and was characterized by the lowest average Internet use time, older age, lower education levels, rural residence, and poor self-rated health. Class 2 (active social engagement, 36.1%) demonstrated intermediate characteristics across most measures. Class 3 (high comprehensive digital engagement, 12.5%) showed the lowest depressive symptoms scores, the longest Internet use time, the youngest mean age, the highest educational level, and the highest proportion of urban residence.

Our findings reveal that Internet use patterns have a significant impact on depressive symptoms. Classes 2 and 3 show a stronger negative association with depressive symptoms compared to Class 1. The results align with prior research.^25–27 Additionally, compared to Class 2, Class 3 was significantly associated with lower odds of depressive symptoms. This may be because Internet users engage in a wider variety of online activities, including watching short videos and sharing content on WeChat moments, which can enrich daily life and help alleviate feelings of loneliness. Moreover, as Qu et al.²⁸ observed, using digital platforms like WeChat to maintain contact with family and friends when geographically separated increased older adults’ opportunities for social exchange, reduced social isolation, and ultimately improved their mental health. The relationship between Internet use time and depressive symptoms showed a nonlinear pattern, with the strongest benefits observed in the first 2 hours of daily use. This aligns with findings by Mu et al.,²⁹ who observed that longer durations of Internet use were more likely to be depressed. This may be because excessive Internet use disconnects older adults from fundamental social interactions and activities, thereby causing feelings of depression.

Notably, the heterogeneity analysis revealed important demographic variations in the relationship between Internet use patterns and depressive symptoms. First, the protective effect of Internet use against depressive symptoms was more pronounced among individuals under 70 years old. Chen et al.³⁰ found that Internet use showed a significant positive association with the mental health of older adults aged 60–69, while no similar effects were observed for those aged 70 and older. This maybe because individuals under 70 generally have better cognitive adaptability and learning ability, making them easily acquire digital skills and more comfortable with digital devices.³¹ In contrast, those over 70 may face greater physical and cognitive barriers, which can limit the mental health benefits of Internet use. Second, males showed stronger benefits from Internet use compared to females. This aligns with the findings of Ma,³² which indicate that Internet use is more likely to alleviate depression in males than in females. It is probably because older female adults generally have more social connections and engage in offline activities, such as square dancing and community volunteering,³³ which enhance their well-being and life satisfaction independently of Internet use. Given that older male adults tend to engage less in offline social activities, Internet use may play a critical role by meeting their social and entertainment needs, which in turn can help alleviate depressive symptoms. Third, rural older adults demonstrated stronger benefits from Internet use, while urban residents showed no such significant associations. This is consistent with the findings of Sun et al.,³⁴ which suggest that rural older adults tend to gain greater benefits from Internet use compared to their urban counterparts. This may be because urban residents can easily access various services and social connections through face-to-face interactions, while rural residents often face difficulties due to geographical isolation. Digital technology helps rural older adults overcome these barriers, enabling them to access essential services and social support online, thus effectively alleviating depression.

The present study has several strengths. Utilizing the LCA approach, the study identifies three distinct patterns of Internet use, providing a more comprehensive understanding of older adults’ digital engagement. Another notable advantage is that our study uses the most recently released, large-scale, nationally representative CFPS 2022 data, offering the latest dynamics of the relationship and thereby enhancing the generalizability of the findings.

Despite these strengths, this study has some limitations. Firstly, a cross-sectional design was used, which prevented us from establishing a causal relationship between Internet use and depressive symptoms. Secondly, the study did not include measures of Internet use intensity and motivation—factors that are essential for understanding how and why Internet use may relate to depressive symptoms. Future studies should use longitudinal data with multiple waves of data to provide stronger evidence on the relationship between Internet use and depressive symptoms. Additionally, taking into account measures intensity and motivation of Internet use could elucidate both the level and underlying motives of Internet use, thereby deepening understanding of its impact on mental health.

Conclusion

This study reveals significant relationships between Internet use patterns and depressive symptoms among older Chinese adults. Through LCA, we identified three Internet use patterns (low digital engagement: 51.4%, active social engagement: 36.1%, and high comprehensive digital engagement: 12.5%), finding that Class 3 (high comprehensive digital engagement) has a stronger protective effect against depressive symptoms.

Our heterogeneity analysis further revealed that Internet use benefits vary across demographic groups, with stronger protective effects observed among individuals under 70 years old, males, and rural residents. These findings suggest the need for targeted digital inclusion strategies that consider individual characteristics and circumstances.

Footnotes

Acknowledgements

We express our gratitude to the CFPS research team for providing us with the data utilized in this study.

ORCID iD

Weichao Chen

Ethical approval

CFPS has been approved by the Biomedical Ethics Review Committee of Peking University (ID: IRB00001052-14010).

Consent for publication

Not applicable.

Authors’ contributions

Weichao Chen and Rong Ji: study conceptualization and design; Weichao Chen and Yuqian Sheng: data analysis; Weichao Chen and Rong Ji: writing; Rong Ji and Caiqi Zheng: proofreading and revision; and all the authors: approval of the final version for publication.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research is funded by the Key Project of Hunan Provincial Department of Education (23A0062).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The data of the 2020 China Family Panel Studies (CFPS) is publicly available at accessed on 18 May 2025.

References

Patel

Chisholm

Parikh

, et al. Addressing the burden of mental, neurological, and substance use disorders: key messages from Disease Control Priorities. Lancet 2016; 387: 1672–1685.

Rodda

Walker

Carter

. Depression in older adults. Br Med J 2011; 343: d5219.

Santomauro

Herrera

AMM

Shadid

, et al. Global prevalence and burden of depressive and anxiety disorders in 204 countries and territories in 2020 due to the COVID-19 pandemic. Lancet 2021; 398: 1700–1712.

Marwaha

Palmer

Suppes

, et al. Novel and emerging treatments for major depression. Lancet 2023; 401: 141–153.

Amichai-Hamburger

Ben-Artzi

. The relationship between extraversion & neuroticism & the different uses of the Internet. Comput Human Behav 2000; 16: 441–449.

Schehl

Leukel

Sugumaran

. Understanding differentiated internet use in older adults: a study of informational, social, and instrumental online activities. Comput Human Behav 2019; 97: 222–230.

Long

Casey

Bhar

, et al. Understanding perspectives of older adults on the role of technology in the wider context of their social relationships. Ageing Soc 2024; 44: 1453–1476.

Jing

Jin

Guo

, et al. The association between constant and new Internet use and depressive symptoms among older adults in China: the role of structural social capital. Comput Human Behav 2023; 138: 107480.

Guo

Cheng

, et al. Impact of Internet usage on depression among older adults: comprehensive study. J Med Internet Res 2025; 27: e65399.

10.

Zhang

Liang

. Association between WeChat use and mental health among middle-aged and older adults: a secondary data analysis of the 2020 China family panel studies database. BMJ Open 2023; 13: e073553.

11.

Breck

Dennis

Leedahl

. Implementing reverse mentoring to address social isolation among older adults. J Gerontol Soc Work 2018; 61: 513–525.

12.

Shete

Mahajan

Garkal

. Smart phone addiction and reaction time in geriatric population. Natl J Integr Res Med 2020; 11: 1.

13.

Liu

Guo

. The impact of mobile internet application (APP) use on the physical and mental health of the elderly. Popul Dev 2021; 27: 117–128. https://qikan.cqvip.com/Qikan/Article/Detail?id=7106291766

14.

Muthén

. Integrating person-centered and variable-centered analyses: growth mixture modeling with latent trajectory classes. Alcohol Clin Exp Res 2000; 24: 882–891.

15.

Moreno

Gower

Pham

, et al. Adolescent media use, parent involvement and health outcomes: a latent class analysis approach. Inf Commun Soc 2023; 26: 746–763.

16.

Greenberger

Chen

Tally

, et al. Family, peer, and individual correlates of depressive symptomatology among U.S. and Chinese adolescents. J Consult Clin Psychol 2000; 68: 209–219.

17.

Rankin

Galbraith

. Reliability and validity data for a Chinese translation of the center for epidemiological studies-depression. Psychol Rep 1993; 73: 1291–1298.

18.

Shi

, et al. Adverse childhood experiences and mental health disorder in China: a nationwide study from CHARLS. J Affect Disord 2024; 355: 22–30.

19.

Liu

Qiu

Tang

, et al. Association of adverse childhood experiences with poor health condition among middle-aged and elderly adults in the United States: a nationally retrospective cohort study. Psychiatry Res 2024; 338: 115977.

20.

Hou

Luo

, et al. Associations of four sensitization patterns revealed by latent class analysis with clinical symptoms: a multi-center study of China. EClinicalMedicine 2022; 46: 101349.

21.

Wang

, et al. The association between lifestyles and health conditions and the choice of traditional Chinese medical treatment in China: a latent class analysis. Medicine (Baltimore) 2022; 101: e32422.

22.

Gao

Yao

, et al. Effect of comorbidity classes on survival of patients with gastrointestinal tract cancer. BMC Cancer 2025; 25: 225.

23.

Tein

Coxe

Cham

. Statistical power to detect the correct number of classes in latent profile analysis. Struct Equation Model Multidiscip J 2013; 20: 640–657.

24.

Seymour

Kennedy

Wang

, et al. Derivation, validation, and potential treatment implications of novel clinical phenotypes for sepsis. JAMA 2019; 321: 2003–2017.

25.

Liu

Wang

. Internet use, physical activity and depressive symptoms in older adults: a cross-lagged panel analysis. J Affect Disord 2024; 350: 937–945.

26.

Chiou

. The effects of social media intergenerational program on depressive symptoms, intergenerational relationships, social support, and well-being in older adults. Geriatr Nurs 2023; 52: 31–39.

27.

Yang

Tng

GYQ

Khoo

, et al. Smartphone-use profiles and cognitive and socioemotional outcomes in middle-aged and older adults: a latent profile analysis. Curr Psychol 2024; 43: 3197–3209.

28.

Houser

Zhang

, et al. Association between using social media WeChat and depressive symptoms among middle-aged and older people: findings from a national survey. BMC Geriatr 2022; 22: 51.

29.

Yuan

Liu

. Internet use and depressive symptoms among Chinese older adults: two sides of internet use. Front Public Health 2023; 11: 1149872.

30.

Chen

Yang

Wang

. Internet use, cultural engagement, and multi-dimensional health of older adults: a cross-sectional study in China. Front Public Health 2022; 10: 887840.

31.

Chen

Wang

. The contribution of audiobook use on the mental health of older adults: the mediating effects of social adaptation. Aging Ment Health 2024; 28: 754–761.

32.

. The causal effect of Internet use on rural middle-aged and older adults’ depression: a propensity score matching analysis. Digital Health 2025; 11: 20552076241310041.

33.

Gilmour

. Social participation and the health and well-being of Canadian seniors. Health Rep 2012; 23: 23–32.

34.

Sun

Zhao

Tao

, et al. Examining urban-rural differences in the impact of Internet use on older adults’ depression: evidence from China. Data Sci Manage 2022; 5: 13–20.