Internet use and frailty in Chinese middle-aged and older adults: empirical evidence from CHARLS 2020

Abstract

Background

At present, the impact of Internet use on health remains controversial. This article verifies the impact of the Internet on the health of middle-aged and older adults from a reverse perspective through the relationship between Internet use and frailty. The key research focuses on whether Internet use is associated with reducing the risk of frailty, and how its levels and multiple types affect this risk.

Methods

We conducted our analysis using data from the 2020 wave of China Health and Retirement Longitudinal Study (CHARLS), selecting a total of 10,727 Chinese adults aged 45 years or older. To examine the relationship between Internet use and frailty, we employed logit regression and addressed endogeneity by employing propensity score matching to match Internet-using and non-Internet-using middle-aged and older adults. Additionally, we tested the robustness of our results by replacing the independent variables with WeChat.

Results

The logit regression results indicated a significant negative effect of Internet use, level of internet use and type of internet use on frailty (p<0.001). The propensity score matching test further confirmed that Internet use had a negative effect on frailty (p<0.001). Moreover, when WeChat was used as a replacement for the independent variable, it also exhibited a negative effect on frailty (P<0.001). Lastly, we conducted a heterogeneity test for gender, age and household registration factors, which revealed persistent heterogeneity in Internet use across these three variables.

Conclusion

Our findings suggest that Internet use, level of internet use and type of internet use has a negative effect on frailty among middle-aged and older adults, scilicet promoting geriatric health. It is recommended that middle-aged and older adults be encouraged to access the Internet, receive guidance on how to use electronic devices such as cell phones and tablets, and engage in social and recreational activities through online media. By employing the Internet in a rational manner, the physical and mental health of middle-aged and elderly individuals in China can be enhanced.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-25419-9.

Introduction

With the acceleration of population aging, the health of older adults has become a pressing concern in both developed and developing countries [1, 2]. In China, the proportion of individuals aged 60 and above is projected to exceed 22.3% by 2035, bringing unprecedented challenges to healthcare systems, social security, and long-term care provision [3]. A study conducted in the United Kingdom has shown that individuals with pre-frailty incur twice as much healthcare expenditure annually as their healthy counterparts, while those with moderate or severe frailty have their healthcare costs increased by 3 to 4 times [4]. With advancing age, factors related to physiology, psychology, and society accelerate the progression of frailty in individuals. In turn, frailty will further evolve into a critical public health issue [4]. Therefore, delaying and addressing frailty is imperative for advancing the construction of “Healthy China”.

Among various aging-related health issues, frailty has attracted considerable attention due to its ability to predict adverse health outcomes such as disability, hospitalization, and mortality. Frailty is commonly defined as a multidimensional syndrome characterized by reduced physiological reserve, increased vulnerability to stressors, and decline in multiple domains, including physical functioning, cognitive ability, and psychological well-being [5]. Compared with disease-specific indicators, frailty offers a more comprehensive reflection of the cumulative burden of aging and has been widely used in geriatric health research and clinical screening [6]. Given its potential reversibility, identifying modifiable risk factors of frailty is essential for developing early interventions that promote healthy aging [7]. It is worth noting that the health of adults aged 45 and above begins to gradually experience degenerative changes in their physiological functions, making this a critical intervention window for the prevention of frailty. Therefore, this article focuses on the population aged 45 and above.

With the rapid expansion of digital infrastructure in China, Internet use has become increasingly prevalent among middle-aged and older adults [8]. According to the China Internet Network Information Center(CNNIC), by June 2023, Internet users aged 50 and above accounted for 16.9% of all users [9]. For middle-aged and older adults, Internet use not only facilitates information acquisition and access to health resources but also enables social interaction and emotional support via digital platforms [10]. A growing body of literature suggests that Internet engagement may be positively associated with better health outcomes, such as reduced depressive symptoms, improved cognitive function, and enhanced life satisfaction [11]. However, empirical evidence on the relationship between Internet use and frailty remains limited and inconclusive [12].

On the one hand, Internet use may act as a protective factor against frailty by enhancing middle-aged and older adults’ social connectedness, promoting active lifestyles, and providing access to health-related knowledge [13]. On the other hand, concerns have been raised regarding digital exclusion, sedentary behavior, and the possible negative psychological effects of excessive screen time [14]. Furthermore, few studies have examined the heterogeneous effects of different levels or types of Internet use (e.g., social, entertainment, financial management) on frailty, nor have they addressed the potential self-selection bias inherent in observational studies of digital behavior [15].

To fill these research gaps, this study aims to explore the association between Internet use and frailty among Chinese middle-aged and older adults using nationally representative data from the 2020 wave of the China Health and Retirement Longitudinal Study (CHARLS). Specifically, we examine whether Internet use is associated with a lower risk of frailty, and how the levels and types of Internet activities relate to frailty status. The significance of this study lies in three aspects. First, internet use is the core form of digital progress. It enriches the growing but still limited literature on digital development and frailty by focusing on Internet use as a potentially modifiable factor for healthy aging [16]. Second, unlike previous studies that often emphasized depressive symptoms or cognitive outcomes, this research highlights frailty—a comprehensive indicator. Third, the use of advanced methods such as propensity score matching and robustness checks strengthens the credibility of our findings and provides methodological references for future studies. These contributions make the study particularly relevant in the context of China’s rapid digitalization and aging population, offering evidence-based insights for policies promoting digital inclusion and geriatric health [17].

Methods

To ensure consistency with our research aims, each analytic approach was aligned with a specific research question. Specifically, binary logit regression was used to test the basic association between Internet use and frailty (RQ1), propensity score matching (PSM) was employed to address potential endogeneity and strengthen causal inference (RQ1 & RQ2), robustness checks using WeChat were conducted to verify the stability of the findings (RQ2), and heterogeneity analyses by gender, age and household registration were performed to examine subgroup differences (RQ3).

Research design and sample selection

This study analyzes cross-sectional data from the latest edition of the CHARLS, which is a survey that collects microdata on individuals and households in China’s middle-aged and older adults population aged 45 years or older. The wave 5 round of the survey was conducted in 2020 and provides information on respondents’ basic information, health status, medical services and utilization, medical insurance, income and consumption, and pension insurance. The study population covers 150 county-level units and 450 village-level units, and multi-stage sampling and probability proportional to size (PPS) sampling methods were used to ensure national representativeness of the sample. Individual identification codes were used to match each component (such as individual basic information, health status, and so on.), and the target population was selected as the middle-aged and elderly population over 45 years old. After excluding missing values, duplicate values, and invalid data (“don’t know” and “refused to answer”) for relevant variables, the study included a total of 10,727 middle-aged and older adults. The sample screening process is shown in Fig. 1.

Fig. 1.

Flowchart of the participant selection process

Variable selection

Frailty

In this paper, the Frailty Index (FI) is selected as the outcome variable. According to Rockwood’s definition and measurement standards [18]– [19], the FI mainly contains health defects such as disease, disability, comorbidity, and depression, with at least 30 defective features included [20]. In this paper, 52 health deficits variables were selected, including 15 chronic disease variables, ADL (6), IADL (6), self-assessed health (1), depression scores CESD-10 (10), and cognitive scores (14)(Supplementary Table S1). Based on previous studies, the FI was calculated by summing the scores of all deficits and dividing by the total number of deficits to derive an FI ranging from 0 to 1. The deficit variables were categorized using a defined threshold of 0.25 (Non-frailty or Pre-frail : <0.25; Frailty: 0.25–1.00) [21–26].

Internet use

The chosen explanatory variable for this study is Internet use. In the 2020 CHARLS questionnaire, respondents were asked, “Have you used the Internet in the past month?” Respondents provided a binary response of “yes” or “no”. We categorized the responses into a dichotomous variable: “Yes = 1” and “No = 0”, following previous research. Additionally, the responses to the question “What do you usually do on the Internet? were categorized as follows: chatting, watching news, watching videos, playing games, managing money, and other activities. Building upon the treatment suggested by scholars [27], we further categorized the level of Internet use among the elderly as follows: “No participation = 0”, “Participation in 1 kind = 1”, “Participation in 2 kinds = 2”, “Involved in 3 kinds = 3”, and so on, with scores ranging from 0 to 6. A higher score indicates a higher level of Internet use. Similarly, the types of Internet use were classified into four categories: socialization (chatting), entertainment (reading news, watching videos, playing games), financial management (money management), and other activities. Respondents were awarded one point for participating in each activity, and cumulative scores were used to measure the level of participation in these four categories.

Covariates

This study includes several variables that may be related to frailty. Grossman’s health needs model suggests that socio-economic, behavioral, and environmental factors can influence health levels [28]. Hence, the main control variables include demographic characteristics (gender, age, marriage, place of residence, household registration, level of education), health behaviors (smoking, drinking, fall down, sleep situation), and socio-economic factors (health insurance, pension insurance, employment status). For age, we selected the middle-aged and older age groups above 45 (45–60 and above 60). Regarding marital status, we re-categorized the questionnaire responses so that temporary separation and cohabitation among married individuals were considered as married, while unmarried, divorced, widowed, and separated individuals (not living as a spouse) were categorized as unmarried. Place of residence and household registration were used to reflect urban-rural inequality more comprehensively. China’s unique household registration system is closely linked to personal life aspects, including healthcare, housing, education, and social welfare [29, 30]. Place of residence was categorized as follows: city or town center = 1, urban-rural integration area = 2, rural area = 3, special area = 4. Household registration was coded as agricultural household, Urban household, and Unified Residency. Education was categorized as elementary school and below = 1, junior high school = 2, senior high school = 3, and college and above = 4. Sleep situation was defined as less than 7 h of sleep at night, following the guidelines of the Healthy China Plan. Smoking, alcohol consumption, fall down, medical insurance, pension insurance, and work status were treated as binary variables (no = 0, yes = 1). Specific variable coding can be found in Supplementary Table S2.

Statistical analysis

In this study, the variables used are all categorical variables. To describe the sample frequency and percentage of each variable, we employed the one-way chi-square test, which visually represents the difference in the number of people who use the Internet or not. As the dependent variables in this study are binary categorical variables, we utilized binary Logit regression models to analyze the data. It is important to note that Internet use may be influenced by personal, family, and environmental factors, and there may be a problem of sample self-selection. To address this issue, we employed propensity score matching (PSM) to match subjects. We used various matching techniques, including nearest neighbor matching (1:2), Mahalanobis distance, kernel matching, and radius matching, to identify covariates and score analysis. Finally, to test the robustness of the results, we replaced the core explanatory variables. Additionally, we analyzed the heterogeneity between gender age and household registration based on binary logit regression models to further explore whether there are differences in the effects of the Internet on frailty across different characteristics of middle-aged and older adults.

Results

Sample description

The distribution of the respondents is presented in Table 1. The study included a total of 10,727 middle-aged and older adults, with 7,438 (69.3%) categorized as Pre-frail ones and 3,289 (30.7%) as Frail ones. Among the respondents, 5,786 (53.9%) were middle-aged and older adults who used the Internet and 4,941 (46.1%) did not. Of the respondents, 4,952 (46.2%) were female and 5,775 (53.8%) were male. In terms of age, 5,557 (51.8%) were in the middle-aged and elderly group aged 45 to 60, while 5,170 (48.2%) were over 60. Moreover, 69.2% of the respondents belonged to Agriculture household registration, and 57.8% lived in rural areas. Most respondents had a low level of education (51.6%). In terms of socio-economic aspects, the coverage rate of medical insurance was high (96.8%), as was the coverage rate of pension insurance (86.9%). Furthermore, 69.0% of the elderly were still working. Regarding health behaviors, a relatively small percentage of older adults had experienced fall down (15.3%), smoked (32.5%), drank alcohol (42.1%), or had irregular sleep patterns (39.7%). To explore the subgroup types of Internet use among middle-aged and older adults and the differences in Internet use between urban and rural areas, the supplementary content is presented in Supplementary Table 3, Supplementary Tables 4, and Supplementary Fig. 1.

Table 1.

Descriptive statistics

Variable	Values	Total(n = 10727)/(%)	Non-Internet Use(n = 4941)/(%)	Internet Use(n = 5786)/(%)	P-value
Frailty	Pre-frail	7438	2877	4561	< 0.001
		69.3%	58.2%	78.8%
	Frail	3289	2064	1225
		30.7%	41.8%	21.2%
Age	45 ~ 60	5557	1494	4063	< 0.001
		51.8%	30.2%	70.2%
	60 ~ 95	5170	3447	1723
		48.2%	69.8%	29.8%
Gender	Female	4952	2208	2744	0.005
		46.2%	44.7%	47.4%
	Male	5775	2733	3042
		53.8%	55.3%	52.6%
Marital status	Unmarried	1201	738	463	< 0.001
		11.2%	14.9%	8.0%
	Married	9526	4203	5323
		88.8%	85.1%	92.0%
Household registration	Agriculture	7419	3802	3617	< 0.001
		69.2%	76.9%	62.5%
	Urban	2028	635	1393
		18.9%	12.9%	24.1%
	Unified Residency	1280	504	776
		11.9%	10.2%	13.4%
Residence	Central of City/Town	3101	936	2165	< 0.001
		28.9%	18.9%	37.4%
	Urban-Rural Integration Zone	1413	589	824
		13.2%	11.9%	14.2%
	Rural	6200	3410	2790
		57.8%	69.0%	48.2%
	Special Zone	13	6	7
		0.1%	0.1%	0.1%
Education	Elementary School	5536	3389	2147	< 0.001
		51.6%	68.6%	37.1%
	Middle School	3218	1129	2089
		30.0%	22.8%	36.1%
	High School	1629	389	1240
		15.2%	7.9%	21.4%
	College or above	344	34	310
		3.2%	0.7%	5.4%
Work status	No	3324	1603	1721	0.003
		31.0%	32.4%	29.7%
	Yes	7403	3338	4065
		69.0%	67.6%	70.3%
Medical insurance	No	342	187	155	0.001
		3.2%	3.8%	2.7%
	Yes	10,385	4754	5631
		96.8%	96.2%	97.3%
Pension insurance	No	1406	689	717	0.019
		13.1%	13.9%	12.4%
	Yes	9321	4252	5069
		86.9%	86.1%	87.6%
Fall down	No	9088	4104	4984	< 0.001
		84.7%	83.1%	86.1%
	Yes	1639	837	802
		15.3%	16.9%	13.9%
Sleep situation	Non-enough sleep	4258	2028	2230	0.008
		39.7%	41.0%	38.5%
	Enough sleep	6469	2913	3556
		60.3%	59.0%	61.5%
Smoke	No	7238	3284	3954	0.041
		67.5%	66.5%	68.3%
	Yes	3489	1657	1832
		32.5%	33.5%	31.7%
Drink	No	6212	3140	3072	< 0.001
		57.9%	63.5%	53.1%
	Yes	4515	1801	2714
		42.1%	36.5%	46.9%

Table 3.

The average treatment effect of internet on frailty

Variable	Matched	Treated	Control	Standard deviation(%)	Deviation reduction(%)	T-value	P-value
Work Status	U	0.703	0.676	5.800		3.010	0.003**
	M	0.700	0.709	−2.100	63.600	−1.150	0.250
Education	U	1.950	1.407	69.100		35.290	<0.000***
	M	1.917	1.914	0.400	99.400	0.210	0.836
Gender	U	0.526	0.553	−5.500		−2.840	0.005**
	M	0.527	0.534	−1.400	74.000	−0.760	0.447
Marital status	U	0.920	0.851	21.900		11.420	<0.000***
	M	0.919	0.926	−2.200	90.100	−1.360	0.173
Fall down	U	0.139	0.169	−8.500		−4.420	<0.000***
	M	0.140	0.133	1.900	77.500	1.080	0.282
Drink	U	0.469	0.364	21.300		10.990	<0.000***
	M	0.464	0.459	0.900	95.700	0.480	0.632
Pension Insurance	U	0.876	0.861	4.600		2.380	0.018
	M	0.874	0.874	0.000	99.400	−0.010	0.989
Household registration	U	1.509	1.333	25.700		13.210	<0.000***
	M	1.497	1.458	5.700	78.000	2.940	0.003**
Residence	U	2.110	2.503	−45.600		−23.430	<0.000***
	M	2.128	2.135	−0.800	98.100	−0.420	0.673
Smoke	U	0.317	0.335	−4.000		−2.060	0.039**
	M	0.318	0.320	−0.300	92.500	−0.160	0.872
Sleep situation	U	0.385	0.410	−5.100		−2.640	0.008**
	M	0.388	0.393	−0.900	82.500	−0.480	0.631

Table 4.

PSM analysis of the effects of internet use on the frailty of middle-aged and older adults

Dependent variable: internet	treated	controls	Diff-Att	Standard Deviation	t-value
Nearest neighbor matching (1:2)	0.215	0.386	−0.171***	0.026	−6.66
Mahalanobis metric matching	0.212	0.311	−0.099***	0.033	−2.99
Kernel matching	0.215	0.332	−0.117***	0.011	−10.3
Radius matching	0.215	0.327	−0.112***	0.012	−9.03

(Values in parentheses indicate the standard deviation of the estimated coefficients, and ***, ** and * indicate that the estimated coefficients are significant at the 1%, 5% and 10% levels, respectively)

Comparing those who did not use the Internet with those who did, older adults who used the Internet had higher levels of frailty (P < 0.001), age (P < 0.001), gender (P = 0.005), marriage (P < 0.001), household registration (P < 0.001), residence (P < 0.001), education (P < 0.001), work status (P = 0.003), medical insurance (P = 0.001), pension insurance (P = 0.019), fall down (P < 0.001), sleep situation (P = 0.008), smoking (P = 0.041), and drinking (P < 0.001), which were significantly different.

Baseline regression results

These analyses correspond to RQ1, examining whether Internet use is associated with a lower risk of frailty.Table 2 presents the results of the binary logit regression models (Model 1 through Model 3) to examine the association between Internet use (its levels and types) and frailty among middle-aged and older adults. In Step 1, only explanatory variables were included, while subsequent Steps gradually incorporated basic demographic characteristics, health behaviors, and socio-economic factors of the elderly. Model 2 and Model 3 are analyzed in the same steps as Model 1.The results indicate that Internet use is associated with a decrease in the level of frailty among middle-aged and older adults, with a coefficient of −0.468 (P < 0.001). Specifically, age (β = 0.443, P < 0.001), place of residence (β = 0.200, P < 0.001), and fall down (β = 0.567, P < 0.001) significantly contribute to increased frailty in old age. On the other hand, gender (β=−0.120, P < 0.1), marriage (β=−0.445, P < 0.001), household registration (β=−0.144, P < 0.001), education level (β=−0.492, P < 0.001), sleep situation (β=−0.430, P < 0.001), drinking (β=−0.335, P < 0.001), and work status (β= −0.346, P < 0.001) have a significant negative effect on senile frailty. Furthermore, as more covariates were added, the negative effect of Internet use on frailty remained significant. The level of Internet use (β=−0.172, P < 0.001) and the type of Internet use (β=−0.150, P < 0.001) both had a significant negative impact on frailty. Overall, these findings suggest that Internet use, along with various demographic, health-related, and socio-economic factors, plays a crucial role in reducing frailty among middle-aged and older adults.

Table 2.

Regression analysis of internet use, levels, types, and frailty in middle-aged and older adults

	Model 1				Model 2				Model 3
	Step 1	Step 2	Step 3	Step 4	Step 1	Step 2	Step 3	Step 4	Step 1	Step 2	Step 3	Step 4
Internet use	−0.982***	−0.473***	−0.469***	−0.468***
	(−22.73)	(−9.40)	(−9.21)	(−9.18)
Levels					−0.370***	−0.174***	−0.173***	−0.172***
					(−20.76)	(−8.69)	(−8.54)	(−8.48)
Type									−0.333***	−0.152***	−0.151***	−0.150***
									(−19.82)	(−8.04)	(−7.87)	(−7.84)
Age		0.556***	0.514***	0.443***		0.590***	0.547***	0.478***		0.593***	0.551***	0.480***
		(11.15)	(10.18)	(8.53)		(12.05)	(11.04)	(9.35)		(12.04)	(11.04)	(9.35)
Gender		−0.359***	−0.148*	−0.120*		−0.363***	−0.153**	−0.125*		−0.370***	−0.160**	−0.132*
		(−7.83)	(−2.52)	(−2.04)		(−7.91)	(−2.61)	(−2.12)		(−8.06)	(−2.73)	(−2.25)
Marital status		−0.520***	−0.486***	−0.445***		−0.526***	−0.491***	−0.450***		−0.527***	−0.492***	−0.450***
		(−7.74)	(−7.13)	(−6.47)		(−7.83)	(−7.20)	(−6.55)		(−7.85)	(−7.22)	(−6.56)
Household registration		−0.119**	−0.116**	−0.144***		−0.116**	−0.113**	−0.141***		−0.120**	−0.117**	−0.145***
		(−3.09)	(−2.98)	(−3.64)		(−3.01)	(−2.91)	(−3.56)		(−3.12)	(−3.01)	(−3.66)
Residence		0.137***	0.144***	0.200***		0.141***	0.148***	0.204***		0.146***	0.153***	0.209***
		(4.52)	(4.71)	(6.24)		(4.67)	(4.84)	(6.36)		(4.83)	(5.01)	(6.52)
Education		−0.489***	−0.485***	−0.492***		−0.492***	−0.488***	−0.496***		−0.504***	−0.499***	−0.506***
		(−14.64)	(−14.38)	(−14.50)		(−14.74)	(−14.48)	(−14.61)		(−15.15)	(−14.87)	(−14.99)
fall down			0.567***	0.567***			0.569***	0.570***			0.569***	0.570***
			(9.59)	(9.57)			(9.64)	(9.62)			(9.64)	(9.62)
Sleep situation			−0.425***	−0.430***			−0.424***	−0.429***			−0.423***	−0.428***
			(−9.03)	(−9.12)			(−9.03)	(−9.11)			(−9.01)	(−9.09)
Smoke			−0.0228	−0.0163			−0.0194	−0.0128			−0.0181	−0.0116
			(−0.39)	(−0.28)			(−0.33)	(−0.22)			(−0.31)	(−0.20)
Drink			−0.361***	−0.335***			−0.361***	−0.336***			−0.363***	−0.338***
			(−7.01)	(−6.48)			(−7.01)	(−6.50)			(−7.05)	(−6.53)
Work Status				−0.346***				−0.344***				−0.345***
				(−6.28)				(−6.24)				(−6.27)
Medical Insurance				−0.144				−0.142				−0.144
				(−1.16)				(−1.15)				(−1.16)
Pension Insurance				−0.0463				−0.0420				−0.0456
				(−0.70)				(−0.64)				(−0.69)
Constant	−0.332***	0.395**	0.463**	0.769***	−0.420***	0.330*	0.397**	0.697***	−0.431***	0.336*	0.402**	0.707***
	(−11.51)	(2.84)	(3.25)	(4.11)	(−15.46)	(2.39)	(2.82)	(3.75)	(−15.65)	(2.43)	(2.84)	(3.80)

(Values in parentheses indicate the standard deviation of the estimated coefficients, and ***, ** and * indicate that the estimated coefficients are significant at the 1%, 5% and 10% levels, respectively)

Endogeneity

These analyses correspond to RQ1, examining whether Internet use is associated with a lower risk of frailty. To address issues of endogeneity, we used the nearest neighbor method for matching. This allowed us to select covariates that achieved appropriate balance between Internet-using and non-using middle-aged and older adults, as demonstrated by the great likelihood values of different models [31]– [32]. After first and second order forms of screening, a total of 11 variables were selected, including work status, education, gender, marriage, fall down, drinking, pension insurance, household registration, smoking, and sleep situation. Figure 2 presents the kernel density plot, which shows that before matching, the kernel density curves had larger deviations and the individual characteristics of Internet-using and non-using middle-aged and older adultsdiffered more significantly. However, after PSM matching, the two curves were noticeably closer, indicating a better matching effect.Overall, these results suggest that our matching procedures were effective in reducing potential endogeneity biases and producing balanced samples for our analyses.

Fig. 2.

kernel density plot

Table 3 presents the basic characteristics of the sample after matching, and the results of the balance test show that there were no statistically significant differences between the treatment group and control group in most of the characteristic variables (P > 0.1), except for household registration (P < 0.05). Additionally, the standardized deviation is much less than 20%.Overall, these findings suggest that our matching procedure effectively reduced potential biases between the treatment and control groups, resulting in a better matching effect as indicated by the greatly reduced standardized deviation of all covariates after matching.

(Before matching, PS R2, LR chi2, and the mean and median of the standard devation were 0.122, 1808, 0, and 19.70, respetivel; Ater matching, the coresponding values are 0.001, 13.99, 0.233, 1.500, so the matching effect is good. Values in parentheses indicate the standard deviation of the estimated coefficients, and ***, ** and * indicate that the estimated coefficients are significant at the 1%, 5% and 10% levels, respectively)

Table 4 presents the estimation results of the average treatment effect using four different matching methods: nearest neighbor matching (1:2), Mahalanobis metric matching, kernel matching, and radius matching. The results show that the average treatment effect of Internet use on the control group was still significantly negative (P < 0.001), which supports the conclusion of the benchmark regression. These findings suggest a strong inhibitory effect of Internet use on the frailty of Chinese elderly individuals.

Robustness check

To validate the consistency of our findings (RQ2), we replaced the key explanatory variable with WeChat use and obtained similar results. To test the robustness of our results, we replaced the independent variables in the benchmark regression with the question “Do you use WeChat?” from the 2020 CHARLS questionnaire [33]. Table 5 steps 1 to 4 represent the gradual inclusion of more control variables. The results show that when basic demographic characteristics are added, the effect of using WeChat on frailty decreases slightly but remains significant (P < 0.001). However, when health behaviors and socioeconomic variables are added, the effect stabilizes. These findings suggest that using WeChat is beneficial to the health of middle-aged and older adults and can decrease their frailty.

Table 5.

Impact of Wechat on frailty in middle-aged and older adults

Variable	Model 4
	Step 1	Step 2	Step 3	Step 4
Wechat	−0.992***	−0.483***	−0.476***	−0.476***
	(−22.61)	(−9.48)	(−9.21)	(−9.21)
Demographic	uncontrolled	controlled	controlled	controlled
Health behavior	uncontrolled	uncontrolled	controlled	controlled
socio-economic	uncontrolled	uncontrolled	uncontrolled	controlled
Constant	−0.371***	0.385**	0.449**	0.764***
	(−13.38)	(2.77)	(3.16)	(4.08)

(Values in parentheses indicate the standard deviation of the estimated coefficients, and ***, ** and * indicate that the estimated coefficients are significant at the 1%, 5% and 10% levels, respectively)

Heterogeneity

To validate the consistency of our findings (RQ2), we replaced the key explanatory variable with WeChat use and obtained similar results. In terms of heterogeneity analysis, Table 6 presents regression results for gender, age and household registration (Model 5, Model 6, Model 7). The results show that Internet use had a significant negative effect on frailty for both males and females (P < 0.001). Moreover, the effect of Internet use on decreasing frailty was stronger for men (β=−0.492) compared to women (β=−0.454). Across age subgroups, Internet use had a significant negative effect (P < 0.001) on frailty in both lower and higher age groups. Additionally, the negative effect of Internet use on frailty was stronger for older adults as they aged. Finally, in the Household registration grouping, Internet use has a significant negative impact on the frailty of all groups (P < 0.001). The response in the Unified Residency is the strongest (β=−0.711), which may be attributed to the fact that this region not only enjoys the digital convenience of urban areas but also retains a certain degree of rural social support.

Table 6.

Heterogeneous effects of internet use on frailty in middle-aged and older adults (Gender, age and household registration)

	Model 5		Model 6		Model 7
	Female	Male	45 ~ 60	60 ~ 95	Agriculture	Urban	Unified Residency
Internet use	−0.454***	−0.492***	−0.469***	−0.478***	−0.413***	−0.557***	−0.711***
	(−6.23)	(−6.84)	(−6.44)	(−6.66)	(−6.88)	(−4.38)	(−4.47)
Age	0.468***	0.399***			0.475***	0.305*	0.518**
	(6.29)	(5.38)			(7.86)	(2.19)	(3.03)
Gender			−0.156	−0.120	−0.166*	0.119	−0.0963
			(−1.62)	(−1.58)	(0.81)	(−0.53)
Marital status	−0.420***	−0.484***	−0.661***	−0.363***	−0.447***	−0.318	−0.610***
	(−4.58)	(−4.55)	(−5.18)	(−4.47)	(−5.28)	(−1.94)	(−3.47)
Household registration	−0.157**	−0.121*	−0.146*	−0.136**
	(−2.79)	(−2.16)	(−2.32)	(−2.65)
Residence	0.195***	0.193***	0.140**	0.256***	0.185***	0.166*	0.202*
	(4.27)	(4.24)	(3.02)	(5.75)	(4.64)	(2.06)	(2.16)
Education	−0.643***	−0.387***	−0.625***	−0.392***	−0.561***	−0.391***	−0.323***
	(−11.82)	(−8.89)	(−12.04)	(−8.66)	(−12.33)	(−5.83)	(−3.56)
Fall down	−0.185*	−0.534***	−0.332***	−0.368***	−0.383***	−0.429**	−0.132
	(−2.42)	(−6.64)	(−3.67)	(−5.21)	(−5.86)	(−3.09)	(−0.77)
Sleep situation	−0.371*	0.166	−0.170	−0.118	−0.0670	−0.128	−0.978*
	(−2.26)	(0.85)	(−0.93)	(−0.70)	(−0.47)	(−0.40)	(−2.20)
Smoke	0.0748	−0.154	0.0354	−0.110	−0.0605	0.00374	0.0672
	(0.79)	(−1.67)	(0.38)	(−1.18)	(−0.81)	(0.02)	(0.27)
Drink	0.479***	0.679***	0.496***	0.614***	0.580***	0.672***	0.368*
	(5.88)	(7.80)	(5.53)	(7.72)	(8.36)	(4.58)	(1.96)
Work Status	−0.629***	−0.240***	−0.450***	−0.415***	−0.419***	−0.512***	−0.418**
	(−9.06)	(−3.70)	(−6.36)	(−6.54)	(−7.70)	(−4.03)	(−2.81)
Medical Insurance	0.225	−0.0499	0.0250	−0.0364	−0.0333	0.0900	0.00480
	(1.59)	(−0.78)	(0.26)	(−0.49)	(−0.49)	(0.61)	(0.03)
Pension Insurance	−0.260**	−0.353***	−0.238**	−0.404***	−0.346***	−0.328*	−0.244
	(−2.99)	(−5.49)	(−3.00)	(−5.89)	(−5.70)	(−2.56)	(−1.51)
Constant	1.045***	0.407	1.259***	0.923***	0.724***	0.127	0.883
	(4.03)	(1.41)	(4.42)	(3.78)	(3.50)	(0.31)	(1.58)
N	4952	5775	5557	5170	7419	2028	1280

(Values in parentheses indicate the standard deviation of the estimated coefficients, and ***, ** and * indicate that the estimated coefficients are significant at the 1%, 5% and 10% levels, respectively)

Discussion

In this study, we used the latest wave of CHARLS 2020 data to investigate Internet use among middle-aged and older adults in China and its effect on frailty. Our results suggest that Internet use has a significant negative impact on frailty in middle-aged and older adults. The specific empirical research results indicate that moderate Internet use (β=−0.468, P < 0.001), the level of Internet use (β=−0.172, P < 0.001), and the type of Internet use (β=−0.150, P < 0.001) all exert a significant impact on frailty among middle-aged and older adults. Specifically, Internet use may reduce frailty in middle-aged and older adults. The richer the level and type of Internet use, the lower the degree of frailty in this population. This is consistent with the findings of other scholars’ studies on the elderly population [34–36]. The accessibility and convenience of the Internet facilitate rapid social connections, enabling relationship maintenance at low cost, expanding social networks, and promoting emotional communication, thereby alleviating loneliness and psychological stress [37]. Social platforms such as WeChat can provide emotional outlets for older adults with mobility impairments, reducing their attention to and perception of physical inconveniences. Similarly, diverse Internet use can inhibit the progression of frailty through both physical and psychological pathways, such as accessing health information and maintaining cognitive activity. Loneliness, cognition, psychology, and emotion are critical components of frailty-related variables.

Descriptive statistical analysis reveals that the Internet usage rate among middle - aged and elderly groups is 53.9%. Moreover, the proportion of middle - aged individuals using the Internet (70.2%) is much higher than that of the elderly (29.8%), indicating that younger - aged groups have a stronger willingness to accept and use the Internet. According to the data released by the CNNIC in June 2025, the number of Internet - using elderly people aged 60 and above in China reached 161 million. The scale of Internet - using elderly people in China is expected to maintain a high-growth pattern in the future. However, there is still a certain gap between the Internet usage proportion of the elderly in China and that in developed countries such as the United Kingdom, the United States, and European countries [38]. A possible reason is that the elderly population in China is still mainly composed of rural residents, and there are significant differences in living conditions, economic levels, and social support compared with urban elderly people. In addition, in terms of the level and type of Internet use, the types of Internet use among middle - aged and elderly people are relatively concentrated. Most of them only use mobile phones for chatting, watching videos, and reading news.

It is worth noting that the frailty level of middle-aged and elderly groups mainly focuses on pre-frailty (7,438 people, accounting for 69.3%). This is somewhat similar to the frailty situation of middle - aged and elderly groups described by previous scholars [39]. Most middle - aged and elderly people are in a state of mild frailty, indicating that the vast majority of them have the opportunity to improve or delay the development of frailty through a healthy lifestyle and management.

Furthermore, we explored the heterogeneity caused by gender, age and household registration differences by testing the impact of Internet use on frailty in middle-aged and older adults. The study findings indicate a significant negative association between Internet use and frailty in different subgroups. These results align with previous research that has examined the influence of gender, age and household on health outcomes [40, 41]. Notably, the effect of Internet use on frailty was more pronounced in males, older adults and unified Residency. This may be attributed to females having a stronger inclination towards real-world socialization and a lower reliance on Internet-based social interactions [42]. Additionally, older adults may face more barriers to interpersonal interaction, communication, and mobility, leading to a greater dependence on Internet use and subsequently experiencing greater mental health benefits [40].

Interestingly, our research has revealed that urban Chinese older adults tend to be more active in terms of both the frequency and type of Internet use than their rural counterparts, as supported by relevant surveys [37]. Education and socio-economic status may present barriers to Internet use among the rural elderly. It is noteworthy that smoking exerted no significant effect on frailty in this study. A plausible explanation is that other confounding variables may have moderated the relationship between smoking and frailty. Additionally, existing studies have revealed that the association between smoking and frailty was not significant in the CHARLS database, whereas it showed a significant impact in the HRS (Health and Retirement Study) and ELSA (English Longitudinal Study of Ageing) databases [43]. Thus, the non-significant relationship observed herein may also be associated with the specific database (CHARLS) used.

To validate the consistency of our findings (RQ2), we replaced the key explanatory variable with WeChat use and obtained similar results. Given that internet use among older adults is characterized by self-selection, it is necessary to address the issue of endogeneity by controlling for potential heterogeneous effects and reducing sample selection bias through Propensity Score Matching (PSM) estimation. After conducting various methods of PSM, the results consistently demonstrate a strong correlation between Internet use and a reduction in frailty among older adults. To ensure the reliability of our findings, we replaced the original explanatory variable, Internet use, with WeChat in the benchmark regression model. The results of this study also reveal a negative effect of WeChat use on elderly frailty, which is consistent with the benchmark regression results.

Drawing on the aforementioned study, the following recommendations are put forward: (1) Implement Internet-based intervention strategies tailored to distinct population groups. For individuals with mild frailty, community-based health guidance WeChat groups ought to be established. Through these groups, individuals can consult family physicians or access online healthcare platforms, engage in communication with doctors or pharmacists, and obtain personalized health recommendations [44]. Furthermore, authoritative WeChat official accounts dedicated to health management should be recommended to middle-aged and older adults, facilitating their acquisition of knowledge regarding fall prevention and chronic disease management. For middle-aged and older adults with moderate frailty or mobility impairments, the combination of Internet technology and home care services can be used, including real-time transmission of physiological data to the medical platform through health monitoring equipment, online access to rehabilitation training videos, and remote consultation with family doctors, which can solve the problem of inconvenient medical travel for middle-aged and elderly people. For instance, promotional videos on rehabilitation training can be regularly distributed within communities, with family doctors providing supplementary demonstrations to enhance middle-aged and older adults’ adherence to training protocols. Additionally, middle-aged and older adults can be equipped with health monitoring watches, enabling real-time tracking of their heart rate, sleep patterns, and daily living conditions via a data platform. (2) Construct an “age-adapted digital environment” to bridge the digital divide. Efforts should be made to address the issue of digital exclusion among older adults in rural areas and the advanced-aged population, optimize digital infrastructure in rural or remote regions, and deploy free WiFi services and public digital devices [45]. (3) To help middle-aged and older adults bridge the digital divide, communities should organize regular smartphone training sessions, using demonstrations to teach essential skills like WeChat video calls and mobile payments. Meanwhile, younger family members should proactively guide their parents in daily internet use, maintain frequent online communication to enhance proficiency, and stay attentive to their online habits and health, ensuring safe, confident engagement with the digital world [9].

Despite the valuable insights gained from this study, there are certain limitations that need to be acknowledged. Firstly, due to discrepancies in cognitive scales and Internet use variables between the 2018 and 2020 CHARLS datasets, only cross-sectional data from the most recent period were utilized, restricting our ability to establish a causal relationship between the Internet and frailty. Secondly, although we made efforts to refine the covariates and employ PSM to enhance the robustness of the results, the determinants of frailty are multifaceted, and there may be unobserved latent variables at play. Lastly, the limitations of the survey database prevented us from exploring specific details such as the duration of Internet use. Nonetheless, this study holds practical implications. Firstly, by utilizing emerging frailty variables, we were able to comprehensively assess the health status of middle-aged and older adults across subjective, objective, and other dimensions, surpassing previous studies that focused on specific health indicators. Secondly, considering the ongoing debate among scholars regarding the impact of the Internet on the health of middle-aged and older adults, our research highlights the potential benefits of Internet use for the health and well-being of elderly people in China. Through endogeneity and robustness tests, we further strengthened the evidence supporting the interrelationship between Internet use and frailty. These findings suggest that promoting Internet use can enhance the physical and mental well-being of older adults and foster healthy lifestyles among middle-aged and older populations. Further research is needed in the future to explore the mechanisms underlying these effects and to develop interventions that promote Internet use among elderly individuals, particularly those who are socially isolated or living in rural areas.

Conclusion

By conducting an analysis of Internet use among Chinese middle-aged and older adults, it was observed that a significant portion of this demographic is experiencing pre-frailty, with a smaller percentage classified as severely frail. Additionally, our findings indicate that the level and type of Internet use have a detrimental impact on the frailty of middle-aged and older adults. Specifically, upon substituting the independent variables, it was determined that the utilization of WeChat has a similar effect on the frailty experienced by the elderly. Furthermore, through heterogeneity testing, it was discovered that both gender, age and household registration are significantly correlated with senile frailty. Taken together, these findings address the stated research aims and demonstrate a consistent alignment between research purpose, applied methods, and observed results.

Abbreviations

CHARLS

China Health and Retirement Longitudinal Study

CNNIC

China Internet Network Information Center

RQ

Research question

Authors’ contributions

QW conducted data analysis, conceptualized research, and participated in manuscript writing. YLG collected and organized the materials. YT contributed to the research conceptualization.All authors have reviewed and approved the final version of the manuscript.

Funding

The research was supported by the Guiding Plan Project of Xinjiang Production and Construction Corps(2024ZD066).

Data availability

The study conducted an analysis using datasets that are publicly available. These datasets can be accessed at the following link: http://charls.pku.edu.cn/.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki (for human subjects research) and relevant national/institutional guidelines. All study procedures were ethically reviewed and approved by the Biomedical Ethics Review Board of Peking University. (IRB Approval Number: IRB00001052–11015). The research was conducted in strict accordance with relevant guidelines and regulations. Prior to data collection, written informed consent was obtained from all participants. This ensures full compliance with ethical standards throughout the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.