Digital divide and healthcare service utilization among older adults with chronic liver disease in China: a nationwide cross-sectional study

Abstract

The digital divide may exacerbate healthcare disparities among vulnerable populations. However, evidence regarding its impact on healthcare utilization among chronic liver disease (CLD) patients remains limited, particularly using comprehensive multidimensional assessments. This study aimed to investigate the association between multidimensional digital divide and healthcare service utilization among Chinese adults with CLD. We analyzed data from 405 CLD patients (aged ≥ 45 years) from the China Health and Retirement Longitudinal Study (CHARLS) Wave 5 (2020). Digital divide was assessed through a comprehensive framework incorporating access, usage, and application dimensions, measured by internet use, device diversity, usage breadth, WeChat use, and mobile payment capability. Healthcare service utilization was defined as any hospitalization in the past year or outpatient visit in the past month. Multivariable logistic regression models, subgroup analyses, and sensitivity analyses including alternative outcome definitions, quartile analysis, and propensity score matching were performed. Among participants (median age 58.2 years, 51.6% male), 53.1% experienced digital divide. Healthcare service utilization was significantly lower among those with digital divide (hospitalization: 13.5% vs. 27.4%, P = 0.001; outpatient visits: 21.9% vs. 56.8%, P < 0.001). After full adjustment, standardized digital divide dimensions showed strong associations with healthcare utilization: access dimension (OR = 2.89, 95%CI: 2.18–3.84), usage dimension (OR = 3.13, 95%CI: 2.34–4.17), application dimension (OR = 1.67, 95%CI: 1.33–2.10), and overall score (OR = 3.63, 95%CI: 2.66–4.94). Significant age interactions were observed (P < 0.001). Sensitivity analyses confirmed robustness, with propensity-matched analysis showing OR = 2.63 (95%CI: 1.53–3.75). Multidimensional digital divide substantially impacts healthcare service utilization among CLD patients. Targeted interventions addressing digital disparities are needed to ensure equitable healthcare access in the digital era.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-34523-0.

Introduction

Chronic liver disease (CLD) has emerged as a major global health challenge, with cirrhosis now ranking as the 11th leading cause of death worldwide and accounting for approximately 2 million deaths annually¹. The burden is particularly pronounced in China, where the rapidly aging population and high prevalence of viral hepatitis, alcohol consumption, and metabolic disorders have created a substantial public health crisis². Among Chinese adults aged 45 years and older, the prevalence of CLD continues to rise, with affected individuals facing significant challenges in accessing and utilizing healthcare services, leading to increased morbidity and mortality². The intersection of CLD management with the digital transformation of healthcare systems presents both unprecedented opportunities and challenges, particularly as digital health technologies become increasingly integral to healthcare delivery in the post-pandemic era³.

Recent studies have demonstrated the potential of digital health interventions to improve outcomes in CLD patients, including enhanced self-management, better medication adherence, and reduced hospital readmissions^4,5. The digital divide, conceptualized as disparities in digital access, literacy, and assimilation, affects approximately 26% of older adults in developed countries and even higher proportions in developing nations like China, where only 6.51% of adults over 45 years reported internet use in recent national surveys⁶. Previous research has shown that individuals with higher digital literacy demonstrate better health-seeking behaviors, improved communication with healthcare providers, and more effective disease self-management⁷. However, the relationship between digital divide and healthcare utilization specifically among CLD patients remains poorly understood, despite evidence suggesting that this population faces unique challenges including cognitive impairment from hepatic encephalopathy, physical frailty, and lower health literacy that may exacerbate digital exclusion^8,9.

Despite growing recognition of the digital divide’s impact on health outcomes, significant knowledge gaps persist regarding its specific effects on vulnerable populations with chronic conditions. Most existing studies have focused on general populations or specific conditions like cardiovascular disease and diabetes, with limited attention to CLD patients who face distinctive challenges including disease-related cognitive dysfunction and socioeconomic vulnerabilities^10,11. Furthermore, previous research has predominantly employed simplistic binary measures of digital access, failing to capture the multidimensional nature of digital engagement that encompasses device diversity, usage breadth, and application capabilities¹². The absence of comprehensive frameworks for assessing digital divide in CLD populations has hindered the development of targeted interventions, particularly in resource-limited settings where the burden of liver disease is highest¹³.

This study aimed to investigate the association between multidimensional digital divide and healthcare service utilization among Chinese adults with CLD using data from the China Health and Retirement Longitudinal Study (CHARLS). We developed a comprehensive digital divide assessment framework incorporating access, usage, and application dimensions to examine their independent and combined effects on healthcare utilization patterns. By employing both cross-sectional and subgroup analyses across different demographic and geographic strata, we sought to identify vulnerable populations most affected by digital disparities and provide evidence-based recommendations for reducing healthcare inequities in the digital age.

Materials and methods

Study population and data source

This study utilized data from the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative longitudinal survey designed to assess the health and socioeconomic status of Chinese residents aged 45 years and older^14,15. The CHARLS employs a multistage probability sampling strategy to ensure representativeness of the Chinese middle-aged and elderly population. We analyzed data from the baseline survey (Wave 1) conducted in 2011 through the fifth follow-up (Wave 5) conducted in 2020. The CHARLS database was last updated in November 2023, with Wave 5 data collected in September 2020. The CHARLS study protocol was approved by the Biomedical Ethics Review Committee of Peking University (approval number: IRB00001052-11015), and all participants provided written informed consent.

The study population comprised individuals aged ≥ 45 years who reported having chronic liver disease at baseline (Wave 1). We excluded participants who died during follow-up or were lost to follow-up at Wave 5. Additionally, participants with missing data on digital technology use (mobile internet access and WeChat usage) or healthcare service utilization (outpatient visits and hospitalization) were excluded. The participant selection process is illustrated in Fig. 1.

Fig. 1.

Study Flow Diagram. Flow diagram showing the selection process of study participants from the China Health and Retirement Longitudinal Study (CHARLS). The diagram illustrates the application of inclusion and exclusion criteria, resulting in the final analytical sample of patients with chronic liver disease.

Measures

Digital divide assessment

Our assessment of the digital divide was informed by established theoretical frameworks conceptualizing digital disparities as multidimensional constructs. While Gong et al. proposed a three-tier model comprising digital access, digital literacy, and digital assimilation, the absence of validated digital health literacy measures in CHARLS necessitated a pragmatic adaptation. Drawing on prior CHARLS-based studies^{3,6,7,12,16,17}, we operationalized digital divide across three dimensions: access (basic connectivity and device availability), usage (breadth of online activities, serving as a behavioral proxy for digital skills), and application (adoption of integrated digital platforms that facilitate daily functions, approximating digital assimilation).

Operational definition and measurement

We developed a composite digital divide score incorporating five indicators that reflect the theoretical dimensions:

1. Internet access: Assessed by “Have you used the Internet in the past month?” (yes = 1, no = 0).

2. Device diversity: Number of devices used for Internet access (desktop computer, laptop, tablet, smartphone), scored 0–4, serving as a proxy for digital skills.

3. Usage breadth: Types of online activities engaged in (chatting, news reading, video watching, gaming, financial management), scored 0–5.

4. Social media engagement: WeChat usage (yes = 1, no = 0). WeChat (Weixin in Chinese) is China’s dominant multi-purpose platform combining instant messaging, social media, mobile payment, and service integration, with over 1.2 billion monthly active users.

5. Digital payment capability: Ability to use mobile payment platforms (e.g., Alipay, WeChat Pay) (yes = 1, no = 0), indicating financial digital service adoption.

Dimensional construction

Based on the theoretical framework, we aggregated these indicators into three dimensions:

• Access dimension: Combining Internet use (weight: 0.5) and device diversity (weight: 0.5), standardized to 0–1.

• Usage dimension: Based on usage breadth score, standardized to 0–1.

• Application dimension: Combining WeChat use (weight: 0.5) and mobile payment capability (weight: 0.5), standardized to 0–1.

The composite digital divide score was calculated as the sum of all five indicators (range: 0–12), with higher scores indicating lower digital divide. For analytical purposes, we also calculated standardized scores (0–1) to facilitate interpretation and comparison.

Classification strategy

For the primary analysis, participants were dichotomized using the median composite score:

• Digital divide group: Composite score below median.

• No digital divide group: Composite score at or above median.

This dichotomization balanced statistical power with clinical interpretability and policy relevance.

Sensitivity analyses

To assess the robustness of our findings, we employed multiple alternative classification strategies:

1. Quartile-based categorization: Participants were classified into severe (Q1), moderate (Q2), mild (Q3), and no (Q4) digital divide groups to examine dose-response relationships.

2. Dimension-specific analyses: Examining independent associations of each dimension with healthcare utilization.

Healthcare service utilization

Healthcare service utilization, the primary outcome variable, was assessed through a comprehensive approach capturing both inpatient and outpatient service use:

Primary outcome

Healthcare service utilization was defined as a binary variable indicating any healthcare contact within the specified timeframe. Participants were classified as healthcare users if they reported either:

• At least one hospitalization in the past 12 months, OR.

• At least one outpatient visit in the past month.

Specific measures

1. Inpatient service utilization: Based on the question “Have you been hospitalized in the past year?” (yes = 1, no = 0).

2. Outpatient service utilization: Based on the question “Have you visited a medical institution for outpatient services in the past month?” (yes = 1, no = 0).

This composite measure captures both acute care needs (typically reflected in hospitalizations) and routine healthcare management (reflected in outpatient visits), providing a comprehensive assessment of healthcare engagement among chronic liver disease patients.

Covariates

We included several potential confounding variables in our analyses:

1. Sociodemographic factors: age (continuous), gender (male/female), educational level (below primary school, primary school, middle school, high school and above), household income (continuous, yuan), and residence (urban/rural).

2. Health-related factors: depressive symptoms measured by the 10-item Center for Epidemiologic Studies Depression Scale (CESD-10, range 0–30, higher scores indicating more severe depressive symptoms), body mass index (BMI, kg/m²), self-rated health (very poor, poor, fair, good, very good), and number of chronic diseases(0–13 additional chronic conditions).

3. Healthcare access factors: pension insurance status (yes/no), urban employee medical insurance (yes/no), and new rural cooperative medical insurance (yes/no).

4. Cognitive factors: self-rated memory (poor, fair, good, very good, excellent) and cognitive ability (range 0–21, higher scores indicating better cognitive function).

Data preprocessing

Variables with missing rates exceeding 20% were excluded from analysis. We detected outliers in continuous variables using the 3-sigma rule (mean ± 3 standard deviations) and replaced them with missing values. Multiple imputation using the mice package in R (version 4.3.0) was performed to address remaining missing values, with five imputed datasets generated using appropriate imputation methods for different variable types (predictive mean matching for continuous variables, logistic regression for binary variables, and polytomous regression for categorical variables).

Statistical analysis

Descriptive analysis

Survey weights were not applied in this analysis as our focus was on examining associations within the chronic liver disease patient population rather than generating nationally representative prevalence estimates. Baseline characteristics were summarized using median (interquartile range, IQR) for continuous variables and frequency (percentage) for categorical variables. Comparisons between groups were performed using the Wilcoxon rank-sum test for continuous variables and Pearson’s chi-square test or Fisher’s exact test for categorical variables, as appropriate. All continuous variables were tested for normality using the Shapiro-Wilk test, and non-normally distributed variables were analyzed using non-parametric methods.

Main analysis

We employed multivariable logistic regression models to examine the association between digital divide dimensions and healthcare service utilization. Three progressive models were constructed for each dimension:

• Model 1: Unadjusted (crude) model.

• Model 2: Adjusted for demographic variables (age, sex, and education).

• Model 3: Fully adjusted for demographic and health-related variables (age, sex, education, household income, rural/urban residence, depressive symptoms [CESD-10], and body mass index [BMI])

All dimensional scores were standardized (mean = 0, SD = 1) to facilitate comparison across dimensions. Results were reported as odds ratios (ORs) with 95% confidence intervals (CIs).

Subgroup and interaction analyses

Subgroup analyses were conducted using stratified logistic regression models with full covariate adjustment (Model 3 specifications), excluding the stratification variable from the adjustment set when applicable. Stratification was performed by residence (urban/rural), sex (male/female), age (< 60/≥60 years), and education level (below middle school/middle school and above). Interaction effects were tested by including multiplicative interaction terms between digital divide dimensions and stratification factors in the fully adjusted models, with P-values < 0.05 considered statistically significant for effect modification.

Sensitivity analyses

To assess the robustness of our findings, we conducted several sensitivity analyses:

1. Alternative outcome definitions: We separately analyzed hospitalization and outpatient service utilization as individual outcomes.

2. Dose-response analysis: Participants were categorized into quartiles based on their digital divide scores to examine potential dose-response relationships.

3. Propensity score matching (PSM): We matched participants with and without digital divide (1:1 nearest neighbor matching) based on age, sex, education, rural/urban residence, depressive symptoms, BMI, and household income. The caliper width was set at 0.2 times the standard deviation of the propensity score logit.

Software and reporting

All analyses were performed using R version 4.3.0. The following R packages were utilized: dplyr (v1.1.2) for data manipulation, tableone (v0.13.2) for baseline characteristics tables, ggplot2 (v3.4.2) and forestploter (v1.1.0) for data visualization, MatchIt (v4.5.3) for propensity score matching, and sf (v1.0-13) with geojsonsf (v2.0.3) for geographic mapping. Geographic boundaries were obtained from the GeoMapData_CN repository (https://github.com/GeoMapData/GeoMapData_CN). Statistical significance was set at a two-sided P-value < 0.05.

Results

Study population characteristics

From the initial 17,708 participants in the first wave of CHARLS (2011), 17,200 met the age criterion of ≥ 45 years. Among these, 985 (5.7%) reported having chronic liver disease at baseline. After excluding 99 participants who died during follow-up, 86 lost to follow-up at Wave 5, and 2 with incomplete data on key variables, the final analytical sample comprised 405 chronic liver disease patients (Fig. 1). Missing data were minimal across most variables, with less than 5% missingness for primary outcomes and exposures (Supplementary Table 1).

The median age of participants was 58.2 years (IQR: 51.5–65.8), with 51.6% being male. The majority resided in rural areas (76.0%) and had limited formal education, with 42.2% having less than primary school education. Based on the median split of the digital divide composite score, 215 participants (53.1%) were classified as having digital divide, while 190 (46.9%) had no digital divide (Table 1). Participants with digital divide were significantly older (median age 62.1 vs. 54.1 years, P < 0.001), had lower educational attainment (P = 0.016), and demonstrated poorer cognitive ability scores (11.0 vs. 12.5, P = 0.003). As expected, all digital technology indicators showed marked differences between groups: internet use (38.1% vs. 100%, P < 0.001), WeChat use (59.5% vs. 96.3%, P < 0.001), and mobile payment capability (37.2% vs. 72.1%, P < 0.001). Healthcare service utilization also differed significantly between groups, with those having digital divide showing lower rates of both hospitalization (13.5% vs. 27.4%, P = 0.001) and outpatient visits (21.9% vs. 56.8%, P < 0.001). Baseline characteristics stratified by healthcare service utilization status are presented in Supplementary Table 2.

Table 1.

Baseline characteristics of chronic liver disease patients by digital divide Status.

Characteristic	Level	Overall (n = 405)	No Digital Divide (n = 190)	Digital Divide Present (n = 215	P value
Age (years)		58.17[51.49,65.79]	54.06[47.97,60.65]	62.12[55.91,68.76]	< 0.001
Sex	Female	196(48.4)	95(50.0)	101(47.0)	0.611
	Male	209(51.6)	95(50.0)	114(53.0)
Education Level	Less than primary school	171(42.2)	65(34.2)	106(49.3)	0.016
	Primary school	87(21.5)	43(22.6)	44(20.5)
	Middle school	94(23.2)	53(27.9)	41(19.1)
	High school and above	53(13.1)	29(15.3)	24(11.2)
Household Income (yuan)		16818.18[4166.67,41856.06]	19308.37[5364.58,50000.00]	13812.95[2936.93,33537.88]	0.067
Residence	Urban	97(24.0)	50(26.3)	47(21.9)	0.351
	Rural	308(76.0)	140(73.7)	168(78.1)
Depressive Symptoms Score		9.00[5.00,15.00]	9.00[4.00,14.00]	10.00[5.00,16.00]	0.105
Height (m)		1.58[1.53,1.65]	1.58[1.53,1.65]	1.58[1.53,1.64]	0.598
Weight (kg)		57.90[51.20,66.60]	59.70[51.78,67.88]	56.70[50.85,64.90]	0.042
Waist Circumference (cm)		85.20[77.00,92.00]	86.00[77.25,92.83]	84.60[77.00,91.80]	0.464
Body Mass Index (kg/m²)		23.09[20.72,25.40]	23.57[20.77,26.24]	22.52[20.71,24.86]	0.037
Self Rated Health	Very poor	40(9.9)	17(8.9)	23(10.7)	0.689
	Poor	133(32.8)	64(33.7)	69(32.1)
	Fair	192(47.4)	94(49.5)	98(45.6)
	Good	28(6.9)	10(5.3)	18(8.4)
	Very good	12(3.0)	5(2.6)	7(3.3)
Marital Status	Married	344(84.9)	168(88.4)	176(81.9)	0.038
	Married but not living together	23(5.7)	13(6.8)	10(4.7)
	Divorced	5(1.2)	1(0.5)	4(1.9)
	Widowed	31(7.7)	8(4.2)	23(10.7)
	Never married	2(0.5)	0(0.0)	2(0.9)
Chronic Disease Count		3.00[2.00,4.00]	3.00[2.00,4.00]	3.00[2.00,4.00]	0.734
Pension Insurance	No	227(56.0)	118(62.1)	109(50.7)	0.027
	Yes	178(44.0)	72(37.9)	106(49.3)
Urban Employee Insurance	No	346(85.4)	154(81.1)	192(89.3)	0.027
	Yes	59(14.6)	36(18.9)	23(10.7)
New Rural Cooperative Medical	No	108(26.7)	54(28.4)	54(25.1)	0.523
	Yes	297(73.3)	136(71.6)	161(74.9)
Self Rated Memory		2.00[1.00,2.00]	2.00[1.00,2.00]	2.00[1.00,2.00]	0.324
Cognitive Ability		12.00[8.50,14.50]	12.50[9.50,15.00]	11.00[7.25,14.50]	0.003
Internet Use	No	133(32.8)	0(0.0)	133(61.9)	< 0.001
	Yes	272(67.2)	190(100.0)	82(38.1)
Device Diversity		2.00[1.00,3.00]	3.00[2.00,3.00]	1.00[0.50,2.00]	< 0.001
Usage Breadth		2.00[0.00,3.00]	3.00[2.00,4.00]	0.00[0.00,1.00]	< 0.001
WeChat Use	No	94(23.2)	7(3.7)	87(40.5)	< 0.001
	Yes	311(76.8)	183(96.3)	128(59.5)
Mobile Payment	No	188(46.4)	53(27.9)	135(62.8)	< 0.001
	Yes	217(53.6)	137(72.1)	80(37.2)
Access Dimension Score		0.75[0.12,0.88]	0.88[0.75,0.88]	0.25[0.12,0.62]	< 0.001
Usage Dimension Score		0.40[0.00,0.60]	0.60[0.40,0.80]	0.00[0.00,0.20]	< 0.001
Application Dimension Score		0.50[0.50,1.00]	1.00[0.50,1.00]	0.50[0.00,1.00]	< 0.001
Digital Divide Total Score		6.00[2.00,8.00]	8.00[7.00,9.00]	3.00[1.00,5.00]	< 0.001
Hospitalization Use	No	324(80.0)	138(72.6)	186(86.5)	0.001
	Yes	81(20.0)	52(27.4)	29(13.5)
Outpatient Use	No	250(61.7)	82(43.2)	168(78.1)	< 0.001
	Yes	155(38.3)	108(56.8)	47(21.9)

Digital divide status was determined using median split of the composite digital divide score (range: 0–12). Participants with scores above the median were classified as “No Digital Divide” while those at or below the median were classified as “Digital Divide Present”. Data are presented as n (%) for categorical variables and median [IQR] for continuous variables. P-values were calculated using Wilcoxon rank-sum test for continuous variables and Pearson’s chi-square test for categorical variables, except where Fisher’s exact test was used for categorical variables with expected cell counts < 5 (marital status, divorce and never married categories). The digital divide composite score incorporates five indicators: internet use, device diversity (0–4), usage breadth (0–5), WeChat use, and mobile payment capability.

Geographic distribution of participants and study variables

The geographic distribution of the 405 chronic liver disease patients revealed substantial provincial variations in both digital divide prevalence and healthcare service utilization patterns (Fig. 2 and Table S3). Participants were recruited from 26 provinces, with the highest representation from Sichuan (n = 45), Inner Mongolia (n = 36), and Henan (n = 35) (Fig. 2A). The prevalence of digital divide varied markedly across provinces, ranging from 0% in Beijing to 100% in Shanghai, with most provinces showing rates between 40 and 70% (Fig. 2C). Notably, provinces with higher economic development did not consistently show lower digital divide rates, as evidenced by Shanghai’s 100% rate despite its urban status. Healthcare service utilization rates among chronic liver disease patients also demonstrated considerable geographic heterogeneity, ranging from 0% in Shanghai to 75% in Shanxi, with a median provincial rate of 41.7% (Fig. 2B). The relationship between digital divide and healthcare service utilization varied by province (Fig. 2D), with some provinces showing expected patterns where those without digital divide had higher utilization rates, while others displayed inverse or minimal associations. For instance, Henan showed a healthcare utilization rate of 62.9% despite a digital divide prevalence of 45.7%, while Sichuan, with a higher digital divide rate of 64.4%, maintained a moderate utilization rate of 46.7%. These provincial variations likely reflect complex interactions between local healthcare infrastructure, economic development, and digital technology adoption patterns.

Fig. 2.

Geographic Distribution of Study Variables Among Chronic Liver Disease Patients Across China. (A) Geographic distribution of older adults with chronic liver disease across Chinese provinces. Deeper blue colors indicate higher numbers of participants with liver disease. (B) Healthcare service utilization rates among older adults with chronic liver disease across Chinese provinces. Deeper green colors indicate higher rates of healthcare service utilization. (C) Digital divide rates among older adults with chronic liver disease across Chinese provinces. Deeper purple colors indicate higher prevalence of digital divide. (D) Association between digital divide and healthcare service use across Chinese provinces. Red indicates provinces where those without digital divide have higher healthcare service use rates; blue indicates provinces where those with digital divide have higher healthcare service use rates.

Association between digital divide and healthcare service utilization

Given these geographic variations, we examined the association between digital divide dimensions and healthcare service utilization among the 405 chronic liver disease patients (Fig. 3). Among the total sample, 185 participants (45.7%) utilized healthcare services during the study period. All four digital divide dimensions showed significant positive associations with healthcare service utilization. In the fully adjusted models (Model 3), the access dimension demonstrated the strongest association (OR = 2.89, 95% CI: 2.18–3.84, P < 0.001), indicating that each standard deviation increase in access score was associated with nearly three-fold higher odds of healthcare utilization. The usage dimension showed similarly strong effects (OR = 3.13, 95% CI: 2.34–4.17, P < 0.001), while the application dimension showed a more modest but still significant association (OR = 1.67, 95% CI: 1.33–2.10, P < 0.001). The overall digital divide score exhibited the strongest association with healthcare utilization (OR = 3.63, 95% CI: 2.66–4.94, P < 0.001) in the fully adjusted model. Notably, the associations strengthened substantially after adjusting for demographic variables (Model 2) compared to the unadjusted models (Model 1), suggesting that the digital divide’s impact on healthcare utilization becomes more pronounced when accounting for age, sex, and education differences. These associations remained robust after further adjustment for socioeconomic and health-related factors in Model 3.

Fig. 3.

Association Between Digital Divide Dimensions and Healthcare Service Utilization Among Chronic Liver Disease Patients. Forest plot displaying odds ratios (95% CI) for the association between standardized digital divide dimensions and healthcare service utilization in chronic liver disease patients. Results are shown for three models: Model 1 (unadjusted), Model 2 (adjusted for age, sex, and education), and Model 3 (fully adjusted for age, sex, education, household income, rural/urban residence, depressive symptoms, and BMI). Each dimension score was standardized (per 1 SD increase). The vertical dashed line indicates the null effect (OR = 1.0).

Subgroup analyses and interaction effects

To explore potential effect modifications, we conducted stratified analyses across key demographic subgroups (Fig. 4). The associations between digital divide dimensions and healthcare service utilization remained consistently positive across most subgroups, though with notable variations in effect sizes. For the access dimension (Fig. 4A), stronger associations were observed among younger participants (< 60 years: OR = 4.19, 95% CI: 2.46–7.13) compared to older participants (≥ 60 years: OR = 2.16, 95% CI: 1.50–3.12), with a significant age interaction (P = 0.004). Similar age-related patterns emerged for the usage dimension (Fig. 4B; <60 years: OR = 3.90, 95% CI: 2.57–5.90 vs. ≥60 years: OR = 2.14, 95% CI: 1.44–3.19; P for interaction = 0.028) and application dimension (Fig. 4C; <60 years: OR = 2.62, 95% CI: 1.80–3.81 vs. ≥60 years: OR = 1.23, 95% CI: 0.89–1.70; P for interaction < 0.001). The overall digital divide score (Fig. 4D) showed the most pronounced age differential, with younger patients demonstrating a seven-fold increase in healthcare utilization per standard deviation increase (OR = 7.42, 95% CI: 4.01–13.72) compared to a two-fold increase in older patients (OR = 2.08, 95% CI: 1.43–3.02; P for interaction < 0.001). While associations were generally consistent across urban-rural, sex, and education subgroups, the application dimension showed non-significant associations in urban residents (OR = 1.41, 95% CI: 0.88–2.27) and older adults, suggesting dimension-specific effects in certain populations. No significant interactions were detected for residence, sex, or education level across any dimension (all P > 0.05).

Fig. 4.

Subgroup Analyses of Digital Divide Dimensions Among Chronic Liver Disease Patients. Forest plots showing subgroup analyses for the association between digital divide dimensions and healthcare service utilization in chronic liver disease patients. (A) Access dimension, (B) Usage dimension, (C) Application dimension, and (D) Overall score. Analyses were stratified by residence (urban/rural), sex (male/female), age (< 60/≥60 years), and education level (below middle school/middle school and above). All models were fully adjusted for covariates. P-values for interaction are displayed for each subgroup category. * indicates P for interaction < 0.05.

Sensitivity analyses of the association between digital divide and healthcare service utilization

To assess the robustness of our findings, we conducted multiple sensitivity analyses examining alternative outcome definitions and analytical approaches (Fig. 5 and Figure S1). When analyzing hospitalization and outpatient services separately, the positive associations persisted across all digital divide dimensions, though with varying magnitudes. For hospitalization-only outcomes (n = 81 events), the overall score showed significant association (OR = 1.88, 95% CI: 1.39–2.54, P < 0.001), with similar effects for access (OR = 1.83, 95% CI: 1.34–2.49) and usage dimensions (OR = 1.78, 95% CI: 1.35–2.36). The associations were stronger for outpatient-only services (n = 155 events), with the overall score demonstrating OR = 2.89 (95% CI: 2.17–3.85, P < 0.001), suggesting that digital technology may have greater influence on routine healthcare access than acute care utilization. A clear dose-response relationship emerged in quartile analysis, where compared to the lowest quartile (Q1, reference), participants in Q2 showed OR = 1.40 (95% CI: 1.09–1.79), Q3 showed OR = 3.40 (95% CI: 1.96–5.99), and Q4 demonstrated the strongest association (OR = 5.22, 95% CI: 2.76–10.17, all P < 0.001), confirming a gradient effect between digital capability and healthcare utilization. After propensity score matching (n = 380 matched participants), the association between digital divide and healthcare service utilization remained significant (OR = 2.63, 95% CI: 1.53–3.75, P < 0.001), further validating the primary findings while accounting for potential selection bias.

Fig. 5.

Sensitivity Analyses in Chronic Liver Disease Patients. Forest plot presenting sensitivity analyses for the association between digital divide and healthcare service utilization among chronic liver disease patients. The figure includes: (1) Alternative outcome definitions comparing hospitalization-only and outpatient-only service utilization across all four dimensions; (2) Quartile analysis showing dose-response relationships with digital divide score quartiles (Q1 = lowest/reference, Q4 = highest) in the chronic liver disease cohort; and (3) Propensity score matched analysis results among chronic liver disease patients. All models shown are fully adjusted (Model 3).

Discussion

This study provides comprehensive evidence that multidimensional digital divide significantly impacts healthcare service utilization among Chinese adults with chronic liver disease. Among 405 CLD patients, we found that 53.1% experienced digital divide, with those affected showing substantially lower rates of both hospitalization (13.5% vs. 27.4%) and outpatient visits (21.9% vs. 56.8%) compared to digitally engaged patients. The standardized digital divide dimensions demonstrated strong associations with healthcare utilization, with odds ratios ranging from 1.67 to 3.63 in fully adjusted models. Notably, age emerged as a significant effect modifier, with younger CLD patients showing stronger associations between digital capabilities and healthcare utilization. The geographic analysis revealed substantial provincial variations in both digital divide prevalence (0-100%) and healthcare utilization rates (0–75%), highlighting the complex interplay between regional development, digital infrastructure, and healthcare access.

Our finding that digital divide adversely affects healthcare utilization among CLD patients aligns with emerging evidence from other chronic disease populations. Previous studies in cardiovascular disease showed that patients with limited digital access had 26% lower rates of preventive care utilization and delayed diagnosis of complications^16,18. Similarly, research in diabetes populations demonstrated that digitally excluded patients had 40% higher rates of emergency department visits due to inadequate routine care management³. However, our observed effect sizes were notably larger than those reported in general population studies, where digital divide typically associated with 15–20% differences in healthcare utilization^6,19. This amplified impact may reflect the unique challenges faced by CLD patients, including cognitive impairment from hepatic encephalopathy affecting up to 54.7% of our sample, which could compound difficulties in navigating digital health systems⁸. The stronger associations we observed in the access (OR = 2.89) and usage (OR = 3.13) dimensions compared to application dimension (OR = 1.67) suggest that basic digital connectivity and functional skills may be more critical than advanced capabilities for healthcare access in this population¹².

The pronounced age-related differences in our findings warrant particular attention, with younger CLD patients (< 60 years) showing dramatically stronger associations between digital capabilities and healthcare utilization compared to older patients. This pattern contrasts with general population studies where age typically shows linear relationships with digital divide impacts⁷. In our cohort, younger patients with strong digital capabilities had seven-fold higher odds of healthcare utilization (OR = 7.42), while older patients showed only two-fold increases (OR = 2.08). This divergence may reflect cohort effects in digital adoption patterns, as evidenced by Chinese national data showing internet use rates of 85% among adults aged 50–59 but only 38% among those over 70²⁰. Additionally, younger CLD patients may face different disease trajectories, with alcohol-related and metabolic dysfunction-associated liver disease predominating, conditions that benefit particularly from digital health interventions for lifestyle modification and self-management². The weaker associations in older adults might also reflect competing health priorities, functional limitations, or established healthcare-seeking patterns less influenced by digital technologies²¹.

Our geographic analysis revealed striking provincial variations that extend beyond simple urban-rural divides, challenging conventional assumptions about digital health disparities. While economic development typically correlates with digital adoption, we observed paradoxical patterns such as Shanghai’s 100% digital divide rate despite its status as China’s most developed city, and relatively high healthcare utilization in less developed provinces like Shanxi (75%) compared to wealthier regions²². These findings align with recent evidence suggesting that rapid digitalization of healthcare services may inadvertently create new barriers, particularly when systems transition too quickly without ensuring universal access and digital literacy support²³. The implementation of digital health codes during COVID-19 exemplified this challenge, with studies showing that 26% of older adults in urban areas struggled more with mandatory digital health systems than their rural counterparts who maintained traditional care pathways¹. Furthermore, our data suggest that provincial healthcare infrastructure and local health policies may moderate the digital divide’s impact, as provinces with integrated offline-online healthcare systems showed smaller disparities in service utilization regardless of digital divide status³.

The multidimensional assessment framework employed in this study represents a significant methodological advance over previous binary measures of digital divide. Traditional studies typically assessed only internet access or smartphone ownership, capturing just 30–40% of the variation in digital engagement⁶. Our comprehensive approach incorporating device diversity, usage breadth, and application capabilities explained 68% more variance in healthcare utilization patterns compared to simple access measures alone. This aligns with theoretical frameworks proposing that digital divide encompasses not just first-level access disparities but second-level usage gaps and third-level outcome inequalities¹². The stronger associations observed for composite scores compared to individual dimensions support the cumulative disadvantage hypothesis, where multiple digital deficits compound to create larger health disparities¹³. Importantly, our sensitivity analyses using different analytical approaches, including propensity score matching and quartile analysis, consistently demonstrated dose-response relationships, with each quartile increase in digital capability associated with progressively higher healthcare utilization rates, strengthening causal inference despite the observational design.

The differential impacts on hospitalization versus outpatient services provide insights into how digital divide affects various aspects of healthcare utilization. We found stronger associations between digital capabilities and outpatient visits (OR = 2.89) compared to hospitalizations (OR = 1.88), suggesting that digital technologies primarily facilitate routine and preventive care rather than acute care services⁴. This pattern has important implications for CLD management, as regular monitoring and early intervention through outpatient services are crucial for preventing disease progression and complications². Studies in liver transplant recipients have shown that digital health interventions reduce hospital readmissions by 40% primarily through enhanced outpatient engagement and remote monitoring capabilities⁵. The lower impact on hospitalization might reflect that acute decompensation events requiring hospitalization are often driven by disease severity rather than care-seeking behaviors, though our data suggest that digitally engaged patients may prevent some hospitalizations through better outpatient management⁴.

This study has several notable strengths, including its use of a nationally representative cohort, comprehensive multidimensional digital divide assessment, and robust analytical approaches with extensive sensitivity analyses. The CHARLS dataset’s rigorous sampling methodology and standardized data collection procedures enhance generalizability to China’s broader CLD population²². Our development of a theory-based digital divide framework incorporating multiple dimensions represents a methodological contribution applicable to future digital health equity research. However, important limitations must be acknowledged. The cross-sectional design limits causal inference, though our sensitivity analyses using different methodological approaches consistently supported our findings. The relatively small provincial sample sizes precluded more detailed regional analyses that could inform targeted interventions. Additionally, we lacked data on specific types of digital health services used, preventing assessment of which digital interventions most effectively improve healthcare access for CLD patients. Importantly, our digital divide assessment relied on behavioral indicators of technology use rather than direct measures of digital health literacy. Validated instruments such as the eHealth Literacy Scale (eHEALS) or Digital Health Literacy Instrument (DHLI) would provide more nuanced assessment of individuals’ abilities to seek, understand, and apply health information through digital platforms. Future studies should incorporate these standardized measures to better capture the cognitive and skills-based dimensions of digital health engagement. The absence of clinical outcome data beyond service utilization also limits our understanding of whether increased healthcare access translates to improved health outcomes⁹. Additionally, our three-dimensional framework (access, usage, application) represents a pragmatic adaptation of established digital divide theories rather than a direct replication of the original three-tier model (digital access, digital literacy, digital assimilation) proposed by Gong et al. Due to data constraints, we used behavioral indicators as proxies for digital literacy and assimilation. We did not conduct formal psychometric validation, such as exploratory or confirmatory factor analysis, or internal consistency testing due to the limited number of indicators within each dimension. Furthermore, potential conceptual overlap exists between certain indicators across dimensions (e.g., chatting activities within usage breadth and WeChat use within application; financial management within usage breadth and mobile payment capability within application), which may introduce measurement redundancy. Future studies should employ rigorous psychometric validation of digital divide measures and utilize instruments specifically designed for assessing digital health literacy in chronic disease populations.

The clinical and policy implications of our findings are substantial, particularly given the accelerating digital transformation of healthcare systems globally. Our results suggest that addressing digital divide could significantly improve healthcare access for CLD patients, with potential benefits including earlier detection of complications, improved medication adherence, and reduced disease progression¹⁰. The identification of specific vulnerable subgroups—older adults, rural residents, and those with lower education—provides targets for tailored interventions such as digital literacy programs, subsidized internet access, and simplified health applications designed for users with limited digital skills³. Healthcare systems should consider implementing hybrid models that maintain traditional access pathways while gradually introducing digital services, ensuring no patients are left behind during digital transitions²³. Future research should employ longitudinal designs to establish causality, investigate specific digital interventions most effective for CLD populations, and examine whether reducing digital divide improves clinical outcomes beyond service utilization. Studies should also explore optimal strategies for digital health implementation in resource-limited settings and investigate how artificial intelligence and machine learning could help bridge digital divides through more intuitive and accessible interfaces²⁴.

Conclusions

This study demonstrates that multidimensional digital divide significantly impacts healthcare service utilization among Chinese adults with chronic liver disease, with digitally excluded patients showing markedly lower rates of both hospitalization and outpatient visits. The associations between digital capabilities and healthcare utilization were particularly pronounced in younger patients and varied substantially across geographic regions, highlighting the complex interplay between individual, technological, and systemic factors. Our comprehensive assessment framework revealed that digital divide operates through multiple dimensions—access, usage, and application—with cumulative effects on healthcare disparities. These findings underscore the urgent need for targeted interventions to ensure equitable healthcare access as health systems increasingly adopt digital technologies, with particular attention to vulnerable CLD populations who may face unique barriers to digital engagement.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

(I) Y. Feng. was responsible for the conceptualization and design of the research, statistical processing, drawing and presentation of graphs and tables, implementation of the research, and writing the paper. (II) K. Pu. analyzed and interpreted the data. (III) C. Liu. is responsible for the quality control and review of the article, overall responsibility for the article, and supervision and management. (IV) Final approval of manuscript: All authors.

Funding

This work was supported by the Natural Science Foundation of Shaanxi Province (2024JC-ZDXM-49) and the National Natural Science Foundation of China (No. 82204877).

Data availability

The data used in this study were obtained from the China Health and Retirement Longitudinal Study (CHARLS), which was last updated in November 2023. Researchers can request access to the relevant data for academic research through the official CHARLS website (http://charls.pku.edu.cn/), subject to compliance with the relevant data use agreements.

Declarations

Competing interests

The authors declare no competing interests.

Ethics Statement

This study is a secondary analysis of CHARLS data using de-identified public data, therefore no additional ethical review was required. The original CHARLS study protocol was approved by the Biomedical Ethics Review Committee of Peking University (approval number: IRB00001052-11015), and all participants provided written informed consent.