Remember me
The internet has become crucial for health information dissemination in the digital era [,]. Online health information seeking (OHIS) has gained popularity due to its accessibility, wide information coverage, ease of use, affordability, and anonymity [,], especially in upper-middle-income countries such as China, where health care resources and in-person medical appointments are limited [,]. As of December 2022, China’s internet penetration rate has reached 75.6%, with 1.067 billion internet users accessing health information on the web []. The shift to digital health is essential for improving health knowledge, increasing confidence in managing health issues, and promoting healthy behaviors [,]. Nevertheless, previous research has noted that information alone might not be sufficient to affect optimal health-related well-being []. The notion of health literacy, first introduced in China in 2005 and widely acknowledged as a cost-effective measure for improving public health [], has evolved into eHealth literacy, which is a foundational skill set that underpins the use of information and communication technologies for health [-]. While research has examined digital health literacy and health information seeking among specific user segments within China, such as community-dwelling older population [], rural residents [], or active social media users [], and some studies have leveraged nationwide survey samples [,], a comprehensive analysis of the broader population’s engagement with online health resources is needed.
OHIS and Health BehaviorsHealth behavior, distinct from medical treatment, concerns how health interventions and societal norms affect the health of individuals’ lifestyles []. While previous research has extensively focused on information seeking and literacy in patient-centered health communication [,], OHIS is not limited to individuals facing health threats []. More commonly, the general public turns to the internet to find information on leading healthier lifestyles, which warrants examination into how the wider population uses the internet for health-related decisions []. By providing access to more health-relevant information, OHIS might enable a more accurate assessment of health status, disease outbreak severity, and the need for health protection measures []. It could contribute to individuals’ perceived control over health threats and reduce negative emotions associated with uncertainty []. As most existing literature has acknowledged the potential of OHIS to contribute to better health outcomes, we first examine the direct path.
Hypothesis 1: OHIS is positively associated with individuals’ engagement in health behaviors.Linking eHealth Literacy With OHIS and Health BehaviorsAmong existing scholarship on internet-facilitated health communication and promotion, the integrative model of eHealth use underscores that macrolevel social disparities, often known as the digital divide in the realm of health communication [,], manifest as microlevel individual variations in the orientation and efficacy of OHIS, influencing people’s internet use for health-related purposes and ultimately their health outcomes []. For instance, previous research found that female, older, married, and better-educated Chinese internet users engage in OHIS more frequently [] and that people’s health status could alter their health behavioral engagement [].
As the integrative model of eHealth use suggests, online health resources can only improve health outcomes if the public has adequate eHealth literacy and avoids low-quality and harmful content []. eHealth literacy refers to a person’s perceived ability to (1) have access to health information on the web and (2) understand the health information accessed []. It not only affects individuals’ OHIS behaviors [-] but also directly and indirectly engages with sequential health outcomes [,]. Centering on the cognitive mechanisms linking OHIS and health behaviors, the study delved into how 2 interrelated dimensions of transactional eHealth literacy influence the process. Functional eHealth literacy refers to individual users’ perceived ability to acquire online health information, while critical eHealth literacy involves more advanced cognitive processes related to information appraisal, including evaluating the reliability, validity, credibility, and applicability of health information []. We first examine the impact of Chinese internet users’ sociodemographic factors and health status on their eHealth literacy.
Research question 1: How are Chinese netizens’ sociodemographic factors and health status associated with their (1) functional and (2) critical eHealth literacy, respectively?Higher eHealth literacy might encourage users to resort to the internet, leading to a stronger likelihood of OHIS under health motivation [,]. With better functional and critical literacy, individuals would have more ability to access health information and make efficient use of online searching tools and technological devices [,]. Conversely, people with less knowledge and confidence in using digital devices are found less likely to conduct OHIS [,]. While an association between eHealth literacy and actual eHealth use is recognized, previous research has lent mixed support for the direction of the relationship, and we cannot rule out the possibility of a bidirectional association. Moreover, individuals’ personal traits may be associated with both eHealth literacy as well as OHIS [,], thus further confounding the process. In this study, we deployed structural equation modeling (SEM) to assess the possible bidirectional relationship between the 2 types of eHealth literacy and OHIS. High critical literacy might contribute to better well-being and engagement in healthy behaviors [,], as it can increase levels of self-engagement, initiative, and control over health concerns and self-care management []. Therefore, we propose 2 hypotheses.
Hypothesis 2: Chinese internet users’ (1) functional and (2) critical eHealth literacy have reciprocal associations with their OHIS.Hypothesis 3: Chinese internet users’ (1) functional and (2) critical eHealth literacy are positively associated with their engagement in health behaviors.Mediator: Perceived Quality of Online Health InformationApart from eHealth literacy, users’ evaluation and satisfaction with obtained health information or experiences using different sources play an important role in health outcomes []. Extant theoretical models suggest that the effectiveness of interactive media use for health information is conditioned on users’ information processing while engaged with the media and content, which may or may not lead to optimal health outcomes. For instance, the 3-stage model of health promotion using interactive technology proposed that the use of interactive online health resources functions through the interplay of the characteristics of user, media, and health message, potentially contributing to users’ health maintenance and improvement [,]. Our study specifically focuses on users’ attitudes toward the quality of health sources and the information itself [], as a previous study denoted that content-related indicators and criteria were used the most in credibility evaluation across different information sources compared with the other functionalities []. Specifically, we measure perceptions of online health information in terms of scientific rigor, timeliness, accuracy, objectiveness, credibility, applicability, as well as potential harm. We propose that the psychological mechanism underlying health information processing is contingent on individuals’ evaluation of the quality of online health information.
Hypothesis 4: Chinese internet users’ perceived quality of online health information will mediate the relationship between OHIS and health behavioral engagement.ObjectivesThis study aimed to examine the underlying mechanism by which OHIS influences the engagement in health behaviors of Chinese internet users across various sociodemographic groups. Our conceptual framework proposes a structural model highlighting the interrelationships among users’ OHIS behavior, their self-assessed ability to acquire information on the web (functional literacy), and their ability to critically appraise acquired health information (critical literacy), which potentially affect their health outcomes. In addition, the study accounts for the mediating effect of users’ perceptions of online health information quality on the pathways.
A cross-sectional online survey was conducted from November 1, 2021, to November 26, 2021, among Chinese internet users accessing health information on the web. The reporting follows the Checklist for Reporting Results of Internet e-Surveys (CHERRIES; [,]).
Our questionnaire was developed with input from media, communication, and public health experts to ensure content relevance and clarity. A group of subject matter experts reviewed the original questionnaire and the adapted scales for eHealth literacy and health behavior. The sample size was determined to balance statistical power with budgetary constraints. We targeted a 95% CI with a 1% margin of error for China’s internet population (1.011 billion) as of June 2021. In addition, we followed the rule of thumb for the minimum sample size for SEM based on our proposed model []. We used quota sampling, considering national representativeness based on sex, age groups, as well as urban and rural residence. Quotas were determined using data from the 48th Statistical Report on Internet Development of China conducted by the China Internet Network Information Center []. The survey was distributed through a professional Chinese online survey platform, IDiaoYan (Zhongyan Technology) []. We conducted a closed survey accessible exclusively to registered panelists.
Our study initially reached 96,335 registered users of the platform, among which 17.41% (n=16,774) users clicked into the first page of the survey. In a prescreening question, we asked whether respondents used the internet for health information, and 27.64% (4637/16,774) of the participants who answered “no” to this question were screened out. Of the 12,137 remaining responses, 7.34% (n=891) were incomplete and 8.36% (n=1015) participants were unable to submit responses due to quota sampling limitations, where response collection ceases once the quota for a specific group is reached. We also excluded invalid responses (231/12,137, 1.9%) based on the following criteria: (1) completion time <3 minutes, (2) failed logic checks, or (3) filled-in demographic information misaligned with the registered profile in the panel. Consequently, the final sample included 10,000 complete and valid responses. The survey development is illustrated in the flowchart of , and the questionnaire in English and original language (Chinese) are included in .
Figure 1. Flowchart illustrating survey development and data screening process. Ethical ConsiderationsThe research protocol for this study was reviewed and approved by the School of Humanities of Tsinghua University and Chinese Academy of Cyberspace Studies. At the beginning of the survey, participants were presented with information about the purpose and procedures of the study, as well as how the data would be handled. They were only allowed to proceed with the study after reading the information and providing informed consent. For participants aged <18 years, parents or guardians provided consent on their behalf, and these participants were required to complete the survey under parental guidance. Participation was on a voluntary basis. All collected data were anonymized, removing any identifiers that could directly or indirectly link the data to individual participants. The data collection and storage protocols were in full compliance with the Personal Information Protection Law of China. Respondents who submitted valid responses were rewarded with 5000 bonus points, equivalent to CNY 5 (US $0.8) through the survey platform’s loyalty points program.
MeasurementsFrequency of OHISDrawing on previous literature [,], we operationalized individuals’ OHIS as the frequency of participants’ engagement in health information seeking through different online channels. Participants were asked to indicate how often they sought health information from (1) mainstream media, (2) professional health media, (3) aggregator news platforms, (4) web portals, (5) open forums, (6) online support forums, (7) search engines, and (8) individual social media accounts. Examples of each information source were provided, as shown in and . Responses ranged from “1=never” to “5=always,” indicating the frequency of OHIS through each channel. Our intended latent construct of OHIS showed good reliability (Cronbach α=0.85; mean 3.30, SD 0.77).
Figure 2. Frequency of use and perceived credibility of different health information channels (N=10,000). Credibility of Information SourceWe calculated a credibility score for each information channel by reverse coding the credibility ranks given by participants. Specifically, participants ranked their top 3 information sources in order of perceived credibility. We later assigned a weight of 3 to the most credible, 2 to the second most credible, and 1 to the third most credible source. We then averaged these weighted rankings across participants to examine the credibility score for each of the 8 sources assessed. The resulting credibility scores ranged from 0 to 3, where a higher score signified greater perceived credibility.
eHealth LiteracyeHealth literacy was assessed using a modified version of the eHealth Literacy scale, which had been translated into Chinese and demonstrated good reliability and validity [,]. Our study introduced a 2-factor model based on functional and critical literacy. Specifically, we assessed functional literacy by asking participants to what extent they were capable of obtaining and accessing health information in the Web 2.0 context [,]. Sample items include “I know what health resources are available on the Internet” and “I am capable of using mobile devices to search for health information.” Critical literacy focused on how participants assessed and evaluated health information obtained from the internet to make health decisions [,], using items such as “I can tell high-quality from low-quality health resources on the Internet.” Response options for both variables included a 5-point Likert scale ranging from “1=totally disagree” to “5=totally agree.” The scores were then averaged to generate 2 indices of users’ functional literacy (Cronbach α=0.804; mean 3.84, SD 0.63) and critical literacy (Cronbach α=0.73; mean 3.55, SD 0.68), showing good reliabilities of the adapted eHealth literacy scales.
Health BehaviorOur study measured health maintenance behaviors using 4 self-reported items that represent people’s commitment to reduce negative health outcomes and facilitate psychological and behavioral well-being []. The scale was revised based on previous literature []. Examples of the items include mental health management as well as good hygiene habits. Participants rated these 4 items on a 5-point scale ranging from “1=not at all” to “5=extremely well.” Cronbach α for this variable was reasonably reliable (Cronbach α=0.74; mean 3.77, SD 0.69).
Perceived Quality of Online Health InformationWe measured participants’ perceived quality of online health information using 5 items assessing the reliability, accuracy, and applicability of health-related online content [,]. Respondents were asked to evaluate the quality of internet health information based on to what degree they think that online health information (1) is supported by reliable scientific evidence; (2) is credible and reliable; (3) does not pose a risk to their personal health and well-being; (4) aligns with the latest advancements and consensus within the medical science community; and (5) is actionable and applicable in real-life scenarios. Responses were scored on a 5-point scale from “1=strongly disagree” to “5=strongly agree” (Cronbach α=0.77; mean 3.51, SD 0.65).
Control VariablesWe collected participants’ sex, age groups, education levels, monthly income, residential areas, and provinces. In addition, participants self-evaluated health status was also measured using scales validated by a previous study [].
Data AnalysisIn our study, we first conducted univariate and multivariate analyses, partial correlation, and regression analysis using SPSS software (version 29.0; SPSS Inc). To assess the normality of eHealth literacy, OHIS, health behavior, and perceived quality, we applied the Kolmogorov-Smirnov test. Non-normal variables were reported as mean (SD) as well as median (IQR). Categorical variables were reported as frequency and percentages. Differences between sex (female and male) and residence types (rural and urban) were examined using the Mann-Whitney U test, while the Kruskal-Wallis test was used for comparison across various age groups, educational levels, income brackets, regions, and health status. We considered P values <.05 (2-sided) to be statistically significant. Partial correlations were examined among functional and critical eHealth literacy, OHIS, evaluation, and health behavior while controlling age, sex, education, income, health status, residence, and regional distribution. We then conducted stepwise linear regression with OHIS and health behavior as dependent variables, respectively, and categorical user characteristics, eHealth literacy, and perceived health information quality as independent variables. Using a stepwise method (P<.05 as the criterion for entry and P>.10 as the criterion for exclusion), all possible combinations of variables were first tested, and the best combinations were selected based on model fit and significance. In addition, we assessed multicollinearity in our regression by examining the variance inflation factor.
To test our hypotheses, we used a 2-step approach following previous research [,]. First, to assess the reliability and validity of the latent variables in our model, we performed a confirmatory factor analysis (CFA). This step evaluated the proposed measurement model. Given that we have a relatively large sample size (N=10,000) that strongly influences the result of the chi-square test, we did not refer to the cutoff value 3 of normalized chi-square but the other absolute and incremental model fit indices (eg, root mean square error of approximation [RMSEA], comparative fit index [CFI], and standardized root mean square residual [SRMR]). After the CFA, we applied SEM to examine the pathways among our key variables while controlling for sex, age groups, residence, educational levels, income, and health status. We used a Bootstrap analysis with a 95% CI to estimate the parameters and their associated SEs. Both the CFA and SEM analyses were conducted using the Lavaan package [] in R software (version 4.2.2; R Foundation for Statistical Computing).
Our study sample included 10,000 participants from 31 provinces in mainland China. The details of participants’ demographic distribution and self-reported health status are presented in . Table S1 in contains detailed descriptive statistics of eHealth literacy across groups, and Table S2 of contains provincial distribution of participants’ eHealth literacy. The sampling followed the 48th Statistical Report on Internet Development of China, where most Chinese internet users resided in urban areas (7060/10,000, 70.6%), were aged between 20 and 49 years (5640/10,000, 56.4%), and were relatively evenly distributed between female (4880/10,000, 48.8%) and male (5120/10,000, 51.2%) users. It should be noted that our respondents were mostly well educated, with about 60% holding a bachelor (3393/10,000, 33.9%) or associated degree (2588/10,000, 25.9%). Furthermore, more than half (5628/10,000, 56.3%) of the participants were from China’s eastern provinces. Users’ health status varied, with most self-reporting good health condition (3777/10,000, 37.8%) and 26.6% (2661/10,000) experiencing subhealth symptoms, such as fatigue and poor appetite. Severe health conditions, such as cancer, were relatively rare among all participants (46/10,000, 0.5%).
Table 1. Demographic characteristics, health status, and distribution of eHealth literacy (N=10,000).DemographicsParticipants, n (%)Test statistics for functional literacyaP valueTest statistics for critical literacyP valueSex3.121.002–0.973.33aTest statistics were reported for dichotomous variables (z score) and multicategorical variables (χ2) alongside P value.
bA currency exchanged rate of 1¥=US $0.16 is applicable.
We measured OHIS by examining the frequency of use of online sources. As shown in , search engines, such as Baidu, emerged as primary tools for health-related inquiries (“often”: 4130/10,000, 41.3% and “always”: 1980/10,000, 19.8%). Conversely, emerging specialized online support forums, such as the bulletin board system forum for patients with diabetes, saw less frequent engagement, with 16% (1600/10,000) of the participants indicating that they “never” have used such platforms. Regarding source credibility, among the 8 channels assessed, professional health media outlets were perceived as the most credible. State media outlets, such as Xinhua News and People’s Daily, were also rated as relatively credible. However, social networking (question and answer) forums and online support communities were regarded as less credible.
Univariate Analysis of eHealth Literacy Across Different Sociodemographic GroupsOur analyses revealed differences in both functional and critical eHealth literacy across various demographic, socioeconomic, and health-related variables. The detailed descriptive statistics across groups can be found in Table S1 in . The Mann-Whitney U test indicated that female users have a significantly higher functional literacy than their male counterparts (z score=3.12; P=.002). No significant difference was found in critical literacy (z score=–0.97; P=.33). A digital divide was observed between urban and rural residents, evident in both functional literacy (z score=–10.29; P<.001) and critical literacy (z score=–3.74; P<.001). Multicategorical group comparison showed variations in literacy levels across age cohorts in functional literacy (χ25=242.3; P<.001) and critical literacy (χ25=116.5; P<.001). Internet users with different income levels demonstrated variations in both functional literacy (χ26=128.3; P<.001) and critical literacy (χ26=226.9; P<.001). Similarly, respondents with different educational backgrounds also showed varied levels of functional literacy (χ25=192.3; P<.001) and critical literacy (χ25=62.9; P<.001). Furthermore, there were marked differences based on participants’ health conditions in functional literacy (χ24=74.5; P<.001) and critical literacy (χ24=17.5; P<.001). Notably, we did not observe significant regional differences in internet users’ functional literacy (χ23=3.6; P=.21); however, users’ critical literacy demonstrated significant regional variations across East, Central, West, and Northeast China (χ23=49.2; P<.001).
Multivariate Analysis on Social and Individual Differences of OHIS and Health BehaviorPartial correlation analysis, as detailed in , revealed significant positive associations among the 5 key variables: functional literacy, critical literacy, OHIS, perceived information quality, and health behavior, controlling for covariates (all P<.001).
Table 2. Partial correlation analysis (Pearson r and 2-tailed P value) among key variablesa.aPartial correlation coefficients were calculated with sex, age groups, residence, region, education, income, and health status as control variables.
bOHIS: online health information seeking.
cNot applicable.
The results of stepwise linear regression investigated the associations between user-oriented characteristics and their OHIS and health behavioral engagement. Users’ sex and regional differences were excluded in the final model predicting OHIS (), while users’ residence types and regional differences were excluded in the final model predicting health behavior ().
Aside from the excluded factors, all considered sociodemographic factors, eHealth literacy, and perceived information quality showed statistically significant main effects on both OHIS and health behaviors (all P<.001). Furthermore, the variance inflation factor values were <2 across the model, suggesting no collinearity issues and thereby affirming the reliability of the regression analyses.
Table 3. Stepwise linear regression predicting online health information seeking (OHIS)a.Included predictorsbb (SE)βt test (df)P valueVIFcIncome0.10 (0.01).2017.38 (9991)<.0011.60Age0.04 (0.01).076.13 (9991)<.0011.73Health condition–0.04 (0.01)–.06–6.11 (9991)<.0011.35Residence0.11 (0.02).065.97 (9991)<.0011.35Education level0.05 (0.01).075.85 (9991)<.0011.73Functional literacy–0.07 (0.01)–.06–4.86 (9991)<.0011.66Critical literacy0.20 (0.01).1713.85 (9991)<.0011.92Perceived quality0.23 (0.01).2017.75 (9991)<.0011.51aThe result shows the final model in stepwise regression predicting OHIS: R2=0.187, adjusted R2=0.186, F1,9991=315.21 (P<.001). Sex and regional distribution were excluded.
bb: unstandardized coefficient.
cVIF: variance inflation factor.
Table 4. Stepwise linear regression predicting health behaviora.Included predictorsbb (SE)βt test (df)P valueVIFcIncome0.03 (0.01).065.62 (9991)<.0011.54Age0.06 (0.01).1512.75 (9991)<.0011.71Education level0.04 (0.01).076.07 (9991)<.0011.62Health condition0.05 (0.01).087.22 (9991)<.0011.35Sex0.06 (0.01).055.08 (9991)<.0011.03Functional literacy0.22 (0.01).2016.84 (9991)<.0011.66Critical literacy0.12 (0.01).129.81 (9991)<.0011.92Perceived quality0.15 (0.01).1412.69 (9991)<.0011.51aThe result shows the final model in stepwise regression predicting health behavior: R2=0.187, adjusted R2=0.187, F1,9991=160.99 (P<.001). Residence and regional distribution were excluded.
bb: unstandardized coefficient.
cVIF: variance inflation factor.
Hypothesis TestingOur research model posited that participants’ functional and critical eHealth levels exhibited reciprocal relationships with OHIS. This, in turn, was posited to enhance their perceptions of the quality of online health information, thereby increasing engagement in health-promoting behaviors. The model mapped all direct and indirect pathways linking 5 key variables, with the literacy-OHIS pathways modeled as bidirectional. We also estimated the covariance between these 5 latent variables in our model, considering their correlations. To control for confounding effects, we included sex, age group, residential area, educational background, income, and health status as covariates in our structural model. Notably, based on the results of stepwise linear regression analyses, internet users’ regional difference was not incorporated as a covariate in the final model.
Measurement ModelAbove all, our measurement model () demonstrated excellent model fit (χ2289=2889.03, χ2/df=10, P<.001; RMSEA=0.030, 95% CI 0.029-0.031; SRMR=0.029; CFI=0.968), underscoring the structural integrity of the 5 key constructs. One item from OHIS (ie, seeking information via search engine) was dropped due to low factor loading. All remaining factor loadings were above the recommended threshold of 0.5, showing acceptable indicator reliability. The constructs also exhibited satisfactory content reliability, with Cronbach α coefficients ranging from 0.73 to 0.86 and composite reliability values spanning 0.73 to 0.86. While our average variance extracted (AVE) fell slightly below the standard 0.5 threshold, Fornell and Larcker [] have discussed that given that AVE is a more stringent measurement, researchers might still conclude the establishment of convergent validity with satisfactory composite reliability. This notion is particularly relevant to our study’s tailored adaptation of existing scales. Given these considerations and supported by supplementary research [], we concluded that the measurement model has established convergent validity based on (1) AVE values marginally <0.5; (2) factor loadings all >0.5, showing strong item-to-construct relationships; and (3) composite reliability values >0.7 across all constructs, and thus collectively affirming the convergent validity of the model.
Table 5. Statistical outcomes of confirmatory factor analysisa.ConstructsStandardized factor loadingz scoreAVEbCRcOHISd (Cronbach α=0.86)0.4670.86aModel fit: χ2289=2889.03, χ2/289=10, P<.001; root mean square error of approximation of 0.030; standardized root mean square residual of 0.029; comparative fit index value of 0.968.
bAVE: average variance extracted.
cCR: composite reliability.
dOHIS: online health information seeking.
eItems constrained for identification purposes.
fQ and A: question and answer.
Structural ModelThe result of the SEM is presented in and (both unstandardized and standardized coefficients are reported). Several model fit measures suggest that our final model was found to be a good fit to the data with χ2404=4183.6, χ2/df=10.36 (P<.001), RMSEA=0.031 (95% CI 0.030-0.031), SRMR=0.029, and CFI=0.955.
Comments (0)