This cross-sectional study was conducted from June 19, 2023, to October 20, 2024. A multistage stratified sampling method was used. First, 2 provinces (autonomous regions and municipalities) were selected from each of the eastern, central, and western regions of China to ensure national geographical representativeness. The surveyed provinces included Gansu, Chongqing, Hubei, Henan, Zhejiang, and Shandong. Within each province, counties were stratified by economic development level; the top and bottom 3% of counties were selected as sampling units, totaling 12 counties.

Trained investigators conducted door-to-door interviews with permanent residents (defined as individuals who had resided in the household for ≥6 months in the previous year), administering paper questionnaires to individuals aged 18 to 70 years. A standardized training manual and interview script were used (Multimedia Appendix 1). All investigators underwent a 2-day centralized training session covering research ethics, questionnaire explanation, and interview techniques. Role-playing and pilot tests were conducted to ensure uniform understanding and delivery of questions. Regular supervision (weekly team meetings to review completed questionnaires) and random callback checks (10% of participants were recontacted via telephone within 1 week to verify responses) were performed during data collection to maintain consistency and quality. The questionnaire was interviewer administered: investigators read each item aloud and recorded responses to minimize literacy-related barriers.

Each questionnaire included a written informed consent form. The study excluded minors. Of the 7200 returned questionnaires, 7004 (97.3%) valid responses were collected after excluding incomplete or inconsistent submissions. Inclusion criteria were (1) medical resource use within the previous 2 years, (2) language communication ability, (3) no major cognitive or mental disorder, and (4) voluntary participation.

Potential confounding variables were controlled based on established methodological approaches [12], encompassing age, educational attainment, annual household income, employment status, chronic disease status, self-rated health, weekly breakfast frequency, smoking status, alcohol consumption, exercise frequency, and subjectively measured sleep quality.

Data entry and database development were performed using EpiData (version 4.6; EpiData Association) with double entry verification. All statistical analyses were conducted in R (version 4.3.1; R Foundation for Statistical Computing) using the lme4 package for fixed-effects modeling and glmnet for least absolute shrinkage and selection operator (LASSO) regression, with statistical significance defined as a 2-sided P value of less than .05. Baseline regression analyses incorporated regional fixed effects, and SEs were clustered at the county level to account for intracounty correlation, providing more conservative statistical inferences. Robustness checks were implemented through LASSO regression, a machine learning method used for variable selection and shrinkage to prevent overfitting in models with many potential covariates. LASSO helps identify a parsimonious set of predictors by shrinking less important coefficients toward zero, thereby addressing potential multicollinearity.

To examine whether physician-patient communication efficacy mediated the association between OHISB and patient adherence, we conducted a mediation analysis using the bootstrap method with 5000 resamples [13]. The indirect effect (product of the coefficient for OHISB to efficacy and efficacy to adherence) and its 95% CI were calculated. Mediation was considered statistically significant if the CI did not include 0.

This study was approved by the ethics committee of Tongji Medical College of Huazhong University of Science and Technology (IORG0003571). All procedures performed were in accordance with the ethical standards of the institutional research committee and with the 1964 Declaration of Helsinki and its later amendments. Participation was entirely voluntary and anonymous; no personally identifiable information was collected. All data were kept strictly confidential and used solely for research purposes. Electronic informed consent was obtained from all individual participants included in the study. No financial compensation was provided to the participants.

The final sample included 7004 rural participants. OHISB distribution was 12.4% (867/7004) frequent users, 18.3% (1280/7004) occasional users, and 69.3% (4857/7004) nonusers. Mean scores were 3.43 (SD 1.41) for patient adherence and 4.32 (SD 0.98) for physician-patient communication efficacy. Regarding health information platforms (percentages calculated among OHISB users: 2147/7004, 30.7%), short video apps showed the highest use (1597/2147, 74.4%), followed by WeChat official accounts (575/2147, 26.8%), whereas specialized platforms demonstrated minimal adoption (hospital-affiliated portals: 39/2147, 1.8%; dedicated medical platforms: 47/2147, 2.2%).

Table 1 presents the demographic and health-related characteristics of the 7004 participants, including age, educational attainment, income, employment status, region, chronic disease status, self-rated health, and lifestyle factors (breakfast frequency, smoking, drinking, exercise, and sleep quality).

Table 2 summarizes the distribution of the core study variables, including OHISB frequency, patient adherence scores, physician-patient communication efficacy scores, and the use of different platform types for health information seeking (percentages calculated among OHISB users in this case).

As shown in Table 3, OHISB was significantly and positively associated with patient adherence across all models.

In the unadjusted model (model 1), the association was strong (β=0.356; P<.001). Incorporating regional fixed effects in model 2 yielded consistent estimates (β=0.354; P<.001), confirming model robustness. Regional fixed effects were applied at the province level. Model 3 further adjusted for covariates, including age, income, and educational attainment, maintaining statistical significance (β=0.260; P<.001). This corresponds to a 0.260-unit increase in adherence per 1-unit increment in OHISB frequency. Significant covariates associated with higher adherence included older age, higher educational attainment, better self-rated health, consistent breakfast habits, and greater exercise frequency.

Model 4 used LASSO regression for robustness checks. The regularization procedure retained significant covariates: age, educational level, health (EQ-5D), weekly frequency of breakfast, and exercise frequency. OHISB showed a consistent positive association with patient adherence (β=0.268; P<.001) relative to fixed-effects models. Control variable coefficients maintained directional and statistical alignment with baseline estimates.

The results of the 4 models demonstrated the robustness of the association between OHISB and patient adherence.

To examine the mechanism through which OHISB may be associated with patient adherence, a mediation analysis was conducted using the bootstrap method with 5000 resamples. The results, presented in Table 4, delineate the direct, indirect, and total associations.

Path a (OHISB–communication efficacy) was negative and significant (β=−0.083; P<.001, 95% CI −0.117 to –0.048). Path b (communication efficacy–adherence) was positive and significant (β=0.310; P<.001, 95% CI 0.278-0.342). The indirect effect (a × b) was negative and significant (β=−0.026; P<.001, 95% CI −0.039 to −0.013), accounting for 9.29% of the total association. The direct effect (c’: OHISB–adherence, controlling for efficacy) remained positive and significant (β=0.301; P<.001, 95% CI 0.255-0.348), indicating inconsistent mediation.

This indirect association accounted for approximately 9.29% of the total association between OHISB and adherence, indicating an inconsistent mediation pattern (ie, the indirect association opposes the direction of the direct positive association) [13,14]. Figure 1 shows the mediation model with standardized path coefficients.

All interaction terms attained statistical significance (Table 5). Positive moderating associations were observed for hospital-affiliated portals (β=0.099; P=.04), specialized medical platforms (β=0.081; P=.04), and WeChat official accounts (β=0.032; P=.03), with hospital-affiliated portals showing the largest effect size. Negative moderating associations were observed for short video apps (β=−0.034; P=.006) and search engines (β=−0.204; P<.001), where search engines demonstrated the most pronounced attenuating pattern.

The reference category for each interaction term was nonuse of the respective platform type. For example, the coefficient for OHISB × short video apps (β=−0.034) indicates that, compared with nonusers of short video platforms, users of these platforms showed a weaker positive association between OHISB and adherence.

OHISB frequency was positively associated with patient adherence, consistent with prior research [15,16]. This finding supports the HBM [17], suggesting that information seeking may enhance perceived treatment benefits and self-efficacy, directly promoting adherence intentions. The adoption of OHISB is significantly lower in rural populations compared with reported rates in urban settings, and the joint use of hospital-affiliated platforms and professional health information platforms is relatively low in these populations [2,18]. This disparity likely stems from overlapping digital divides: deficient rural infrastructure, technological barriers among older adults, and terminology-dense content that raises cognitive thresholds [19]. The high health literacy demands of professional platforms further limit their potential benefits [20]. Future initiatives must optimize platform usability and deliver comprehensible, tailored health messaging to rural populations.

Our analysis suggests a dual-process pattern: while OHISB was directly and positively associated with adherence, a simultaneous negative indirect association was observed through reduced physician-patient communication efficacy [21]. This finding must be interpreted with nuance; it does not imply that online information seeking is inherently harmful [22]. Rather, it highlights a critical contingency: the net benefit of OHISB may be weaker when the information environment or the clinical encounter creates specific risks. Reduced communication efficacy may be more likely under circumstances where (1) information from online sources is fragmented, contradictory, or of low credibility, which may contribute to patient confusion or misplaced confidence [23]; (2) patients experience “information overload,” a state that increases anxiety and hinders effective information processing during consultations [23]; or (3) a power dynamic is activated wherein clinicians perceive patient-initiated information as challenging their authority, prompting defensive communication [24]. For health system design, these findings suggest that promoting patient empowerment through digital tools should be coupled with parallel interventions to improve information quality; scaffold patient-clinician dialogue; and train clinicians in collaborative, information-integrated consultation styles. The indirect effect, while statistically significant, accounted for only 9.29% of the total association, suggesting that the direct positive association predominated.

The moderating effects of platform types exhibited significant heterogeneity. Hospital-affiliated portals, specialized medical platforms, and WeChat official accounts strengthened the positive association between OHISB and patient adherence, whereas short video apps and search engines weakened it, corroborating previous findings [25]. Mechanistically, this divergence likely stems from differential information credibility, content quality, and interaction paradigms. Hospital-affiliated and specialized platforms enforce rigorous content curation to deliver evidence-based structured information congruent with clinical standards, enabling patients to integrate information into shared decision-making [26]. WeChat official accounts further facilitate sustained physician-patient interaction that contextualizes health information. Conversely, despite dominating use in this study (1597/2147, 74.4%), short video and search platforms propagate fragmentary, entertainment-focused content (eg, treatment claims emphasizing efficacy while omitting risks). Their recommendation algorithms may create echo chambers that increase exposure to homogeneous viewpoints and amplify cognitive biases [24]. Concurrently, their attention-grabbing formats trivialize complex health topics, whereas inconsistent information quality exposes users to contradictory perspectives, generating information overload [22]. This complexity may induce decisional uncertainty that ultimately attenuates the benefits of information seeking.

However, the observed platform-specific differences should be interpreted with caution as they may partly reflect unmeasured confounding factors. For instance, users of different platforms may differ systematically in digital literacy, health literacy, access to alternative information sources, or underlying health status. Short video platforms, which dominate rural use, may attract users with lower digital literacy or less engagement with formal health care, whereas those who actively seek out hospital-affiliated portals may have higher health literacy or more complex health needs. Such self-selection could contribute to the differential associations observed. Future research should measure and adjust for these potential confounders to isolate the unique moderating effects of platform characteristics.

This study has several limitations. First, its cross-sectional design precludes causal inference; longitudinal or experimental studies are needed to establish temporal sequences and strengthen causal claims. Consequently, terms such as “effect,” “impact,” and “pathway” are used descriptively to reflect statistical associations rather than causal relationships. Second, the use of a single-item, 3-point scale to measure OHISB, while practical, lacks granularity and conceptual specificity. Third, patient adherence was measured using a single item assessing behavioral intention rather than actual behavior, which may overestimate adherence due to social desirability bias. Test-retest reliability was not assessed, which represents an additional limitation. Future work should use validated, multidimensional scales to capture this complexity and provide a more nuanced understanding of the OHISB and patient adherence construct.

Drawing on cross-sectional survey data collected during 2023 and 2024, this study investigated OHISB and patient adherence, with particular emphasis on the mediating role of physician-patient communication efficacy and moderating associations with platform type. Three principal findings emerged.

First, OHISB frequency was positively associated with patient adherence, wherein higher seeking frequency corresponded to greater adherence. Second, physician-patient communication efficacy was observed as a negative indirect association whereby OHISB was associated with reduced communication efficacy, which in turn was associated with lower adherence—partially offsetting the direct positive association. However, this indirect association accounted for only 9.29% of the total association, indicating that the direct positive association predominated. Third, the association between information seeking and adherence varied significantly by platform type: hospital-affiliated portals, specialized medical platforms, and WeChat official accounts showed positive moderating associations, whereas short video apps and search engines showed attenuating associations that diminished overall empowerment returns.

These findings align with the HBM. The direct positive association supports the HBM’s premise that information seeking enhances perceived benefits and self-efficacy. The negative indirect association is consistent with the concept of perceived barriers: information overload and uncertainty may impede effective patient-clinician interaction. Platform-specific moderation suggests that information quality and credibility shape whether seeking behavior reduces or amplifies these barriers.

These findings suggest several implications for policy and practice, although these should be considered suggestive rather than definitive given the observational study design. First, to leverage the direct positive association and address low adoption rates, digital health equity initiatives should include the co-design of culturally and linguistically adapted health information content (eg, developing plain-language, dialect-compatible health materials for low-literacy users) with rural residents. Second, to address the negative indirect association, health systems might consider dual-focused interventions: (1) for patients, brief, scalable “health information literacy” modules to build skills for critical evaluation of online content; and (2) for clinicians, training in communication strategies to acknowledge, discuss, and integrate patient-sourced information constructively. Third, to respond to the heterogeneous platform effects, a graded regulatory framework could be considered, including accreditation incentives for authoritative content on professional platforms and algorithmic governance measures for short video and search platforms (eg, mandatory source disclosure for health claims and downgrading of unverified medical advice).

These findings and implications are grounded in a rural Chinese context and may not be directly generalizable to other settings. Future research should use longitudinal designs, validated multidimensional measures, and diverse populations to extend and refine these observations.