Digital Health Engagement Regarding Lumbar Disc Herniation in Terms of Exercise Adherence: Integrating Variable-Centered and Person-Centered Perspectives

Abstract

Background

Exercise adherence is pivotal for Lumbar Disc Herniation (LDH) rehabilitation, yet remains suboptimal. Digital health offers support, but patient engagement varies significantly. Current research lacks an integrated perspective examining variable-centered general mechanisms and person-centered heterogeneity in digital health engagement and adherence.

Aim

This study aimed to: (1) elucidate the variable-centered mechanism by testing the mediating role of health information seeking behavior (HISB) between eHealth literacy and exercise adherence; and (2) identify person-centered latent profiles of digital health engagement to predict adherence levels.

Methods

This cross-sectional study (from July to November 2025) involved 346 LDH patients. Digital health engagement was operationalized using the eHealth Literacy Scale (eHEALS) and the Health Information Seeking Behavior Scale (HISB). Exercise adherence was assessed using a validated LDH-specific scale. Data were analyzed using Structural Equation Modeling (SEM) and Latent Profile Analysis (LPA).

Results

SEM confirmed eHealth literacy positively predicted adherence directly and indirectly via HISB (mediation effect: 30.6%). LPA identified four profiles: “Digital Disengaged” (36.4%), “Passive Spectators” (36.1%), “Digital Champions” (22.3%), and “Distress-Driven Seekers” (5.2%). While youth and higher socioeconomic status predicted “Digital Champions”, severe functional disability (ODI) was the strongest predictor for “Distress-Driven Seekers” (OR=2.76, P<0.05). “Distress-Driven Seekers” achieved high adherence, comparable to “Digital Champions” (P>0.05) and superior to others (P<0.001).

Conclusion

HISB serves as a crucial mechanism linking eHealth literacy to rehabilitation adherence. Severe disability distress drives high-intensity information seeking, compensating for low literacy and enabling high adherence in “Distress-Driven Seekers”. Personalized digital health interventions should be developed to accommodate diverse engagement patterns. Notably, “Distress-Driven Seekers” require targeted clinical guidance and “information prescriptions” to mitigate risks of medical misinformation, health anxiety, and potential secondary injuries caused by inappropriate self-management.

Background

Lumbar Disc Herniation (LDH) represents a prevalent musculoskeletal disorder and a leading cause of years lived with disability (YLDs) globally.¹ With approximately 266 million individuals diagnosed with degenerative lumbar spine disease and low back pain annually,² it imposes a substantial personal and healthcare burden, constituting a significant global public health challenge.^1,2 Effective conservative management encompasses a diverse array of modalities, ranging from advanced physical interventions like high- and low-intensity laser therapy³ to structured physical exercise rehabilitation. While exercise is widely recognized as the gold standard for alleviating symptoms and preventing recurrence, its therapeutic efficacy is heavily contingent upon patients’ long-term adherence.⁴ However, exercise adherence among LDH patients remains far from ideal. Most patients struggle to maintain regular exercise, with adherence rates varying widely from 8% to 91%.⁵ In clinical practice, exercise therapy is predominantly conducted in home settings, rendering patients highly susceptible to impediments such as kinesiophobia, lack of professional guidance, and inadequate dynamic support.⁶ This not only disrupts the rehabilitation process but also constitutes a critical bottleneck leading to recurrent symptom flare-ups and chronicity. Consequently, enhancing patient adherence to rehabilitation exercise regimens has emerged as an urgent priority within the field of spinal disorder rehabilitation.

With the rapid development of mobile internet and digital technologies, digital health has emerged as a key strategy for optimizing global healthcare resource allocation, demonstrating significant advantages in the field of musculoskeletal pain and rehabilitation. Systematic reviews and meta-analyses indicate that such interventions, characterized by high reach, low cost, and ease of access, are effective in alleviating pain, improving physical function, and promoting self-management.⁷ Recent randomized controlled trials have further confirmed that AI-assisted multimodal telerehabilitation programs not only effectively resolve the challenge of lacking professional supervision outside clinical settings but also yield superior therapeutic outcomes compared to conventional exercise telerehabilitation.⁸ Moreover, digital health interventions have been shown to achieve clinical efficacy comparable to in-person physiotherapy, with significantly lower patient dropout rates.⁹ By leveraging smartphone applications, wearable devices, and online platforms, these tools provide personalized exercise guidance and real-time feedback, thereby empowering patients to participate more actively in their rehabilitation. This paradigm shift highlights the concept of “digital health engagement”, which encompasses not merely the passive reception of services but the active utilization of digital resources for health management.¹⁰ As patients increasingly rely on the internet for rehabilitation guidance, understanding how they interact with these digital environments is critical for optimizing adherence outcomes.

However, effective digital health engagement does not occur automatically; personal capacity and literacy are fundamental prerequisites for such participation.¹⁰ Defined as “the ability to seek, find, understand, and appraise health information from electronic sources”,¹¹ eHealth literacy is a cornerstone of digital empowerment. Although existing evidence confirms that higher eHealth literacy facilitates the accessibility and adoption of digital technologies,^12,13 literacy essentially represents a “competence” or potential skill rather than actual “performance”.¹⁴ Classic research by Van der Vaart et al revealed a significant gap between patients’ self-reported literacy levels and their actual performance in executing internet health tasks.¹⁵ Furthermore, the theory of the “third-level digital divide” by Scheerder et al emphasizes that possessing skills does not automatically translate into beneficial usage outcomes.¹⁶ This suggests that digital skills alone do not guarantee that patients will actively utilize technology to support their exercise routines, underscoring the existence of intermediary behavioral mechanisms bridging this gap.

To bridge the gap between competence and actual performance, Health Information Seeking Behavior (HISB) is identified as a critical pathway connecting this chain. According to Anker et al, HISB is not merely a concrete manifestation of digital literacy but a dynamic process wherein patients purposefully acquire knowledge to cope with health challenges.¹⁷ Online health information seeking specifically refers to individuals using the internet to search for information regarding health, disease risks, and protective behaviors.¹⁸ The rapid proliferation of internet-based health resources and social media has fundamentally transformed how the public accesses and shares health-related information.¹⁹ Particularly in the management of musculoskeletal conditions such as lumbar disc herniation, this active seeking behavior constitutes a core strategy for patients to cope with pain and reduce uncertainty about their disease.^20,21 The Information-Motivation-Behavioral Skills (IMB) model provides a solid theoretical foundation for understanding this translational mechanism.²² Within this framework, eHealth literacy serves as a foundational capacity that triggers active information seeking (Information), enabling patients to obtain precise rehabilitation guidance and alleviate fears regarding exercise, thereby promoting adherence behaviors. Existing empirical research supports this mediation effect; Li et al found that HISB effectively transforms static literacy into dynamic health management capabilities,²³ confirming that active seeking acts as a pivotal bridge translating knowing into doing. Therefore, analyzing the mediating pathway of HISB holds significant theoretical and practical value for unlocking how digital capabilities translate into clinical adherence.

Although online health information seeking is generally considered a facilitator of health behaviors, existing evidence regarding its association with treatment adherence exhibits significant heterogeneity.^24,25 Specifically, the direction of this association can be both positive and negative. While appropriate information seeking empowers patients and reduces uncertainty, excessive or unguided searching can lead to information overload and exacerbate health anxiety, often associated with cyberchondria, which may subsequently result in avoidance behaviors and poor adherence.²⁶ This strongly suggests the presence of latent moderating effects or unobserved subgroup differences. However, traditional “variable-centered” studies often rely on mean-based assumptions of uniform linear laws, thereby masking the “person-centered” heterogeneity objectively present within patient populations.²⁷ In clinical reality, patients’ literacy and behaviors do not exist in isolation but are intertwined in complex patterns. Empirical evidence, ranging from decision-making subgroups in stroke patients^28,29 and technology “ Adopters” stratification in diabetes³⁰ to the digital divide in broader populations,^31,32 confirms that multidimensional feature combinations (rather than single variables) determine actual behavior. Notably, Shen et al highlighted that unbalanced development of eHealth literacy dimensions leads to significantly reduced health behavior performance.³³ Although Latent Profile Analysis (LPA), a person-centered taxonomic method, has proven to be a powerful tool for uncovering such heterogeneity in the aforementioned fields,³⁴ there is a paucity of research applying it to LDH rehabilitation. Whether distinct latent profiles of digital engagement exist within this population, and how these complex feature configurations differentially shape exercise adherence, remains an empirical gap urgently needing to be filled.

Therefore, the present study aims to integrate both variable-centered and person-centered perspectives to provide a comprehensive understanding of digital health engagement in LDH rehabilitation. Specifically, the objectives are to: (1) elucidate the behavioral mechanism by testing the mediating role of HISB in the relationship between eHealth literacy and exercise adherence using Structural Equation Modeling (SEM); (2) characterize the heterogeneity of digital engagement by identifying distinct latent profiles based on patients’ literacy and seeking behaviors using LPA; and (3) examine the association between these profiles and exercise adherence, while exploring the sociodemographic and clinical determinants of profile membership. Findings from this study will provide empirical evidence for developing tailored, precision digital interventions to optimize rehabilitation outcomes.

Methods

Study Design and Participants

This study was a web-based cross-sectional survey aimed at exploring associations between variables and population heterogeneity, rather than establishing causal relationships. Using convenience sampling, patients with lumbar disc herniation attending the outpatient clinics of Tuina (Chinese Massage), Orthopedics, and Rehabilitation at a tertiary hospital in Zhejiang Province were recruited between July and November 2025. It should be explicitly noted that while this single-center convenience sampling approach facilitated standardized data collection, it inherently limits the generalizability of our findings to broader patient populations. The study procedure adhered to the principles of the Declaration of Helsinki and was approved by the Ethics Board of Hangzhou TCM Hospital Affiliated to Zhejiang Chinese Medical University (No. 2025KLL186). Recruitment was conducted in the outpatient waiting areas by uniformly trained research assistants, who explained the study’s purpose and requirements in detail and emphasized the anonymity of responses. Patients provided informed consent and participated voluntarily. Upon obtaining consent, patients completed the electronic questionnaire either independently or with non-directive assistance from the research assistants. Questionnaire data were uploaded directly to a secure cloud database via an encrypted network, ensuring privacy and confidentiality throughout the data collection process. Data were screened to exclude incomplete questionnaires, resulting in a final sample size of 346.

Inclusion Criteria: (1) Age ≥ 18 years; (2) Clinically diagnosed with LDH via imaging,³⁵ currently in the non-surgical conservative treatment or postoperative rehabilitation phase (≥ 3 months); (3) Prescribed a specific home-based rehabilitation exercise plan by a physician; (4) Capable of using a smartphone and understanding the survey content; and (5) Provided informed consent. This combined cohort was purposefully selected to capture a comprehensive spectrum of digital health engagement behaviors across the continuum of LDH care, as both patient groups face parallel psychological and behavioral challenges regarding home-based exercise adherence and online health information seeking. Exclusion Criteria: (1) Comorbid spinal pathologies (eg, lumbar spinal stenosis, instability, fractures, tuberculosis, tumors, or ankylosing spondylitis); (2) Cognitive impairment or history of psychiatric disorders preventing questionnaire completion; (3) Severe somatic comorbidities (eg, malignancy, cardiopulmonary failure) contraindicating exercise; (4) Concurrent participation in other clinical trials.

Sample size estimation integrated a priori power analysis with methodological guidelines for latent variable modeling. First, an a priori power analysis was conducted using G*Power 3.1 software. Based on a multiple linear regression model, the parameters were set as follows: a significance level (α) of 0.05, a statistical power (1-β) of 0.95, and a medium effect size (f²) of 0.15. The model included 11 predictor variables, covering sociodemographics, disease-specific clinical characteristics, eHealth literacy, and health information-seeking behaviors. The calculation indicated a minimum sample size of 178 for regression analysis (increasing to 192 after accounting for a 10% invalid response rate). However, given that this study involves LPA, methodological recommendations suggest a minimum sample size of 300 to ensure model estimation stability.³⁶ Taking the higher threshold into account to guarantee statistical robustness, and adjusting for a 10% potential invalid response rate, the final target sample size was set at a minimum of 334. Ultimately, 350 questionnaires were collected. After excluding 4 responses containing logical inconsistencies, 346 valid questionnaires were retained, exceeding the minimum sample size requirement.

Measures

General Information

A self-designed questionnaire collected demographic and clinical data, including gender, age, BMI, marital status, education, economic status, occupational status, duration of diagnosis, and functional disability (Oswestry Disability Index, ODI).

eHealth Literacy

eHealth literacy was assessed using the Chinese version of the eHealth Literacy Scale (eHEALS), originally developed by Norman et al³⁷ and subsequently translated and validated by Guo et al³⁸ This unidimensional instrument measures an individual’s combined skill set in using information technology for health purposes, specifically focusing on the ability to find, understand, and evaluate electronic health information. The scale comprises 8 items (eg, “I know how to use the Internet to answer my health questions”) rated on a 5-point Likert scale ranging from 1 (“Strongly Disagree”) to 5 (“Strongly Agree”). The total score ranges from 8 to 40, with higher scores indicating a higher level of perceived eHealth literacy. In the present study, the scale demonstrated excellent internal consistency with a Cronbach’s α of 0.983.

Health Information Seeking Behavior

Health Information Seeking Behavior (HISB) was assessed using the scale originally developed by Zamani et al³⁹ and subsequently translated and validated in Chinese by Sun et al⁴⁰ The scale comprises 43 items distributed across four dimensions: attitude toward information seeking (6 items), information needs (14 items), information sources (15 items), and low barriers (8 items). Items are rated on a 5-point Likert scale ranging from 1 (“Very Unimportant”) to 5 (“Very Important”). Items within the “barriers” dimension (items 36–43) were reverse-coded to ensure that higher values reflect lower perceived barriers. The final score is derived from the mean of all items, with higher scores indicating a more active level of health information seeking behavior. In the present study, the scale demonstrated excellent internal consistency (Cronbach’s α = 0.968).

Exercise Adherence

Exercise adherence was assessed using the Exercise Compliance Scale for Lumbar Disc Herniation Patients, developed and validated by Gao et al (2025) specifically for patients undergoing conservative treatment.⁴¹ It is worth noting that while the original scale is titled “compliance”, its specific dimensions, comprising preparation, exercise, supervision, and recommendation adherence, align closely with the concept of “adherence”, which emphasizes active patient participation. Therefore, consistent with modern rehabilitation terminology, the term “exercise adherence” is adopted throughout this study. Constructed upon the Health Belief Model and the conceptual framework of rehabilitation training adherence, the scale consists of 20 items across the four aforementioned dimensions. Items are rated on a 5-point Likert scale ranging from 1 (“Completely unable to do”) to 5 (“Completely able to do”). The total score ranges from 20 to 100, with higher scores indicating a stronger level of exercise adherence.

Although this instrument was originally developed for patients undergoing conservative treatment, it focuses on generalized behavioral constructs of compliance (eg, execution frequency and barrier management) rather than specific biomechanical movements. Given that all participants in this study followed a prescribed home-based regimen, the scale is conceptually applicable across our combined cohort. To empirically validate its use, a sensitivity analysis demonstrated excellent internal consistency across both clinical subgroups (Cronbach’s α = 0.956 for the conservative subgroup and 0.967 for the postoperative subgroup). In the present study, the scale demonstrated excellent internal consistency, with a Cronbach’s α of 0.962.

Data Collection and Quality Control

Investigators underwent standardized training before guiding patients to complete the electronic survey via the “Wenjuanxing” platform. To ensure data quality, all items were mandatory with embedded attention checks, and submissions were restricted to one per IP/WeChat account. On-site investigators provided non-directive assistance and immediate completeness verification. Participants received health educational materials as compensation. Post-collection, data underwent dual cross-checking and logical screening to exclude invalid responses, strictly adhering to anonymity and confidentiality standards throughout the process.

Data Analysis

Analyses were performed using IBM SPSS Statistics 27.0 and R 4.3.3. The significance level was set at two-tailed p < 0.05.

1. Descriptive and Correlation Analysis: Continuous variables were presented as Mean ± Standard Deviation (M±SD) or Median (Interquartile Range, IQR) based on normality, while categorical variables were expressed as frequencies and percentages. Pearson correlation analysis was conducted to examine the bivariate associations among eHEALS, HISB, and exercise adherence.

2. Mediation Analysis (Variable-Centered): To ensure methodological robustness, structural equation modeling (SEM) was conducted using the lavaan package in R. A path analysis model was constructed to test the mediating role of HISB between eHealth literacy and exercise adherence, adjusting for sociodemographic and clinical covariates (eg, age, ODI). Parameters were estimated using Maximum Likelihood. The significance of indirect effects was assessed using bias-corrected bootstrapping with 5000 resamples, where a 95% Confidence Interval (CI) not including 0 indicated statistical significance.

3. Latent Profile Analysis (Person-Centered): LPA was performed using the tidyLPA package in R to identify distinct patient subgroups based on the standardized Z-scores of eHEALS and HISB total scores. Models ranging from 1 to 5 profiles were fitted with specifications of equal variances and zero covariances. Model selection was guided by a combination of fit indices: lower Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and sample-size adjusted BIC (aBIC); higher Entropy (indicating classification precision); significant p-values for the Lo-Mendell-Rubin Likelihood Ratio Test (LMR-LRT) and Bootstrapped Likelihood Ratio Test (BLRT); and a theoretical interpretability with a minimum profile size > 5%.

4. Profile Predictors and Outcome Comparison: First, univariate analyses (χ² tests for categorical variables) were used to identify sociodemographic and clinical characteristics that differed across profiles. Significant variables were then entered into a multinomial logistic regression model to identify predictors of profile membership, utilizing the largest profile as the reference category. Odds Ratios (OR) and 95% CIs were calculated. Finally, the impact of profile membership on exercise adherence was examined. One-way Analysis of Variance (ANOVA) was employed to assess differences in adherence scores across the identified latent profiles. Post-hoc pairwise comparisons were subsequently conducted using Tukey’s Honest Significant Difference (HSD) test.

Results

Common Method Bias Test

Given that the data were derived from patient self-reports, Harman’s single-factor test was employed to assess potential common method bias. All measurement items of the core variables were subjected to an unrotated exploratory factor analysis. The results yielded seven factors with eigenvalues greater than 1. The first unrotated factor accounted for 32.68% of the total variance, which is well below the critical threshold of 40%. Consequently, common method bias was not considered a significant issue in this study.

Demographic Information of Participants

A total of 346 patients with LDH were included in the final analysis. The mean age of the participants was 51.30±16.85 years, with females accounting for 52.9%. Regarding sociodemographic characteristics, the majority held a college or undergraduate degree (57.5%), 72.5% were currently employed, and more than half (55.8%) reported a medium self-perceived economic status. In terms of functional disability based on ODI scores, 51.7% were classified as having minimal disability, 32.9% as moderate disability, and 15.3% as severe disability and above.

Univariate analysis revealed significant differences in exercise adherence across various groups: patients who were younger (18–64 years), had higher educational attainment, reported better economic status, were employed, and presented with higher ODI scores (indicating more severe functional disability) demonstrated significantly higher levels of exercise adherence (all P < 0.05). Specifically, participants aged 18–64 years showed significantly higher exercise adherence than those aged ≥65 years (72.03±10.84 vs 60.44±7.96, P < 0.001). Regarding LDH-specific clinical characteristics, duration of LDH was not significantly associated with exercise adherence (P = 0.440), suggesting that disease chronicity alone may not explain adherence differences in this cohort. In contrast, exercise adherence differed significantly across ODI-based disability levels (P < 0.001). Patients with severe disability and above reported the highest adherence scores, followed by those with moderate disability and minimal disability (73.20±9.83, 67.46±11.45, and 60.25±9.78, respectively). Conversely, variables such as gender, BMI, marital status, and duration of LDH showed no significant association with exercise adherence (all P > 0.05), suggesting that disease chronicity was not significantly related to adherence differences in this study. Detailed results are presented in Table 1.

Table 1.

Demographic Characteristics of Participants (n=346)

Variable	Number (%)	Exercise Adherence (M±SD)	t/F	P
Gender			1.194	0.233
Male	163 (47.1)	70.09±11.25
Female	183 (52.9)	68.63±11.42
Age(years)			8.912	<0.001
18~64	265 (76.6)	72.03±10.84
≥65	81 (23.4)	60.44±7.96
BMI (kg/m2)			0.563	0.640
Underweight	73 (21.1)	70.12±12.81
Normal weight	145 (41.9)	68.66±10.73
Overweight	85 (24.6)	68.99±10.86
Obesity	43 (12.4)	70.84±11.97
Marital Status			0.938	0.392
Single	45 (13.0)	70.27±13.12
Married	269 (77.7)	69.46±10.95
Divorced or Widowed	32 (9.3)	66.84±12.0
Education			92.806	<0.001
High school or below	100 (28.9)	58.91±7.48
Associate degree or bachelor’s degree	199 (57.5)	72.97±10.07
Master’s degree and above	47 (13.6)	76.02±8.32
Economic Status			19.220	<0.001
Weak	48 (13.9)	60.75±7.64
Medium	193 (55.8)	69.87±10.74
Good	105 (30.3)	72.24±12.03
Occupational status			2.388	0.017
Employed	251 (72.5)	70.21±11.23
Unemployed	95 (27.5)	66.97±11.37
Duration of LDH(years)			0.902	0.440
<3	171 (49.4)	68.68±11.64
3~9	99 (28.6)	70.78±10.79
10~19	24 (6.9)	70.08±10.06
≥20	52 (15.0)	69.32±11.35
ODI (%)			34.552	<0.001
Minimal Disability	179	60.25±9.78
Moderate Disability	114	67.46±11.45
Severe Disability and above	53	73.20±9.83

Note: The bold values indicate statistically significant results (P < 0.05).

Abbreviations: BMI, body mass index; ODI, Oswestry Disability Index.

Variable-Centered Perspective: Correlation and Mediation Analysis

Descriptive statistics for the core study variables were as follows: the mean score for eHealth literacy was 24.53±5.39, for HISB was 3.42±0.23, and for exercise adherence was 69.32±11.35. According to the criteria proposed by Kline (2023), the skewness and kurtosis values for all core variables fell within the acceptable range for normal distribution. As shown in Table 2, Pearson correlation analysis revealed significant positive associations among eHealth literacy, HISB, and exercise adherence (all P < 0.001).

Table 2.

Variable Correlation Analysis

Variables	1. eHEALS	2. HISB	3. Exercise Adherence
1. eHEALS	1
2. HISB	0.428***	1
3. Exercise Adherence	0.483***	0.549***	1
M±SD	24.53±5.39	3.42±0.23	69.32±11.35
Kurtosis	−0.826	0.623	−0.912
Skewness	0.234	−0.140	0.588

Note: ***P < 0.001.

A mediation analysis was conducted within the SEM framework using the lavaan package in R, treating eHealth literacy as the independent variable, HISB as the mediator, and exercise adherence as the dependent variable. The model controlled for covariates including age, educational attainment, economic status, occupational status, and functional disability (ODI). Path analysis indicated an excellent model fit (CFI = 1.000, TLI = 1.000, RMSEA = 0.000, SRMR = 0.000). The results of the path coefficients are detailed in Figure 1 and Table 3. eHealth literacy significantly and positively predicted HISB through path a (β = 0.320, SE = 0.002, P < 0.001). After including HISB as the mediator, eHealth literacy still significantly predicted exercise adherence through the direct path c′ (β = 0.186, SE = 0.089, P < 0.001) Additionally, HISB significantly and positively predicted exercise adherence through path b (β = 0.256, SE = 1.497, P < 0.001). Bias-corrected bootstrapping analysis (5000 resamples) revealed that the indirect effect of eHealth literacy on exercise adherence via HISB was significant (unstandardized effect = 0.172, 95% CI [0.114, 0.238]). Given that the direct effect of eHealth literacy on adherence remained significant, HISB was identified as a partial mediator. The total effect of the model was 0.563. The mediation effect accounted for 30.6% of the total effect, suggesting that approximately one-third of the influence of eHealth literacy on adherence is realized through active information-seeking behaviors.

Figure 1.

The mediating effect path coefficient of health information seeking behavior.

Notes: X = eHealth literacy; M = health information seeking behavior; Y = exercise adherence. Path a represents the effect of eHealth literacy on health information seeking behavior; path b represents the effect of health information seeking behavior on exercise adherence after controlling for eHealth literacy; path c′ represents the direct effect of eHealth literacy on exercise adherence after including the mediator. *** P < 0.001. The model controlled for age, education, economic status, occupational status, and ODI.

Table 3.

Direct, Indirect, and Total Effects in the Mediation Model

Pathway	Unstandardized B	SE	Standardized β	95% CI
Direct Paths
eHEALS → HISB (a)	0.014	0.002	0.320***	[0.010, 0.017]
HISB → Adherence (b)	12.584	1.497	0.256***	[9.769, 15.525]
eHEALS → Adherence (c’)	0.391	0.089	0.186***	[0.221, 0.566]
Effects
Total Effect	0.563	0.090	0.268***	[0.386, 0.736]
Indirect Effect (Mediation)	0.172	0.032	0.082***	[0.114, 0.238]

Notes: Path a represents the effect of eHealth literacy on HISB; path b represents the effect of HISB on exercise adherence after controlling for eHealth literacy; path c’ represents the direct effect of eHealth literacy on exercise adherence after including HISB as the mediator. The indirect effect was calculated as a×b, and the total effect was calculated as c’+ a×b. *** P < 0.001. Analysis controlled for age, education, economic status, occupational status, and ODI.

Abbreviations: SE, Standard Error; CI, Confidence Interval; HISB, Health Information Seeking Behavior.

Person-Centered Perspective

Latent Profile of Digital Health Engagement Among LDH Participants

LPA was conducted using the standardized scores (Z-scores) of eHEALS and HISB. As presented in Table 4, the values for AIC, BIC, and aBIC demonstrated a decreasing trend as the number of profiles increased. After comprehensively considering both fit indices and classification interpretability, the 4-profile model was identified as the optimal solution. This model exhibited a relatively low BIC value (1881.26), the highest Entropy value (0.77), and a significant BLRT (P = 0.01), supporting its superiority over the 3-profile model. Although the LMR test was non-significant (P = 0.80), the 4-profile model was retained following a rigorous sensitivity evaluation. First, simulation studies have consistently demonstrated that the BLRT possesses greater power than the LMRT for identifying the correct number of classes.⁴² Second, the 4-profile solution exhibited excellent classification stability (verified via 5000 random starts) with average posterior probabilities for all classes exceeding 0.90. Third, and most importantly, prioritizing clinical interpretability was paramount. Retaining the 4-profile model successfully isolated the “Distress-Driven Seekers” (Profile 4), a clinically critical subgroup characterized by high distress and high information-seeking, which would have been obscured in a 3-profile solution, thereby reducing clinical utility. Furthermore, the smallest profile accounted for 5.2% of the sample, exceeding the conventional 5% threshold for adequate representation. Consequently, the 4-profile model was selected as the optimal solution.

Table 4.

Model Fit Statistics for Latent Profile Analysis

Models	AIC	BIC	aBIC	Entropy	LMR(P)	BLRT (P)	Profile Probability (%)
1	1969.81	1985.19	1972.50	1	–	–	100
2	1875.35	1902.27	1880.07	0.67	0.86	0.01	63.0/37.0
3	1881.8	1920.26	1888.54	0.47	0.17	0.98	53.8/37.6/8.7
4	1831.25	1881.26	1840.02	0.77	0.80	0.01	36.4/36.1/22.3/5.2
5	1825.83	1887.37	1836.62	0.72	0.74	0.02	29.5/22.5/22.0/21.1/4.9

The 4-profile model classified LDH patients into four distinct latent profiles based on their digital health engagement, with Figure 2 illustrating the standardized score characteristics of each. These profiles were identified as: “Digital Disengaged” (Low Literacy-Low Seeking, 36.4%, n=126), characterized by the lowest scores on both eHEALS and HISB; “Passive Spectators” (Adequate Literacy-Low Seeking, 36.1%, n=125), who displayed average eHealth literacy levels but relatively low seeking behavior, reflecting a knowing-doing gap; “Digital Champions” (High Literacy-High Seeking, 22.3%, n=77), demonstrating excellent performance on both indicators; and a unique subgroup labeled “Distress-Driven Seekers” (Low Literacy-High Seeking, 5.2%, n=18), which featured low eHealth literacy yet extremely high-intensity information seeking behavior, forming a notable divergent pattern. Furthermore, multidimensional feature analysis (Figure 3) revealed that these “Distress-Driven Seekers” scored the highest on the “Low Barriers” dimension of the HISB (Z > 2.0), indicating that despite their lower literacy levels, they perceived the lowest threshold for accessing health information.

Figure 2.

Standardized mean scores of eHEALS and HISB across 4 profiles.

Notes: The y-axis represents the standardized Z-scores calculated from the total scores of eHEALS and HISB. The lines illustrate the distinct patterns of digital engagement for each latent profile.

Figure 3.

Characterizing profiles by eHealth Literacy and HISB sub-dimensions.

Notes: In the figure, “Disengaged” refers to Digital Disengaged, “Passive” refers to Passive Spectators, “Champions”‘refers to Digital Champions, and “Distress-Driven” refers to Distress-Driven Seekers.

Factors Influencing Latent Digital Health Engagement Profiles in LDH Patients

Univariate analysis results (Table 5) indicated significant differences among the latent profiles regarding age, educational level, economic status, and ODI scores (all P < 0.05). Subsequently, multinomial logistic regression was employed to analyze the impact of sociodemographic and disease characteristics on profile membership, utilizing Profile 1 (“Digital Disengaged”) as the reference category (Table 6). Results revealed that for “Digital Champions” (Profile 3), age (OR=0.96, P<0.01), educational level (OR=3.05 for Level 2), and economic status (OR=5.67 for Level 2) were significant predictors, indicating that younger patients with higher socioeconomic status are more likely to be classified into this group. For “Distress-Driven Seekers” (Profile 4), the most prominent predictor was the ODI (OR=2.76, P=0.042). It is noteworthy that while Profile 2 also exhibited a significantly elevated ODI risk compared to the reference group (OR=1.52, P=0.017), the effect size for Profile 4 (OR=2.76) was considerably higher. This suggests that severe physical functional limitation is a critical driver compelling patients with low literacy to engage in high-intensity information seeking. Finally, due to the small size of Profile 4 (n=18), complete separation occurred, with zero observations in the reference categories for education and income. Consequently, odds ratios for these variables cannot be estimated (marked as NA). Rather than implying a lack of effect, this statistical artifact reflects an association so strong that all members of this profile belong to higher socioeconomic categories. While these NA values do not bias the model’s other estimates, they limit the exact quantification of socioeconomic predictors for this subgroup. Therefore, these results should be interpreted with caution, and replication in larger samples is warranted.

Table 5.

Sample Characteristics Across Digital Health Engagement Profiles (N = 346)

Variable	P1 (n=126)	P2 (n=125)	P3 (n=77)	P4 (n=18)	χ2	P
Gender					3.613	0.306
Male	58 (46.0%)	66 (52.8%)	33 (42.9%)	6 (33.3%)
Female	68 (54.0%)	59 (47.2%)	44 (57.1%)	12 (66.7%)
Age(years)					22.554	<0.001
18~64	87 (69.0%)	90 (72.0%)	71 (92.2%)	17 (94.4%)
≥65	39 (31.0%)	35 (28.0%)	6 (7.8%)	1 (5.6%)
BMI (kg/m2)					10.661	0.300
Underweight	21 (16.7%)	27 (21.6%)	20 (26.0%)	5 (27.8%)
Normal weight	55 (43.6%)	59 (47.2%)	27 (35.0%)	4 (22.2%)
Overweight	33 (26.2%)	29 (23.2%)	18 (23.4%)	5 (27.8%)
Obesity	17 (13.5%)	10 (8.0%)	12 (15.6%)	4 (22.2%)
Marital Status					5.890	0.436
Single	12 (9.5%)	15 (12.0%)	15 (19.5%)	3 (16.7%)
Married	99 (78.6%)	100 (80.0%)	57 (74.0%)	13 (72.2%)
Divorced or Widowed	15 (11.9)	10 (8.0%)	5 (6.5%)	2 (11.1%)
Education					45.045	<0.001
High school or below	53 (42.1%)	41 (32.8%)	6 (7.8%)	0 (0.0%)
Associate degree or bachelor’s degree	57 (45.2%)	70 (56.0%)	57 (74.0%)	15 (83.3%)
Master’s degree and above	16 (12.7%)	14 (11.2%)	14 (18.2%)	3 (16.7%)
Economic Status
Weak	27 (21.4%)	19 (15.2%)	2 (2.6%)	0 (0.0%)	22.475	<0.001
Medium	62 (49.2%)	72 (57.6%)	44 (57.1%)	15 (83.3%)
Good	37 (29.4%)	34 (27.2%)	31 (40.3%)	3 (16.7%)
Occupational status					0.732	0.866
Employed	92 (73.0%)	89 (71.2%)	58 (75.3%)	12 (66.7%)
Unemployed	34 (27.0%)	36 (28.8%)	19 (24.7%)	6 (33.3%)
Duration of LDH(years)					7.980	0.536
<3	66 (52.4%)	61 (48.8%)	37 (48.0%)	7 (38.9%)
3~9	29 (23.0%)	37 (29.6%)	28 (36.4%)	5 (27.8%)
10~19	12 (9.5%)	7 (5.6%)	3 (3.9%)	2 (11.1%)
≥20	19 (15.1%)	20 (16.0%)	9 (11.7%)	4 (22.2%)
ODI (%)					29.728	<0.001
Minimal Disability	30 (23.8%)	18 (14.4%)	5 (6.5%)	0 (0.0%)
Moderate Disability	49 (38.9%)	41 (32.8%)	19 (24.7%)	5 (27.8%)
Severe Disability and above	47 (37.3%)	66 (52.8%)	53 (68.8%)	13 (72.2%)

Notes: The result in the table is n (%). The bold values indicate statistically significant results (P < 0.05).

Abbreviations: BMI, body mass index; P1-3, Profile 1–3.

Table 6.

Multinomial Logistic Regression Predicting Profile Membership (n=346)

Predictors	Profile 2 vs. 1	Profile 3 vs. 1	Profile 4 vs. 1
	OR (95% CI)	OR (95% CI)	OR (95% CI)
Age	1.00 (0.98, 1.02)	0.96 (0.93, 0.98)**	0.96 (0.91, 1.01)
ODI (Disability)	1.52 (1.08, 2.16)*	2.16 (1.32, 3.55)**	2.76 (1.04, 7.33)*
Education	–	–	–
Level 1 (Ref)	1.00	1.00	1.00
Level 2	1.45 (0.79, 2.65)	3.05 (1.09, 8.53)*	NA^a
Level 3	0.94 (0.39, 2.28)	2.64 (0.80, 8.71)	NA^a
Economic Status	–	–	–
Level 1 (Ref)	1.00	1.00	1.00
Level 2	1.48 (0.74, 2.98)	5.67 (1.21, 26.65)*	NA^a
Level 3	1.13 (0.52, 2.48)	5.44 (1.12, 26.49)*	NA^a

Notes: The model controlled for age, education, economic status, and ODI. Significance: *P < 0.05, **P < 0.01. ^a Estimates for “Distress-Driven Seekers” (Profile 4) were not calculated due to complete separation caused by zero cell counts in specific categories (refer to Table 5).

Abbreviations: OR, Odds Ratio; CI, Confidence Interval; NA, Not Applicable.

Predicting Exercise Compliance Through Latent Profile Membership

ANOVA revealed significant overall differences in exercise adherence scores among the four digital health engagement profiles (F = 47.79, P < 0.01). Tukey’s HSD post-hoc tests elucidated specific patterns of difference (see Figure 4): the “Distress-Driven Seekers” (Profile 4) achieved the highest adherence scores (80.5±3.6), which were not significantly different from those of the “Digital Champions” (Profile 3) (77.9±8.8, P = 0.738). However, both groups demonstrated significantly higher adherence compared to “Passive Spectators” (68.8±9.2) and “Digital Disengaged” (63.0±10.9) (all P < 0.05). These findings indicate that despite possessing lower eHealth literacy, “Distress-Driven Seekers” attained a high level of adherence—equivalent to that of “Digital Champions”—through high-intensity behavioral engagement in information seeking.

Figure 4.

Comparison of exercise adherence scores across latent profiles.

Notes: In the figure, “Disengaged” refers to Digital Disengaged, “Passive” refers to Passive Spectators, “Champions”‘refers to Digital Champions, and “Distress-Driven” refers to Distress-Driven Seekers. Data are presented as Mean ± SE. Different lowercase letters (a, b, c) indicate statistically significant differences (p < 0.05) based on Tukey’s HSD post-hoc test. Groups sharing the same letter are not significantly different.

Discussion

To the best of our knowledge, this is the first study to elucidate the relationship between digital health engagement and exercise adherence in LDH patients by integrating variable-centered and person-centered perspectives. Our study initially confirmed the suboptimal status of exercise adherence among LDH patients, underscoring the urgency of investigating its underlying drivers. Building on this context, our findings offer two primary insights that extend current understanding. First, at the mechanistic level, we demonstrated that eHealth literacy is not merely a static capability but influences adherence primarily through the mediating process of active information seeking. This suggests that the translation of digital skills into health behaviors is heavily dependent on the patient’s proactive engagement with information. Second, and perhaps more significantly, our person-centered analysis uncovered substantial heterogeneity regarding digital health engagement within the patient population. While we identified typical profiles such as “Digital Disengaged”, “Digital Champions”, and “Passive Spectators”, the most notable discovery was the “Distress-Driven Seekers” subgroup. This finding highlights a counter-intuitive phenomenon where patients with lower digital literacy, driven by severe physical distress, utilize high-intensity searching as a compensatory strategy to achieve adherence levels comparable to their high-literacy peers. Collectively, these findings resonate with the person-based approach to intervention development,⁴³ providing a scientific basis for identifying heterogeneous subpopulations and determining tailored behavioral targets. This holds significant clinical value for facilitating the transition from generic education to precision self-management in disease rehabilitation.⁴⁴

From a variable-centered perspective, this study confirms the significant positive predictive effect of eHealth literacy on exercise adherence among LDH patients. This finding aligns with previous results in chronic disease management,^45,46 underscoring digital literacy as a foundational capability for maintaining health behavior adherence. However, the deeper contribution of our study lies in elucidating the intrinsic mechanism by which literacy promotes adherence, specifically through the mediating role of health information-seeking behavior. The mediation model indicates that active information acquisition acts as a critical bridge translating static digital literacy into dynamic health behaviors, a finding consistent with Lu et al’s research on patient adherence in online communities.⁴⁷ As noted in recent literature, individuals with higher eHealth literacy tend to possess greater digital self-efficacy, which motivates them to invest sufficient cognitive effort to seek and utilize health resources effectively.⁴⁸ Specifically, high-literacy patients are more inclined to proactively use online resources to resolve uncertainties during rehabilitation. Consistent with findings by Dong et al,⁴⁹ web-based health information seeking significantly reduces patients’ uncertainty regarding their condition and treatment, thereby fostering the confidence necessary to execute rehabilitation plans. Furthermore, the finding of partial mediation aligns with the complex understanding of the multidimensional role of eHealth literacy. Beyond driving information-seeking behaviors, the intrinsic components of literacy, including high-level information judgment and critical thinking, directly enhance adherence quality. As demonstrated by Park and Lee⁵⁰ in their study on chronic disease management, eHealth literacy is independently associated with various components of self-care adherence, enabling patients to accurately comprehend and apply rehabilitation guidelines even in the absence of additional searching.

From a person-centered perspective, our latent profile analysis unveils a critical reality: a “digital divide” predicated on disparities in digital literacy and behavior remains prevalent among patients with LDH.⁵¹ Specifically, Profile 1 (Digital Disengaged) and Profile 2 (Passive Spectators) collectively account for 72.5% of the sample, constituting the majority of the patient population. This high prevalence of low engagement aligns closely with recent findings by Oudbier et al, which highlight that despite the growing accessibility of digital health tools, actual adoption rates among the general patient population remain limited, revealing a significant disconnect between technological supply and actual patient integration.⁵² This reaffirms the perspective of digital inclusion as a social determinant of health, suggesting that structural sociodemographic factors, such as older age, lower educational attainment, and marginalized socioeconomic status, profoundly constrain the effective utilization of digital resources, potentially exacerbating health inequalities.^19,31

Of particular interest is Profile 2 (Passive Spectators), which exhibits a “High Literacy–Low Seeking” paradox: despite possessing adequate eHealth literacy, these patients fail to translate this capability into active information-seeking behavior. Similarly, Oudbier et al noted that many non-users actually harbor a strong motivation to access health information but are hindered by various barriers from converting this intent into actual usage.⁵² One plausible explanation for this discrepancy lies in a deficit of intrinsic behavioral motivation. Patients with chronic pain may develop learned helplessness due to the prolonged disease course, diminishing their expectancy of improving their condition through personal effort. This aligns with the theoretical premises of the IMB model, which posits that information capability alone, without motivational drive, is insufficient to generate behavior;²² this is further corroborated by qualitative evidence identifying lack of motivation and support as a primary barrier to the uptake of digital interventions.⁵³ Alternatively, this reluctance may reflect psychological defensiveness toward the digital environment, specifically a generalized mistrust regarding the quality of online information and data privacy. Previous studies have confirmed that trust serves as a crucial mediator linking seeking behavior to uncertainty management; a lack of trust can lead patients to adopt a strategy of digital silence, actively avoiding deep engagement despite having the capability.⁴⁹ Consequently, from a behavioral change perspective, merely enhancing literacy is insufficient to drive behavior; future interventions must concurrently address the psychological mechanisms of motivation activation and trust rebuilding.

Secondly, Profile 3 (Digital Champions) represents an ideal model of digital health engagement. Composed primarily of younger and well-educated individuals, this group exhibits the dual characteristics of high eHealth literacy and high information-seeking behavior. This profile aligns with the value-oriented segment identified in research on telemedicine consumption in China, as these individuals tend to view digital health as a holistic health investment rather than merely a convenience tool.⁵⁴ Furthermore, our univariate analysis revealed that younger patients (18–64 years) exhibited significantly higher exercise adherence than their older counterparts (≥ 65 years). This finding is consistent with the profile characteristics of the “Digital Champions”, who were more likely to be younger and to have higher levels of digital health engagement. One possible explanation is that younger patients may have greater familiarity with digital resources and stronger capacity to translate online rehabilitation information into sustained exercise practice. In contrast, older adults may face greater barriers to digital inclusion, a finding that is consistent with previous evidence identifying older age and lower education as important barriers to digital participation.⁵⁵ The behavioral patterns of this group strongly support the Knowledge-Attitude-Practice model in the digital era. Consistent with the study by Liu et al, eHealth literacy not only directly predicts health behaviors but also operates through the mediating pathway of online information seeking.⁴⁸ Importantly, the Digital Champions in this study achieved an effective transition from cognition to behavior. As confirmed by Park and Lee in their research on chronic diseases, high literacy is strongly and positively correlated with all aspects of self-care adherence.⁵⁰ The intrinsic mechanism of this translation can be explained by the recent findings of Tan et al, which suggest that high-quality digital experiences can enhance patient self-efficacy, thereby driving treatment adherence.⁵⁶ This indicates that LDH patients in the Digital Champions group form a virtuous cycle: the improvement of eHealth literacy leads to successful information-seeking behaviors, which in turn enhances self-efficacy and ultimately sustains exercise adherence. Consequently, this finding not only defines the ideal target audience for digital interventions but also provides a critical basis for understanding the dynamic mechanisms of digital health behavior, sociodemographic impacts, and the design of inclusive strategies.⁵⁷ The future challenge lies in how to draw upon the success factors of this ideal model to activate and empower a broader patient population.

However, the most illuminating finding of this study lies in the identification of the atypical subgroup, Profile 4 (Distress-Driven Seekers). This discovery effectively challenges the traditional convention that low literacy inevitably leads to low engagement. In sharp contrast to the “Passive Avoidance” profile discussed earlier, which is characterized by a lack of motivation, this group demonstrated an exceptionally high willingness to seek information despite having lower eHealth literacy. Integrating the logistic regression results, which showed this group had the highest scores for physical functional disability, the Protection Motivation Theory provides a robust explanatory framework for this behavioral paradox.⁵⁸ The theory posits that when the perceived health threat is sufficiently severe, the resulting sense of urgency can override concerns about one’s own lack of ability.⁵⁸ This finding aligns with the recent study by Chen and Gao on patients with type 2 diabetes, which identified that patients characterized by high disease threat perception (high sensitivity) combined with high coping efficacy consistently demonstrate the strongest adherence to health behaviors.⁵⁹ In the present study, severe physical pain and functional limitations constitute a high-intensity threat perception, which activates the patients’ coping instinct and compels them to view information seeking as a necessary means to reduce uncertainty. As noted by Patel and Shepherd in their recent study on the low back pain population, increasing disease burden is a core driver compelling patients to utilize online resources for help.²² Furthermore, a systematic review by Wang et al indicates that such high-intensity seeking behavior often serves as a psychological compensatory strategy to cope with health anxiety and reduce uncertainty.⁶⁰ This pain-driven sense of crisis compels them to adopt a broad-spectrum approach to information seeking in an attempt to regain control. Consequently, despite lacking sophisticated retrieval skills, they compensate for their competency deficits through high-intensity investment, ultimately achieving high adherence levels comparable to the “Digital Champions”.

Nevertheless, we must recognize that the high adherence of “Distress-Driven Seekers” is built on a fragile foundation. In stark contrast to the adherence of the “Digital Champions”, which is grounded in rational judgment, the motivation of this group often stems from unverified panic or one-sided cognition. The combination of low critical literacy and high-frequency information exposure constitutes a highly insidious clinical risk signal. Lacking the core ability to discern the authenticity of information, these patients are extremely prone to falling into error traps during the search process. Research by Zada et al indicates that with the increasing prevalence of AI-assisted self-diagnosis, patients with low literacy struggle to identify erroneous medical advice generated by large language models, significantly increasing the risk of misdiagnosis and improper treatment.⁶¹ Even more concerning is the warning from Altorfer et al that generative AI can act as a deceptive “false friend”, leading patients to blindly trust fabricated scientific evidence and execute harmful rehabilitation protocols.⁶² This implies that their high adherence may be founded on misconceptions, such as causing secondary injury by strictly performing exercises unsuitable for their condition. Therefore, this group is by no means a simple model of adherence; on the contrary, they represent the highest-risk population in clinical practice. Intervention strategies for them should not merely focus on encouragement but rather prioritize active risk identification and information correction. Specifically, drawing on the recommendations of Chan et al, we can introduce professional-led or peer-support models to channel their intense information-seeking intent into a standardized scientific track, thereby mitigating potential harms.⁶³

The findings of this study strongly advocate for a paradigm shift in clinical practice from a traditional “one-size-fits-all” health education model toward precision interventions based on latent profile characteristics, thereby maximizing resource efficiency.⁶⁴ Specifically, for “Digital Disengaged” patients who are predominantly older with minimal digital literacy, clinicians should prioritize traditional offline education and actively incorporate family proxy mechanisms to encourage younger family members to assist in accessing digital health resources.⁶⁵ For “Passive Spectators” who possess adequate literacy but lack active engagement, healthcare providers may consider Motivational Interviewing to identify psychological barriers and stimulate intrinsic motivation, thereby helping to bridge the intention–behavior gap.⁶⁶ Nevertheless, this recommendation is based on theoretical fit and indirect evidence rather than LDH-specific intervention trials. Future prospective studies are needed to verify whether Motivational Interviewing can effectively improve exercise adherence in this subgroup. Conversely, “Digital Champions” should be viewed as potential community assets; thus, clinicians should adopt an empowerment strategy by inviting them to serve as expert patients or peer leaders to leverage the role-model effect for other groups.⁶³ Finally, particularly in the context of the prevalence of generative AI, “Distress-Driven Seekers” are highly vulnerable to medical misinformation when utilizing AI tools for self-diagnosis.⁶¹ Consequently, physicians should proactively issue information prescriptions by directly recommending professionally certified digital resources or authoritative health expert channels. This strategy effectively satisfies their information needs and guides them from blind searching to precise acquisition, thereby mitigating the risk of being misled due to a lack of critical judgment.^61–63

Beyond these patient-level precision interventions, fundamentally reducing digital inequalities requires macro-level structural adjustments within healthcare systems. For digitally disadvantaged groups—especially the highly vulnerable “Digital Disengaged” cohort—health systems and technology developers must prioritize age-friendly digital interfaces featuring simplified navigation, larger typography, and multimodal support (eg, voice assistance).⁶⁵ Furthermore, dismantling systemic barriers necessitates infrastructural investments, such as providing free and stable internet access in clinical and community settings, to ensure equitable access to digital rehabilitation resources regardless of socioeconomic status.

This study possesses two strengths compared to prior literature. First, unlike previous studies that predominantly focused on linear relationships using a single methodological perspective,^47,48,50,56 we innovatively integrated both variable-centered (SEM) and person-centered (LPA) approaches. This mixed-method strategy not only elucidates the general mechanisms linking eHealth literacy to adherence but also captures the multidimensional heterogeneity within the patient population. By establishing a two-dimensional taxonomy based on ability (eHealth literacy) and behavior (health information-seeking behavior), this study effectively circumvents the risk of masking individual differences with statistical averages. Second, the identification of the unique “Distress-Driven Seekers” subgroup offers a novel contribution that challenges the traditional stereotype that low literacy inevitably leads to low engagement. By revealing the compensatory nature of information-seeking behavior driven by severe functional disability, our findings provide a more nuanced insight into the digital divide, effectively supplementing existing theoretical frameworks.

Several limitations of this study should be acknowledged. First, the cross-sectional design precludes the determination of causal relationships among eHealth literacy, HISB, and exercise adherence. Future longitudinal studies should monitor changes in digital engagement indicators, including eHealth literacy, HISB intensity, and objective digital application logs, together with clinical and behavioral outcomes such as ODI, pain intensity, health anxiety, exercise adherence, and symptom recurrence. Such designs would help clarify the temporal pathways through which digital engagement patterns influence rehabilitation adherence and clinical outcomes. Second, relying entirely on self-reported data introduces potential measurement biases. While we applied procedural remedies (eg, anonymity) and Harman’s single-factor test to mitigate common method variance, the absence of an a priori marker variable precluded more robust statistical controls.⁶⁷ Specifically, operationalizing “digital health engagement” solely through the eHEALS and HISB scales has inherent limitations, as it lacks objective measures such as the actual frequency of digital tool use or the duration of interaction. Furthermore, the comprehensive nature of the 43-item HISB scale may have introduced response fatigue that potentially compromised data reliability, notwithstanding the non-directive support provided by research assistants. Future studies are recommended to incorporate objective monitoring tools, such as digital application logs and wearable devices, and consider utilizing validated short-form versions of these assessment tools to obtain more accurate data while minimizing patient fatigue. Third, the inclusion of both conservatively treated and postoperative patients introduces clinical heterogeneity. Although we utilized the ODI to statistically account for variances in functional impairment across the cohort, future research should stratify these clinical subgroups to explore potential differences in their specific digital engagement trajectories. Fourth, the “Distress-Driven Seekers” profile was relatively small (n=18, 5.2%), limiting the precision and generalizability of findings for this subgroup. Although ODI significantly predicted membership in this profile, the wide confidence interval (OR = 2.76, 95% CI: 1.04–7.33) suggests uncertainty in the effect estimate. The small subgroup size also contributed to complete separation for education and income, rendering some odds ratios unestimable. Therefore, findings related to this profile should be interpreted cautiously and verified in larger, socioeconomically diverse samples. Finally, the reliance on convenience sampling from a single-center survey in Zhejiang Province, China, introduces potential selection bias. Although socioeconomic covariates were controlled, regional disparities in digital infrastructure and healthcare resources may limit the generalizability of our findings to broader patient populations, particularly those in rural areas or distinct cultural contexts.⁶⁸

Conclusion

This study demonstrates that exercise adherence among patients with LDH is shaped by both general behavioral mechanisms and distinct digital engagement profiles. HISB served as an important pathway linking eHealth literacy to exercise adherence, while LPA identified four profiles: “Digital Disengaged”, “Passive Spectators”, “Digital Champions”, and “Distress-Driven Seekers”. Notably, the “Distress-Driven Seekers” profile indicates that high adherence does not necessarily reflect safe or well-informed self-management. Patients in this subgroup may be vulnerable to medical misinformation, health anxiety, and secondary injury caused by inappropriate rehabilitation guidance. These findings support a shift from uniform health education to profile-based, resource-sensitive precision interventions. However, the feasibility, acceptability, and cost-effectiveness of such interventions should be evaluated before wider clinical implementation. Future research should prioritize prospective validation of these profiles and randomized or pragmatic trials of specific strategies, particularly clinician-led information prescriptions for “Distress-Driven Seekers” and activation-oriented interventions for “Passive Spectators”, to improve safe and sustainable rehabilitation adherence in LDH patients.

Funding Statement

The study was supported by Medical Scientific Research Foundation of Zhejiang Province, China (Grant No.2025HY0745), Zhejiang Traditional Medicine and Technology Program, China (Grant No.2026ZL0526, 2021ZB196), and the Construction Fund of Key Medical Disciplines of Hangzhou (No.2025HZPY06).

Abbreviations

LDH, lumbar disc herniation; Eheals, eHealth Literacy Scale; HISB, Health Information Seeking Behavior Scale; SEM, Structural Equation Modeling; LPA, Latent Profile Analysis; ODI, Oswestry Disability Index.

Data Sharing Statement

The datasets generated and analysed during the current study are not publicly available but are available from the corresponding author upon reasonable request.

Ethics Approval and Consent to Participate

This study was approved by the Ethics Board of Hangzhou TCM Hospital Affiliated to Zhejiang Chinese Medical University (NO.2025KLL186). All participants gave written informed consent. The research protocol was established according to the ethical guidelines of the Declaration of Helsinki.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors confirm that there are no conflicts of interest for this work.