Digital health literacy, proactive health behaviours, the digital healthcare environment and health-related quality of life among older adults with type 2 diabetes in Chinese primary care: a cross-sectional study

Abstract

Background

Digital health literacy (DHL) may be associated with health-related quality of life (HRQoL) in older adults with type 2 diabetes, but the behavioural and contextual correlates of this association remain unclear. This study examined that association, the indirect association through proactive health behaviours (PHB), and potential effect modification by the digital healthcare environment.

Methods

A cross-sectional survey was conducted from July to August 2025 among older adults with type 2 diabetes recruited by convenience sampling from primary healthcare chronic disease management registries in Huai’an, China. DHL, PHB and four single-item digital healthcare environment indicators were assessed using structured instruments, and HRQoL using the EuroQol five-dimensional three-level questionnaire (EQ-5D-3L) and EuroQol Visual Analogue Scale (EQ-VAS). Adjusted indirect-association (5000 bias-corrected bootstrap resamples) and effect-modification analyses used heteroscedasticity-consistent type 3 (HC3) standard errors.

Results

Among 1053 participants, the mean DHL score was 21.24±10.43, the mean PHB score 76.75±9.88 and the mean EQ-5D-3L utility score 0.968±0.069. The adjusted total association between DHL and utility was significant (B=0.000569, 95% CI 0.000228 to 0.000910), whereas the direct association was not (B=0.000152; p=0.338). The indirect association through PHB was significant (B=0.000418, bias-corrected 95% CI 0.000260 to 0.000619) and accounted for 73.5% of the total. There was no robust evidence of effect modification for EQ-5D-3L utility: interactions with remote service use and online health management use reached significance only under HC3 inference and were not reproduced in sensitivity analyses. An exploratory positive PHB-by-remote-service-use interaction was observed for the supplementary EQ-VAS outcome.

Conclusions

Higher digital health literacy was associated with better HRQoL among older adults with type 2 diabetes, with a significant cross-sectional indirect association observed through proactive health behaviours. No robust effect modification was identified for the primary EQ-5D-3L outcome. These findings support further longitudinal and intervention research on digital health capabilities and proactive health behaviours in this population.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study used a theory-informed analytical framework to examine the association between digital health literacy and health-related quality of life, including indirect-association and effect-modification analyses.

• A relatively large community-based sample of older adults with type 2 diabetes was recruited through primary healthcare chronic disease management registries.

• Standardised instruments were used to assess key constructs, including EuroQol five-dimensional three-level questionnaire (EQ-5D-3L)/EuroQol Visual Analogue Scale and structured scales for proactive health behaviours and digital health literacy.

• The cross-sectional design does not establish temporal ordering among the study constructs, so the observed indirect association cannot be interpreted as a causal pathway.

• Registry-based identification combined with convenience recruitment limits generalisability; in addition, low exposure to digital services and high EQ-5D-3L utility scores may have reduced sensitivity for detecting moderation effects.

Introduction

Diabetes is one of the most common chronic non-communicable diseases worldwide, with type 2 diabetes accounting for the vast majority of cases. According to the International Diabetes Federation (IDF), approximately 589 million adults aged 20 to 79 years were living with diabetes globally in 2024, and this number is projected to increase to 853 million by 2050.¹ Diabetes is particularly prevalent among older adults, affecting approximately 20% of this population. It has been identified as the third leading cause of disability-adjusted life years (DALYs) among adults aged 50 to 75 years and the second leading cause among those over 75 years.^{2 3} Older adults with type 2 diabetes not only face a higher risk of cardiovascular disease and mortality but also often experience a longer disease course, a greater burden of complications and comorbidities, and more rapid functional decline.^4–6 These interacting physiological and pathological factors directly affect individuals’ perceptions of their physical, psychological and social well-being, thereby significantly reducing their health-related quality of life (HRQoL).^{2 7 8} Therefore, in older adults with type 2 diabetes, improving HRQoL has become an important goal of disease management, particularly in the context of ageing, multimorbidity and functional vulnerability.^{1 8}

In the evolving landscape of diabetes care, the increasing use of digital healthcare technologies has created new opportunities for older patients. In recent years, alongside growing attention to social issues such as the digital divide and technology-related inequalities, the concept of digital health literacy (DHL) has gained prominence. Many scholars have regarded eHealth literacy and DHL as equivalent concepts.⁹ However, in 2023, Wang and Luan distinguished between the two, arguing that DHL places greater emphasis on the influence of digital technologies and the broader context of digital information.¹⁰ Against the backdrop of rapid digital health development, DHL has become a critical factor affecting the effectiveness of chronic disease management. DHL encompasses not only the ability to obtain and understand digital health information but also the capacity to evaluate information, engage in interactive participation and translate digital information into health-related actions.¹¹ Previous studies have demonstrated that low health literacy is closely associated with poorer disease knowledge, weaker self-management behaviours and worse health outcomes in individuals with diabetes.^{12 13} Meanwhile, research on digital health adoption among older adults has shown that limited digital competence, complex platforms, low self-efficacy and insufficient social support may restrict the use of and benefits from digital health resources, thereby negatively impacting the HRQoL of older adults with type 2 diabetes.^14–18 Importantly, DHL is not synonymous with the actual use of digital health services or the tangible benefits derived from them. A review has indicated that the utilisation of digital health services is influenced not only by knowledge and skills but also by factors such as educational attainment, platform usability and attitudes toward technology.¹⁶ Among older adults with chronic conditions, the adoption of digital health technologies is marked by significant inequalities, which may also be shaped by urban–rural disparities, gender and contextual support conditions.¹⁴ Although the predictive role of digital literacy in HRQoL has been preliminarily recognised, research specifically focusing on DHL among older adults with type 2 diabetes remains limited. Most previous studies have examined this issue from the perspective of general health literacy or eHealth literacy,^{19 20} which may not fully capture patients’ actual capabilities in digital health application contexts. To address this limitation, a more targeted assessment framework is needed to integrate the characteristics of digital interaction in chronic disease management among older adults.¹⁴ To better identify the features of digital health capability in this population, our team previously developed a digital health literacy assessment scale for patients with chronic diseases based on the Transaction Model of eHealth Literacy (TMeHL), which was applied in the present study. Based on a more precise assessment of digital capability, further exploration of how DHL is associated with quality of life in relation to specific behavioural factors may provide important evidence for the development of age-friendly digital health interventions.

Existing studies examining the pathways linking DHL and HRQoL have most commonly treated self-management behaviour as a mediating factor.^21–23 However, self-management often focuses primarily on patients’ adherence to medical advice, and previous research has shown that its effects on improving long-term physiological indicators and sustaining health outcomes are not always consistent. This suggests that behavioural factors may help explain the observed association between DHL and quality of life.^{4 5 7 24} According to the capability, opportunity, motivation and behaviour (COM-B) model, health behaviour is jointly driven by capability, opportunity and motivation. As a key capability-related resource, DHL may enhance older patients’ sense of control over their health and stimulate intrinsic motivation by improving their ability to access and evaluate digital information,²⁵ thereby facilitating a shift from passive disease management to more active health engagement. Compared with self-management alone, proactive health behaviour (PHB) refers to sustained behavioural activities through which individuals consciously assume responsibility for their own health, make full use of health resources and actively maintain, restore and improve health from the perspective that they themselves are primarily responsible for their health.^{26 27} Among older adults with type 2 diabetes, PHB is not a single action but rather a sustained pattern of behaviours involving health responsibility, exercise, diet, stress management and support utilisation. Existing systematic reviews have shown that interventions targeting PHB can improve quality of life, clinical outcomes and psychological outcomes in people with diabetes.^28–30 Additionally, studies among older adults have suggested that health-promoting behaviours or healthy lifestyles may mediate the association between eHealth literacy and HRQoL.^{31 32} These findings suggest that DHL may be associated with quality of life less directly than through its relationship with more active PHB.

In addition to individual capabilities and behaviours, the external service environment may also influence this process. Digital healthcare services—such as remote consultations, online health management and digital follow-up—may shape the association between PHB and health outcomes by increasing feedback frequency, providing interactive support and improving service accessibility. However, existing evidence among older adults with chronic diseases remains inconsistent. Some studies indicate that digital health interventions effectively improve behavioural outcomes, such as physical activity, but their impact on overall quality of life is not always significant.^{33 34} Furthermore, the digital healthcare environment is typically examined separately as a background factor or intervention component and has rarely been examined in relation to the associations among DHL, PHB and HRQoL. Whether the digital healthcare environment moderates the association between PHB and HRQoL remains unclear, particularly among older adults with type 2 diabetes.

Against this background, the present study focused on older adults with type 2 diabetes to examine the association between DHL and HRQoL, the cross-sectional indirect association through PHB and potential effect modification by the digital healthcare environment. The findings may provide evidence to inform future research on digital health management and age-friendly health promotion strategies for older adults with type 2 diabetes.

Methods

Study design

This cross-sectional study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement.³⁵ See online supplemental file 2.

Setting and sample

This cross-sectional study used a multisite, registry-based convenience sampling approach. Participants were recruited from primary healthcare institutions in Huai’an Industrial Park, the Bailuhu area of Qingjiangpu District, and Lianshui County, Huai’an City, Jiangsu Province, China. At the participating sites, potentially eligible older adults with type 2 diabetes were identified from existing chronic disease management registries or follow-up records. Primary healthcare staff contacted eligible individuals according to the survey schedule and invited them to participate voluntarily.

Inclusion criteria: (1) met the diagnostic criteria for type 2 diabetes according to the Clinical Guideline for the Prevention and Treatment of Type 2 Diabetes in the Elderly in China (2022 edition)³⁶; (2) were aged 60 years or older; (3) had been included in the chronic disease management records or follow-up system of the selected primary healthcare institutions and had resided in the community for at least 6 months; (4) were mentally alert, able to communicate their feelings and capable of completing the survey; (5) provided written informed consent and voluntarily participated in the study. Exclusion criteria: (1) had severe diabetic complications, other serious physical illnesses or psychiatric disorders; (2) had experienced major life events or traumatic events recently.

Sample size calculation

As this was an analytical cross-sectional study, the sample size was estimated using the formula specific to such studies³⁷:

$${\displaystyle {\displaystyle n=\frac{u_{\alpha /2}^2\pi (1-\pi )}{{\delta }^2}}}$$

The significance level was set at α=0.05 for a two-sided test, with a critical value of 1.96. The allowable error (δ) was specified as a relative error equal to 20% of the expected proportion (δ=0.2π). Since the target population comprised older adults with type 2 diabetes and the study sample was drawn from primary healthcare chronic disease management records, no prevalence data exactly matching the registered prevalence of older adults with type 2 diabetes in the study setting were available. A conservative estimate was therefore derived from provincial epidemiological data: according to findings reported in The BMJ, the prevalence of diabetes among adults in Jiangsu Province was 11.5%,³⁸ of which more than 90% was type 2 diabetes. Accordingly, π=0.104 and δ=0.0208 were used, yielding a calculated sample size of 827.42, which was rounded up to 828 participants. Allowing for an anticipated 15% rate of invalid or incomplete questionnaires, the target sample size was 974. The final analytic sample comprised 1053 participants and therefore exceeded this target.

Research instruments

Sociodemographic and disease-related characteristics questionnaire

Sociodemographic characteristics included age, sex, ethnicity, place of residence, marital status, education level and living arrangement. Age was recorded in completed years and analysed as a continuous variable; the age bands presented (60–69, 70–79 and ≥80 years) were used for descriptive purposes only. Residence was classified as urban or rural. Marital status was grouped as currently married versus unmarried/divorced/widowed. Education level was categorised as primary school or below, junior or senior high school, and college or above. Living arrangement was classified as living with spouse, living with children/grandchildren, living with spouse and children/grandchildren, living alone, or other.

Disease-related characteristics included duration of chronic disease, treatment modality, family history of diabetes and comorbidity burden. Duration of chronic disease was categorised as <5 years, 5–9 years and ≥10 years. Treatment modality was categorised as oral glucose-lowering agents only, insulin with or without oral agents, no glucose-lowering treatment, or other/unspecified treatment. Family history of diabetes was recorded as yes or no. Comorbidity burden was classified as no comorbidity, one comorbidity, and two or more comorbidities based on participants’ self-reported chronic conditions.

Proactive health behaviour scale for patients with chronic diseases

The Proactive Health Behaviours Scale for patients with chronic diseases was developed by Chen et al³⁹ in 2024. The scale consists of 20 items across five dimensions: individual health responsibility, psychological stress coping, physical activity, nutritional behaviours and interpersonal support. Each item is rated on a 5-point Likert scale, ranging from 1 (‘strongly disagree’) to 5 (‘strongly agree’), with higher scores indicating a greater level of proactive health behaviours. The Cronbach’s α coefficients for the five dimensions ranged from 0.804 to 0.870. In the present study, the overall Cronbach’s α coefficient for the scale was 0.763.

Health-related quality of life

HRQoL was assessed using the EuroQol five-dimensional three-level questionnaire (EQ-5D-3L) and the EuroQol Visual Analogue Scale (EQ-VAS). The EQ-5D, developed by the EuroQol Group, is a widely used international instrument for measuring HRQoL.⁴⁰ In this study, the Mainland Chinese version of the questionnaire was employed. The EQ-5D-3L descriptive system comprises five dimensions: mobility, self-care, usual activities, pain/discomfort and anxiety/depression, each with three levels. Health states derived from the EQ-5D-3L were converted into utility scores using the Chinese time trade-off (TTO) value set.⁴¹ Higher utility scores indicate better health-related quality of life. The EQ-VAS is a vertical visual analogue scale ranging from 0 to 100, where 0 represents the ‘worst imaginable health state’ and 100 represents the ‘best imaginable health state.’ Higher scores indicate better self-rated health status. Values outside the prespecified 0–100 range were considered invalid because they cannot represent valid EQ-VAS responses. One recorded value of 754 was identified as out of range and was treated as missing in EQ-VAS-based analyses. As an internationally standardised instrument, the EQ-5D-3L has been widely used in Chinese and older populations, and previous studies have demonstrated its good feasibility and structural validity.

Digital health literacy scale for patients with chronic diseases

The Digital Health Literacy Scale for patients with chronic diseases was developed in our previous work based on the TMeHL framework. The scale development process included a literature review, focus group discussions, two rounds of Delphi expert consultations and a small-sample linguistic adaptation, following established guidance for the development and evaluation of patient-reported outcome measures.⁴² Following pilot and formal surveys among patients with chronic diseases, item analysis, as well as reliability and validity testing, was conducted. The final scale consists of 12 items across four dimensions: digital health information acquisition ability, digital health interactive participation ability, digital health information evaluation ability and digital health behaviour transformation ability, with three items per dimension. All items are rated on a 5-point Likert scale, with higher scores indicating a higher level of digital health literacy. The original validation study demonstrated that the scale had good structural validity and internal consistency. Exploratory factor analysis extracted four common factors, with a cumulative variance contribution rate of 78.781%. Confirmatory factor analysis showed a good model fit: χ²/df=2.401, root mean square error of approximation (RMSEA)=0.070, goodness-of-fit index (GFI)=0.940, normed fit index (NFI)=0.952, incremental fit index (IFI)=0.972, Tucker-Lewis index (TLI)=0.960 and comparative fit index (CFI)=0.971. The average variance extracted (AVE) for each dimension ranged from 0.618 to 0.753, and the composite reliability (CR) ranged from 0.828 to 0.901. The scale-level content validity index (S-CVI/Ave) was 0.952, the Cronbach’s α coefficient for the total scale was 0.925 and the split-half reliability coefficient was 0.869. In the present study, the Cronbach’s α coefficient of the scale was 0.917. These psychometric results were obtained in the scale development and validation study conducted by our research group, which is not yet published.

Digital healthcare environment indicators

Digital healthcare environment was assessed using four prespecified indicators drawn from the healthcare accessibility and service continuity sections of the study questionnaire: healthcare accessibility, information continuity, remote service use and online health management use. These indicators captured different aspects of participants’ access to and experience of digitally enabled chronic disease management services and were analysed separately rather than combined into a composite score. Healthcare accessibility was assessed using the item, “How convenient is it for you to obtain chronic disease services, such as medical consultations, follow-up care, or medication refills?” Responses ranged from 1 (‘very difficult’) to 5 (‘very convenient’), with higher scores indicating greater accessibility. Information continuity was assessed by asking whether healthcare professionals at different local healthcare institutions could readily access participants’ previous clinical information, such as medical records or examination reports. Responses ranged from 1 (‘fully able’) to 5 (‘unable’) and were reverse-coded so that higher scores represented better information continuity. Remote service use was assessed by asking whether participants had used remote approaches, such as online or telephone consultations, to receive chronic disease care. Responses ranged from 1 (‘never used or unaware of such services’) to 4 (‘frequently used’). Online health management use was assessed by asking whether participants had received chronic disease management through online tools such as WeChat, mobile applications or text messaging, including health information, medication reminders, follow-up reminders or remote follow-up. Responses used the same four-category frequency scale. The four indicators were treated as distinct observed variables and were not combined into a total score, so the set does not form a scale with dimensions or a summed score. The items were developed by the research team for the present study to capture aspects of digitally enabled chronic disease management relevant to Chinese primary care; they were not drawn from a published instrument and had not previously been used as a validated measurement tool. Because each construct is represented by a single item rather than a multi-item scale, internal consistency coefficients and factor-analytic indices do not apply. The items were instead evaluated through their response distributions and their associations with proactive health behaviours.

Data collection

Eligible participants were identified from chronic disease management registries by the participating primary healthcare institutions and contacted by telephone in batches according to the survey schedule. They were informed of the time, location and relevant precautions and were invited to attend their affiliated community health service centre or township health centre. Before fieldwork, all investigators, who were master’s students, completed standardised training covering the study objectives, interpretation of questionnaire items, interview technique, standardised instructions, informed consent procedures and quality control protocols. Questionnaires were administered face to face on a one-to-one basis. Participants able to complete the questionnaire independently did so, with investigators verifying completeness on site; for those with limited education, reading or writing difficulties or visual impairment, investigators administered the questionnaire item by item following a standardised script to minimise information bias and missing data. Each questionnaire was collected immediately on completion and checked for missing or incomplete responses, which were clarified with the participant on site wherever possible. All questionnaires were subsequently reviewed centrally by the research team for completeness, logical consistency and outliers, and any questionnaire with substantial missing information on key variables or evident logical errors was to be excluded. Of the 1250 eligible individuals contacted and invited, 1053 attended and completed the questionnaire and 197 (15.8%) did not take part, giving a response rate of 84.2%. No returned questionnaire was excluded at central review. Nevertheless, small amounts of item-level missingness remained in the analytic dataset: DHL was missing for 5 participants (0.5%), EQ-VAS for 10 (0.9%), remote service use for 6 (0.6%), online health management use for 13 (1.2%), and treatment modality was missing or unclassifiable for 6 (0.6%). PHB and EQ-5D-3L utility were complete for all 1053 participants. Questionnaires were not coded by primary healthcare institution: to protect participant and institutional confidentiality, all questionnaires were anonymised at the point of collection and assigned a single study-wide serial number. Privacy requirements precluded systematic collection of demographic or health information on non-participants, and reasons for non-participation were not recorded systematically. Informal field feedback suggested that some non-participants were away from the community for extended periods, bedridden or unwilling to participate. Access to the original questionnaires and to the electronic dataset was restricted to authorised project team members, and the paper questionnaires were stored in locked filing cabinets. Each participant received a small token of appreciation valued at approximately US$2.

Data analysis

All analyses were conducted using SPSS version 26.0 and PROCESS Macro version 4.2; equivalent regression models were re-estimated in Python 3.13.5 using statsmodels 0.14.6 for robust inference and sensitivity analyses. Approximately symmetric continuous variables were summarised using means and SD, whereas markedly skewed variables were additionally described using medians and IQRs. Single-item digital healthcare environment indicators were treated as ordinal measures and summarised using medians, IQRs and response distributions where appropriate. Categorical variables were reported as frequencies and percentages. Distributional characteristics were assessed using histograms, Q-Q plots, skewness and excess kurtosis. Pearson correlations were used for primary bivariate analyses, with Spearman correlations as a sensitivity analysis. Regression-based indirect-association and effect-modification models with continuous outcomes were estimated using ordinary least-squares (OLS) linear regression. Three linear regression models were fitted for the indirect-association analysis: PHB regressed on DHL and covariates (a path); EQ-5D-3L utility regressed on DHL and covariates (total association, c path); and EQ-5D-3L utility regressed on both DHL and PHB together with covariates, yielding the direct association (c′ path) and the PHB–utility coefficient (b path). For the effect-modification analyses, EQ-5D-3L utility was regressed on mean-centred PHB, the mean-centred moderator, their product term and the same covariate set, with one model fitted per moderator. Conventional and heteroscedasticity-consistent type 3 (HC3) standard errors were computed for every model and are reported alongside each other for the interaction terms. The cross-sectional indirect association of DHL with EQ-5D-3L utility through PHB was examined using a PROCESS Model 4-equivalent specification with 5000 bias-corrected bootstrap resamples.⁴³ Effect modification by healthcare accessibility, information continuity, remote service use and online health management use was examined separately using Model 1-equivalent specifications, with PHB and moderator scores mean-centred before constructing interaction terms. Model diagnostics included residual plots, the Breusch–Pagan test, variance inflation factors and Cook’s distance. Because EQ-5D-3L utility showed a pronounced ceiling and heteroscedastic residuals, HC3 heteroscedasticity-consistent standard errors were used.⁴⁴ Unstandardised coefficients were reported as primary estimates, with standardised coefficients and a completely standardised indirect effect additionally provided. Because EQ-5D-3L utility is bounded and concentrated at its upper limit,⁴⁵ fractional-logit models⁴⁶ and an alternative Chinese EQ-5D-3L tariff⁴⁷ were used as sensitivity analyses. All multivariable models adjusted a priori for age, sex, residence, marital status, education level, diabetes duration, treatment modality, family history of diabetes and comorbidity burden; age was modelled continuously. EQ-VAS was examined as a supplementary outcome. Missing values were not imputed. For DHL, EQ-VAS, remote service use and online health management use, analyses used complete observations for the variables included in each model. The six participants with missing or unclassifiable treatment modality were retained as a separate indicator category in multivariable models rather than excluded or imputed. Accordingly, the primary indirect-association model included 1048 participants; EQ-VAS analyses included 1043 participants, and moderation analyses involving remote service use and online health management use included 1047 and 1040 participants, respectively. Because reasons and predictors of item-level missingness were not systematically captured, the available data did not permit formal verification that missingness satisfied the missing-at-random (MAR) assumption; therefore, MAR was not assumed. Institution-level recruitment identifiers were not retained in the analytic dataset; therefore, within-institution clustering could not be incorporated into the regression models, and cluster-robust standard errors, generalised estimating equations (GEE) and multilevel models could not be validly estimated. All tests were two-sided, with p<0.05 considered statistically significant.

Patient and public involvement

Patients and/or the public were not involved in the design, conduct, reporting or dissemination plans of this research.

Results

Sample characteristics

A total of 1053 older adults with type 2 diabetes were included in the analytic sample. The mean age of the participants was 70.66 years (SD=6.24). Among them, 44.2% were aged 60–69 years, 47.3% were aged 70–79 years, and 8.5% were aged 80 years or older. Women comprised 51.4% of the sample. Most participants lived in urban areas (74.0%) and were currently married (84.5%). Regarding educational attainment, 49.5% had a primary school education or less. Additionally, 51.5% lived with their spouse, and 10.1% lived with children or grandchildren. Concerning disease-related characteristics, 26.7% had a chronic disease duration of less than 5 years. Detailed participant characteristics are presented in table 1.

Table 1. General characteristics of the participants (n=1053).

Variable	Category	N	Proportion (%)
Sociodemographic characteristics
Age (years)	60–69	465	44.2
	70–79	498	47.3
	≥80	90	8.5
Sex	Men	512	48.6
	Women	541	51.4
Ethnicity	Han	1047	99.4
	Hui	4	0.4
	Other	2	0.2
Residence	Urban residence	779	74.0
	Rural residence	274	26.0
Marital status	Currently married	890	84.5
	Unmarried/divorced/widowed	163	15.5
Education level	Primary school or below	521	49.5
	Junior or senior high school	469	44.5
	College or above	63	6.0
Living arrangement	Living with spouse	542	51.5
	Living with children/grandchildren	106	10.1
	Living with spouse and children/grandchildren	307	29.2
	Living alone	93	8.8
	Other	5	0.5
Disease-related characteristics
Duration of chronic disease	<5 years	281	26.7
	5–9 years	235	22.3
	≥10 years	537	51.0
Treatment modality	Oral agents only	662	62.9
	Insulin with/without oral agents	216	20.5
	No glucose-lowering treatment	120	11.4
	Other/unspecified treatment	49	4.7
Family history of diabetes	Yes	376	35.7
	No	677	64.3
Comorbidity burden	No comorbidity	181	17.2
	One comorbidity	490	46.5
	Two or more comorbidities	382	36.3

Descriptive statistics of the main study variables

Descriptive statistics for the main study variables are presented in table 2, with dimension-level scores in online supplemental table S1 and the EQ-5D-3L response distribution in online supplemental table S2. The mean DHL total score was 21.24 (SD 10.43), with a median of 18.00 (IQR 12.00–24.25). The mean PHB total score was 76.75 (SD 9.88), with physical activity showing the lowest item-mean score (3.01, SD 0.96). EQ-VAS had a mean of 73.44 (SD 14.54; n=1043). EQ-5D-3L utility had a mean of 0.968 (SD 0.069) and a median of 1.000 (IQR 0.973–1.000).

Table 2. Descriptive statistics of the main study variables.

Variable	Mean (SD)	Median (IQR)
Digital health literacy, total score	21.24 (10.43)	18.00 (12.00–24.25)
Proactive health behaviours, total score	76.75 (9.88)	78.00 (70.00–84.00)
Healthcare accessibility	4.58 (0.79)	5.00 (4.00–5.00)
Information continuity	3.90 (1.04)	4.00 (4.00–5.00)
Remote service use	1.17 (0.53)	1.00 (1.00–1.00)
Online health management use	1.16 (0.55)	1.00 (1.00–1.00)
EQ-5D-3L utility score	0.968 (0.069)	1.000 (0.973–1.000)
EQ-VAS	73.44 (14.54)	75.00 (60.00–80.00)

EQ-5D-3L, EuroQol five-dimensional three-level questionnaire; EQ-VAS, EuroQol Visual Analogue Scale.

Correlations among the main study variables

Digital health literacy was positively correlated with proactive health behaviours (r=0.271, p<0.001) and with the EQ-5D-3L utility score (r=0.154, p<0.001). Proactive health behaviours were positively correlated with the utility score (r=0.284, p<0.001) and with EQ-VAS (r=0.264, p<0.001). Proactive health behaviours were also positively correlated with each of the four digital healthcare environment indicators: healthcare accessibility (r=0.131, p<0.001), information continuity (r=0.098, p=0.001), remote service use (r=0.134, p<0.001) and online health management use (r=0.145, p<0.001). All four indicators were answered by almost all participants (valid n=1053 for healthcare accessibility and information continuity, 1047 for remote service use and 1040 for online health management use); healthcare accessibility and information continuity were left-skewed, with 70.6% and 29.6% of participants selecting the highest response category. Spearman rank correlations were examined as a sensitivity analysis; the direction, magnitude and statistical significance of all nine core associations were consistent with the Pearson estimates, with absolute differences of 0.039 or less (online supplemental table S3).

Cross-sectional indirect association through proactive health behaviours

In the adjusted model, DHL was associated with higher PHB (B=0.227, 95% CI 0.162 to 0.292; β=0.240), and PHB was associated with higher EQ-5D-3L utility after adjustment for DHL and covariates (B=0.001840, 95% CI 0.001292 to 0.002388; β=0.261). The total association between DHL and utility was B=0.000569 (95% CI 0.000228 to 0.000910; β=0.085), whereas the direct association after inclusion of PHB was small and not statistically significant (B=0.000152, 95% CI −0.000159 to 0.000462; β=0.023; p=0.338). The bootstrapped indirect association was 0.000418 (bias-corrected 95% CI 0.000260 to 0.000619), with a completely standardised indirect effect of 0.063, and accounted for 73.5% of the total association (table 3). In the supplementary EQ-VAS model, the indirect association through PHB was also statistically significant (B=0.084, bias-corrected 95% CI 0.055 to 0.121; completely standardised indirect effect 0.060).

Table 3. Adjusted cross-sectional indirect association between DHL and EQ-5D-3L utility through PHB (n=1048).

Effect/path	B	HC3 SE	95% CI	P value	Standardised β
Total association: DHL → utility	0.000569	0.000174	0.000228 to 0.000910	0.001	0.085
Path a: DHL → PHB	0.227	0.033	0.162 to 0.292	<0.001	0.240
Path b: PHB → utility	0.001840	0.000279	0.001292 to 0.002388	<0.001	0.261
Direct association: DHL → utility	0.000152	0.000158	−0.000159 to 0.000462	0.338	0.023
Indirect association: DHL → PHB → utility	0.000418	—	0.000260 to 0.000619	—	0.063*

^*Completely standardised indirect effect. The indirect-association CI was estimated using 5000 bias-corrected bootstrap resamples.

DHL, digital health literacy; EQ-5D-3L, EuroQol five-dimensional three-level questionnaire; HC3, heteroscedasticity-consistent type 3; PHB, proactive health behaviours.

Model diagnostics and sensitivity analyses

Distributional assessment showed that PHB and EQ-VAS were approximately symmetric (absolute skewness ≤0.45), whereas DHL showed a right-skewed distribution (skewness 1.48). EQ-5D-3L utility was markedly left-skewed (skewness −4.75; excess kurtosis 32.71) and showed a pronounced ceiling, with 628 of 1053 participants (59.6%) reporting health state 11 111. Remote service use and online health management use were strongly concentrated in the lowest response category (88.2% and 90.6%). Formal normality tests were not used as a decision rule, because at this sample size they reject for departures too small to affect inference; distributional judgements were based on skewness, excess kurtosis, histograms and Q-Q plots, and on the residuals of each fitted model. Spearman sensitivity analyses produced the same direction and statistical-significance pattern as the corresponding Pearson correlations, with absolute differences of 0.039 or less. For the regression models, the PHB model showed no substantial residual-distribution or heteroscedasticity concern (Breusch–Pagan p=0.571; White p=0.850; the Ramsey regression equation specification error test (RESET) (P=0.773), whereas the EQ-5D-3L utility models showed departures from residual normality and evidence of heteroscedasticity (Breusch–Pagan p=0.019 before and p<0.001 after inclusion of PHB) together with significant RESET statistics (p=0.007 and p<0.001), consistent with a bounded outcome concentrated at its upper limit. The maximum variance inflation factor (VIF) was 2.45, and the maximum Cook’s distance was 0.077, with no observation exceeding a Cook’s distance of 1, indicating no substantial multicollinearity or highly influential individual observation.

Effect modification by digital healthcare environment indicators

In HC3 OLS models for EQ-5D-3L utility, the PHB-by-healthcare-accessibility interaction was not statistically significant (B=−0.000132, 95% CI −0.000543 to 0.000280; p=0.530; standardised β=−0.015) nor was the PHB-by-information-continuity interaction (B=0.000090, 95% CI −0.000257 to 0.000436; p=0.612; standardised β=0.013). The PHB-by-remote-service-use and PHB-by-online-health-management interaction terms were small, and their statistical significance depended on the method of inference: both were non-significant under conventional standard errors (p=0.162 and p=0.079) and reached significance only under HC3 standard errors (B=−0.000446, 95% CI −0.000887 to −0.000005; p=0.048; standardised β=−0.034; and B=−0.000596, 95% CI −0.001042 to −0.000151; p=0.009; standardised β=−0.046, respectively). Simple slope probing was uninformative for these two moderators because 88.2% and 90.6% of participants were at the lowest response category, so the 25th percentile, median and 75th percentile all equalled 1 and produced identical conditional slopes. In the supplementary analysis using EQ-VAS as the outcome, the PHB-by-remote-service-use interaction was positive (B=0.193, 95% CI 0.018 to 0.367; p=0.030), whereas the interaction with online health management use was not statistically significant. Sensitivity analyses using fractional-logit models and the alternative Chinese EQ-5D-3L tariff did not reproduce the remote-service-use or online-health-management interactions (table 4, online supplemental table S4). Because these two interactions were significant only under robust inference and were not reproduced in the sensitivity analyses, they are not interpreted as evidence of effect modification.

Table 4. PHB-by-digital-healthcare-environment interaction terms for the EQ-5D-3L utility score.

Moderator	B	Conventional 95% CI	Conventional p value	HC3 95% CI	HC3 p value
Healthcare accessibility	−0.000132	−0.000590 to 0.000326	0.573	−0.000543 to 0.000280	0.530
Information continuity	0.000090	−0.000247 to 0.000426	0.601	−0.000257 to 0.000436	0.612
Remote service use	−0.000446	−0.001070 to 0.000179	0.162	−0.000887 to −0.000005	0.048
Online health management use	−0.000596	−0.001262 to 0.000069	0.079	−0.001042 to −0.000151	0.009

EQ-5D-3L, EuroQol five-dimensional three-level questionnaire; HC3, heteroscedasticity-consistent type 3; PHB, proactive health behaviours.

Discussion

This study examined the association between DHL and HRQoL among older adults with type 2 diabetes, together with the cross-sectional indirect association through PHB and potential effect modification by the digital healthcare environment. Higher DHL was associated with better HRQoL, and a statistically significant indirect association was observed through PHB. The moderation analyses did not provide robust evidence that the measured digital healthcare environment indicators modified the PHB–EQ-5D-3L association across model specifications. A positive PHB-by-remote-service-use interaction was observed only for the supplementary EQ-VAS outcome.

This study found that DHL was generally low among older adults with type 2 diabetes and that it was positively associated with HRQoL, which is consistent with previous research.⁴⁸ Shahid et al reported that health literacy was linked to adverse diabetes outcomes,⁴⁹ and a systematic review by Al Sayah et al further demonstrated that low health literacy was associated with poorer diabetes knowledge and several negative health outcomes.⁵⁰ Given the increasing use of digital technologies in chronic disease management, this relationship can be naturally extended to DHL, as patients’ ability to obtain, understand, evaluate and apply digital health information is increasingly relevant to how they manage their disease and to their quality of life. However, our findings also suggest that DHL does not automatically translate into improved HRQoL. Older adults with type 2 diabetes often face age-related functional decline, accumulated chronic disease burden and barriers to digital technology use.^{19 22} In this context, digital capability may function more as a resource condition than as a sufficient determinant of HRQoL. A review of eHealth adoption among older adults has shown that key factors influencing digital health use include not only knowledge and skills but also self-efficacy, social support, platform usability and the manner in which information is delivered.¹⁵ Similarly, Arcury et al found that among low-income older adults, internet use and eHealth literacy were strongly influenced by health status, social support and technological conditions.⁵¹

A major finding of this study was the statistically significant cross-sectional indirect association between DHL and HRQoL through PHB. Higher DHL was associated with greater engagement in proactive health behaviours, and greater PHB was in turn associated with better HRQoL. This pattern is consistent with a behavioural explanation for the observed association between digital health capability and HRQoL, but it does not establish temporal or causal mediation. This finding aligns with evidence from chronic disease self-management research.^{4 52–54} A systematic review demonstrated that self-management interventions can improve quality of life, self-efficacy and emotional outcomes in patients with chronic diseases, supporting the importance of behavioural engagement in translating health-related knowledge and resources into meaningful outcomes.^{55 56} DHL represents patients’ capacity to obtain, evaluate and apply digital health information, whereas PHB captures more proximal behavioural engagement in areas such as diet, physical activity, medication management, stress coping and health monitoring.^{7 10 11} The present findings therefore position PHB as an important behavioural correlate linking DHL with HRQoL in this population. Alternative temporal ordering remains possible; for example, better health or functional status could be associated with both greater engagement in proactive health behaviours and greater capacity to acquire and use digital health skills. Longitudinal studies are needed to clarify the temporal sequence among these constructs.

The indirect component accounted for 73.5% of the total observed association, suggesting that PHB may represent an important behavioural correlate linking DHL with HRQoL. In absolute terms, a 1-SD difference in DHL corresponded to an indirect difference of approximately 0.004 in EQ-5D utility. This magnitude is small. Minimally important difference estimates for the EQ-5D are derived primarily for within-person change over time, whereas the present estimate is a between-person indirect association obtained in a cross-sectional design, so the two are not directly comparable. Establishing whether an association of this size matters to patients would require longitudinal or intervention studies using prespecified patient-relevant thresholds.

The analyses did not provide robust evidence that the measured digital healthcare environment indicators modified the association between PHB and the primary EQ-5D-3L utility outcome. The original sample-size calculation was not designed specifically for interaction effects, and the study therefore cannot be assumed to have had adequate power to detect small moderation effects. This issue is compounded by the limited variation in the moderators: remote service use and online health management were concentrated in the lowest response categories. The EQ-5D-3L utility distribution also showed a pronounced ceiling, further reducing variation in the primary outcome. Accordingly, the absence of stable interaction effects should not be interpreted as evidence of a true null effect. Rather, the present data provide no robust evidence of effect modification, while small contextual effects may have remained difficult to detect.^{56 57} The supplementary EQ-VAS analysis showed a positive PHB-by-remote-service-use interaction. One possible explanation for the difference between outcomes is their measurement structure. EQ-5D-3L utility is a preference-based measure derived from five discrete health dimensions and showed substantial ceiling concentration in this sample, whereas EQ-VAS provides a continuous global assessment of current health and exhibited greater variability.^{57 58} EQ-VAS may therefore be more sensitive to relatively small differences in perceived health associated with remote service use. Nevertheless, this interaction was observed only for a supplementary outcome and was examined alongside multiple environmental interactions; it should therefore be regarded as exploratory pending replication.

Taken together, these findings extend and refine the existing literature in several respects. Previous studies have primarily examined direct associations between digital or eHealth literacy and health outcomes, or indirect associations through self-management and health-promoting behaviours. By examining PHB as a multidimensional behavioural construct, the present study provides further evidence that the association between DHL and HRQoL is accompanied by an important behavioural component. More importantly, the study moves beyond individual-level associations by explicitly testing whether digital healthcare environment indicators modify the PHB–HRQoL association. The lack of robust effect modification for the primary EQ-5D-3L outcome does not support a simple pattern in which greater access to or use of digital healthcare services uniformly strengthens the association between proactive health behaviours and HRQoL. The differing findings for EQ-5D-3L and EQ-VAS further suggest that contextual moderation may depend on how health outcomes are measured and on the distribution of both environmental exposures and outcomes. Thus, the contribution of the moderation analysis lies in identifying contextual effect modification as a relevant but more conditional and measurement-sensitive question for future research rather than establishing a specific digital healthcare factor as a confirmed moderator.

Clinical practice and future research based on this study

The practical implications of these findings should be considered together with evidence from intervention studies rather than inferred from the present cross-sectional associations alone. Existing systematic reviews indicate that digital and behaviour-support interventions can improve health behaviours among older adults and people with chronic diseases, particularly when digital support is integrated with continued behavioural guidance and community-based or home-based care.^{33 59–61} In this context, age-friendly digital health programmes could combine digital literacy support with goal setting, behavioural guidance, repeated feedback, and family or social support rather than relying on information provision alone. The present findings identify PHB as an important behavioural correlate of both DHL and HRQoL, but they do not establish that increasing DHL or PHB will itself improve HRQoL. Future research should use longitudinal or experimental designs, objective service-use logs, validated multi-item measures of the digital healthcare environment and more inclusive sampling strategies that recruit individuals both within and outside routine chronic disease management systems.

Limitations

First, the cross-sectional design precludes establishment of temporal ordering among DHL, PHB and HRQoL; the observed indirect association should therefore not be interpreted as a causal pathway, and alternative temporal relationships remain possible. Second, the registry-based sampling frame and convenience recruitment limit generalisability. The findings are therefore most applicable to older adults with type 2 diabetes who are registered in and engaged with primary healthcare chronic disease management programmes at the participating sites; the sample should not be considered population-representative of all older adults with type 2 diabetes in Huai’an or in other primary care settings. Individuals who were undiagnosed, unregistered, disengaged from routine care, unable to attend or facing digital-access barriers could not enter the study or were under-represented, and a further 197 eligible individuals (15.8% of those contacted) did not participate. Because no demographic, clinical or digital-access information was available for non-participants, we could not determine whether they differed systematically from respondents or quantify the direction or magnitude of non-response bias. Since these groups are among those most likely to have low digital health literacy and low levels of proactive health behaviour, the observed levels of both variables may have been higher than those in the underlying source population. The direction of any bias in the estimated associations among DHL, PHB and HRQoL cannot, however, be determined from the available data, because non-participation may have depended on health status as well as on digital capability, and we cannot exclude the possibility that the associations take a different form among older adults disengaged from routine care. In addition, institution-level recruitment identifiers were not retained, preventing estimation of within-institution correlation; cluster-robust standard errors, GEE and multilevel models could not therefore be validly estimated, and if outcomes were positively correlated within institutions, the precision of the regression estimates may have been overstated. Third, residual confounding remains possible because potentially relevant factors, including income, cognitive status, depressive symptoms, diabetes complications, glycated haemoglobin (HbA1c), device ownership, internet access and caregiver support, were not measured. Finally, each digital healthcare environment construct was measured with a single item rather than a validated multi-item scale, so measurement error in these variables could not be quantified and may have attenuated the interaction estimates. The limited variation in remote service use and online health management use, together with the pronounced ceiling in EQ-5D-3L utility, may further have reduced sensitivity to small interaction effects, and the original sample-size calculation was not designed for moderation analyses, so adequate power to detect small effect modification cannot be assumed. Longitudinal studies using more diverse samples, objective indicators of digital healthcare use and more comprehensive covariate measurement are needed to confirm these findings.

Conclusions

Higher digital health literacy was associated with better HRQoL among older adults with type 2 diabetes, with a statistically significant cross-sectional indirect association observed through proactive health behaviours. The measured digital healthcare environment indicators did not show robust effect modification of the PHB–EQ-5D-3L association across model specifications. These findings identify digital health literacy and proactive health behaviours as closely related correlates of HRQoL in this population and support further longitudinal and intervention research to determine their temporal and clinical relevance.

Data availability statement

Data are available upon reasonable request.