The study used a convenience sampling method and included adults (≥18 years) with type 1 diabetes who could understand Swedish. Women with type 1 diabetes were excluded if pregnant due to changes in maternal insulin sensitivity during pregnancy, as this may require alterations in their treatment plan [26]. This could indirectly influence other predictor variables like well-being and psychosocial self-efficacy [27,28].

Participants were recruited primarily through advertisements on social media, particularly Facebook (using the marketing feature as well as posting in private groups for people with diabetes in Sweden). In addition, advertisements were placed on the websites of various associations for people with diabetes in Sweden and at a diabetes center in a regional hospital. More details on recruitment methods are available in a previously published paper [29].

The sample size was calculated using SPSS (version 28; IBM Corp). A sample size of 270 participants was required in order to detect an association between the dependent and predictor variables using a regression model. This calculation was based on an F test with 20% predictability using 25 predictors in the full model and 15% predictability with 10 predictors in the nested model at 80% power and a .05 level of significance. To account for potential missing values, we decided to include 300 participants in the study.

Data were collected between September and November 2022 (approximately 2 months) until the desired sample size was reached, primarily through a web-based survey (Survey&Report platform by Artisans Media). The survey could be accessed via a website link or QR code provided in the advertisement flyer. Three screening questions (age, diabetes type, and pregnancy status) at the beginning of the survey helped determine eligibility to participate. The survey closed automatically if any of the exclusion criteria were met. Alternatively, participants could opt to answer a paper-based survey, which was sent to the address they provided (n=6). The survey was in Swedish and was part of a larger study. It had 64 questions in total, and data from 55 questions have been included in this paper. Certain questions were marked as mandatory, requiring participants to answer them before proceeding to the next page. Additionally, questions that were not applicable were hidden based on the participant’s responses to the preceding question. Thus, the number of questions each participant answered varied from 53 to 64. Participants had the option to partially complete the survey and save their progress to finish it at a later time. Therefore, the duration taken to answer the web-based survey varied highly from 5 minutes to 1.5 days. The majority (273/295, 92.5%) answered the web-based survey in 60 minutes, with 15.2% (45/295) answering it in less than 8 minutes.

This study was conducted in accordance with the World Medical Association’s Helsinki Declaration. The study plan was reviewed by the Swedish Ethical Review Authority, and ethics approval (Dnr: 2021-05337-01 and Dnr: 2022-04079-02) was received for this paper before the commencement of data collection. Participation in the survey was voluntary, and informed consent was obtained from all participants either via the survey tool or in written form. The participants did not receive any remuneration or compensation for their participation in the study. To deidentify the data and protect participant privacy, the raw data were pseudonymized either using the web survey tool or using codes and keys (for paper surveys). In addition, the survey tool, cloud storage (Sunet Drive), laptops, and software used in the analysis were procured by Karlstad University, ensuring the European Union’s General Data Protection Regulation.

Data analysis was conducted using SPSS (version 28; IBM Corp). Mean and SD or frequency and percentages were used to describe the characteristics of the included participants. In these data, residuals were found to be normally distributed, homoscedastic, and free from multicollinearity. Nested linear regression models were used to identify the best-fitting model. Predictor variables were grouped into 3 blocks. Block 1 consisted of demographic variables, block 2 comprised disease-specific variables, and block 3 involved well-being. Model 1 included variables from block 1, model 2 included variables from block 1 and block 2, and model 3 encompassed variables from all 3 blocks. Multiple linear regression was run using the enter method to identify the best-fitting model. A P value of <.05 was considered to be statistically significant. No imputations were performed for missing values.

The final sample size achieved was 301. Data from participants with missing values in 1 or more of the predictor variables were excluded from the regression analysis (n=16), resulting in a sample size of 285 participants for analysis. A survey completion rate of 68.4% (301/440) was achieved for the web-based survey. This was calculated by dividing the number of participants who completed the survey and was included in the final sample by the total number of participants who initiated answering the survey (see Figure 1 for more details).

The mean Sw-eHEALS score among this sample was 33.42 (SD 5.32; range 8‐40). A ceiling effect in the Sw-eHEALS score (with the maximum score of 40 achieved by 56/301, 18.6% of participants) was found in this sample (see Figure 2 for more details). The majority of participants completed the survey digitally (295/301, 98%). Participants had a mean age of 43 (SD 16) years, with the majority being women (215/301, 71.4%). See Table 1 for descriptive statistics on variables included in the regression analysis. All participants (301/301, 100%) reported using 1 or more forms of digital device for their diabetes self-care. Digital device use by participants consisted of blood glucose meters (146/301, 48.5%), intermittent scanning CGM (119/301, 39.5%), real-time CGM (156/301, 51.8%), CSII (102/301, 33.9%) pumps, AID (71/301, 23.6%), and smart insulin pens (28/301, 9.3%). See Table 2 for details on the Sw-eHEALS score in relation to DHTs used by the participants.

Nested linear regression models were used to explore the associations between the outcome variable, eHealth literacy, and predictor variables. Model 1, comprising demographic variables alone, accounted for 12.9% of the total variation in eHealth literacy, with age, education level, and income level showing associations with the Sw-eHEALS score. Model 2, involving both demographic and disease-specific variables, explained 31.5% of the total variation in eHealth literacy and was deemed the best-fitting model. In model 2, the predictors’ age, glycated hemoglobin (HbA_1c), and psychosocial self-efficacy showed associations with the Sw-eHEALS score after adjusting for demographic and disease-specific variables. Model 3, involving all predictors (ie, demographic and disease-specific variables and well-being), explained 31.6% of the variance in eHealth literacy. However, the F change for this model was not significant and therefore is not presented here. See Table 3 for detailed results of the regression analyses.

This study explored associations between eHealth literacy and demographic factors, disease-specific factors, and well-being among adults with type 1 diabetes. The sample in this study was slightly younger, predominantly women, and had a shorter duration of diabetes compared to the statistics on adults with type 1 diabetes published by the Swedish National Diabetes Register [6]. The majority of the participants in this study had a university-level education, which is not in line with studies reporting on type 1 diabetes population from Sweden [36] or other countries [37]. The mean Sw-eHEALS score among this sample was higher than that found in other studies among people with type 2 diabetes [38], the general population [30], and older adults [14] in Sweden and in other cultural and language settings [39]. Comparable empirical studies on eHealth literacy among adults with type 1 diabetes were not found. The mean Sw-eHEALS score is slightly higher among participants who use 3 or more digital devices, mHealth app users, and users of various features. However, this difference is too minor to draw a conclusion.

Similar to our results, other studies have found that younger age [14,22] and self-efficacy [18,40] are associated with higher eHealth literacy scores. However, in contrast to our findings, some studies found no association between eHealth literacy scores and age [17,39]. Additionally, some studies did not find any association between eHealth literacy and gender [15,17], education, or income [17], which aligns with this study’s findings when adjusted for disease-specific factors. Conversely, some studies found significant associations of eHealth literacy with gender [14], education level [15], employment status, well-being, and living status [17]. In this study, higher eHealth literacy was associated with lower HbA_1c levels, but similar studies to compare our results were not found. Similar to our findings, studies have found a relationship between HbA_1c levels and health literacy [41,42] or functional health literacy [43]. In contrast, other studies found no association between HbA_1c levels and mobile eHealth literacy [44] or functional health literacy [45]. However, the finding on the association between higher eHealth literacy and lower HbA_1c levels needs to be read with caution, considering the near normal range mean HbA_1c levels, self-reported HbA_1c, ceiling effect in Sw-eHEALS, and various other uncontrolled factors that could influence HbA_1c levels in this sample. Therefore, further studies are needed to determine the clinical relevance of this finding.

eHealth literacy and its association with various factors among people with type 1 diabetes is a less studied area. This study utilized widely used and validated questionnaires to measure eHealth literacy [30,31], psychosocial self-efficacy [32,33], and well-being [46]. Other questions in the survey were pilot-tested to validate their content among the targeted population and health professionals. We achieved a sufficient sample size to perform regression analysis with adequate power. The study also had higher than average completion rates for a web-based survey [47]. The total survey response time of less than 8 minutes, which may indicate insufficient effort responding, was seen in 15.2% (45/295) of the sample who answered the web-based survey, reducing the risk of inflated correlations [48]. However, we have not done an in-depth analysis to detect and eliminate insufficient effort responding. The majority of participants were recruited via social media, allowing for recruitment from all over Sweden, which strengthens the study’s transferability. Additionally, the higher rate of digital survey responses compared to paper format responses may imply that participants with higher eHealth literacy were more likely to volunteer, potentially leading to selection bias. We may also have missed participants who do not use social media. The sample in this study consisted entirely of DHT users, which is not surprising, given that CGM and CSII use is high in Sweden [6], as it is financed through a publicly funded high-cost protection scheme [49].

The outcome variable, the eHealth Literacy Scale (eHEALS) score, is a valid and reliable measure of self-reported eHealth literacy among people with chronic diseases [50]. This instrument has been widely tested, used in diverse populations, and has sufficient moderate quality evidence for comprehensibility [51]. However, the eHEALS instrument has its weaknesses. The original eHEALS measures people’s perceived skills with eHealth and is an indirect measure of eHealth literacy [31]. It is a single-factor scale, which was developed before the time of social media and mHealth, prioritizing ease of administration [52]. Therefore, it is not updated to account for the current dynamicity, interactivity, and multifaceted nature of the internet, social media, and mobile web [51-53]. This has led to the development of newer, more relevant instruments to measure eHealth literacy [53-55]. Findings from this study, therefore, call for further research in this field using newer measures that account for the dynamicity and evolving nature of eHealth literacy.

The ceiling effect in the eHEALS score seen in this study (Figure 2) may have led to an inability to capture true differences between participants achieving the highest possible score, thus reducing the reliability of the results [56]. It may also point toward the outdated content validity of this instrument [56] in the current digital era. However, this ceiling effect has not been previously reported in other studies using the same instrument [30,38,39]. The results of this study, therefore, should be generalized with caution, considering the advanced DHTs currently used by people with type 1 diabetes.

In this study, associations were found between eHealth literacy and age, psychosocial self-efficacy, and HbA_1c levels. People with lower HbA_1c levels had higher eHealth literacy scores, which may indicate their ability to effectively use electronic information and DHT to manage their glucose levels. Interventions to improve eHealth literacy in this population are therefore important for better glucose control. Therefore, further studies focusing on the development and testing of eHealth literacy interventions are recommended. Our results highlight the importance of considering people’s age and psychosocial self-efficacy in acquiring appropriate eHealth literacy. Therefore, eHealth literacy interventions should be tailored to meet the needs of people in varying age groups and with different levels of psychosocial self-efficacy. Further studies in this area are therefore recommended.

The use of nested regression models is a strength of this study, improving data generalizability. However, the results of this paper are to be interpreted with caution, especially due to the ceiling effect observed in the eHealth literacy scores. Further studies in this area using newer eHealth literacy tools are important to validate our findings.