Differential influences of social support on app use for diabetes self-management – a mixed methods approach

Patient self-management, in which patients are decision-makers in control of their daily diabetes management [6], moved to the core of diabetes care when research revealed that patient non-compliance rates were high in care approaches based solely on healthcare professional responsibility (“doctor knows best principle”) [6]. Wilkinson, Whitehead, and Ritchie [7] reported that the ability to self-manage a diabetic condition (e.g., diet adherence, physical activity, medication intake, managing psychological aspects [8, 9]) is influenced by a whole series of factors (e.g., in [10, 11]). Prominently, social support has been reported to influence self-management behaviors and outcomes, associated with improved patient lifestyle and clinical outcomes, and reduced psychosocial disease symptoms [2]. The impact of support has been distinguished between that provided by healthcare professionals [12‐14] and by the personal social patient network like relatives or peers [15‐17]. This impact on diabetes management and health outcomes has been examined extensively, and been compared to one another ( [1, 18]; also see [4]).

First, studies on professional support show the beneficial effects of professional support on diabetes outcomes [19], especially when health professionals empower patients by using shared decision-making and supportive communication. A collaborative process of joint decision-making with an active contribution from both parties and the involvement of patients in treatment decisions (providing choices, considering patient preferences, and encouraging them to ask questions [13, 19, 20]) have been shown to be most supportive in disease care [21].

Second, in terms of non-professional support, family and friends can deliver emotional and practical support daily in diabetes care [22‐25]. However, the personal social patient network has been reported to have a negative impact on diabetes self-management and health outcomes, when interactions like nagging behaviors, a lack in understanding, too much sympathy or pity expressed, or other negative behaviors occur [16, 26‐29]. The concept of “miscarried helping” represents some of these problems [30].

Research has shown that interventions and care approaches need to differentially address support by the professional system and by the patient’s personal network for achieving improved outcomes [1]. This distinction applies similarly to research related to technology-supported diabetes care [31], which includes the use of diabetes apps that are designed to help patients improve their self-management (e.g., logbook apps).

In previous research, the use of (e.g., cloud-based) technology for support by healthcare professionals was reported to mostly result in positive outcomes for diabetes self-management [32‐34]. Wehbe, Curcio, Gajjar, and Yadlapati [35] stated that the integration of technology into diabetes care affects physician-patient relationships positively when discussions are facilitated and when shared decision-making between physicians and patients is improved. However, according to these authors, the higher workload for physicians can also affect the relationship negatively. Abbasgholizadeh, Menear, Robitaille, and Legare [36] reported that health apps have potential for improved patient participation in shared health decision-making, but bear risks like security concerns or increased patient anxiety.

Beneficial effects of technology use on self-management and health outcomes were typically reported in studies that addressed non-professional social support by the patient’s personal networks [37, 38]. The use of social media, online communities, or mobile health (mHealth) has enhanced or improved psychological empowerment [39, 40], self-care information and knowledge [38, 41], blood glucose levels (HbA1c), as well as glucose self-monitoring and physical activity [3, 42], and other health aspects [43‐45]. The inclusion of family members or spouses in technological diabetes management systems was found to improve their understanding of self-management requirements and affected the communication about diabetes and the support of the patient positively [46]. Notably, different studies reported that online social support by non-professionals might not always result in beneficial self-management and health outcomes. For example, this occurs when participation in online communities leads to mismatches between needed and received support, or when support hinders health improvements for patients with high self-efficacy [47].

Existing studies almost exclusively focused on social support as an outcome of technology use, or on technology as a means for delivering social support. However, in order to increase an understanding of factors that influence technology use for self-care, research is still required that examines how professional and non-professional social support affect (mobile) diabetes technology use. It can be assumed that social support influences the use of mobile diabetes technology (e.g., apps) just as much as it influences traditional diabetes self-management behaviors (blood glucose testing, healthy nutrition, exercise, etc.) [4].

The number of studies examining factors influencing mHealth use is steadily rising [48], particularly in the context of disease self-management [49, 50]. While theoretical models of technology adoption [51] are one step towards understanding factors that influence (mobile) technology adoption and use, there is a gap in proving both significant and substantive usage effects for mHealth [52, 53]. Previous studies investigated factors influencing mHealth use in specific target groups (e.g. [54, 55]), for specific diseases or health topics [56‐60], and for sustained mHealth engagement [61]. This study aims to explicate the effect of social support.

Social support has been shown to play a role as an influencing factor on aspects of technology use in mHealth studies. Zhang et al. [59], for instance, found that the theoretical factors of social influence and performance expectancy had the strongest direct influence on behavioral intention to use diabetes apps for disease management. Likewise, Quaosar et al. [55] found that social influence had a significant impact on intention to use mHealth, specifically in populations of higher age. We aim to address the research gap in these studies by comparing differential influences of healthcare professionals and personal patient networks on technology for self-care.

In the first study we conducted 21 semi-structured face-to-face interviews with type 1 (T1DM) and type 2 (T2DM) adult diabetes patients in Singapore. The interview guide [63] included questions on patients’ previous diabetes app (non-) use, their diabetes self-management, their attitudes, the support received from their medical team, relatives, and from others, as well as other relevant factors from self-management [7] and technology adoption theories, such as Unified Theory of Acceptance and Use of Technology (UTAUT) [51].

In addition to an interview guide developed based on these theories, we used a standardized background information questionnaire to collect relevant information about the patient background (e.g., demographic data, disease information) [63]. Open questions collected information about app and technology (non-) use as part of the diabetes self-management, as well as professional and non-professional social support, including shared decision-making in the patients’ treatment, their communication with healthcare professionals, and the support they received from their personal social network. Regarding the latter, we focused on family and friends’ support, being their strong-tie partners [64].

Thematic analysis, as described by Braun and Clarke [65], was used for data analysis, following established standards for qualitative research [66]. A theme was defined as a topic resulting from different interview statements with similar content. First, preliminary categories were developed from the interviews regarding technology use, self-management, and social support using a matrix for data extracts. Broad themes were developed from the categories, which were reviewed against data extracts and the entire data set, before results were analyzed and interpreted (discussion among multiple researchers where results were ambiguous). A diabetes app (non-) user typology summarized the results on emerging themes and to map out differences in social support for different groups.

In the second study we conducted a standardized online survey with 65 type 1 and type 2 diabetes patients. The questionnaire included questions on their diabetes background, self-management behaviors, mHealth use, social support, general attitudes and feelings, psychological empowerment, and demography [63].

Previous diabetes app use or non-use, as well as the length and frequency of previous app use were measured based on the National Survey on Health App Use [67]. Decision-making styles by health professionals and communication with the patient were measured using the Provider Participatory Decision-making Style Scale [13] (PDMstyle, M = 3.25, SD = .08, α = .943) and the Provider Communication Scale [13] (PCOM, M = 3.50, SD = .37, α = .919). Items from these scales, for example, asked to “specify how often the doctors or health care professionals who take care of your diabetes offered you choices in your medical care” (PDMstyle, “none of the time” to “all of the time”) and “tell us how the doctors or health care professionals who take care of your diabetes are at explaining treatment alternatives” (PCOM, “poor” to “excellent”). 5-point Likert scales were used throughout the survey. Social support by the personal social patient network was measured with a scale on support by family and friends from the Diabetes Care Profile (DCP, section V, M = 3.29, SD = .19, α = .931) [68, 69]. Scale items included questions on support delivered by family and friends, for example, “my family or friends help and support me a lot to take my medicine” (DCP, “strongly agree” to “strongly disagree”). Diabetes background and self-management data were adapted from the Diabetes Care Profile [69], two psychological empowerment scales [70, 71], the National Survey on Health App Use [67], and the Summary of Diabetes Self-Care Activities [72].

The survey was pre-tested with sixteen participants with and without diabetes. The online survey was disseminated (snowball method) to former face-to-face interview participants, as well as through social media and diabetes support groups. A lottery participation (monetary incentives: three coupons of 100 SGD each) was offered at the end of the survey.

Binary logistic regression (enter, blockwise) was used to test H1 and H2 on social support by health professionals versus non-professionals (family and friends) influencing diabetes app use, and to compare the strength of both (and other relevant included self-management and diabetes background factors from self-management and technology adoption theory) on diabetes app use. Independent variables were metric or coded as dummy variables, while diabetes app use was used as a binary variable (previous diabetes app use or non-use). Independent factors that included data from both app users and non-users were used in binary logistic regression. Regression models were compared for model fit, prediction success, and the inclusion of significant predictors by starting with all available factors, and then removing factors step by step.

Table 1 summarizes demographic and disease characteristics of the interview (Study 1, N = 21) and the online survey samples (Study 2, N = 65). Overall, the samples were diverse, including diabetes patients from various demographic and disease-related subgroups. The interview sample mostly included active patients managing their condition, with five respondents at high risk for health complications due to lacking diabetes knowledge, lacking motivation and insufficient self-management, unfavorable self-management attitudes, and/or critical health conditions (e.g., high blood glucose values). In both samples there were more type 2 diabetes patients than type 1 diabetes patients, following the global prevalence (~ 90%) of type 2 diabetes [73].

As a foundation for comprehensively examining the influence of (H1) professional and (H2) non-professional support on diabetes app use, and to specify the hypotheses, themes were extracted from the interview data, relating to the aspects “previous app use for diabetes management”, “professional support”, and “non-professional support”. Table 2 summarizes the results in form of a diabetes app (non-) user typology, which displays the differences in social support in differing app user and non-user groups.

In Study 2 we tested hypotheses H1a, H1b, and H2a using binary logistic regression. Checking for autocorrelation of all independent variables, physician decision-making and physician-patient communication were highly correlated with r = .772, p < .01. Therefore, decision-making and communication were recoded into a single variable “physician-patient relationship” [4].

Based on theoretical considerations and the interview results, we began with a binary logistic regression model that included a maximum of independent factors: physician-patient relationship, medical specialty of the physician, family/friend support, and other relevant factors from technology adoption theory [(51] and from self-management theory [7] shown to be relevant predictors of technology use for disease self-management. We then compared different models by reducing independent factors to find the model with the best fit.

A model with good fit included the factors derived from the interviews physician-patient relationship, family/friend support, medical specialty of the physician, as well as the additional UTAUT factors’ perceived app potential (performance and effort expectancy), previous health information seeking online (technological experience), and age; and the self-management factors type of diabetes, length of diabetes, perceived health status, payment problems and insurance coverage (socioeconomic), blood glucose testing adherence (self-management behaviors), interest in innovation (attitudes), perceived diabetes knowledge, program or support group participation, and psychological empowerment. The test of this model against the constant-only model was statistically significant, indicating that the predictors as a set reliably distinguished between diabetes app use and non-use (χ²(16) = 26.752, p < .05). Nagelkerke’s R² = .656 indicated a high relationship between prediction and grouping (goodness-of-fit). Prediction success overall was 90% (80% for app non-use and 96% for app use). The Wald criterion demonstrated that only the family/friend support (Wald (1) = 5.315, p = .021) and the medical specialty of the consulted physician (dummy GP or specialist, Wald (1) = 4.014, p = .045) made a significant contribution to the prediction of diabetes app use. The Exp(β) value indicated that when the family/friend support was increased, the relative probability (odds ratio) that diabetes apps were used decreased with Exp(β) = .044, β = − 3.131. The Exp(β) value indicated that when the patients were supervised by specialist doctors the relative probability (odds ratio) that diabetes apps were used increased with Exp(β) = 9460.805, β = 9.155. In contrast to non-professional support (family/friends), the physician-patient relationship was not found to be a significant predictor in the model.

A model resulting after the removal of the factors interest in innovation, insurance coverage, and program or support group participation (due to lacking significance) showed a slightly lower prediction success with 75% (70% for app non-use and 79% for app use), but overall model significance with χ²(13) = 26.936, p < .05, and a moderate to high relationship between prediction and grouping (goodness of fit) with Nagelkerke’s R² = .509. In this model, the Wald criterion demonstrated that the family/friend support (Wald (1) = 6.617, p = .010) and perceived health status (Wald (1) = 7.839, p = .005) made a significant contribution to the prediction of diabetes app use. Again, the Exp(β) value showed that when support by family/friends was increased, the relative probability (odds ratio) that diabetes apps were used decreased with Exp(β) = .283, β = − 1.261. The Exp(β) value also indicated that when the perceived health status was improved the relative probability (odds ratio) that diabetes apps were used increased with Exp(β) = 8.030, β = 2.083.

Further reducing the independent factors, the models showed similar results to the last model, resulting in family/friend support and the perceived health status being significant predictors of diabetes app use. A further reduction of factors decreased the model fit, yet the medical specialty of the physician nearly reached significance again.

Overall, after testing various models, only the family/friend support, the medical specialty of the supervising physician, and the perceived health status significantly predicted diabetes app use. Less family/friend support was likely leading to a higher chance of diabetes app use, while the use of diabetes specialists or a better perceived health status increased the chance of app use for self-management.

Despite the physician-patient relationship lacking significance for predicting diabetes app use in the models (apart from the medical specialty of the physician as a significant predictor), additional t-test calculations showed that the medical specialty of the physician related to the style of decision-making and the physician-patient communication. There were differences between specialists and GPs, with specialist care positively related to higher shared decision-making and better physician-patient communication as compared to GPs (Decision-making PDMstyle: specialists: M = 3.72, SD = 1.16, n = 37, general practitioners: M = 2.65, SD = 1.30, n = 22; t [57] = − 3.275, p < .01, N = 59; Communication PCOM: specialists: M = 3.82, SD = 1.01, n = 37, general practitioners: M = 3.13, SD = 1.09, n = 22; t [57] = − 2.472, p < .05, N = 59) (compare [4]).

Future research needs to examine aspects of social support influencing mHealth use, also comparing the influence of professional versus non-professional support on mHealth use. Apart from the social influence of health professionals and of family and friends on mHealth use as studied here, previous studies show that “peer” patients also play an important role as sources of information and of empowerment for patients [38, 81]. Thus, firstly, follow-up research should include the role of patient peers. Secondly, the investigation can be extended from physicians to other healthcare professionals involved (e.g., nurse educators). Thirdly, mHealth studies should extend their theoretical foundations beyond technology adoption theories. There are few theories applied in mHealth research, which mainly relies on traditional behavior change theories [82, 83]. To investigate factors influencing and predicting technology use for self-management, a broader theoretical foundation needs to be available, overcoming the limitations of a very narrow range of theories. Fourthly, even though the socioeconomic and demographic factors included in the regression analysis (e.g., age, insurance coverage) did not predict the use of diabetes apps, it has to be further examined if these factors have an impact on the results, like the association between the specialty of the physician and the use of diabetes technology, or the negative association between non-professional support and technology use. Similarly, and fifth, an influence of app characteristics and app quality on diabetes app use is to be expected and follow-up research has to examine how these app characteristics interrelate with other factors influencing app use for diabetes self-management, including aspects of social support.

While our study focused on diabetes apps, these results can equally inform mHealth research in other health domains involving self-management, including various chronic conditions such as heart diseases, obesity, asthma, stroke, cancer, arthritis, hypertension, multiple sclerosis, etc. [84, 85]. Even though there are specific characteristics for each disease, general knowledge about technology use for self-management can be helpful across health conditions to design technology accordingly.

For diabetes policy and practice, the study results suggest that there are notable differences in patient groups regarding diabetes app use, with diabetes patients being a heterogeneous group with varying needs [86]. Segmentation theory needs to be used to tailor diabetes app characteristics specifically for subgroups with varying preferences and needs regarding technology (also considering socioeconomic, cultural, and demographic differences). Moreover, influencing factors need to be considered when designing technology for diabetes care, like the influence of aspects of social support and social influence [87]. Both healthcare professional supervision and family/friend support can influence technology-supported self-management. In particular, the care provided by GPs was perceived as unsatisfactory by most diabetes patients in the Study 1 sample. This resulted in less mobile technology use as compared to specialist supervision, which deficiency should be addressed to achieve equal care for all patients. The inclusion of relatives in (technology-supported) self-care should be further investigated depending on patient preferences and the usefulness of support provided (to avoid misguided support).

Study limitations included the small sample size of the online survey, which limited sophisticated multi-variate data analysis [4] and broad generalizability. Considerable effort was undertaken to achieve a larger sample size (extension of the field phase, seeking contact with official care organizations, seeking cooperation with healthcare professionals, repeated invitations for survey participation), yet without success. Singaporean citizens suffering from diabetes receive a large amount of invitations for study participation since the government declared “the war against diabetes” in 2016 [88], so these are a prized yet scarce resource for research. Due to the cross-sectional nature of the study design, final conclusions about causality could not be drawn. Additionally, due to the recruitment method a certain degree of self-selection was unavoidable (over-representation of support group members and educated participants). The sample was therefore not representative for all Singaporean individuals with diabetes; however, we managed to include participants of the main Singaporean cultural backgrounds in the sample (Chinese, Malay, Indian). These three groups comprise 97% of Singaporeans [89]. Finally, triangulation of the study findings with results on diabetes app features and quality [4, 5] was not undertaken here due to length limitations.