Social exclusion in people with diabetes: cross-sectional and longitudinal results from the German Ageing Survey (DEAS)

As social exclusion can be linked to worse health and overall reduced quality of life, we describe social exclusion in people with diabetes and assess whether diabetes can be considered as a risk factor for social exclusion. We analyzed two waves (2014, 2017, N = 6604) from a survey of community-dwelling people aged > 40 using linear regression, group comparison and generalized estimating equations to explore the association between diabetes, social exclusion, socioeconomic, physical and psychosocial variables. In the entire cohort, diabetes was cross-sectionally associated with social exclusion after adjusting for covariates (p = 0.001). In people with diabetes, social exclusion was further associated with self-esteem (p < 0.001), loneliness (p =  < 0.001), income (p = 0.017), depression (p = 0.001), physical diseases (p = 0.04), and network size (p = 0.043). Longitudinal data revealed that higher levels of social exclusion were already present before the diagnosis of diabetes, and future social exclusion was predicted by self-esteem, loneliness, depression, and income, but not by diabetes (p = .221). We conclude that diabetes is not a driver of social exclusion. Instead, both seem to co-occur as a consequence of health-related and psychosocial variables.


Methods
Sample. The data were taken from the public release of the German Ageing Survey (Deutscher Alterssurvey, DEAS), conducted and provided by the Research Data Centre of the German Centre of Gerontology (Deutsches Zentrum für Altersfragen, DZA) and funded by the Federal Ministry for Family Affairs, Senior Citizens, Women and Youth. The DEAS is a representative cross-sectional and longitudinal survey of the community-dwelling population aged 40 and above in Germany, with the main goal of assessing physical and mental health, living conditions, psychosocial parameters and well-being in middle-aged and older adults 38 . For this purpose, representative population samples were drawn at each wave (cross-sectional data) and participants completed a computer-assisted interview (CAPI) as well as a drop-off self-report questionnaire. Participants of each wave were also invited to complete future waves, leading to a rich longitudinal data register. Therefore, it is well-suited to provide data about social and medical determinants of well-being and provides a multitude of variables relevant to the presented research questions. The DEAS is an ongoing survey with currently six waves  and several shorter surveys during the COVID-19 pandemic (2020-2022) 37 . In accordance with the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG) and the Institute for Applied Social Science (INFAS), no ethics approval was needed for the study as data were collected under pseudonyms and voluntarily, and the study was deemed low-risk, as no work on patients was included. Nonetheless, both institutions approved the study, and data collection was conducted according to the Declaration of Helsinki. Written informed consent was obtained from all participants. For more information, please refer to the homepage of the DZA (https:// www. dza. de/ en/ resea rch/ fdz/ german-ageing-survey) and the related publications [38][39][40][41] . As the questionnaires and variables included in the survey vary between waves (social isolation, for example, was only added to the survey instrument list in 2014), we selected the most recent two waves that contained all the relevant variables (see below). We refrained from using the latest waves due to the potential bias introduced to the data collected during the COVID-19 pandemic, which strongly influenced physical and mental health. Therefore, we used cross-sectional and longitudinal data from the fifth (2014) and sixth wave (2017) (see section Data Availability). The study population included people with and without diabetes in both waves.
Variables. In the CAPI, the presence of diabetes was assessed via self-report (yes/no); participants were asked if they had ever been told by a doctor that they suffered from diabetes -of note, the question did not differentiate between type 1 and type 2 diabetes.
Social exclusion was assessed using the scale by Bude and Lantermann 7 . It consists of four items ranging from 1 = "strongly agree" to 4 = "strongly disagree": "I am worried to be left behind", "I feel like I do not really belong to society", "I feel that I am left out", and "I feel excluded from society". In line with the DEAS guidelines, the sum score of the scale was treated as a continuous variable 42 , with higher values representing higher perceived social exclusion.
In addition, the following sociodemographic, psychosocial, and medical covariates were extracted from the database: Statistical analysis. All analyses were conducted using IBM SPSS statistics (Version 25), JASP (Version 0.16), and R (Version 4.1.1). The statistical significance was determined with p < 0.05. Missing values were treated with pairwise deletion. All cross-sectional analyses were performed based on the sixth wave (2017).
First, descriptive statistics were used to characterize the sample. Normality was assessed with the Shapiro-Wilk test, revealing non-normal distributions (p < 0.001) for all variables. Univariate group comparisons via Mann Whitney U test or Chi 2 test were performed to determine differences between people with and without diabetes. Second, multiple linear regressions were used to analyze the association between social exclusion (dependent variable) and the above mentioned covariates (independent variables) using a stepwise selection algorithm and the AIC as selection criterion. Multicollinearity was assessed using the variance inflation criterion (VIF), revealing values between 1.07 and 2.20 at most. VIF values between 1 and 5 can be considered low to moderate; as it is common practice to remove variables with a VIF ≥ 5, no variables were removed in our models 50 . Finally, the dynamics of social exclusion were studied between the wave 2014 and 2017 by using paired Wilcoxon test and Generalized Estimating Equations (GEE) to account for repeated measures and within-person design.

Results
Factors associated with social exclusion. As a first step, we aimed to understand how participants with diabetes differed from those without diabetes in terms of health, well-being and social exclusion. Of the N = 6604 participants in wave six (2017), 13.5% (N = 897) reported to have been diagnosed diabetes mellitus (mean age 70.7 ± 9.5 years, 42.6% female). Detailed sample characteristics of the respective participants are given in Table 1. Group comparisons between participants with and without diabetes revealed higher levels of social exclusion in participants with diabetes in both waves ( Fig. 1), with small-to-medium effect size (r = -0.153 in 2017 and r = -0.127 in 2014). Likewise, people with diabetes were older (r = -0.257), had a higher BMI (r = -0.372), a higher number of physical diseases (r = -0.477) and worse physical functioning according to the SF36 (r = 0.336). Additionally, people with diabetes had a lower monthly income (r = 0.217), lower self-esteem (r = 0.136) and autonomy (r = 0.109), and more depressive symptoms (r = -0.154) (see Table 1).
When using stepwise regression models to assess social exclusion in the entire cohort (N = 6604), the diagnosis of diabetes was independently associated with social exclusion (p = 0.001) after adjusting for psychosocial and sociodemographic covariates (Table 2a). Therefore, we proceeded to assess the predictors of social exclusion in the entire study population as well as in people with diabetes in particular.
The strongest predictor of social exclusion was self-esteem, both in the entire cohort (p < 0.001, Table 2a) and when repeating the regression analysis in people with diabetes (p < 0.001, Table 2b). Both in people with and without diabetes, social exclusion was additionally linked to loneliness (p < 0.001), income (p < 0.001 in the entire cohort and p = 0.017 in the diabetes cohort), depression (p < 0.001 and p = 0.001), number of physical diseases (p = 0.01 and p = 0.040), and overall satisfaction with life (p = 0.001 and p = 0.047).

When does social exclusion occur?
To understand how social exclusion and diabetes in particular are linked, we used longitudinal data to assess whether newly diagnosed diabetes serves as a predictor of subsequent social exclusion. Overall, 6265 persons were interviewed both in wave five (2014) and six (2017). From 2014 to 2017, 188 people reported new onset of diabetes. Group comparison of baseline parameters in 2014 among 1) people who were newly diagnosed with diabetes between 2014 and 2017, 2) people who already reported diabetes in 2014, and 3) people who did not develop diabetes is given in Table 3a. Of note, the level of social exclusion in people who were newly diagnosed with diabetes in 2017 differed significantly from those who did not report new-onset diabetes, but not from those who already reported diabetes in 2014, indicating that a higher level of social exclusion was already present before the diagnosis of diabetes. People who did not have diabetes in 2014 but developed diabetes by 2017 were already characterized by more chronic disorders, higher BMI, higher levels of both depression and loneliness, lower income, and poorer self-esteem in 2014 (Table 3a). To confirm this pattern, we performed a GEE on the variable social exclusion in wave six (2017) using variables from wave 2014 and 2017. Here, like in the previous models, future social exclusion was mainly predicted by self-esteem, loneliness, depression, and income (

Discussion
In a representative study population of middle-aged and older German adults, we used group comparisons, linear regression and GEE to assess how diabetes and social exclusion are linked. While social exclusion is increased in people with diabetes, our results indicate that diabetes itself only has a weak direct effect on social exclusion. Instead, social exclusion in people with diabetes is mainly driven by Table 1. Characteristics of the entire cohort (wave 2017). *Group comparison performed for people with and without diabetes. P-values based on Mann-Whitney U test for metric and chi 2 test for categorical variables. Effect size for categorical variables = Cramer´s V, Effect size for metric variables = rank biserial correlation. Effect size less than 0.3 indicate a small, between 0.3 and 0.5 a medium, and effect sizes greater than 0.5 a large effect 50 . M Mean, SD Standard Deviation. www.nature.com/scientificreports/ psychosocial aspects such as self-esteem, depression, social network size, and socioeconomic parameters. This is in line with our longitudinal observation that social exclusion precedes the onset of diabetes. These results indicate that the presence of diabetes itself is not necessarily an independent driver of social exclusion; instead, social exclusion in diabetes can be interpreted as a consequence of several other health-related and psychosocial conditions that occur in people with diabetes. Our study therefore extends the understanding of the association between diabetes and social exclusion and the moderating effects of the following various psychosocial factors 10,11 . These findings will be discussed in detail below. We found that the strongest effect between diabetes and social exclusion was evident for self-esteem. In line with an earlier study 26 , people with diabetes reported lower self-esteem than people without diabetes. Self-esteem is the degree to which people have a favorable or unfavorable opinion of themselves and is significantly related to both mental and physical health 51 . It is considered an important psychological factor, even influencing glucose level and eventually the course of diabetes via psycho-neuroendocrine mechanisms or through stress-related unhealthy behavior 26 52 on stigma, which report the perception of being responsible for having diabetes and considering the diagnosis a personal failure as the most prevalent aspects of stigma in diabetes. Our data underline the importance of self-esteem for well-being and social activity in people with diabetes, however, further studies are necessary to determine which aspects of self-esteem are particularly relevant. Moreover, cultural aspects of body-related self-esteem have to be taken into account when clinicians want to encourage positive body image because of its potential health benefits 53 . Interestingly, the BMI was not a significant predictor of social exclusion. This suggests that lower self-esteem cannot be reduced to being overweight or obese in the cohort of people with diabetes. It also shows that not obesity per se, rather than the own view towards the weight, is relevant for exclusion 54 . Additionally, as people with diabetes reported significantly more physical illnesses, the higher BMI Table 2. Linear regression for social exclusion (wave 2017) in A) the entire cohort and B) people with diabetes. a Stepwise selection, AIC. Adjusted R 2 = 0.43. Entered independent variables: age, sex, social network size, Body-Mass-Index (BMI), SF36 Physical functioning (standardized), Education, Total number of physical diseases, Depression Scale ADS, 6-Item Scale for Loneliness, Satisfaction With Life, Generalized Self-Efficacy Scale, Self Esteem Scale, Scale noticed autonomy in older age, Monthly equivalence income, Intra-familiar relationship, diabetes (yes/no). Data from the entire cohort (N = 6604) from wave 2017. b Stepwise selection, AIC. Adjusted R 2 = 0.42. Entered independent variables: age, sex, social network size, Body-Mass-Index (BMI), SF36 Physical functioning (standardized), Total number of physical diseases, Depression Scale ADS, 6-Item Scale for Loneliness, Satisfaction With Life, Generalized Self-Efficacy Scale, Self Esteem Scale, Scale noticed autonomy in older age, Monthly equivalence income, Intra-familiar relationship, Education. Data from people with diabetes (N = 897) from wave 2017. www.nature.com/scientificreports/ may also be a byproduct of worse physical health and thus lack of activity, suggesting that it is not the body image but rather the lack of physical ability to participate that drives social exclusion. The second most important predictor of social exclusion in people with diabetes was loneliness. Patients with diabetes frequently experience moderate loneliness 28 . As in our study, loneliness in people with diabetes was found to be associated with the presence of chronic disorders and younger age. At this point, it is worth noting that social exclusion and loneliness are distinct concepts. Perceived social exclusion describes the feeling that one does not belong to the society, whereas loneliness is the state that an individual's social network is smaller than desired or the resulting support is lower than expected 42 . Loneliness is an emerging issue that is associated with deleterious outcomes and poor health 55 . For example, higher levels of loneliness were associated with subsequent higher levels of functional limitations, and higher levels of functional limitations were in turn associated with subsequent higher levels of loneliness, suggesting that the association between loneliness and functional limitations  www.nature.com/scientificreports/ among people with diabetes is bidirectional 56 . Loneliness is a complex and multidimensional phenomenon, and the utilized De Jong Gierveld scale 45 in the studied dataset is based on a multidimensional perspective containing overall, emotional, and social loneliness 45 .
The third most important predictor of social exclusion in our analysis was depression. In people with diabetes, we found higher levels of depressiveness in comparison to people without diabetes, although the effect size was small. In general, people with diabetes and especially those with obesity and physical inactivity, have an increased risk for depression 57 . Overall, lack of self-esteem, loneliness and poorer physical health are often associated with depression 58 .
In addition to self-esteem, loneliness, and depression, also the income, the number of chronic disorders, the social network size, life satisfaction, and autonomy were found to be associated with social exclusion in our analysis as well as in previous studies on stigma 4,12 . The current study suggests that the social network size mediates the effect on the relationship between diabetes and social exclusion. A large social network was associated with better intra-familiar relationships. One can therefore assume that social support and familial support are critical factors in overcoming social exclusion, as they may buffer the lack of acceptance from society in general 59 . The same can be assumed for a higher income.
Overall, our results show that social exclusion takes place in people with diabetes, and highlight the need to provide psychosocial support to people with diabetes in particular.

Limitations
Our study is not free of limitations. Social aspects depend on cultural and economic characteristics within a society 11 and may differ from country to country. This limits the generalizability of our results and highlights the need for further studies to take cultural aspects into account. Especially in Germany, a universal welfare state that offers insurance and financial support to all citizens, results regarding social exclusion may differ from other countries where such welfare structures are not in place 60 . Still, even in a developed welfare state, there are significant differences in social exclusion, health, and income for people with and without diabetes, but depending on the country and the supportive infrastructures, the predictors of social inclusion may vary. These country-specific limitations also hold for the practical implications of the presented results, as measures to reduce social exclusion depend on the structures already in place (e.g. financial support, caregivers, reduction of stigma in society, insurance, availability of public transport and accessible buildings for handicapped persons).
In addition, different measures for social exclusion exist and may partly explain the mixed findings, as the Bude and Lantermann 7 tool, which was used here, covers only perceived social isolation. Generally, the provided data is based on self-report, which is always subject to bias, such as recall-bias and social desirability 61,62 . However, all instruments used in the surveys are validated and frequently used in the scientific literature, and self-report is required to assess subjective constructs such as life satisfaction, depressive symptomology, or feeling excluded 1,8 . In addition, the use of a nationwide survey limits generalizability of the results, as a potential selection bias cannot be excluded. It is likely that people in nursing homes or hospitals, who may suffer from much more severe levels of both diabetes and social exclusion, are underrepresented in this dataset. Likewise, in the provided dataset, there was no differentiation between type 1 and type 2 diabetes, merely the overall diagnosis of diabetes was assessed. Again, self-report of diabetes may be critical, especially if blood-glucose levels are stable (e.g. due to medication) and people do not 'feel' that they have diabetes. To counteract this risk, the survey explicitly asks "Has a doctor ever told you that you are suffering from [Diabetes]" 39 . Likewise, many studies show that self-report of diabetes as used in survey data is reliable, especially when searching for social implications 63 rather than medical aspects of the disease, where a more detailed assessment may be necessary [64][65][66] . Still, in future studies, it may be fruitful to differentially assess the relevance of social exclusion for the two types of diabetes. As diabetes type 1 often starts earlier in life, its dynamics may be different from type 2, for example due to the use of insulin 67,68 , although people with type 2 diabetes also report social distress 69 . Additionally, due to the nature of the dataset, the current occupational status was not properly represented. Based on the mean age of the participants, it is likely that most participants were retired. Still, as people without diabetes were younger than those with diabetes, it is possible that the occupational status may differ here. To incorporate the role of occupation for social exclusion, we included both education and monthly net income in our analyses, however, these variables cannot fully capture the psychological differences between work and retirement in terms of social roles, network, and daily life structure. Therefore, in future research, it would be of interest to differentially assess the relationship between social exclusion and occupational status, as it has been shown that occupational status influences social network size and well-being, especially mental health [70][71][72][73] .
Although the current analysis is strengthened by the large sample size and longitudinal data collection, it remains an exploratory overview to initially assess whether diabetes and social exclusion are linked at all and which variables contribute to social exclusion in this particular patient population. In future studies, it would be beneficial to understand how exactly diabetes and social exclusion are linked by using structural equation modelling or mediation analysis. This approach may shed further light into the direction of effects, especially concerning the relationship between physical heath/multimorbidity, self-esteem, social exclusion, and diabetes. This allows a deeper insight into the parameters linking social exclusion in diabetes, however, a stronger theoretical foundation of the important variables is necessary first.

Conclusion
Social exclusion is relevant in people with diabetes, however, the illness alone is not a predictor of social exclusion. Instead, social exclusion in people with diabetes is mainly driven by psychosocial and health-related factors that are connected to the illness, which explains the cross-sectional association between both. Longitudinal results shed a light on the occurrence of social exclusion before the diagnosis of diabetes, showcasing that health-related