Diabetes Distress, Depressive Symptoms, and Anxiety Symptoms in People With Type 2 Diabetes: A Network Analysis Approach to Understanding Comorbidity

Type 2 diabetes is associated with a significant mental health burden (1). For instance, depression (2) and symptoms of anxiety (3) have been shown to be prevalent in people with type 2 diabetes. People with type 2 diabetes also risk experiencing diabetes-related distress, a diabetes-specific mental health comorbidity associated with reduced self-management and higher HbA_1c (4,5) Diabetes distress reflects feelings related to physical activity, diet, future complications, and doctor-patient relationships (6). Diabetes distress is commonly measured with the Problem Areas in Diabetes (PAID) questionnaire (7) or the Diabetes Distress Scale (DDS) (6). The conceptual distinction between the PAID and the DDS, and whether they measure the same or different constructs, has been the subject of recent debate (8,9). However, both the PAID and the DDS encapsulate a broad range of emotional problems associated with living with diabetes.

Co-occurrence of diabetes distress, depressive symptoms, and anxiety symptoms within the same person with type 2 diabetes is also common. Fisher et al. (10) found that, among people with type 2 diabetes, ∼30% of those with major depressive disorder and ∼50% of those with generalized anxiety disorder met the criteria for a major depressive disorder/generalized anxiety disorder dual diagnosis. Reported comorbidity between diabetes distress and depressive symptoms has ranged from 4.5% (11) to 22.5–75.4% (12). Longitudinal research with people with type 2 diabetes has shown diabetes distress and depressive symptoms to be cyclically related, influencing each other over time (13). Questions have also been raised as to the overlap between depression and diabetes distress, with it being suggested that depression and diabetes distress are both aspects of the same emotional distress continuum (14) and, conversely, intersecting but distinct concepts (11). There are anxieties that are unique to living with type 2 diabetes (15). Worries about hypoglycemic reactions, high blood glucose, and future complications are among the most frequently endorsed emotional problems for people with diabetes (16,17). Clarifying if and how specific diabetes distress problems, depressive symptoms, and anxiety symptoms interact and reinforce one another in people with type 2 diabetes could provide improved insight into their co-occurrence.

One overarching limitation to the current knowledge on diabetes distress and related psychological symptoms in people with type 2 diabetes relates to the use of summaries of scores on measurement items as an indicator of overall severity (e.g., by calculating an average or total summary score across all items of a depression measurement and interpreting it as an indicator of depression severity). However, this approach assumes, to some extent, that all items are equally indicative of an underlying mental health state and may cloud useful insights (18). To improve our understanding of psychological comorbidity among people with type 2 diabetes, exploring individual symptoms/problems and item-level interactions could be advantageous. For instance, in a sample of adults with bulimia nervosa, binge eating was shown to be an important symptom to bulimia nervosa psychopathology, highly connected to other symptoms and therefore likely to be implicated in the activation of other bulimia nervosa symptoms and the maintenance of bulimia nervosa (19). Illuminating the interplay between individual mental health symptoms and diabetes-specific problems in people with type 2 diabetes, identifying symptoms/problems that cluster together or symptoms/problems that trigger or sustain others may provide a more nuanced picture of the factors associated with the onset and maintenance of mental health comorbidity in people with type 2 diabetes.

Network psychometrics provides an opportunity for the underlying structure of mental health constructs and connections between individual items or symptoms to be investigated and visualized (20). With a network analysis approach, a mental health problem is explained as arising from the dynamic interaction between the symptoms themselves, with individual symptoms reinforcing and inhibiting others, rather than from an unmeasurable latent variable (20). Applying network analysis to mental health constructs allows the relative contribution of symptoms in a network to be investigated and thereby identify influential, highly connected symptoms. The symptoms of multiple constructs can be modeled in the same network, which can enlighten understanding of their interaction and co-occurrence, presenting several opportunities for better understanding comorbidity. Symptoms that connect two mental health constructs (e.g., depression and diabetes distress), known as bridges, increase the likelihood of one construct activating another (21). With the use of network analysis, symptom-level interactions can be examined between symptoms of mental health problems and the individual components of other psychological constructs (e.g., the individual diabetes-specific problems as measured with the 17-item DDS [DDS-17]).

In the current study we aim to use network analysis to gain a fuller estimation of the dynamics between individual diabetes distress problems, as measured with the DDS-17, and individual depressive and anxiety symptoms in people diagnosed with type 2 diabetes in the last 10 years who were insulin naive. To the best of our knowledge this has not yet been investigated. To do so, we examine item-level interactions, and, therefore, the terms diabetes distress problems and depressive and anxiety symptoms refer not to diagnostic entities but, rather, to groups of symptoms or problems. First, a network of diabetes distress is estimated. Second, the interplay between diabetes distress, depressive symptoms, and anxiety symptoms is explored and the network examined for the presence of bridges.

The sample was derived from the Montreal Evaluation of Diabetes Treatment (EDIT) study, a prospective cohort study of 2,033 middle-aged adults with type 2 diabetes living in Quebec, Canada. Participants were recruited by random digit dialing and letters to people with a type 2 diabetes diagnosis registered in the Quebec health insurance system (Régie de l'assurance maladie du Québec). Participants were eligible for the study if they were aged between 40 and 75 years, had been diagnosed with type 2 diabetes by a physician within the past 10 years, and were insulin naive. Details on the EDIT cohort have previously been published (13).

The study sample comprised N = 1,796 participants (88% of the full sample) with complete data on all diabetes distress, depressive, and anxiety items from the baseline EDIT study wave in 2011. In the included sample, mean (SD) age was 60 (8) years (range 40–76), and 49% of the sample was female. The 237 participants with missing data were slightly older (mean [SD] age 62 [8] years; P = 0.001) and more likely to be female (59%; P = 0.007) compared with the included sample.

Diabetes distress was assessed with the DDS-17 (6). The scale includes four subscales addressing different types of distress experienced within the past month: emotional distress (e.g., “diabetes controls my life”), physician-related distress (e.g., “doctor doesn’t give clear directions”), regimen-related distress (e.g., “not motivated to keep up self-management”), and interpersonal distress (e.g., “friends/family not supportive”). Item choices range from 1 (“not a problem”) to 6 (“a very serious problem”). In the present sample, internal consistency was excellent (α = 0.93).

With the nine-item Patient Health Questionnaire (PHQ-9) (22) we assessed the frequency of experiencing nine depressive symptoms within the past 2 weeks. Response options for each item range from 0 (“not at all”) to 3 (“nearly every day”). The PHQ-9 is a screening tool with items based on the DSM-IV diagnostic criteria for major depressive disorder. The internal consistency of PHQ-9 items was good (α = 0.79) in the study sample.

With the seven-item Generalized Anxiety Disorder Assessment (GAD-7) (23) we assessed the frequency of experiencing seven symptoms of generalized anxiety disorder within the past 2 weeks. Response options for each item range from 0 (“not at all”) to 3 (“nearly every day”). The GAD-7 scale items are based on criteria for generalized anxiety disorder from the DSM-IV. The internal consistency of GAD-7 items was excellent (α = 0.84).

A list of the DDS-17, PHQ-9, and GAD-7 items can be found in Supplementary Table 1.

Additionally, sociodemographic (age, sex, marital status, and education), health (smoking status, frequency of physical activity, frequency of alcohol use, BMI, physical health comorbidities, and mental health diagnoses), and diabetes-specific (diabetes duration, diabetes complications, and self-rated diabetes control) characteristics are reported to describe the study sample.

To assess self-rated diabetes control, participants were asked to rate their diabetes control within the previous month as poor, fair, good, very good, or excellent.

Statistical analyses were carried out with R, version 4.0.5. See Supplementary Table 2 for details on statistical analysis and R packages used.

We estimated a first network using only the DDS-17 items to explore the association between the diabetes distress items. A second network was modeled with the 17 items of DDS-17, the 9 items of PHQ-9, and the 7 items of GAD-7 to explore the interconnections between diabetes distress, depressive, and anxiety items. A network consists of nodes, representing the variables of interest (e.g., scores on each symptom/distress item), connected by edges, representing the relationships between nodes (e.g., partial polychoric correlations between item scores, adjusting for all other correlations in the network). For each network, a Gaussian graphical model was estimated with extended Bayesian Information Criterion (EBIC) model selection (24). As the data were ordinal, a polychoric correlation matrix was estimated as input (25). See Supplementary Table 2 for further Information on the estimation procedure, the statistical regularization technique, and packages used.

The network centrality properties of strength and expected influence were examined. Strength refers to the sum of absolute edge weights directly connecting a node to others in the network (25). One-step expected influence is used to evaluate the (nonabsolute) sum of the edges directly connecting a node to other nodes, with positive edges allowed to outweigh negative (26).

For the diabetes distress network, communities/clusters were also examined. Further details on the cluster analysis can be found in Supplementary Table 2.

For the network modeling the items of the DDS-17, PHQ-9, and GAD-7 together, the role of individual items as bridges between constructs was examined. The bridge strength (the absolute sum of the edge weights connecting an item of one mental health construct to items of another mental health construct [21]) and bridge expected influence (the nonabsolute sum of all edges connecting an item to others in a different construct [21]) of each item were investigated. Bridge strength and bridge expected influence were computed for each mental health scale item with the items of another scale separately (i.e., diabetes distress items–depressive items, diabetes distress items–anxiety items, and anxiety items–depressive items).

A post hoc bootstrapping framework was used to investigate the stability and accuracy of the networks and to ensure that the sample size was adequate (25) and is described in more detail in Supplementary Table 2.

The sample characteristics are described in Table 1.

The mean and SD for each item are presented in Supplementary Table 3, and polychoric correlations for all items are reported in Supplementary Tables 4–8. The assessment of item informativeness and redundancy (as explained in Supplementary Table 2) indicated that all items could be used in the analysis.

The network estimated with the DDS-17 items is presented in Fig. 1. A network was estimated with 81.6% (111 of 136) nonzero edges. Figure 1 also presents centrality indices as standardized z scores. Items high in strength are highly connected, and those high in expected influence have more positive connections to other items.

The bootstrapped CIs around the estimated edge weights were wide and overlapping, implying that the order of edge weights should be interpreted with caution. Case-dropping subset bootstrapping indicated that, with a correlation stability (CS) coefficient of (CS[cor = 0.7] = 0.361), node strength is interpretable with caution, and expected influence was highly stable (CS[cor = 0.7] = 0.75).

Figure 2 presents the diabetes distress, depressive, and anxiety network. The network was estimated with 43.6% (230 of 528) nonzero edges. Figure 2 also presents node centrality indices as standardized z scores.

Bootstrapped CIs indicate that edge weights are unlikely to differ significantly from one another. The CS coefficient for strength (CS[cor = 0.7] = 0.439) indicates that order of node strength can be interpreted with caution. Expected influence was extremely stable (CS[cor = 0.7] = 0.75).

Relatively strong bridging connections were observed as follows: between “failure” (dep6) and “not motivated to keep up self-management” (dd16), between “trouble relaxing” (anx4) and “sleep problems” (dep3), and between “restless” (anx5) and “moving/speaking slowly or being restless” (dep8). Bridge strength and bridge expected influence produced identical results, and so only bridge strength is reported in Fig. 3. Bridge strength was highly stable, with a CS coefficient of (CS[cor = 0.7] = 0.75). The higher the standardized z score for an item (e.g., dd1), the greater its connections (strength) and positive connections (expected influence) with items of the other mental health construct (e.g., depression).

To our knowledge, this study is the first to examine the network structure of diabetes distress and the interconnections between individual diabetes distress problems, depressive symptoms, and anxiety symptoms in a group of individuals with type 2 diabetes. With a network analysis of diabetes distress items we identified several highly influential problems. With a second network analysis of diabetes distress, depressive, and anxiety items we identified several influential problems/symptoms and bridges.

The identified pattern of clusterization of diabetes distress items, as reported in Supplementary Table 2 and Supplementary Fig. 1, matched the four subscales of the DDS-17: emotional burden, physician-related distress, regimen-related distress, and interpersonal distress (6).

Problems from the physician-related distress component on the DDS-17, “doctor doesn’t give clear directions” (dd4) and “doctor doesn’t take concerns seriously” (dd9), were identified as being high in node strength and expected influence in the diabetes distress network. The importance of the patient-doctor relationship in effective diabetes care and patient satisfaction is established (27,28). From the regimen-related distress subscale, “not motivated to keep up self-management” (dd16), was also highly connected and influential. Diabetes is a largely self-managed condition. A considerable proportion of people with diabetes taking part in the Diabetes Attitudes, Wishes and Needs (DAWN) study reported that they felt “burnt out” from coping with diabetes and stressed about the responsibility of their care (29). These physician-related and regimen-related problems may be central to activating other diabetes-specific problems and burdens for people with type 2 diabetes and hold particular clinical importance. Highly central nodes in a cross-sectional network of social anxiety symptoms were shown to predict the correlation between change in one node and change in the other network symptoms (30). “Friends/family don’t appreciate difficulty of diabetes” (dd13 [interpersonal distress]) and “overwhelmed by demands of diabetes” (dd14 [emotional distress]), though not high in node strength, were high in expected influence, meaning they had more positive connections to other nodes in the network. In network approaches to psychopathology, nodes high in expected influence are suggested to be important for the development and maintenance of mental health concerns (26). By activating a relatively large number of other symptoms in the network, they may be more likely to trigger and sustain the other symptoms and, therefore, the symptom network.

Our findings suggest that regimen-related and physician-related distress items are highly connected to other items within the DDS-17. There are two commonly used scales available at the present time for measuring diabetes distress, the PAID and the DDS, and differences have been reported between the two scales in terms of item content and psychometric properties (9). Moreover, questions have been raised around the extent to which the individual subscales of the DDS reflect the underlying concept of diabetes distress. For instance, Fenwick et al. (8) (2018) found that the subscale physician-related distress failed to discriminate between levels of distress and therefore suggest that this subscale may not fit within the construct of diabetes distress. Gonzalez et al. (5) (2015) measured the construct of diabetes distress using only the 5-item emotional burden subscale. These debates surrounding the use of the DDS-17 as a measurement of diabetes distress should be considered when interpreting our findings on the connectivity of individual items in the diabetes distress network. However, regardless of whether regimen-related and physician-related distress are viewed as integral to the concept of diabetes distress, items reflecting regimen and physician-related problems were found to be the most highly connected and influential items within the DDS-17 network. This suggests that diabetes-specific regimen and physician-related problems may be triggering other diabetes-specific problems as measured with the DDS-17. Therefore, notwithstanding the aforementioned issues, regimen-related and physician-related problems should be considered as potentially contributing to the development and maintenance of negative emotion and other symptoms of distress in people with type 2 diabetes.

In the second network, the nodes highest in strength and expected influence were “not motivated to keep up self-management” (dd16), “failure” (dep6), and “worrying too much” (anx3). On visual inspection of the combined network, depressive and anxiety items cluster together and share several strong bridging connections, while the diabetes distress items form a separate cluster. The relationship between depression and diabetes distress is complex, and there is some confusion as to whether they are distinct, though overlapping, concepts (11) or both elements of the same underlying emotional distress continuum (14). The combined network provides support for viewing diabetes distress and depression as discrete, though related, constructs. However, the edge weight analysis indicated that the order of edge weights in the network should be interpreted with caution.

One strong connection between diabetes distress and depressive symptoms was observed, between “failure” (dep6) and “not motivated to keep up self-management” (dd16). “Failure” (dep6) was the node highest in bridge strength and bridge expected influence, in the interplay between diabetes distress and depressive symptoms, indicating that it had the most connections and most positive connections with diabetes distress items. Within psychometric research adopting a network approach, the “spreading” of activation from one mental health problem to another, through bridging connections between items, is suggested to be central to explaining comorbidity (21). Research has shown that people with diabetes report experiencing feelings of failure in relation to managing their condition (31) particularly in relation to starting insulin therapy (as it may be viewed as having “failed” at managing diabetes through other means) (32). Feelings of failure may represent a link between the context of living with diabetes and managing the condition and the development of mental health issues, such as depressive symptoms. Equally, people with diabetes who also have depression may be more likely to have negative feelings around how they are managing their diabetes. Diabetes distress and depression have been shown to be bidirectionally (33) and cyclically (13) related. The role of the depressive symptom of “failure” (dep6) in the connection between depressive symptoms and diabetes-specific problems as measured with DDS-17 merits further investigation to tease apart the direction of activation.

For the interplay between depressive and anxiety symptoms, the nodes highest in bridge strength and bridge expected influence were all anxiety symptoms: “worrying too much” (anx3), “uncontrollable worry” (anx2), and “trouble relaxing” (anx4). These findings support the role of symptoms of anxiety in the development and maintenance of depression, particularly the role of sustained worry in activating feelings of sadness or exhaustion. There is evidence that supports anxiety playing a role in the development of depression (34) including anxiety symptoms preceding the development of depressive disorders (35). A strong connection was observed between “trouble relaxing” (anx4) and “sleep problems” (dep3). It is plausible that sleep problems may be linked to persistent agitation and arousal. Similarly, when bridges in the interplay between diabetes distress and anxiety were examined, anxiety symptoms were the most influential. People with type 2 diabetes frequently report experiencing worries that are unique to living with type 2 diabetes (16,17), and these contextual worries and anxieties could be significant in understanding mental health problems and comorbidity in people with diabetes. While causal relations cannot be inferred from cross-sectional data, these findings highlight the potential for network analysis to identify meaningful symptom-level interactions that should be explored further.

This study is based on a large sample of people with type 2 diabetes and assessed diabetes distress, depression, and anxiety with validated measurements. Rather than using psychological scale sum scores and cutoffs, we used a symptom-level approach to examine comorbidity. Investigating individual symptom-level interactions can take account of diversity in mental health symptomology and comorbidity with the ultimate ambition of leading to more personalized treatments (18). The study also had several limitations that should be considered in interpreting the results. First, there are limitations associated with the sample and the generalizability of the findings. The sample was predominately White, and thus results may not generalize to other racial or ethnic groups. The subjects were insulin naive, had been diagnosed with type 2 diabetes in the last 10 years, and were aged between 40 and 75 years. As this sample represents a specific group among people with type 2 diabetes, it would be important to examine item networks in other populations with diabetes, such as those with type 1 diabetes, with longer durations of type 2 diabetes, using insulin treatment, and with more diverse sociodemographic characteristics. Another limitation is that the data were collected in 2011 and so the treatment and management options available for the study sample may differ from those available today. Second, the use of cross-sectional data means that it was not possible to make causal inferences about the symptom-level relationships. For further insight into temporal relationships and which item’s activation precedes the activation of its neighbors, longitudinal analyses is needed. Third, network analysis does not take into account covariates or confounders; the partial correlations between items only control for the other items in the network. It is possible the observed associations between nodes result from an external factor (e.g., stressors, biological factors, diabetes complications) not modeled in the network. Finally, there are limitations associated with the novelty of network analysis for psychological research. Dealing appropriately with ordinal data is an area of current discussion in network psychometrics, and to date there is no “gold standard” method (36). Computing polychoric correlations is a commonly used method of dealing with ordinal data that assumes that a normally distributed latent variable underlies the ordinal data (25). This method was chosen as it retains important information on the severity and order of the data, but it can be problematic with variables that may have a real zero, as could be expected with some items (e.g., items assessing suicidal ideation) (25). In future research investigators should continue to develop and extend methods of using psychological data in networks analysis.

In this study we identified specific psychological factors of potential importance for the understanding of mental health in people with type 2 diabetes. Future symptom-level diabetes–mental health research should include use of prospective data to investigate whether the identified regimen-related and physician-related distress problems precede the development or worsening of diabetes distress. The EDIT study provides 5 years of follow-up data on participants with type 2 diabetes and would thus allow for further exploration of symptom-level connections through, for example, analysis of changes in network structure and connectivity between study waves. In future research investigators should look to use panel data of many subjects at multiple time points and individual time-series data (37) and explore analysis of subgroups (e.g., by age, HbA_1c levels, diabetes duration, or mental health status, such as major depression) to provide a more thorough and nuanced understanding of influential mental health components and interactions in individuals with type 2 diabetes.

According to network theory, interventions that focus on highly connected nodes would have the greatest effect in reducing the severity of the network as a whole (38). Interventions that effectively reduce regimen-related distress and physician-related distress may, therefore, be beneficial for reducing overall diabetes distress severity. For example, a 6-week empowerment-based intervention, focused on setting personally meaningful motivated goals, was found to be successful in reducing regimen-related and physician-related distress in adults with type 2 diabetes (39). Similarly, deactivating bridge symptoms could limit the spread of one mental health problem to another and reduce comorbidity (21). Cognitive behavioral therapy is already a commonly used and effective treatment for depression in people with diabetes (40). The current findings suggest that behavioral activation and cognitive restructuring focused on “failure” (dep6), “worrying too much” (anx3), “uncontrollable worry” (anx2), and “trouble relaxing” (anx4) early in the treatment process may reduce the emergence of comorbid mental health problems for people with type 2 diabetes, though this requires empirical examination.

This study highlights individual psychological symptoms and problems that could play a central role in the development and maintenance of diabetes distress and of comorbidity among diabetes distress, depressive symptoms, and anxiety symptoms in people with type 2 diabetes. Future research is needed to replicate these findings in different samples, explore temporal dynamics, and investigate the role of physician-related and regimen-related diabetes-specific problems and feelings of failure and worry in mental health comorbidity in people with type 2 diabetes. This study’s findings provide a good starting point for further examining the symptom-level interplay between important mental health factors in people with type 2 diabetes.

Acknowledgments. The authors thank the EDIT study participants for giving their time and providing their data.

Funding. A.M.M. is supported by the Ad Astra Fellows PhD Studentship. The EDIT study received funding from a Canadian Institutes of Health Research grant (MOP-106514) and was supported by the BIP Research (Bureau d’intervieweurs professionnels) and Régie de l’assurance maladie du Québec.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.M.M. conceptualized the study, carried out the statistical analysis, and wrote the first draft of the manuscript. S.S.D. conceptualized the study and contributed to the statistical analysis plan. N.L., A.N., N.S., and S.S.D. edited and provided feedback on earlier drafts of the manuscript. A.M.M. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 25th PsychoSocial Aspects of Diabetes (PSAD) Annual Scientific Meeting, 15–16 October 2020.