Comparison of diabetes distress and depression screening results of emerging adults with type 1 diabetes onset at different ages: findings from the German early-onset T1D study and the German Diabetes Study (GDS)

Data were taken from two cohort studies conducted at the German Diabetes Center (DDZ) in Düsseldorf, Germany. One of these studies was the “Clinical Course of Type 1 Diabetes in Children, Adolescents and Young Adults with Disease Onset at Preschool Age” observational study (childhood-onset study group). The study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the ethics committee of the University of Düsseldorf (reference number: 3254). All study participants gave written informed consent. The study design of the first questionnaire survey has been described previously [20], and subsequent surveys were designed similarly. In short, nationwide surveys were conducted in the form of standardised self-administered questionnaires. All study participants had a clinical diagnosis of T1D during their first five years of life and a diabetes duration of at least ten years. Details on the cohort study are given in Additional file 1: Figure S1. For this analysis, data from follow-up surveys conducted in 2012–2013, 2015–2016, and 2018–2019 were used because data on diabetes distress were collected only during these years.

The other study was the German Diabetes Study (GDS), which was a prospective observational cohort study that was initiated at the DDZ and developed into a national multicentre study (Additional file 1: Table S1). This study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the University of Düsseldorf (reference number: 4508) and has been registered with Clinicaltrials.gov (identification number: NCT01055093). The GDS study design was characterised by intensive phenotyping at 5-year intervals and annual telephone interviews. The primary inclusion criterion was a clinical diagnosis of type 1 or type 2 diabetes mellitus within the last 12 months in adults aged ≥ 18 years (adult-onset study group) according to the current recommendations by the American Diabetes Association. The diagnosis of type 1 diabetes was based on diabetes manifestation with ketoacidosis or immediate insulin requirement along with the presence of at least one islet cell-directed autoantibody or C peptide levels below the detection limit. The exclusion criteria at baseline were poor glycaemic control (HbA1c level > 9.0%), severe diseases and psychiatric disorders, among other criteria, which are described elsewhere in detail [21]. For this analysis, data from the full tests conducted from November 2010 to the end of 2021 were eligible because data on diabetes distress were not collected previously.

Emerging adulthood was defined, as usual, as 18 to 30 years of age [22‐25]. Thus, study participants aged 18–30 years were eligible for inclusion in this investigation. A total of 821 participants in the childhood-onset study and 195 participants in the GDS fulfilled this criterion. After the exclusion of persons who were not followed up after diabetes distress was included in the questionnaires, 749 participants were included in the childhood-onset study group (T1D onset in 1993–2005, surveyed in 2012–2019) (Additional file 1: Figure S1), and 163 participants were included in the adult-onset study group (T1D onset in 2007–2020, surveyed in 2010–2021) (Additional file 1: Figure S2 and Table S1). The data from each participant’s most recent follow-up were used.

The primary outcome was diabetes distress, which was assessed using the 20-item Problem Areas in Diabetes (PAID-20) scale [26]. The total score ranged from 0–100, with higher scores indicating more severe diabetes distress. Following common practice, a total score of 40 or more was considered to indicate seriously increased diabetes distress (screening positive) [27]. The nine-item depression module from the full Patient Health Questionnaire (PHQ-9) was used as a diagnostic algorithm for the screening of depression. The PHQ-9 total score (ranging from 0–27) was used as a continuous measure, with higher scores indicating more severe depressive symptoms. According to the standard cut-off, a total score of 10 or more was considered indicative of depression (screening positive) [28, 29]. If only one item was not answered, the missing value was replaced by the mean value of the completed items according to Kroenke et al. [30] (N = 18 with childhood onset and N = 1 with adult onset). The participants of both study groups answered the printed questionnaires at home. Further demographic and health- and diabetes-related data were considered (Table 1). Height and weight for the calculation of body mass index (BMI) and HbA1c levels were self-reported by the childhood-onset study group but sampled and analysed in a standardised way at the study centre for the adult-onset study group.

All variables are described as percentages or means and standard deviations (SDs). For the unadjusted comparisons of the two study groups, the Wilcoxon test was applied for continuous variables, and the chi-squared test was used for categorical variables.

We first compared outcome levels between the two study groups in an unadjusted regression model (Model M0, linear model for continuous outcomes (PAID-20 total score, PHQ-9 total score)), and binary outcomes were compared using a logistic model (screening for diabetes distress and depression). To consider confounding, we applied a doubly robust causal inference approach, which is an inverse probability weighted regression adjustment method that combines inverse probability weighting and regression adjustment to estimate potential outcome means [31]. Within this approach, first, a multivariable logistic model (“treatment/exposure model”) is used to estimate the probability of belonging to the exposure (adult-onset T1D) or control group (childhood-onset T1D), dependent on potential confounders of the exposure-outcome relationship. Standardised differences were used to check the balance between groups regarding confounders after inverse probability weighting of the data. Second, a multivariable model (a linear model for continuous outcomes and a logistic model for binary outcomes) with the outcome as the dependent variable (“outcome model”) is applied, including the exposure and potential confounders, and the inverse probability weights estimated from the exposure model are used to estimate regression coefficients. This approach is said to be doubly robust because it provides unbiased estimates for potential outcome means as long as one of the models is correctly specified [32]. Sex and age were considered potential confounders in Model M1; HbA1c level and education level were additionally considered in Model 2 and Model 3, respectively. The results are presented as adjusted potential outcome means (POM) and respective differences (average causal effect, ACE) with 95% confidence intervals (CIs) and respective p values from a Wald test.

All analyses were performed in complete-case analysis, as the rates of missing values of all variables were at most 5%. Two-sided p values < 0.05 were considered statistically significant. All analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, North Carolina, USA, 2016).

The crude PAID-20 total score and the proportion of participants who screened positive for diabetes distress were similar in both study groups (Table 1). The crude PAID-20 total score difference with the 95% CI including zero indicated that the ratings did not differ between the childhood-onset and adult-onset cohorts. However, considering differences in the age and sex distribution between the childhood-onset and adult-onset cohorts, the adjusted CI estimate of the ACE suggested that the PAID-20 total scores were 5 (95% CI 1, 10) points higher in the adult-onset group than in the childhood-onset group. Additionally, when differences in HbA1c values were considered, the mean difference in PAID-20 total scores increased to 11 (95% CI 7, 15) points (Table 2). The additional inclusion of education in the model did not change this result.

The proportion of participants who screened positive for diabetes distress was similar in both study groups according to the crude difference and the ACE after adjusting for age and sex. However, in Model 2, our results suggested that more emerging adults (18%, 95% CI 8%, 28%) in the adult-onset study group than in the childhood-onset study group screened positive for diabetes distress, and Model 3 confirmed the result (Table 2).

Considering the individual PAID-20 items, the groups differed significantly in four of the 20 items. Study participants in the adult-onset group felt particularly burdened regarding eating (Items 5 and 11), and they felt more scared when thinking about living with diabetes (Item 3). Feelings of guilt or anxiety when getting off track with diabetes management (Item 13) was a more important issue for participants in the childhood-onset study group (Additional file 1: Table S2).

The PHQ-9 total score was slightly higher in the childhood-onset group than in the adult-onset group (Table 1). The crude difference of 1 point in the PHQ-9 total score disappeared after adjustment for additional variables (Table 2). According to the chi-squared test and the doubly robust estimations, the proportion of participants who screened positive for depression did not differ between the groups.

Roughly estimated, 20% to 40% of all people with diabetes are considered to experience distress, with differentially high prevalence rates attributable to sample characteristics, such as diabetes type, disease duration, age, sex, and ethnicity [33]. The two groups of participants examined in this study had a crude proportion of 17–18% of participants who screened positive for diabetes distress. However, considering influencing factors, in particular, HbA1c levels, it became apparent that the adjusted diabetes distress prevalence in emerging adults with adult-onset (short-term) T1D was approximately 18% higher than that in their peers with childhood-onset (long-term) T1D. The marked impact of HbA1c was in line with previous studies that observed diabetes distress to be associated with higher HbA1c values [33]. In addition, our results are consistent with previous studies that have shown that diabetes distress is more strongly associated with HbA1c than depressive symptoms in adolescents [34] and adults [35, 36] with T1D. The lower HbA1c value in the adult-onset study group can be explained not only by the selection criteria (HbA1c ≤ 9%) but also by the fact that some of the volunteers may have been in remission. Data from most subjects in the adult-onset group were collected within 12 months of disease onset, when the probability of remission phase is highest. Recently, it was shown that most patients in the age group 12–30 years experience their remission phase within one year after T1D onset and that the HbA1c level after one year was significantly lower in the remitters [37]. Overall, the hypothesis put forward by Wentzell et al. [2] that an early age of T1D onset has a favourable effect on perceived diabetes distress later in life was confirmed.

Because there is some overlap of diabetes distress with depression, we also screened for the presence of depression [5]. The proportion of participants who screened positive for depressive symptoms was in the lower range of the prevalence observed in a recent analysis [38]. We are aware of only one single-centre study from Japan that investigated the prevalence of screening positive for depression (PHQ-9 scores ≥ 10) in adults (mean age 40 years, 69% women) with different ages at T1D onset. The proportions of participants who screened positive for depression were 21% (0–12 years of age at onset), 18% (13–19 years of age at onset) and 13% (20–40 years of age at onset). As in our adjusted analyses, the groups with positive and negative depression screening results did not differ with respect to age at onset [39].

From our observations, the question arises as to the possible reasons for the greater diabetes distress in emerging adults with a short diabetes duration. It is reasonable to assume that the disease places a particular burden on this group of patients because the life-changing onset of the disease coincides with other changes during this stage of life. Emerging adulthood is a dynamic and changeable period with unique developmental, social, and emotional challenges [2] that requires specialised diabetes care and education [12]. The complexity of the disease experience is individually different, as are the individual resources and problems related to living with diabetes. Perhaps, young adults with long-term diabetes have been more supported by their parents or significant others than their newly diagnosed peers. Unfortunately, we are not aware of any study results on this. Another aspect is that diabetes can both strengthen and hinder one's own definition of identity [3]. Findings from interviews with female emerging adults indicated that they reframed living with diabetes as an opportunity for empowerment and personal growth, even if they experienced distress [40]. Although the generalisability of this study may be limited due to gender differences in diabetes perceptions [41], letting go of habitual patterns of interpretation regarding the disease can be helpful for all people with diabetes. This process of cognitive reframing often happens naturally and unconsciously, but it can also take place as cognitive restructuring under the guidance of a psychologist [42]. In patients with disease onset in adulthood and, consequently, a short disease duration, this process is unlikely to have progressed to the same extent.

Our results imply that the relevant proportion of participants who screened positive for diabetes distress in both the childhood-onset (long-term) and adult-onset (short-term) T1D groups calls for awareness by health care professionals and individual counselling to reduce diabetes distress. Emerging adults with short-term T1D may need more psychosocial and practical support than is usually offered to integrate the disease into their lives. For diabetes researchers, our study may suggest that accounting for age at onset and diabetes duration, respectively, may help explain heterogeneity in the data when psychological factors are examined.

A strength of the study is that it draws attention to diabetes distress, which is an important outcome from the patient's perspective but has not been adequately studied thus far. Another strength is that the analysed outcomes were based on validated and widely used screening questionnaires. Furthermore, we applied a doubly robust causal inference approach to best match the different samples.

The main limitation is that the GDS and the early-onset cohort had different focuses and were not designed for this research question. Hence, analyses had to be limited to the small number of potential confounders that were matched in both studies. Therefore, we applied a doubly robust causal inference approach that was particularly suitable to account for unrecorded influencing factors and provide unbiased estimates even if one of the models was misspecified (e.g., because not all influencing factors were included). Although inverse probability weighting did not optimally balance the confounders (in particular age) between the two study groups (Additional file 1: Table S3), the additional regression adjustment considered an imbalance of confounders beyond weighting and suggested that the effect estimates were likely to be unbiased. In our analyses, the HbA1c value proved to be a very strong influencing factor. We assumed that the influencing factors associated with the HbA1c value were indirectly considered by the chosen method of analysis. In addition, we cannot rule out the possibility that the comparatively small sample size of one study group had an impact on the results in that the confidence intervals of the estimated differences were enlarged and existing differences may not have been demonstrated to be statistically significant. Another limitation is that both studies were not representative and relied on different methods for participant recruitment: one study was population-based and included persons from all over Germany, while the other study mainly recruited participants from the areas surrounding the seven GDS centres with HbA1c < 9%. The study population of the adult-onset sample from centres (in total n = 36) other than the DDZ (n = 127) was too small to consider cluster effects in the analyses. In addition, it is likely that there was a selection bias in both studies, as people who wanted to address their condition and engage in the advancement of diabetes research were more likely to participate. However, despite the lack of representativeness, an association between age at onset and diabetes distress was evident. Finally, it is not possible to distinguish the influence of the age at T1D onset from the influence of the duration of diabetes from our data.