Continuity and Change in Glycemic Control Trajectories From Adolescence to Emerging Adulthood: Relationships with family climate and self-concept in type 1 diabetes

Participants were taken from the German Longitudinal Study on Juvenile Diabetes (2), which received full institutional review board approval. In 1991, 109 patients with type 1 diabetes were recruited from pediatric health care services offering outpatient care in two German cities, Bonn and Freiburg (88% of the families initially contacted agreed to participate). Only the Bonn subsample (n = 72 at time 1; M_age = 13.72; SD 1.46) was assessed eight times and, consequently, constituted the sample for the present study (37 females, 35 males). These patients came from broad socioeconomic strata, with 24% being from low-class families, 49% from middle-class families, and 27% from upper-class families. A total of 87% came from two-parent and 13% from one-parent families, and patients had on average 1.35 brothers or sisters (SD 1.18), with a range from 0 to 6. They were followed-up in 1992, 1993, 1994, 1998, 2000, 2001, and 2002 (M_age at time 8 = 24.94; SD 1.43), hence covering mid-to-late adolescence and emerging adulthood. Mean duration of diabetes at time 1 was 4.79 years (SD 3.78). Mean BMI scores ranged from 20.50 (SD 2.42) at time 1 to 23.44 (SD 2.87) at time 8. Participants were visited at home by research project team members at each data wave and were asked to fill out questionnaires. At the different time points, patients visited their treating physicians to determine A1C values as a criterion for glycemic control. Questionnaires were sent to the physicians to retrieve A1C values from the patients' medical records.

As in most longitudinal studies, data were missing at different time points for different participants. For the present study, 15.62% of the data were missing. To minimize the bias associated with this occasional attrition, we used the expectation maximization algorithm to impute missing data. A nonsignificant Little's (16) Missing Completely At Random test [χ² (376) = 35.99, NS] indicated that missing values could be reliably estimated.

General family climate was assessed at times 1–4 using the system maintenance dimension of the Family Environment Scale (17), tapping into the degree to which family interactions are characterized by organization and control. The latter two subscales assess the importance placed on having a clear structure in family activities and responsibilities and the extent to which clear rules and procedures are used to govern family life. Reliability and validity data of the German version are provided by Schneewind (18).

Positive self-concept was assessed at times 1–4 and 6 with the Offer Self-Image Questionnaire (19). The scales assessing psychological self (i.e., impulse control, emotional tone, and body image) and coping self (i.e., mastery of external world, emotional health, and superior adjustment) were combined to assess general positive self-concept or self-esteem. Sample items read: “Most of the time, I am happy” and “I feel that I am able to make decisions.” A study on German adolescents revealed a Cronbach's α of 0.83 for this combined scale (20).

Pearson product-moment correlation coefficients were calculated among A1C values at the different time points to assess differential stability and change in glycemic control. Latent class growth analysis (21) with M-PLUS 4.0 was used to identify developmental classes of glycemic control. Intercepts and linear slopes were estimated from times 1 through 4 and from times 5 through 8, resulting in two sequential growth curves for each developmental class (capturing development from times 1 through 4 and from times 5 through 8, respectively). Several criteria were used to decide on the number of classes. First, the Bayesian information criterion statistic for a solution with k classes should be lower than for a solution with k-1 classes, suggesting that adding additional classes improves model fit. Second, classification quality was assessed by entropy (E), a standardized summary measure of classification accuracy based on the posterior classification probabilities. Entropy ranges from 0.00 to 1.00, with higher values indicating more accurate classification. Third, we used the bootstrapped likelihood ratio test, which provides a P value that can be used to determine if there is a statistically significant improvement in fit through the inclusion of an additional class. Fourth, we evaluated the substantive usefulness of the classes. Next, one-way univariate ANOVA with subsequent post hoc Tukey's honestly significant differences tests were conducted to investigate whether the classes obtained differed on family climate and self-concept.

All A1C stability coefficients are displayed in Table 1. Stability coefficients between two adjacent time points were all significant at P < 0.001 and ranged from 0.40 to 0.82 (from 0.40 to 0.67 at times 1–4 and from 0.63 to 0.82 at times 5–8). A1C values at times 1 and 2 were not substantially correlated with those at times 5–8, whereas A1C values at times 3 and (especially) 4 were substantially correlated with the latter.

Latent class growth analysis conducted on the A1C values favored a three-class solution (Bayesian information criterion 1,940.84; E = 0.99) over a two-class solution (Bayesian information criterion 2,021.24; E = 0.95), with bootstrapped likelihood ratio test significant at P < 0.001. In the less parsimonious four-class solution, one class consisted only of two individuals. Table 2 presents all intercepts and slopes for the three-class solution. Class 1 (n = 51) was labeled moderate control, class 2 (n = 10) optimal control, and class 3 (n = 11) deteriorating control. Figure 1 gives an overview of the observed mean A1C values for these classes.

Additional χ² analyses indicated that males and females tended to be differently distributed among these three classes [χ² (2) = 4.89; P = 0.087], with 55, 50, and 18% of classes 1, 2, and 3, respectively, being males. Table 2 presents a series of ANOVAs, indicating that from time 3 on, significant class differences in mean A1C values emerged and that from time 5 on, all three classes were differentiated with respect to mean A1C. No substantial differences among the classes were found in (analyses not shown) adolescent family composition, socioeconomic status, mean duration of diabetes, or BMI scores across time.

A series of ANOVAs with developmental class as independent variable and system maintenance and self-concept as dependent variables indicated, as shown in Table 3, that the three classes differed substantially on system maintenance and self-concept at several time points. At times 1, 2, and 4, the optimal control class scored highest on system maintenance. At times 3, 4, and 6, the optimal control class scored highest and the deteriorating control class scored lowest on positive self-concept.

This longitudinal study is the first to examine how continued assessments of general family climate and self-concept throughout adolescence were related to differential development in glycemic control throughout adolescence and emerging adulthood. These assessments well into emerging adulthood constitute an important strength of this study because the late teens and twenties are years of profound change. Adolescents in their late teens increasingly become more independent and engage themselves in identity-related work and explorations, leading to a diversification of life paths (1,4). These developmental trends seem to be reflected in the three developmental classes of glycemic control obtained, that is, the optimal control, moderate control, and deteriorating control classes (with females being somewhat overrepresented in the third class). Indeed, from times 1–4, these classes clearly diversified with respect to mean A1C values and, as expected (4), this trend continued and further intensified throughout emerging adulthood. More specifically, whereas at times 1 and 2, the three classes were not substantially differentiated from each other with respect to A1C values, they were so from time 3 onward. In summary, these developmental classes were not only different with respect to levels (or intercepts) of A1C values, but also with respect to changes (or slopes) in A1C values throughout adolescence and emerging adulthood.

Further, we were able to isolate certain developmental factors that could partially account for these differential developmental classes of glycemic control. From time 3 on, these three classes were differentiated on the basis of their mean self-concept scores, with the optimal control class scoring the highest and the deteriorating control class scoring the lowest on positive self-concept. As such, the present findings complement previous research (13) by demonstrating that self-concept variables are related to longitudinal trajectories of glycemic control in type 1 diabetes. Hence, these findings indicate the need to actively strengthen adolescents' self-concept to possibly yield long-term benefits in the area of glycemic control. With respect to general family climate, a somewhat different picture emerged. Although the three glycemic control classes did not have divergent A1C values at times 1 and 2, these classes were characterized by mean differences in family system maintenance (i.e., the degree of control and organization in family interactions) already at times 1 and 2. Adolescents belonging to the optimal control class experienced more family control and organization already from mid-adolescence on compared with other adolescents. Thus, adequate parental monitoring in the adolescent years seems to be beneficial to optimize glycemic control throughout adolescence and emerging adulthood (11).

The present study underscores the importance of age and developmentally appropriate diabetes education to the entire family unit for adolescents with type 1 diabetes (5). If adolescents with diabetes perceive their family climate as well organized and setting clear rules, they are more likely to follow an optimal pathway of glycemic control throughout adolescence and emerging adulthood. Adolescents whose parents provide guidance and supervision in family life have better glycemic control during adolescence and were able to maintain these levels of glycemic control during emerging adulthood. Apparently, adolescents experiencing an organized family climate seem to internalize the structure and organization that was modeled within their family and seem to apply these skills and competences to their own diabetes management later in life (22).

The differential stability observed in the A1C values throughout the study strengthens the conclusion supporting the need for a thorough diabetes education program early in adolescence (5). Stability coefficients of glycemic control were generally higher in emerging adulthood compared with adolescence, meaning that in adolescence, there seems to be more room for externally induced change in glycemic control compared with emerging adulthood. Once certain young people reach the emerging adult years and become more independent (i.e., start living on their own, find a job, and/or settle into a family of their own), it might be more difficult to change their glycemic control habits compared with when they were adolescents. In sum, the present findings underscore the need to start educating the family and the individual with diabetes already in the adolescent years to optimize the transition to independent self-care in the emerging adult years. Previous research indicated that such education may need to go hand in hand with cognitive-behavioral interventions attending to the special needs of the adolescent or emerging adult population to achieve substantial behavior change in individuals with diabetes (6).

Despite the fact that the generalizability of our findings is somewhat limited (mainly due to the modest sample size with patients all originating from a specific region of the world), the main conclusions of the present study have important implications for diabetes care. There was some degree of continuity between glycemic control status of adolescence—especially of late adolescence and not of mid-adolescence—and that of the postadolescent years (6,23). Three different developmental classes of glycemic control could be empirically identified. A1C values diversified throughout adolescence, and this diversification was strengthened throughout emerging adulthood, as could be expected (4). Both general family climate and self-concept distinguished among these three classes from adolescence onward, with individuals in the optimal control class scoring highest on family organization and positive self-concept. Future research needs to investigate whether these classes differ with respect to treatment adherence, medical outcomes, and complications (6). Further, females were somewhat overrepresented in the deteriorating control class. Previous research indicated that adolescent girls were more distressed by their diabetes, experienced lower self-esteem, and reported more depressive and eating disorder symptoms, the latter also being associated with a less positive family climate among youth with diabetes (3,24,25). Future research should investigate whether these factors can account for the deteriorating control trajectory observed throughout adolescence and emerging adulthood for some females.

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