ESM Methods and Results HbA 1c Responsible laboratories were contacted regarding their reference range to allow for a comparison of a standardized HbA 1c value (formula as previously used by Rose and colleagues 2002 [1]: (patients HbA 1c / norm value of the particular laboratory) * mean norm value of all laboratories; mean norm value for all laboratories was 6.05% (SD 0.24)). Questionnaires To assess a variety of psychological problems and symptoms the Symptom Checklist-90-R (SCL-90-R) [2] was used. It comprises 90 somatic and mental problems allowing for nine symptom dimensions (somatization, obsessivecompulsive, interpersonal sensitivity, depression, anxiety, hostility, phobic anxiety, paranoid ideation, and psychoticism). Of relevance to this article is severity of depression. Diabetes-related distress in daily life was measured using the German questionnaire Alltagsbelastungen bei Diabetes (FBD) [3], which distinguishes the type and extend of daily distress caused by diabetes itself or its treatment. The global score offers information about the extent of the diabetes-related distress in daily life [4]. Diabetes-related quality of life (DQoL) was measured using the Lebensqualität bei Diabetes (LQD) [5], which is an unpublished German questionnaire based on items from the Diabetes Quality of Life questionnaire by the Diabetes Control and Complications Trial (DCCT) research group [6]. Retrospectively (the last 4 weeks), patients had to self-assess how satisfied or burdened they are by their diabetes (0 5; 5 = always), resulting in three scales of quality of life (QoL) in type 1 diabetes: impact, satisfaction and impact of blood glucose. Adequacy of perceived familial
support was measured by the revised perceived family support and communication questionnaire (PFSQ-R) [7], which is the short version (26 items) of the questionnaire of perceived familial support and communication. Amongst others, it results in a global score of adequacy of perceived familial support which is evaluated here. With exception of the FBD, which was not applied at baseline, data from all mentioned questionnaires were gathered at all six time points (t0 t5). GMM We applied growth mixture models with linear, quadratic and cubic growth factors to gain a variety of straight and curved trajectories. To identify the number of latent classes, we increased the number of classes stepwise and examined the fit statistics using Akaike Information Criteria (AIC), Bayesian Information Criteria (BIC), Sample- Size Adjusted BIC (lower values imply a better fit), and Entropy (higher values (near one) imply a better distinction of classes) [8, 9]. The Lo-Mendell-Rubin Likelihood Ratio Test (LMR-LRT) was used to compare the currently tested class-model (k) against a class-model with one less class (k 1). A statistically significant p-value of the LMR-LRT indicates the rejection of a k 1 class model in favor of the k class model. To obtain the proper solution, model convergence was tested using different start values. Comparing the fit indices for different classes and different growth factors, we selected the model with best fit that allowed clinically relevant conclusions (e.g. we did not test more than six classes as number of cases would be unreasonably small). Test statistics of GMM allowed for three possible models of trajectories. The two class and the five class models were rejected due to the higher clinical relevance of a three class model. For individual trajectories of each class see figures 1, 2 and 3.
References 1. Rose M, Fliege H, Hildebrandt M, Schirop T, Klapp F (2002) The network of psychological variables in patients with diabetes and their importance for quality of life and metabolic control. Diabetes Care 25:35-42 2. Franke GH (1995) Die Symptom-Checkliste von Derogatis Deutsche Version. Beltz Test Gesellschaft, Göttingen [published in German] 3. Waadt S, Duran G, Herschbach P (1995) Klinische Diagnose psychosozialer Belastungen: Der Fragebogen zu Alltagsbelastungen bei Diabetes mellitus. In: Kohlmann CW (ed) Diabetes und Psychologie: Diagnostische Ansätze. Verlag Hans Huber, Bern, pp 17-33 [published in German] 4. Bott U (2000) Die Messung der Lebensqualität. In: Berger M (ed) Diabetes Mellitus. Urban Fischer, München, pp 106-119 5. Hirsch A, Bartholomae C, Volmer T (2000) Dimensions of quality of life in people with non-insulin-dependent diabetes. Qual Life Res 9:207-218 6. Diabetes Control and Complications Trial Research Group (1988) Reliability and validity of a diabetes quality-of-life measure for the diabetes control and complications trial (DCCT). Diabetes Care 11:725-732 7. Aymanns P (1992) Krebserkrankung und Familie: Zur Rolle familialer Unterstützung im Prozess der Krankheitsbewältigung. Huber, Bern [published in German] 8. Asparouhov T, Muthen B (2014) Variable-Specfic Entropy Contribution. http://www.statmodel.com/download/univariateentropy.pdf, accessed 12 March 2015 9. Jung T, Wickrama KAS (2008) An introduction to latent class growth analysis and growth mixture modeling. Social and Personality Psychology Compass 2:302 317
ESM Fig. 1: GMM of individual trajectories of depression based on the depression subscale of the SCL-90-R class 1 (nod)
ESM Fig. 2: GMM of individual trajectories of depression based on the depression subscale of the SCL-90-R class 2 (improved)
ESM Fig. 3: GMM of individual trajectories of depression based on the depression subscale of the SCL-90-R class 3 (worsed)