Family structures and parents’ occupational models: its impact on children’s diabetes

The current study is based on data from the multicentre German, Austrian, Swiss, and Luxembourgian Diabetes Prospective Follow-up Registry (i.e., Diabetes-Patienten-Verlaufsdokumentation [DPV]) [9]. This study, ‘Family structure and adherence of type I diabetes mellitus’, received approval from the Ethics Committee in October 2020 (Vote #185/20, department of medicine Justus-Liebig University Gießen). As of August 2023, 490 diabetes centres (hospitals and outpatient centres) and 601 200 patients with diabetes are included in the DPV registry. The data are pseudonymized and transmitted for central plausibility checks and analyses to the Institute of Epidemiology and Medical Biometry of Ulm University (Ulm, Germany). After reporting inconsistent data back to participating centres for validation and correction, the data are then completely anonymised prior to analysis.

Out of the DPV registry, patients who first presented with diabetes before 2000 or after 2018 were excluded (Fig. 1). We included patients with a clinical diagnosis of T1DM between six months and 18 years at the time of diagnosis with information on family structure, parents’ employment characteristics and HbA_1c. We included the data of 15,340 patients from 301 diabetes centres in our study sample (286 German centres, 14 Austrian centres, 1 Luxembourgian centre). The data pool was aggregated for a 2-year period prior and following the last psychosocial data collection.

Demographic data included age, age at diabetes onset, duration of diabetes, sex, and migratory background (patient or at least one parent born outside Germany, Austria, Switzerland, or Luxembourg).

Clinical parameters were evaluated 2 years prior and following the last psychosocial entry. They included body-mass index standard deviation score (BMI SDS), the daily dose of insulin (IE/kg), the use of continuous or flash glucose monitoring, the type of insulin therapy (multiple daily injections or insulin pump treatment) and HbA_1c (percentage; mmol/mol). Further outcome measures included rates of severe hypoglycaemia (with or without coma) and diabetic ketoacidosis, rates of diabetes-associated hospital admissions (i.e., hospitalization), and inpatient days in hospital each year due to diabetes. BMI SDS was calculated using the national KiGGS (German Health Interview and Examination Survey for Children and Adolescents) reference data in Germany [10]. HbA_1c values were measured locally and standardized mathematically to the Diabetes Control and Complications Trial (DCCT) reference range (4.05–6.05%) using the multiple of the mean (MOM) method [11]. Severe hypoglycaemia (SH)—an event requiring external assistance by another person; hypoglycaemic coma (SHC)—severe hypoglycaemia associated with seizure or loss of consciousness; and an event of diabetic ketoacidosis (DKA)—a pH of less than 7.3 or serum bicarbonate of less than 15 mmol/L or both were defined according to the guidelines of International Society for Pediatric and Adolescent Diabetes (ISPAD) [12, 13]. Rates of severe hypoglycaemia, DKA, and hospital admission rates were presented as events per 100 patient years. Length of inpatient treatment was given as the number of inpatient days per patient and year.

The database includes information on family structure and working time models (WTM) of the children’s caretakers and was stratified according to these categories. We defined living with both biological parents as a two-parents household (TPH) and living with one biological parent as a one-parent household (OPH). Blended family households (BFH) were defined as living with one biological parent and his/her partner in the household. Other living conditions (e.g., living with other relatives, adoptive parents, boarding school, foster care homes of the child welfare service, etc.) were categorized as living without a biological parent (no parent household, NPH). Working time models differentiate into working full-time (A), part-time (B), homemaker, being in training, receiving a pension (C) and being unemployed (D). The categorical system focuses on the father regarding two parent households and the mother in terms of one-parent households (Fig. 2).

The data evaluation and statistical analysis was performed using SAS 9.4 (SAS Institute, Cary, NC, USA). Linear regression models were applied to calculate and compare HbA_1c (% or mmol/mol) between family structures and working time models. Rates of SH, SHC, DKA, and hospital admission rates and length of inpatient treatment were analysed via negative binomial regression using individual time under risk as offset. All models were adjusted for age group (< 10, 10 to 15, ≥ 15 years), sex, diabetes duration group (< 2, 2 to 5, ≥ 5 years) and migratory background (yes, no). Results are presented as least-square means with 95% confidence intervals. The Tukey–Kramer method was used for multiple testing in unbalanced data. Goodness-of-fit was assessed via R-Squared. Models were tested for each dependent variable using analysis of variance.

Patients living in a two-parents household had significantly lower HbA_1c levels compared to other living conditions (7.70; 60.70 vs. 8.06; 64.63 for OPHs; 8.07; 64.66 for BFHs; 8.21; 66.22 for NPHs) as well as lower rates of DKA (1.79) than subjects living in one-parent households (3.41), blended family-households (4.00), and no parent households (7.19) (see Table 1).

The rates of hypoglycaemic coma were higher for children living in no-parent households (6.29), compared with children living in two-parent households (3.15), single-parent households (3.68) and blended family households (4.36). Rates of severe hypoglycaemia did not differ significantly between these categories. Length of inpatient treatment and hospital admission rates were lowest for adolescents living in a two-parent household (12.82; 57.46 vs. 16.07; 76.02 for OPHs; 16.10; 80.69 for BFHs; 19.01; 106.29 for NPHs). Table 2 compares the effects between two-parent and single-parent households, with overall poorer outcomes for children living in two-parent households.

Goodness-of-fit analysis showed 14% of the variability concerning the HbA_1C parameters can be explained by family structure (R² = 0.14, F_{(3, 15,330)} = 92.38, p < 0.0001) (see Table 4).

The adjusted outcomes indicated significantly higher HbA_1c parameters in children of parents with working time models of category C (7.88; 62.58) and D (7.97; 63.55) versus categories Aa (7.62; 59.82) and Ab (7.61; 59.71) as well as category C vs category Ac (7.70; 60.60). Lower rates of hypoglycaemic coma were recorded in children from couples of category Ab compared to D (2.48 vs. 5.12). Rates of severe hypoglycaemia and DKA did not differ significantly between working time models (Table 1). In group comparison, the length of inpatient treatment differed between categories with longer inpatient treatment for C (14.73), D (16.60), Aa (15.41) and shorter treatment for Ab (11.97), Ac (12.28) and B (12.14). Higher hospital admission rates were reported for categories C (75.76) and D (72.16) as well as for Aa versus Ab (60.58 vs. 53.66) and Ab vs Ad (53.66 vs. 66.29). Table 3 contrasts the findings for the categories Ab and D regarding the working time models. Living with a full-time working father and a part-time working mother came with overall lower outcomes compared to living with an unemployed father.

Goodness-of-fit analysis indicated 13% of the variability concerning the HbA_1C parameters can be explained by working time models in two-parent households (R² = 0.13, F_{(6, 9842)} = 7.01, p < 0.0001) (see Table 4).

According to goodness-of-fit analysis, there was no additional significant information gained regarding the working time models in one-parent households for HbA_1c levels, DKA, severe hypoglycaemia and hypoglycaemic coma. Hospital admission rates were higher for category C (84.50) compared to A (72.77) and B (72.25). Length of inpatient treatment differed between category B (14.63) and D (18.36) (see Table 4).

An obvious strength of the study is the large sample size. To date, most studies included less participants and were conducted in a single country or culture [8, 22, 23, 28]. This study combines data from three different countries, the data collection is well established, and the database has proven to be a matrix of intensive publication activity.

The interpretation of R² can be limited so we complemented analyses of variance for all tested models. Nevertheless, we did not include further validation techniques like residual analysis to obtain an even more thorough understanding of the model’s performance.

The study is limited to the categorical system given in the data mask. Overall, the chosen classification reflects a family-related aspect of social status and can be associated with parental education, working hours, occupational status and household income [14]. The influence of parental income and education should be investigated in future research. The data were limited by the information provided and we had to exclude patients who did not provide information on family structure. In addition, future studies should include more diverse modern family forms (e.g., different forms of blended or rainbow families) in their research. This could be helpful in describing coping potential and identifying factors of clinical importance for better treatment options.