Trajectories of Childhood Adversity and Type 1 Diabetes: A Nationwide Study of One Million Children

Type 1 diabetes is a chronic condition that typically presents in childhood or early adulthood (1). Autoimmune destruction of the insulin-producing pancreatic β-cells triggered by factors in the environment in genetically susceptible individuals usually precedes the clinical onset of type 1 diabetes (1). The environmental factors responsible and the mechanisms through which they act are largely unknown.

The experience of stressful adversities in childhood has been suggested to be a risk factor for type 1 diabetes that may act via several intertwined pathways (2,3). The rapid development of the human stress response system makes early life a critical period in which excessive exposure to adversities may result in enduring alterations in the way the brain and body respond to stress (4,5). Alterations in the stress response system may, in turn, disrupt the immunological balance, leading to a more inflammatory phenotype and a higher risk of developing immune-mediated diseases (4,6), including type 1 diabetes (2). Hyperactivation of the stress response may also lead to insulin resistance due to increased peripheral cortisol levels, which adds pressure on the β-cells and thereby make them more susceptible to autoimmune attack (7,8). Type 1 diabetes incidence is known to peak in puberty (9), and exposure to stressful adversities may contribute to β-cell stress that is caused by increased insulin demand with the rapid physical growth and substantial hormonal influence that takes place during puberty (7,8). Exposure to adversity in early life and during puberty may, therefore, be particularly important for type 1 diabetes development.

Most prior studies have found a positive association between childhood adversities and type 1 diabetes (10–13), with effect estimates indicating up to three times higher risk of type 1 diabetes after exposure to at least one adverse experience (14). However, these studies have been limited by retrospective recall of adverse events, substantial loss to follow-up, and small numbers of case subjects with type 1 diabetes. In a recent prospective population-based cohort study, we found no or negligible effects of accumulation of childhood adversities on type 1 diabetes risk (15). Only females who experienced a very high degree of adversity showed a higher risk of developing type 1 diabetes. Timing of exposure to adversities and age at onset of type 1 diabetes have not been thoroughly addressed, which may contribute to the contradictory results of previous studies.

The timing of exposure to adversities such as parental divorce, alcohol abuse, and psychiatric illness is hard to determine because the onset is often gradual. For example, the stress associated with having a parent with severe psychiatric illness does not start on the date of the parent’s clinical diagnosis; it was likely initiated long before. Studies assessing the importance of timing of exposure to childhood adversities for type 1 diabetes development are, therefore, extremely difficult to operationalize.

We hypothesize that the effect of childhood adversities on the risk of type 1 diabetes depends on the timing of exposure. To test this hypothesis, we aim to describe the age-specific incidence of type 1 diabetes in the five most common trajectory groups of adversity experienced before the age of 16 years among all children born in Denmark in the period 1980–1998 (16). We will perform the analyses separately for males and females to account for the well-known differences in age at onset of type 1 diabetes (9) and potential sex differences in the effect of childhood adversities on type 1 diabetes suggested in previous studies (15,17).

The DANish LIFE course (DANLIFE) cohort is a register-based life course cohort, which includes all children born in Denmark between 1 January 1980 and 31 December 2015, totaling 2,223,927 children (18). The unique personal identification number given to all Danish residents enabled linkage of information from nationwide registers on demographic, socioeconomic, and health-related factors in DANLIFE. The unique identifier also enabled linkage to parents and siblings. The family linkage was used to identify 12 objectively measured childhood adversities (i.e., family poverty, long-term parental unemployment, death of a parent or a sibling, parental and sibling somatic illness, foster care, parental and sibling psychiatric illness, parental alcohol and drug abuse, and maternal separation). A detailed description of the definitions of the childhood adversities included in DANLIFE can be found in the cohort profile (18).

In a previous study, five trajectory groups based on the annual rate of adversity exposure across three dimensions (i.e., material deprivation, loss or threat of loss in the family, and family dynamics) were identified in DANLIFE from birth until 16 years of age using a group-based multitrajectory model (16). The aim was to cover entire childhood trajectories (i.e., 0–16 years of age). Therefore, 1,064,864 people born after 1998 who could not be followed until their 16th birthday, 11,161 people who died before their 16th birthday, and 50,274 people who emigrated before their 16th birthday were excluded, resulting in a sample of 1,097,628 individuals. Supplementary Table 1 presents the specific adversities included in each of the three dimensions and their definitions. All 12 adversities were recorded once annually for each parent or sibling and were summed by age and dimension. Thus, the trajectory groups cover timing, frequency, and accumulation of the adversity exposures using the full capacity of the prospective and repeatedly measured information in the Danish registers. The robustness of the trajectory groups was ensured by multiple sensitivity analyses. Details on the identification of the five trajectory groups have been published previously (16). The identified five trajectory groups are characterized by:

1. Low adversity (54%): a very low annual rate of adversity across childhood and adolescence.

2. Early life material deprivation (20%): a high annual rate of material deprivation in the first 4–5 years of life.

3. Persistent material deprivation (13%): a high annual rate of material deprivation across childhood and adolescence.

4. Loss or threat of loss (10%): a high and increasing annual rate of severe somatic illness or death within the family across childhood and adolescence.

5. High adversity (3%): a very high and increasing annual rate of adversity across all three dimensions throughout childhood and adolescence. The annual rate of adversities was especially high and increasing in the family dynamics dimension, especially during adolescence, when children on average had almost one adversity every year.

Supplementary Figure 1 presents the characteristics of the trajectory groups. In the current study, we build on this previous work by using the information on allocation to the five trajectory groups in the DANLIFE cohort.

Since the definition of “parental somatic illness” included parental type 1 diabetes, we excluded those who had a parent with type 1 diabetes (n = 15,635) in order not to confuse the effect of genetic predisposition to type 1 diabetes with the effect of childhood adversities. Thus, the final study population included 1,081,993 people without parental type 1 diabetes.

Identification of type 1 diabetes and date of diagnosis in the DANLIFE cohort using registers have been described in detail elsewhere (15). Briefly, we used information from the Danish Registry of Childhood and Adolescent Diabetes (1980–2015) (19), the Danish Adult Diabetes Registry (2005–2015) (20), and the Danish National Patient Register (1980–2015) (21), supplemented by information on second purchase of prescribed oral antidiabetic drugs before 15 years of age or second purchase of prescribed insulin before 30 years of age from the Danish National Prescription Registry (1995–2015) (22). The first record of a diabetes diagnosis or the date of the second purchase of antidiabetic drugs or insulin defined the date of diagnosis. The same registers and criteria were used to identify parental and sibling type 1 diabetes.

Potential confounders were selected a priori using the method of directed acyclic graphs (23) (Supplementary Fig. 2). They included age, date of birth, birth weight (grams), birth order (1, 2, 3, or ≥4), maternal age at birth (years), parental education at birth (low: ≤9 years; middle: 10–12 years; or high: >12 years), parental area of origin based on father’s (or, if missing, mother’s) nationality (European origin [including Europe, North America, Australia, and New Zealand] or other), and sibling type 1 diabetes (yes/no). The specific registers used to identify the potential confounders can be found in the DANLIFE cohort profile (18). Sibling type 1 diabetes was used as a proxy for genetic predisposition to type 1 diabetes acquired at conception. Therefore, the time point of the type 1 diabetes diagnosis of a sibling was irrelevant and could occur at any time before 1 January 2016. All potential confounders, including sibling type 1 diabetes, were added in supplementary analyses as time-fixed variables at baseline except for age, which was used as the underlying time scale. Date of birth, maternal age at birth, and birth weight were treated as continuous variables.

We used the high time resolution of the register-based data to model exposure to childhood adversities and development of type 1 diabetes as two parallel processes. We first determined the trajectory groups of adversity, and secondly, we modeled the age-specific incidence of type 1 diabetes in each of these groups. We specified a Poisson regression model for the age-specific type 1 diabetes incidence with natural splines (with knots at age 2, 8, 14, 20, and 30 years) for males and females separately in each of the five trajectory groups. Knots were selected to provide a flexible curve in younger ages at which it is known that incidence rates of type 1 diabetes vary more than in older ages. Follow-up was from the date of birth until 31 December 2015. We then assessed the rate ratio between the age-specific incidence rates of type 1 diabetes in each trajectory group relative to the incidence rate in the low adversity group. We also estimated both crude and adjusted incidence rate ratios (IRRs) and 95% CIs of type 1 diabetes in three age groups divided into childhood (0–10 years), puberty (11–15 years), and early adulthood (≥16 years) using the low adversity trajectory group as reference and age as the underlying time scale in a subsample of the study population with complete information on all covariates (n = 1,066,122; 98.5%). All analyses were performed separately for males and females.

A total of 5,619 people developed type 1 diabetes during follow-up. Among these, 1,797 people were diagnosed after their 16th birthday. Table 1 presents the background characteristics of the study population by trajectory group. The proportions of people with low birth weight, low parental education, and teenage mothers were markedly higher in the persistent material deprivation and the high adversity groups compared with the low adversity group.

The top panel of Fig. 1 shows the age-specific incidence rates of type 1 diabetes for males and females in each of the five trajectory groups of childhood adversity. The rates in the low adversity group, including more than half (54%) of the study population, resembled the well-known age-specific incidences of type 1 diabetes with a peak in puberty at ∼10 years of age for females and 14 years of age for males (9). Generally, we saw no clear differences when visually comparing the age-specific incidence rates for males and females in the other four trajectory groups with the well-known pattern for males and females in the low adversity trajectory group (top panel of Fig. 1). The only clear exception was that females in the high versus low adversity trajectory group appeared to have a higher risk of being diagnosed with type 1 diabetes in early adulthood. Inclusion of the effect of calendar time in the model did not change the results for the five trajectory groups.

The bottom panel of Fig. 1 presents the rate ratios between the age-specific incidences of type 1 diabetes for males and females in each trajectory group relative to males and females in the low adversity group. The horizontal line (at a rate ratio equal to 1) represents no relative difference in age-specific incidence rates in the two trajectory groups being compared. For the large majority of the study population, there were no time-specific differences in the age-specific incidence rates when compared with the low adversity group. The only clear differences were seen in the high adversity group, in which males showed a higher incidence rate in childhood (<11 years of age) and females showed a higher incidence rate in early adulthood (≥16 years of age) when compared with males and females in the low adversity group. There were also some minor variations in age-specific incidence in the other trajectory groups compared with the low adversity group (i.e., a lower risk among females before 10 years of age in the early life material deprivation group, a higher risk among females after 16 years of age in the persistent material deprivation and the loss or threat of loss groups, and a higher risk among males between 11 and 15 years of age in the loss or threat of loss group).

Among the 98.5% of the study population with complete information on all covariates (n = 1,066,122), we estimated crude and adjusted IRRs of type 1 diabetes in childhood (0–10 years), puberty (11–15 years), and early adulthood (≥16 years) for males and females separately in each of the trajectory groups using the low adversity group as reference (Table 2). In the high adversity group, the incidence rate of type 1 diabetes was 80% higher in childhood among males (adjusted IRR 1.78 [95% CI 1.31–2.42]) and twice as high in early adulthood among females (adjusted IRR 2.19 [95% CI 1.57–3.07]) compared with males and females in the low adversity group. In the loss or threat of loss group, the incidence rates of type 1 diabetes were also slightly higher in puberty (adjusted IRR 1.30 [95% CI 1.07–1.58]) among males and in childhood (adjusted IRR 1.36 [95% CI 1.11–1.67]) and early adulthood (adjusted IRR 1.39 [95% CI 1.07–1.81]) among females compared with males and females in the low adversity group.

In an unselected cohort study covering all Danish children born in the period 1980–1998, we found no clear associations between trajectories of adversity and type 1 diabetes incidence in the vast majority of the study population; the only exception being a small proportion of people in the high adversity trajectory group (3%) who experienced high and increasing levels of adversity across social, health, and family-related dimensions. This group seemed to have a different age-specific incidence of type 1 diabetes compared with the low adversity group. In this group, a higher incidence rate was found in childhood (before 11 years of age) among males compared with males in the low adversity group. The age-specific pattern was different for females, in whom a higher incidence rate was found from 16 years of age onward compared with females in the low adversity group.

“Parental somatic illness” was one of the adversities included in the loss or threat of loss trajectory group and was counted every year a parent was hospitalized with one of the illnesses in the Charlson comorbidity index, which includes type 1 diabetes and a few other autoimmune diseases (e.g., connective tissue disease). The slightly higher incidence rates in the loss or threat of loss group compared with the low adversity group may, therefore, be a result of residual confounding by genetic predisposition to autoimmune disease, even though we excluded people with parental type 1 diabetes, and this result should, therefore, be interpreted with caution.

In a recent study using the entire DANLIFE cohort born in the period 1980–2015 (N = 2,223,927), we found that exposure to accumulation of childhood adversities, measured as a score from 0 to 7+, had no or negligible effect on overall type 1 diabetes risk (15). The only exception was a small group of females (0.2%) exposed to a very high degree of adversity (7+) who had a higher risk of developing type 1 diabetes compared with unexposed females. We add to this evidence by showing that exposure to a high degree of adversity appears to be specifically related to a higher incidence of type 1 diabetes from 16 years of age onward among females compared with females in the low adversity trajectory group. We are unable to compare this result to previous studies because they have been based on much smaller samples and generally have lacked the power to assess accumulation of adversities, timing of exposure, and age at onset of type 1 diabetes in males and females separately. However, Dube et al. (17) found that exposure to accumulation of adversities in childhood was associated with a higher risk of developing an autoimmune disease in adulthood and that the association was strongest among females.

In our previous study mentioned above, we found no effect of accumulation of childhood adversities among males (15). The current study reveals that a small proportion of males exposed to a high degree of adversity have a higher incidence of type 1 diabetes with onset in childhood (before 11 years of age) compared with males in the low adversity trajectory group. This association was concealed in our previous study because we did not assess the importance of timing of exposure or age at onset of type 1 diabetes. Our present result corroborates previous studies assessing exposure to adversity early in life and type 1 diabetes among males and females combined (10,24), although not consistently (25).

The mechanisms underlying the observed differences in age at onset of type 1 diabetes between males and females in the high adversity group are difficult to determine and should be explored further. Evidence suggests that females experience higher levels of stress in their relationships with family and friends compared with males and that these differences become more salient in adolescence (26). This could contribute to the higher risk of type 1 diabetes incidence in early adulthood observed between females but not between males in the high versus low adversity groups. We found similar, although small, peaks in early adulthood among females in the persistent material deprivation and the loss or threat of loss trajectory groups, which further supports this hypothesis. The observed higher incidence of type 1 diabetes found among males in the high adversity group in childhood compared with males in the low adversity group fits the hypothesis that exposure to adversity in early life may be important for the development of type 1 diabetes since it may affect the physiological stress response (5). However, if this were true, we would expect to see the same pattern among females, which we did not. Thus, our results suggest that the mechanisms underlying the association between high exposure to adversity and type 1 diabetes differ between males and females, but the explanation for this remains unclear.

No specific tests for interaction between age and sex were made in this study because it is well-established that incidence rates of type 1 diabetes have substantially different shapes for males and females (9). Neither did we perform any tests for interaction between age and trajectory group because the main purpose of the study was to provide a description of IRRs between the different trajectory groups. Also, formal P values for interaction are highly dependent on the available sample size, and the value of such P values is therefore limited.

The nationwide Danish registers provided an unselected study population and the opportunity to precisely assess the age-specific incidence of type 1 diabetes across trajectories of childhood adversities for males and females separately. The adversity trajectories were identified in the same study population and included all children born in Denmark in the period 1980–1998. Thus, we were able to assess age at onset of type 1 diabetes in five groups defined by the five most common patterns of timing of exposure to adversities in this nationwide population sample.

Several drawbacks of using register data should also be noted. First, the adversities recorded in the registers are limited. For example, we were unable to cover adversities related to physical and sexual abuse because there are no registers covering child maltreatment in Denmark. However, the adversity covering foster care will capture severe cases of child maltreatment and neglect to some extent. Second, information on some of the adversities, such as alcohol abuse and psychiatric illness, is incomplete since many cases are never registered. Third, the intensity of stress exposure and the time it takes for the stress to wear off likely vary between the specific adversities. By assuming that the most stressful period is the year the adversity occurs, we may have underestimated the long-term effect of specific adversities (e.g., the death of a parent or a sibling). However, by including multiple and repeated measures of adversity, we believe to have captured important patterns of adversity in childhood.

An additional limitation of our study is potential misclassification of diabetes type among people diagnosed in early adult life. People with early-onset type 2 diabetes may be wrongly classified as having type 1 diabetes due to their young age (27). This may have inflated the incidence rate of type 1 diabetes in the high adversity group in young adulthood since both adversity and type 2 diabetes are known to have a strong socioeconomic gradient. While this may partly explain the higher risk of type 1 diabetes among females in early adulthood, the same pattern was not observed among males, making this potential bias less likely to explain our findings. Also, since the information on diabetes type is based on repeated clinical assessments, we assume that the misclassification is modest.

We used high-resolution data to model trajectories of childhood adversity in parallel with type 1 diabetes development to understand the intertwined and accumulated effects of social- and health-related factors over time in line with the syndemics framework (28). This parallel modeling precludes a traditional ordering of causes and effects, and our findings before 16 years of age may be partly affected by reverse causality since there are some indications that having a child with type 1 diabetes may lead to adversities such as divorce (29) and parental psychiatric illness (30,31). Having a child with type 1 diabetes may also be harder on families with fewer resources, resulting in a stronger relationship between type 1 diabetes and adversities in deprived families. This may have contributed to the higher incidence of type 1 diabetes seen among males diagnosed in childhood in the high adversity trajectory group.

Finally, we included only those with complete follow-up until their 16th birthday and thus conditioned on being alive and resident in Denmark until this date. Since both childhood adversity and type 1 diabetes are associated with mortality (32,33), we might have underestimated a positive association between childhood adversities and type 1 diabetes before 16 years of age due to selection bias (34). However, the mortality rate in Denmark is very low in this age group, and, therefore, we expect the magnitude of this bias to be small.

In conclusion, exposure to adversity in childhood is generally not associated with type 1 diabetes when taking into account timing, frequency, and accumulation of exposure and age at onset. Experiencing a very high degree of adversity may affect the risk of type 1 diabetes development among both males and females, and the effect seems to depend on age at onset.

Acknowledgments and Funding. The authors thank the Innovation Fund Denmark (grant 5189-00083B) and Helsefonden (grant 17-B-0102) for providing financial support and the Danish Clinical Registries for providing access to the Danish Registry of Childhood and Adolescent Diabetes and the Danish Adult Diabetes Registry, making this study possible.

Duality of Interest. B.C., J.S., and M.E.J. own shares in Novo Nordisk A/S. J.S. serves as an adviser to Medtronic, Janssen, and Novo Nordisk and has received fees for speaking on behalf of Medtronic, Sanofi, Novo Nordisk, and Bayer AG. M.E.J. has received research grants from AstraZeneca, Amgen, Sanofi, and Boehringer Ingelheim. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.B. contributed to the design of the study and was responsible for data management, data analyses, interpretation of the results, and writing the first draft of the manuscript. A.R. contributed to data management and data interpretation and reviewed and edited the manuscript. B.C. contributed to the design of the study, oversaw the data analyses and interpretation, and reviewed and edited the manuscript. J.S., M.E.J., and N.H.R. contributed to study design, interpretation of the results, and reviewed and edited the manuscript. J.B. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This work was presented in poster form at the 56th Annual Meeting of the European Association for the Study of Diabetes, 21–25 September 2020.