Introduction
In 2015, Miller et al described a cohort from the Type 1 Diabetes Exchange, demonstrating a substantial rise in Hemoglobin A1C (A1C) levels in adolescence and emerging adulthood.1 A follow-up study showed that this longitudinal rise in A1C is even more striking with chronologic time.2 The factors that drive this rise in A1C are poorly understood, but there appears to be an effect of sex on A1C trajectory.3 For example, while there is no difference in the mean A1C between girls and boys up until age 15, females have a significantly higher mean A1C compared with males at age 16 and beyond.3 In fact, A1Cs have been shown to be higher among females compared with males at diagnosis and over the course of their lifespans as assessed through different study designs, including cross-sectional studies, longitudinal cohort studies, and registry studies.4
Other than female sex, a number of different factors are associated with elevated A1C trajectory, including younger age at type 1 diabetes diagnosis,5 less frequent blood glucose monitoring,6 minority ethnicity,78 lack of insulin pump use38 and higher maternal conflict scores.6 Yet another factor that may impact A1C trajectory is body mass index (BMI), which is known to rise substantially among post-pubertal females, especially when compared with post-pubertal males.59 Higher BMI is associated with a less favourable A1C trajectory in some studies,510 but not in others.3 As such, the differences in A1C trajectory by sex and BMI among individuals diagnosed with type 1 diabetes in childhood and adolescence remain incompletely understood.3
The purpose of this study was to describe A1C trajectory by sex among individuals living with type 1 diabetes diagnosed in childhood and adolescence using a paediatric diabetes registry and to determine if there is effect modification of sex by BMI categories on A1C trajectory. Given that BMI trajectory rises among adolescent females,59 we hypothesised that there would be differences in A1C trajectory by BMI categories, which would be particularly pronounced in females. Research findings supporting this hypothesis would prompt further study in weight assessment and management in the optimal treatment of type 1 diabetes during adolescence and into early adulthood.
Methods
Study design and data source
This was a retrospective cohort study using data from a provincial diabetes registry from British Columbia (BC), Canada, known as the British Columbia Pediatric Diabetes Registry (BC-PDR), which was established in 2017. Details regarding the creation of this provincial registry are described elsewhere.11 In short, recruitment into this diabetes registry began at the diabetes clinic at BC Children’s Hospital in Vancouver, Canada in 2017, with expansion to other areas of the province starting in 2019.11 The BC-PDR collects two types of data: patient-level data collected at recruitment (eg, sex, date of birth, date of diabetes diagnosis) and visit-level data collected at each diabetes clinic visit (eg, height, weight, BMI and A1C), which could be every 3–6 months depending on the needs of the patient and local practice patterns.11 Study investigators had full access to registry data. Data (eg, A1C) from up to 2 years prior to recruitment until the last paediatric diabetes visit (typically around age 18 years) were extracted from the BC-PDR in July 2023.11 The current study includes data from 30 June 2016 to 30 June 2023.
Data linked to other databases were not used in this study.
Patient and public involvement
There was no patient or public involvement in the design or conduct of this study.
Exposure, outcome, and other variables
Our primary exposures were biological sex assigned at birth (female and male) and patient age (in years). Our primary outcome was age-specific A1C (according to the National Glycohemoglobin Standardisation Programme). We excluded A1C values within 3 months of diabetes diagnosis, as we were interested in longitudinal A1C trajectory rather than A1C at diagnosis.
We considered BMI as a possible effect modifier. Age- and sex-specific BMI percentiles were calculated per WHO standards and were categorised as underweight/normal (<85th percentile), overweight (≥85th and <95th percentile) and obese (≥95th percentile).12 Other variables ascertained from the BC-PDR included age at diagnosis, age at study recruitment, A1C at diagnosis, A1C at study recruitment, ethnicity, presentation of diabetes at onset, presence of chronic comorbidities, blood pressure (BP), number of injection time points per day, insulin pump use, total daily dose (TDD) of insulin, number of glucometer checks per day, continuous glucose monitor (CGM) use, recent severe hypoglycaemia and recent hospitalisation for diabetic ketoacidosis (DKA) (see online supplemental table 1 for data dictionary). These variables were used to characterise the study population, as well as for model adjustment, as described below.
Statistical analysis
The study sample was restricted to individuals aged 0–20 years old with type 1 diabetes and at least one A1C≥3 months following diagnosis. For cohort characteristics, we calculated medians and IQRs for continuous variables and counts and proportions for categorical variables, stratified by biological sex. We also calculated the average A1C with SD for females and males at each age.
We used linear generalised additive models (GAMs) to examine longitudinal A1C values by yearly age and compared trajectories between females and males. GAMs use flexible penalised functions to allow for possible non-linear associations between exposure and outcome.13 Age was modelled with a thin plate regression spline. Random effects were included to account for repeated measures among patients over time.
To examine differences between females and males, biological sex assigned at birth was included in the model as a parametric term and as an interaction with age. We plotted the fitted age- and sex-specific values from these models and computed age-specific mean differences between sexes with corresponding 95% CIs, calculated using the delta method. We used the likelihood ratio test, comparing models with and without the age-sex interaction to assess if the A1C trajectories of females and males differed significantly from one another.
To examine possible patient factors that contributed to observed divergence in A1C between females and males between age 15 and 18 we summarised the cohort by sex at clinic visits closest to age 15 (between age 14 and 16) and age 18 (between age 17 and 19). As such, the sample of this sensitivity analysis was restricted to individuals with A1C tests at ages 14–16 and/or 17–19, respectively. Based on these results, we adjusted the above GAMs for BMI, TDD of insulin and presence of chronic comorbidities to examine if this changed the age-specific differences in A1C (ie, if any of these factors informally mediated some of the total age-specific sex effect).
Finally, we assessed the possibility of effect modification by BMI by stratifying our age-sex specific models into underweight/normal, overweight and obese. We estimated the conditional effects of sex within strata of BMI (female vs male) and the joint effects (sex plus BMI), compared with males who were underweight/normal BMI. These models were run both unadjusted and adjusted for TDD of insulin and presence of chronic comorbidities.
Missing data
Descriptive analyses present counts for missing data. GAMs use imputed missing values for BMI and TDD of insulin at specific visits by using patient information from previous or subsequent visits. Specifically, if the first or last BMI or TDD was missing, the subsequent or prior value (if within 1 year) was used to impute.
If an intermediate visit’s BMI or TDD value was missing, we used a time-weighted average (ie, closer time points given higher weight) of the previous and subsequent record, provided they were both within 1 year of the missing visit (ie, 2-year window). In cases where both the previous and next non-missing records were >1 year from the missing visit, we did not impute.
Results
Descriptive statistics
There were 7635 A1C records among 674 individuals with type 1 diabetes included in this study. Baseline characteristics of study participants are summarised in table 1. Demographic and clinical characteristics were similar between females and males at recruitment (table 1).
Each patient contributed a median of 5 years of data (IQR 3–6) and 11 A1C measurements (IQR 6–15), with no differences between females and males. At 5 years of age, both males and females had an average A1C of 7.5% (SD=1.0% for both sexes), which increased to 8.2% (SD=1.6%) in females and 8.3% (SD=1.6%) in males at age 13. By age 17, females had an average A1C of 8.7% (SD=1.9%), while average A1C in males decreased to 8.0% (SD=1.7%) (table 2).
Generalised additive models
Estimated A1C trajectories from GAMs by sex are depicted in figure 1a. There was a significant interaction between age-specific A1C trajectories and sex (p<0.001). Namely, females had a consistent increase in A1Cs with age, while males had a rise in A1Cs during childhood, peaking at age 13, then decreasing after that.
From age 5 to 13 years, the mean A1C difference (female – male) was essentially zero (eg, at age 8, it was −0.02% (95% CI −0.12% to 0.16%, p=0.77)) (figure 1b). Beginning at age 14, there was a statistically significant difference between mean A1C in females and males (eg, at age 14, it was 0.15% (95% CI 0.04% to 0.25%, p=0.008) (figure 1b). This difference continued to increase through the adolescent years into emerging adulthood, such that the mean A1C difference between females and males at ages 16, 18 and 20 were 0.51% (95% CI 0.40% to 0.62%, p<0.001), 0.77% (95% CI 0.59% to 0.94%, p<0.001) and 1.03% (95% CI 0.62% to 1.44%, p<0.001), respectively (table 2).
Analyses adjusted for BMI, TDD of insulin and presence of chronic comorbidity were similar, with and without imputation, though females had slightly lower mean A1C from ages 8 to 11 (online supplemental table 2) and online supplemental figure 1).
Descriptive characteristics at the clinic visit closest to age 15 and 18 are shown in online supplemental table 3) and were similar between sexes for all covariates except for BMI. At age 15, the proportion of females with underweight/normal, overweight and obese BMI was 47.9%, 28.5% and 23.6%, respectively, but among males, it was 69.9%, 14.5% and 15.5%. At age 18, the differences in proportion of individuals in each BMI category by sex persisted; the proportion of females with underweight/normal, overweight and obese BMI was 45.7%, 25.9% and 28.4%, respectively, but among males, it was 69.9%, 16.8% and 13.3%.
Figure 2 shows the estimated A1C trajectories by sex and BMI. Figure 3 depicts the mean differences in A1C between BMI-sex strata (vs underweight/normal BMI males). There was clear effect modification by BMI, with differences in sex being driven predominantly by overweight and obese females (figure 2).
There was no difference in A1C trajectory between normal BMI males and normal BMI females (eg, at age 18, mean A1C difference was 0.10%, 95% CI −0.19% to 0.40%, p=0.50), while for overweight and obese females there were large A1C differences compared with normal weight males (eg, for obese females at age 18, mean A1C difference was 0.89%, 95% CI 0.64% to 1.13%, p<0.001) (figure 3). Additionally, there were lower A1Cs among overweight and obese males compared with normal weight BMI males beginning at age 14 (eg, for overweight males at age 18, mean A1C difference was −0.74%, 95% CI −1.14% to −0.34%, p<0.001) (figure 3). All results were similar with and without adjustment (figure 3).
Missing data and imputation
Proportions of missing data are shown in table 1. Over the 7-year study, 3160/7635 (41.4%) BMI measures were missing and 2671/3160 (84.5%) of missing values were imputed. The median age is 12 years (IQR 9–15) among those not missing BMI and 14 years (IQR 10–16) among those missing BMI.
TDD of insulin measures were missing in 3972/7635 (52.0%) records and imputed in 2790/3972 (70.2%) of those missing. The median age is 12 years (IQR 9–15) among those not missing TDD of insulin and 13 years (IQR 10–16) among those missing TDD of insulin. Sensitivity analyses were performed using a complete case approach (ie, no imputation) for comparison.
All results were similar with and without imputation (online supplemental figures 2 and 3).
Conclusions
Summary of findings
The A1C trajectories of females and males living with type 1 diabetes are overlapping during childhood. However, starting at age 14, the A1C trajectory of females progressively rises compared with males. We observed a difference in BMI distribution between females and males at age 15 and 18 years.
The difference in A1C trajectory between females and males is not explained by differences in BMI, TDD of insulin or presence of chronic comorbidities, but is substantially altered in different BMI classes. Specifically, this sex difference is driven by A1C increases in overweight and obese females starting at age 14, with no differences in A1C trajectory between females and males among those with underweight/normal BMI.
Sex differences in A1C trajectory
Although two studies did not find a sex difference in A1C trajectory,814 many more have observed a sex difference in A1C, whereby females have higher A1Cs.3–6915 We found that the age at which A1C trajectories between females and males began to separate was at 14 years. Other studies are consistent with this, with divergence starting between 12 years9 and 16 years.3 This suggests that the departure between A1C trajectories in females and males starts during puberty. Several reasons have been proposed for increased A1C among pubertal females, which may be broadly grouped into biological factors and psychosocial factors.414
Biological factors
There are multiple physiologic mechanisms that may mediate differential glycaemic control between post-pubertal females and males, including sex steroid effects, hormonal signalling and immune system functioning.16 For example, IGF-1 levels are decreased in type 1 diabetes due to insulinopenia, resulting in higher compensatory GH levels.17 Due to the role of oestrogen in stimulating GH release,16 adolescent females have higher levels of GH compared with adolescent males, which increases insulin resistance and thus glucose levels.18 During exercise, oestrogen in females mediates a preference for lipid oxidation, while the lack of oestrogen in males mediates a preference for carbohydrate oxidation, which may be another reason for higher glucose levels in females.16 Testosterone may also facilitate less adipose tissue deposition, less insulin resistance and lower glucose levels in adolescent males.18 Females with type 1 diabetes have two times the leptin levels compared with males during puberty, indicating leptin resistance and downstream effects on appetite, fat mass and insulin sensitivity.19 Finally, levels of endothelial progenitor cells (especially CD133+KDR+ and CD34+KDR+ CD133+ cells) are associated with improved glycaemic control and are significantly lower in females compared with males.2021 Further studies are needed to determine the relative contribution of the above physiologic mechanisms and uncover other yet unknown processes.
Psychosocial factors
Adolescent females have significantly higher diabetes distress scores compared with males in all seven subscales of the Type 1 Diabetes Distress Scale (powerlessness, management distress, hypoglycaemia distress, negative social perceptions, eating distress, physician distress and family/friend distress), with powerlessness and management distress being the primary drivers of the positive correlation between diabetes distress and A1C.22 Female youth living with type 1 diabetes also report higher psychological distress, higher perceived stress, more eating disorder symptoms, lower levels of resilience and lower well-being scores, all of which may impact glycaemic control.23 A higher proportion of female youth report fear of hypoglycaemia compared with male youth,24 which can cause higher glucose levels through less aggressive insulin dosing, over-treating hypoglycaemic events and consumption of meals with higher carbohydrate content.25 Some studies have also shown that adolescent females have higher rates of smoking, which is associated with poorer glycaemic control.2426 Lastly, females living with diabetes incur higher direct healthcare expenditures, such as prescription costs, compared with males, which may discourage adequate treatment of diabetes.27
Effect modification of sex by BMI on A1C trajectory
We found that there was a higher proportion of females in the overweight and obese categories at post-pubertal ages, which is consistent with other studies.59 When we adjusted the A1C trajectory analyses for BMI, females still had higher A1Cs starting at age 14. Stratified analysis demonstrated that this higher A1C trajectory was driven by overweight and obese females. These findings suggest that BMI alone is not the main factor involved in elevated A1Cs among overweight and obese females, but rather a synergistic factor with female sex.
Interestingly, we also found that post-pubertal males who were in the overweight and obese BMI categories had lower A1Cs compared with those in the underweight and normal weight BMI categories, representing a stark difference in the effect of BMI on A1C trajectory according to sex.
Like for the sex difference in A1C trajectory above, we presume that there are both biological and psychosocial mechanisms behind the differential effect of sex on A1C according to BMI categories.5 From a biological perspective, a paediatric study found that females with type 1 diabetes under the age of 15 years had higher A1Cs at 2–5 years post-diagnosis compared with males, but this was not the case in type 2 diabetes. Additionally, females with type 1 diabetes had higher BMI at 5 years post-diagnosis, but no sex effect on BMI among children with type 2 diabetes.9 This suggests that disease-specific processes differ between type 1 diabetes (eg, autoimmunity and insulin deficiency) and type 2 diabetes (eg, insulin resistance) such that the sex-BMI interaction may differ depending on the underlying pathophysiology of diabetes.9 Another study involving children, adolescents and young adults (aged 7–24 years) with type 1 diabetes demonstrated a higher A1C trajectory among females starting at age 16, but no overall A1C difference between those of normal weight and overweight/obese BMI.3 However, those with overweight/obese BMI interestingly had lower A1Cs in the age 14–18 year and 19–24 year categories (but not the age 7–13 year category), although no interaction analysis was done to determine if there was a sex-BMI effect.3 We suspect that there may have been a sex-related effect of BMI on A1C trajectory, similar to that shown in this study. From a psychosocial perspective, numerous processes have been suggested, such as need for independence, sense of immortality, peer influence, depression and disordered eating, although evidence for these is lacking.14 Further study is needed to confirm the sex-BMI interaction on A1C and to elucidate the mechanistic pathways of these observations.
Clinical implications
Recent evidence has shown that there is substantial disease heterogeneity within type 1 diabetes, which may be impacted by genetics, immunologic behaviour, c-peptide status, sex hormone status, BMI and other factors.2829 There has been a suggestion to introduce the concept of different ‘endotypes’ within type 1 diabetes, where each endotype has unique and individualised modalities of care optimisation.2930 Our study suggests that post-pubertal females with overweight or obese BMI may be a distinct endotype that warrants particular attention because of the dramatic rise in A1C trajectory starting at age 14 years. We wonder if the introduction of weight management strategies for this endotype might be an intervention that warrants further study. Some of these strategies may include counselling for lifestyle modification, the use of oral adjunctive medications (eg, metformin) or perhaps the use of injectable adjunctive medications (eg, GLP-1 agonists).31 We note that comorbid elevations in glucose and BMI contribute to risk for insulin resistance, metabolic syndrome and future cardiovascular complications, highlighting the need for treatment strategies for both.9 More research is needed to unravel the seemingly complex associations between age, sex, BMI and glycaemic parameters, since the interactions between these are not fully understood.14
Strengths and limitations
A major strength of this study is the use of a provincial diabetes registry that started collecting data in 2017. Additionally, the use of GAMs allowed us to flexibly compare A1C trajectories, while using all available measures. Several studies have observed longitudinal sex differences in A1C trajectory and BMI trajectory separately,59 but our study includes analyses that stratify A1C trajectory according to sex and BMI, allowing us to uncover the interaction between these parameters as described.
Our study has several limitations. Our data are observational and unmeasured variables, such as socioeconomic status, presence of other autoimmune conditions and presence of psychiatric comorbidities may have provided further insights.10 However, given that the two primary exposures (sex and age) do not have any theoretical confounders (ie, causes), our estimates of sex- and age-specific A1C (or their differences) are unbiased. That said, additional data on variables such as c-peptide, autoantibody titre, diabetes distress, nutrition, physical activity, health literacy and frequency of diabetes follow-up would have allowed us to better understand the relationship between sex, BMI and A1C levels.32 The data were from only one provincial jurisdiction in Canada, which may limit generalisability.
Despite this, our findings are congruent with other studies showing a rise in A1C values starting in adolescence.12 Lastly, there are significant proportions of missing data for some characteristics. In the GAMs, we imputed data for BMI and TDD of insulin using fixed values for individuals within 1 year. Though we do not expect drastic changes in these measures within 1 year, this approach requires strong assumptions and may underestimate uncertainty.
Sensitivity analyses using a complete case approach led to very similar results. In general, missing data were comparable between sexes, which suggests a lower risk of bias in sex-specific differences.
Conclusions
There is an increase in A1C values beginning in adolescence that is more pronounced in females compared with males. This sex difference in A1C trajectory appears to be driven by females with a BMI in the overweight and obese category. Further study is needed to clarify the relationship between age, sex, BMI and glycaemic parameters in hopes of understanding the best ways to help stabilise glycaemic control during this vulnerable life stage.
Supplementary material