Joint 30-year HbA1c and lipid trajectories and mortality in type 1 diabetes

Abstract

Aims:

Higher HbA1c has been associated with dyslipidemia in type 1 diabetes, but it is unknown whether there is heterogeneity in this association. Thus we assessed the longitudinal association between HbA1c and lipids over 30 years in a type 1 diabetes cohort and examined whether variation in such longitudinal patterns was associated with total and cause-specific mortality.

Methods:

Data were from the Pittsburgh Epidemiology of Diabetes Complications study (n=581 with ≥2 visits, 51% male, baseline mean age 27, diabetes duration 19 years). Longitudinal associations between HbA1c and lipids were assessed in mixed models. Group-based multi-trajectory models identified simultaneous trajectories of HbA1c and lipids.

Results:

Longitudinal HbA1c was associated with Non-HDLc (p<0.0001) and triglycerides (p<0.0001), but not HDLc (men: p=0.72, women: p=0.76). There was heterogeneity in the HbA1c-Non-HDLc association only, with five HbA1c-Non-HDLc groups identified. One group (20%) had an unexpected combination of high HbA1c but normal Non-HDLc and had only moderately increased cardiovascular mortality (rate ratio [RR]=2.80, 95% CI 1.31–6.00) and kidney disease mortality (RR=2.30, 95% CI 0.97–5.50) compared to Low HbA1c-Normal Non-HDLc.

Conclusions:

These results suggest there is a subgroup with type 1 diabetes who, despite poor glycemic control, has a relatively good prognosis, perhaps related to good Non-HDLc.

Introduction

Poor glycemic control has been associated with non-favorable lipid levels in type 1 diabetes starting as early as childhood (1–9). This hyperglycaemia-associated dyslipidemia may contribute to the elevated risk of vascular disease and mortality experienced by people with type 1 diabetes. Most prior studies of the HbA1c – lipid association were cross-sectional or had longitudinal follow-up of less than ten years (1–10). One long-term study was conducted by the Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications (DCCT/EDIC) study investigators, in which they examined what they termed “coprogression” of HbA1c and lipids over 30 years (11). They reported a significant longitudinal association between higher HbA1c and worse lipid profile, including higher LDLc and triglycerides and lower HDLc. However, given the original aims of the DCCT, individuals with key risk factors, including hypercholesterolemia, hypertension and greater than minimal retinopathy, were excluded from the study (12), thus their results may not be applicable to the entire type 1 diabetes population. We hypothesized that there is heterogeneity in the HbA1c-lipid association and that specific longitudinal patterns may be associated with worse prognosis. Thus, our objective in the current study was to assess the 30-year longitudinal association between HbA1c and Non-HDLc, HDLc, and triglycerides in an observational type 1 diabetes cohort, both overall and by identifying heterogeneous joint HbA1c – lipid trajectories. We also examined the association between these joint trajectories and all-cause and cause-specific mortality.

Subjects, Materials, and Methods

Study Population

The Pittsburgh Epidemiology of Diabetes Complications (EDC) Study is a prospective cohort study of childhood-onset (<17 years old) type 1 diabetes. All participants (n=658) were diagnosed, or seen within one year of diagnosis, at Children’s Hospital of Pittsburgh between 1950 and 1980. The cohort has been described in detail elsewhere (13,14); the study timeline is shown in Figure S1. In brief, participants have been followed since 1986–1988, initially with biennial examinations for ten years and thereafter with biennial questionnaires and further examinations at 18, 25, and 30 years. The current analyses included participants with ≥2 HbA1c and cholesterol measurements over follow-up (n=581). Research protocols were approved by the University of Pittsburgh institutional review board and all participants provided written informed consent. The reported investigations have been carried out in accordance with the Declaration of Helsinki.

HbA1c and Lipid Measures

HbA1 was measured at six study visits (1986–88, 1988–90, 1990–92, 1992–94, 1994–96, 1996–98) and HbA1c was measured at three visits (2004–06, 2011–13, 2016–2018). Thus, each participant could have a maximum of nine HbA1/HbA1c measures over 30-years (Figure S1). During the first six visits, HbA1 was measured using automated high-performance liquid chromatography (Diamat; Biorad, Hercules, CA). The two assays had high agreement (r=0.95; Diamat HbA1=−0.18+1.00[Isolab HbA1]). HbA1 values were converted to DCCT-aligned HbA1c values using a regression equation derived from duplicate assays (DCCT HbA1c=0.14 + 0.83[EDC HbA1]) (15). Beginning in 2004, HbA1c was measured using the DCA 2000 analyzer (Bayer Healthcare LLC. Elkhart, IN) and converted to DCCT-aligned HbA1c with the equation: DCCT HbA1c=(EDC HbA1c-1.13)/0.81 (15).

Lipid measurements were obtained at the same time points as HbA1c, as described above (Figure S1). From baseline (1986–88) through the 10-year (1996–98) examination, serum total cholesterol and triglycerides were determined enzymatically (16,17). HDL cholesterol (HDLc) was determined using a modified precipitation technique (18) based on the Lipid Research Clinics method (19). Beginning in 2004, serum lipids were measured using the Cholestech LDX (Cholestech Corp., Hayward, CA). Rather than using calculated LDL cholesterol in the current analysis, we decided a priori to examine Non-HDL cholesterol (Non-HDLc). Non-HDLc, calculated by subtracting HDLc from total cholesterol, is the sum of cholesterol in low-density lipoprotein, intermediate-density lipoprotein, very-low-density lipoprotein, lipoprotein (a), and chylomicrons, and has been shown to be a stronger predictor of vascular disease risk than LDLc alone (20). Additionally, Non-HDLc measurements are less affected by fasting state than LDLc (21), thus using Non-HDLc in analysis helps to maximize both sample size and generalizability, as participants did not need to be excluded due to non-fasting. The inability of a participant to fast prior to their blood draw may be related to concerns regarding hypoglycaemia or other health considerations, so results could be biased if those participants were excluded.

Baseline Clinical Risk Factors

Blood pressure (BP) was measured according to the Hypertension Detection and Follow-Up protocol (22) with a random-zero sphygmomanometer. Hypertension was defined as BP ≥140/90 mmHg or use of BP-lowering medication. Urinary albumin was measured by immunonephelometry (23) in three timed urine samples (24-hour, overnight, and 4-hour collections) obtained over a two-week period. Albumin excretion rate (AER) was calculated for each of the three urine samples; the median of the three AERs was used in analyses. Glomerular filtration rate (eGFR) was estimated by the CKD-EPI creatinine equation (24). Serum creatinine was measured using an Ectachem 400 Analyzer (Eastman Kodak Co., Rochester, NY). Height and weight were measured using standard methods to calculate BMI. Waist and hip circumference were measured at least twice; the average of each was used to calculate the waist-hip ratio. Estimated glucose disposal rate (eGDR), a validated estimate of insulin sensitivity (25), was calculated using the following equation: eGDR (mg/kg/min)=24.395-(12.971*Waist-Hip Ratio)-(3.388*Hypertension)-(0.601*HbA1c).

Smoking history, insulin regimen, physical activity, hypoglycaemia, and medication use were obtained by self-administered questionnaire. Insulin dose was calculated as total insulin units per day/body weight (kg). Physical activity was estimated using the Paffenbarger questionnaire (26). Hypoglycemia requiring assistance was defined as at least one episode of hypoglycemia in the past 2 years resulting in unconsciousness and/or hospitalization or an episode of hypoglycemia in the past 12 months that was not recognized by the participant (i.e., someone else had to tell or help the participant).

Mortality Ascertainment and Adjudication

Vital status was determined as of 31 January 2020. Deaths were identified via next of kin or searches in both the Social Security Death Index and the National Death Index. Causes of death were classified by a committee of physicians using the Diabetes Epidemiology Reporting International (DERI) protocol (27). The adjudication process included review of death certificates, hospital records, autopsy reports, and/or interviews with next-of-kin. The primary underlying cause of death and rank ordered secondary contributors were identified. The following cause-of-death classifications were examined in the current analyses: 1) cardiovascular disease (CVD), 2) diabetic kidney disease (DKD), 3) other chronic diabetes-related complications (e.g., infection/sepsis, chronic malnutrition, brain damage secondary to diabetes, etc.), 4) acute diabetes complication mortality (i.e., hypoglycaemia, hyperglycaemia, DKA), and 5) Other non-diabetes related causes (e.g., cancer, accident, suicide, etc.).

Statistical Methods

Longitudinal associations between continuous HbA1c and each lipid were assessed separately using linear mixed models, with random intercepts and slopes. The lipid was the dependent variable and HbA1c was entered as a fixed effect independent variable. Models were adjusted for type 1 diabetes duration at baseline and sex as fixed covariates and lipid-lowering medication use status at each study visit as a time-varying covariate. For HDLc, the analysis was performed separately by sex.

Group-based multi-trajectory analysis is a data-driven method that facilitates the identification of latent clusters of individuals with similar longitudinal trajectories of multiple continuous risk factors (28). For each lipid with a significant overall longitudinal association with HbA1c, we assessed evidence of heterogeneity of the association by identifying HbA1c-lipid multi-trajectories using a two-stage method in a censored normal model (29). In the first stage of modelling, we identified the number of multi-trajectories describing the latent patterns of HbA1c and lipids in our data set and in the second stage, determined the optimal functional form of those trajectories. This two-stage method is a robust approach because the primary aim of trajectory analysis is to identify the number of trajectory groups, while specific functional form is of secondary importance. In the first stage, we assumed all trajectories were linear (i.e., the simplest functional form allowing for increase or decrease over time). Models with increasing number of trajectories were compared using the Bayesian Information Criterion (Table S2). We required, a priori, that for a model to be selected, all trajectories must comprise ≥5% of the cohort. In the second stage, the optimal functional form (constant, linear, quadratic, or cubic) of each trajectory identified in stage 1 was assessed using the significance level of each polynomial term in the equation. In this step, all trajectories were first modelled as cubic functions (i.e., 3^rd order polynomial, β₀ + βt + βt² + βt³) and any term with p>0.05 was eliminated. The multi-trajectory analysis was performed using the PROC TRAJ macro (SAS 9.4, SAS Institute Inc., Cary, NC) (30).

For each lipid measure with evidence of heterogeneity, pairwise comparisons of baseline characteristics between the most favorable multi-trajectory group (e.g., low HbA1c, low lipid) as the reference group and each other trajectory group were compared using t-tests (or Wilcoxon’s rank sum test) or chi-square tests, as appropriate. Frequency of total mortality and specific underlying (primary) cause of death were compared between the reference trajectory group and each other trajectory using chi-square tests. We applied a Bonferroni-adjusted significance level of p<0.01 (p=0.05 divided by 5 groups).

To further examine the two major causes of death in this cohort (i.e., CVD and DKD), mortality rates per 1000 person years were calculated for: 1) CVD as the underlying (primary) cause of death and 2) DKD as the underlying (primary) cause of death. Two less restrictive definitions were also compared: 1) CVD as the underlying or a contributing cause of death and 2) DKD as the underlying or a contributing cause of death. Rate ratios and 95% confidence intervals for each definition were calculated in diabetes duration and sex-adjusted Poisson models for each trajectory group compared to the most favorable reference trajectory. All analyses were performed using SAS 9.4 (SAS Institute Inc., Cary, NC).

Results

Overall Longitudinal Associations between HbA1c and Lipids

Of 9 possible HbA1c and lipid measures during the follow-up period, the median number of measures per participant was 6 [IQR 4–8]. Mean HbA1c and lipid patterns over 30 years are shown in Figure 1. There was a significant longitudinal association between HbA1c and Non-HDLc (Figure 1, Panel A), with each 1-unit (%) higher HbA1c associated with 6.6±0.4 mg/dl higher Non-HDLc (p<0.0001), after adjustment for type 1 diabetes duration at baseline and sex as fixed covariates and lipid-lowering medication use status at each study visit as a time-varying covariate. There was no significant HbA1c x time interaction (i.e., no evidence that the HbA1c – Non-HDLc association changed over time). There was also a significant longitudinal association between HbA1c and triglycerides (Figure 1, Panel B), with each 1-unit higher HbA1c associated with 9.4±0.9 mg/dl greater triglycerides (p<0.0001). In contrast, there was no association between HbA1c and HDLc over time in either sex (Figure 1, Panels C and D; men: p=0.72, women: p=0.76, respectively).

Figure 1.

Overall 30-Year longitudinal associations between HbA1c and Lipids in the EDC cohort. Panels: A) Non-HDLc, B) Triglycerides, C) HDLc in men, D) HDLc in women. Values are means (geometric mean for triglycerides). Error bars are standard errors.

HbA1c – Lipid Multi-Trajectory Analysis

Given the overall association between HbA1c and both Non-HDLc and triglycerides, we further examined these associations for evidence of heterogeneity. For Non-HDLc, we identified five HbA1c – Non-HDLc combined groups (fitted trajectories Figure 2, observed mean HbA1c and Non-HDLc by group Figure S2). Four of the groups, totalling 80% of the cohort, had the expected HbA1c – Non-HDLc longitudinal relationship based on the overall association: Group 1 (Low HbA1c, Normal Non-HDLc), Group 3 (High HbA1c, High Non-HDLc), Group 4 (Moderate improving HbA1c, High improving Non-HDLc), and Group 5 (High Improving HbA1c, Very High Improving Non-HDLc). One group (Group 2), comprising 20% of the cohort, had an unexpected combination of High HbA1c, Normal Non-HDLc. On the contrary, there was no evidence of meaningful heterogeneity in the HbA1c – Triglyceride association, with just two concordant groups identified: 1) Low-moderate HbA1c – Low-Moderate Triglycerides (60%) and 2) High HbA1c – High Triglycerides (40%) (data not shown).

Figure 2.

30-Year HbA1c – Non-HDLc multi-trajectory groups identified in the EDC cohort

Baseline characteristics of each HbA1c – Non-HDLc trajectory group are shown in Table 1. Groups 1 and 2 had different HbA1c trajectories, but similar Non-HDLc trajectories and had few differences in other characteristics. A smaller proportion of Group 2 was using blood pressure medications (p=0.006) and eGDR was slightly lower (p<0.0001) compared to Group 1. Notably, Groups 1 and 2 had similar albumin excretion rate and Group 2 had higher average eGFR (p<0.0001) than Group 1. Group 3 had lower HDLc, higher triglycerides, larger waist-hip ratio, lower eGDR, higher albumin excretion rate, higher white blood cell count, and a greater proportion of ever and current smokers compared to Group 1. Groups 4 and 5 were older with longer diabetes duration; both groups had baseline risk factor profiles that were generally worse than Group 1, with more extreme differences observed in Group 5 and intermediate differences in Group 4. Very few participants were using lipid-lowering medications at baseline, but use increased over time, as shown in Figure S3. Completeness of lipid-lowering medication data was high with a mean of only 1.2% of those with clinical data missing lipid medication use status at each visit (range 0–3.5%). There was no difference in lipid-lowering medication use over follow-up between Groups 1 and 2 (p=0.72), while Group 3 had marginally higher lipid medication use (p=0.05), and Groups 4 and 5 had significantly higher use (p=0.0008 and <0.0001, respectively) compared to Group 1.

Table 1.

Baseline characteristics by HbA1c – Non-HDLc multi-trajectory group

	HbA1c – Non-HDLc Multi-Trajectory Group
	1: Low HbA1c, Normal Non-HDLc (n=204)	2: High HbA1c, Normal Non-HDLc (n=118)	3: High HbA1c, High Non-HDLc (n=69)	4: Moderate Improving HbA1c, High Improving Non-HDLc (n=162)	5: High Improving HbA1c, Very High Improving Non-HDLc (n=28)
HbA1c, %	7.9 (1.0)	9.2 (1.5) *	10.1 (1.8) *	8.7 (1.3) *	9.4 (1.5) *
HbA1c, mmol/mol	63 (10.9)	77 (15.8) *	87 (19.6) *	72 (14.4) *	80 (16.0) *
Non-HDLc, mg/dl	108.5 (20.7)	115.1 (21.4) *	151.7 (28.3) *	161.7 (30.8) *	225.3 (49.0) *
Type 1 diabetes duration, years	18.3 (7.1)	16.5 (7.2)	18.2 (7.7)	22.2 (7.2) *	22.0 (7.2) *
Age, years	26.4 (7.4)	24.4 (7.7)	26.4 (7.7)	30.7 (7.2) *	30.7 (6.9) *
Age at type 1 diabetes onset, years	8.2 (4.3)	7.8 (4.0)	8.2 (3.9)	8.6 (3.8)	8.7 (3.8)
Female sex, % (n)	53.4% (109)	51.7% (61)	40.6% (28)	46.9% (76)	46.4% (13)
HDL-c, mg/dl	56.1 (13.1)	56.0 (12.1)	51.7 (11.0) *	52.3 (11.6) *	47.0 (9.3) *
Triglycerides, mg/dl**	66.5 (49.5–88.5)	72.0 (57.0–94.0)	101.5 (78.0–174.0) *	102.0 (75.0–154.0) *	220.5 (134.0–313.0) *
Lipid-lowering medication use, % (n)	0	0	0	1.9% (3)	0
Systolic BP, mmHg	111.3 (15.6)	108.6 (11.5)	112.9 (12.4)	117.3 (17.3) *	124.2 (16.9) *
Diastolic BP, mmHg	70.9 (10.6)	70.4 (9.7)	73.7 (8.8)	74.8 (11.9) *	78.9 (12.0) *
Blood pressure-lowering medication use, % (n)	8.2% (16)	0.8% (1) **	10.5% (7)	18.2% (29) **	32.1% (9) *
Hypertension, % (n)	13.7% (28)	2.5% (3)	14.5% (10)	27.2% (44) *	42.9% (12) *
BMI, kg/m2	23.0 (3.1)	23.3 (3.3)	23.3 (3.3)	24.2 (3.2) *	25.6 (3.6) *
Waist-Hip Ratio	0.80 (0.07)	0.82 (0.07)	0.84 (0.06) *	0.84 (0.07) *	0.88 (0.09) *
Estimated Glucose Disposal Rate, mg/kg/min	8.79 (1.57)	8.02 (1.31) *	6.85 (159) *	7.23 (1.96) *	5.73 (2.64) *
Insulin Dose, units/kg	0.80 (0.27)	0.82 (0.24)	0.81 (0.25)	0.75 (0.20)	0.82 (0.35)
MDI or Insulin Pump, % (n)	9.7% (19)	7.9% (9)	3.0% (2)	5.1% (8)	10.7% (3)
Albumin Excretion Rate, µg/min**	10.6 (6.0–62.9)	10.6 (6.3–27.1)	44.4 (11.0–337.0) *	30.9 (9.33–297.9) *	405.8 (36.7–1664.6) *
eGFR, ml/min	103.4 (30.3)	116.9 (22.8) *	106.6 (31.9)	93.6 (31.6) *	82.1 (45.5) *
WBC, x109	6.0 (1.6)	6.2 (1.6)	7.3 (2.1) *	6.9 (1.9) *	7.8 (2.5) *
Ever Smoker, % (n)	27.5% (56)	26.3% (31)	46.4% (32) *	51.9% (84) *	46.4% (13)
Current Smoker, % (n)	14.7% (30)	18.6% (22)	37.7% (26) *	27.2% (44) *	21.4% (6)
Physical Activity, kcal/wk**	1590 (701–2973)	1568 (616–3196)	2296 (804–3876)	1232 (560–2510)	1484 (308–2436)
Hypoglycemia Requiring Assistance, % (n)	43.3% (84)	45.6% (52)	35.4% (23)	38.2% (60)	42.9% (12)
First-Degree Family History of Myocardial Infarction, % (n)	16.2% (33)	9.3% (11)	11.6% (8)	22.8% (37)	28.6% (8)

Values are mean (SD) unless otherwise indicated.

^*p<0.01 compared to Group 1 (bold text highlights statistically significance differences)

^**Median (p25, p75).

Mortality Associations

Proportions of total mortality and underlying (primary) cause of death are shown in Figure 3. Compared to Group 1 (Low HbA1c, Normal Non-HDLc), total mortality was slightly higher in Group 2 (High HbA1c, Normal Non-HDLc; 24.6% vs. 16.7%, p=0.01), while all other groups had substantially greater mortality, ranging from 46 to 68% (p<0.0001 vs. Group 1 for all). Patterns were similar for CVD mortality, though the comparison between Group 1 and Group 2 (p=0.04) did not reach the Bonferroni-adjusted significance level. The proportion with DKD mortality was low in all groups, except Group 5 (High Improving HbA1c, Very High Improving Non-HDLc), of which nearly 18% had DKD as the underlying cause of death (p=0.004 vs. Group 1). There were no differences across groups for death from other chronic diabetes-related complications. Group 4 (Moderate Improving HbA1c, High Improving Non-HDLc) had a slightly greater proportion of deaths from acute complications (4.3% vs. 1.5% in Group 1, p=0.05), while Group 2 had a somewhat greater proportion of deaths from non-diabetes causes (6.8% vs. 2.9% in Group 1, p=0.05).

Figure 3.

Total and Underlying Cause-Specific Mortality by 30-Year HbA1c – Non-HDLc multi-trajectory groups. Comparisons are with the Low HbA1c – Normal Non-HDLc trajectory group (reference), adjusted for type 1 diabetes duration and sex. Group 1=Low HbA1c – Normal Non-HDLc, 2=High HbA1c – Normal Non-HDLc, 3=High HbA1c – High Non-HDLc, 4=Moderate, Improving HbA1c – High Improving Non-HDLc, 5=High, Improving HbA1c – Very High, Improving Non-HDLc

We further examined the two most common causes of death, CVD and DKD, by calculating rate ratios of incidence rates by person time, using Group 1 (Low HbA1c, Normal Non-HDLc) as the reference group. Diabetes duration-adjusted rate ratios and 95% confidence intervals are shown in Figure 4. Compared to Group 1, Group 2 only had increased mortality for CVD as the underlying or a contributing cause (Rate Ratio [RR]=2.8, 95% CI 1.3–6.0), but did not have increased mortality for CVD as the primary underlying cause alone or for either DKD mortality definition. Rate ratios for Group 4 reached statistical significance for both CVD mortality definitions and DKD as the underlying or contributing cause, but the magnitude of the rate ratios were similar to those observed for Group 2. Groups 3 and 5 had similar rate ratios for both CVD mortality definitions. Only Group 5 had significantly increased mortality from DKD as the primary underlying cause alone (RR=12.7, 95% CI 3.4–47), while both Groups 3 and 5 had increased mortality from DKD as the underlying or a contributing cause. Adjustment for potential confounding factors including smoking, insulin dose, BMI, hypertension status, and albumin excretion rate did not meaningfully alter the observed rate ratios, except for Group 5: High Improving HbA1c, Very High Improving NonHDLc for which the rate ratios were reduced after adjustment but still significantly increased compared to Group 1 (Supplemental Figure S4).

Figure 4.

Cardiovascular (CVD) and Diabetic Kidney Disease (DKD) mortality rate ratios by HbA1c – Non-HDLc trajectory group adjusted for type 1 diabetes duration. Low HbA1c – Normal Non-HDLc trajectory group was the reference for comparisons.

Discussion

Over 30 years, worse glycemic control was associated with higher Non-HDLc and triglycerides in the EDC cohort overall, consistent with previous reports in children and adults with type 1 diabetes (1–11). However, to our knowledge, our study is the first to report evidence of heterogeneity in the HbA1c – Non-HDLc association, with 20% of the cohort having normal Non-HDLc despite chronically elevated HbA1c (Group 2). Importantly, that discordant HbA1c – Non-HDLc subgroup had a risk factor profile similar to the subgroup with the most favorable trajectories (Low HbA1c – Normal Non-HDLc). Concomitant with their relatively good risk factor profile, the discordant High HbA1c – Normal Non-HDLc group also appears to carry some protection against long-term CVD and DKD mortality despite their poor glycemic control.

In type 1 diabetes, the association between glycemic control and lipids has been most extensively studied in children and young adults, in an effort to better understand the earlier CVD risk observed in the type 1 diabetes population (1–8). Our finding that HbA1c is strongly associated with Non-HDLc and triglycerides but not with HDLc over time is consistent with prior longitudinal studies in youth (4–9). In adults, the CACTI study likewise observed that worse glycemic control was associated with higher Non-HDLc and triglycerides, but not with HDLc, over six years of follow-up (10). The DCCT/EDIC study reported on 30-year associations between HbA1c and lipids, observing strong longitudinal associations between all examined HbA1c exposure definitions (time-varying current, DCCT/EDIC follow-up time-weighted updated mean, and DCCT updated mean) and both LDLc and triglycerides (11). In that study, HDLc was associated with time-varying current HbA1c (but not the updated mean definitions), which is in contrast to our study where no HbA1c – HDLc association was observed.

We have previously reported on HbA1c trajectories alone in the EDC cohort and identified a group with chronically elevated HbA1c who had a favorable risk factor profile (31). Our current results extend those findings, identifying a latent subgroup who maintains good Non-HDLc over the long-term, despite high HbA1c. Furthermore, we also show that this group seems to be somewhat protected from increased CVD and DKD mortality over 30 years, compared to the other groups with high HbA1c. Importantly, their maintenance of normal Non-HDLc despite poor glycemic control was not explained by lipid-lowering medication use. We also observe that the group with moderate, but improving HbA1c and high, but improving Non-HDLc (Group 4) had intermediate mortality risk, suggesting some benefit from their improvements in both glycemic control and lipids, especially compared to those whose HbA1c and Non-HDLc remained high over follow-up (Group 3). In contrast, Group 5 remained at high mortality risk, suggesting that their improvements in glycaemia and Non-HDLc may be the result of compensatory mechanisms or intensified treatment regimens in response to complication development.

Given the close association between HbA1c and lipids consistently observed across type 1 diabetes studies, a key unanswered question is: does poor glycemic control cause dyslipidemia per se or does their association reflect a common underlying mechanism? Our identification of a discordant HbA1c – Non-HDLc subgroup comprising 20% of the EDC cohort seems to support the latter hypothesis. Lower insulin sensitivity or, conversely, greater insulin resistance has been proposed as a possible mechanism linking poor glycemic control to dyslipidemia (32). In the current study, estimated insulin sensitivity (eGDR) was only slightly lower in the High HbA1c – Normal Non-HDLc group compared to the Low-Normal group. However, that difference was driven by their higher HbA1c alone, as the other two components included in the calculation of eGDR (waist-hip ratio and hypertension) did not differ. Thus, the relatively good insulin sensitivity observed in the High HbA1c – Normal Non-HDLc may possibly explain their lack of dyslipidemia in the presence of poor glycemic control.

One of the most striking observations in the current analyses is that the High HbA1c – Normal Non-HDLc group had similar AER and less hypertension at baseline than the Low – Normal group. Thus, there is little evidence of early kidney damage or renin-angiotensin-aldosterone system dysregulation in the High HbA1c – Normal Non-HDLc group, despite their poor glycemic control. Their lack of kidney damage is unexpected given the well-established association between HbA1c and higher AER, particularly in the microalbuminuria range (33–36). Our observation of a generally favorable risk factor profile in that group, combined with their comparable CVD and DKD mortality to the Low-Normal group suggests that those individuals may have some protection against the detrimental effects of chronic hyperglycemia. One possible reason for this observation is individuals in that group have genetically or epigenetically modulated resistance to glycemia-induced metabolic derangement and vascular damage. Another possibility is that their consistent, albeit high, long-term glycemic control also reflects more stable short-term glucose levels. In vitro studies have shown lower variability of glucose levels to be associated with less oxidative stress and reduced kidney injury even under high glucose conditions, though data have been less consistent in clinical studies (37). A third possibility is the High HbA1c – Normal Non-HDLc group has a large proportion of “high glycators” (38), thus their HbA1c overestimated their actual blood glucose exposure. More research is needed to determine whether such a subgroup with high HbA1c but normal Non-HDLc can be identified in other type 1 diabetes cohorts and to study specific factors that may explain their protection in the face of high glycemic exposure.

The strengths of our study include the use of a well-characterized, exclusively type 1 diabetes cohort with long-term follow-up that has been shown to be epidemiologically representative of childhood onset type 1 diabetes (39,40). The EDC study did not exclude participants based on clinical factors at baseline, increasing generalizability. The cohort has been followed for 30 years to ascertain mortality status and risk factors. Causes of death were adjudicated systematically by a committee of physician epidemiologists. A specific strength of the current analyses is that the multi-trajectory approach facilitates the simultaneous assessment of multiple latent longitudinal risk factor classes, providing data-driven evidence of heterogeneity in the HbA1c – Non-HDLc association within the EDC cohort.

An inherent limitation of long-term cohort studies, including EDC, is that treatment regimens during the early study period may not represent current therapeutic experiences of more recently diagnosed people with type 1 diabetes. However, despite improvements in treatment, recent US data show that average glycemic control in contemporary youth/young adults with type 1 diabetes remains similar (HbA1c~8%) to that observed in the EDC cohort at baseline (41). Another limitation is the potential for “survivor bias”, as participants with <2 HbA1c and lipid measurements (n=77), including n=13 who died prior to the second study visit, were excluded from this analysis. As described in the Methods section, the methodology for measuring HbA1c and lipids changed during follow-up, with the first six study visits using different assays than the last three visits, however, any variation introduced due to changing methods would be likely to bias results toward the null rather than lead to spurious findings. Additionally, HbA1c units were not IFCC standardized, but are aligned to DCCT. A further limitation is that the EDC cohort is 98% Non-Hispanic White, reflecting the demographics of Allegheny County, Pennsylvania (83% White, 14% Black, 3% other racial/ethnic groups) and lower incidence of type 1 diabetes among Black patients during the period when the cohort was diagnosed. Thus, it is unknown whether our findings apply to populations with greater racial/ethnic diversity.

In conclusion, while higher HbA1c was associated with both higher Non-HDLc and triglycerides overall, we observed heterogeneity in the longitudinal HbA1c – Non-HDLc association in this type 1 diabetes cohort. Specifically, we found evidence of a subgroup of people with type 1 diabetes who, despite consistently high HbA1c for 30 years, had relatively low CVD and DKD cause-specific mortality rates, perhaps related to favorable levels of Non-HDLc. This subgroup warrants further study to identify potential protective factors.

Abbreviations:

CVD

cardiovascular disease

DKD

diabetic kidney disease

eGDR

estimated glucose disposal rate