Sex-Specific Blood Pressure Trajectories and Cardiovascular Disease in Type 1 Diabetes: 32-Year Follow-up of the Pittsburgh Epidemiology of Diabetes Complications Cohort Free

Women with diabetes experience a loss of the relative protection against cardiovascular disease (CVD) afforded to premenopausal women without diabetes (1), but specific reasons for that loss of protection have not yet been identified. There is some evidence that established clinical risk factors may relate to diabetes complication risk differently in women and men (2–4). While lipids, particularly HDL cholesterol, have well-established sex differences (5), there is increasing evidence that blood pressure (BP) may also differ in women and men across the life course (6). In type 1 diabetes, BP may increase CVD risk at lower levels than current clinical targets (7), but sex-specific data are limited. Most data on sex differences in BP in people living with type 1 diabetes are from studies of children and adolescents (8,9), so associations with long-term outcomes could not be analyzed. One exception is a recent report from the Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications (DCCT/EDIC) study examining data over 35 years, which showed that, while women with type 1 diabetes had more favorable cardiometabolic risk factor profiles compared with men, including lower BP, there was no corresponding CVD risk benefit (4). The DCCT/EDIC analysis used a two-stage approach, which first characterized longitudinal risk factors and then separately examined the association between time-varying most recent risk factors and CVD incidence. Thus, there remains a need to directly model associations between longitudinal trajectories of cardiometabolic risk factor exposures and CVD risk separately by sex.

Joint models (10) are a statistical approach to simultaneously model longitudinal risk factors and a time-to-event outcome, while accounting for differential follow-up time, thus reducing potential for bias (11,12). We previously used joint models to examine longitudinal HbA_1c in the Pittsburgh Epidemiology of Diabetes Complications (EDC) childhood-onset type 1 diabetes cohort and found no sex differences either in the mean trajectory of HbA_1c over time or in the relationship between longitudinal HbA_1c and CVD by sex (13). Motivated by the need to identify other clinical risk factor targets that may explain the greater relative risk of CVD in women compared with men with type 1 diabetes and the prior evidence suggesting important BP differences by sex, we now use the joint models approach to study associations between BP and CVD. We examined three CVD end point definitions separately by sex: total CVD, major adverse cardiovascular events (MACE), and hard coronary artery disease (hCAD). We hypothesized there are differences in longitudinal trajectories of BP between women and men, as well as differences in BP associations with incidence of CVD end points. Thus, our objective was to characterize longitudinal trajectories of systolic BP (SBP) and diastolic BP (DBP), pulse pressure (PP), and mean arterial pressure (MAP) and their associations with 32-year CVD incidence separately in women and men with type 1 diabetes. As glycemic control and nephropathy are strongly related to both BP (14,15) and CVD in type 1 diabetes (16), we also examined whether the associations were independent of longitudinal trajectories of HbA_1c and overt nephropathy, as well as other established CVD risk factors.

EDC is a prospective cohort study of childhood-onset (<17 years old) type 1 diabetes. All participants (n = 658) were diagnosed with type 1 diabetes, or seen within 1 year of their diagnosis, at Children’s Hospital of Pittsburgh between 1950 and 1980. The cohort has been described in detail (17,18). Briefly, participants have been followed since the baseline examination in 1986–1988. The current analysis includes data collected through 31 December 2020. Clinical examinations were conducted biennially for the first 10 years of follow-up and at 18, 25, and 30 years. Questionnaires were administered biennially over the entire follow-up. Research protocols were approved by the University of Pittsburgh Institutional Review Board (Pittsburgh, PA) (approval #19040065); all participants provided written informed consent. The current analyses were preplanned as part of an ancillary study (ADA grant #7–23-ICTSWH-19, institutional review board approval #23030147) and examined the cohort stratified by sex assigned at birth, abstracted from medical records at cohort inception (n = 325 women, n = 333 men).

Blood pressure was measured at baseline and 2, 4, 6, 8, 10, 18, 25, and 30 years’ follow-up according to the Hypertension Detection and Follow-up protocol (19), using a random-zero sphygmomanometer from baseline through year 10 and an aneroid sphygmomanometer at 18, 25, and 30 years. Three measures were taken; the average of the second and third measures was used in analysis. PP was calculated as PP = SBP − DBP. MAP was calculated as MAP = SBP + (2*DBP)/3. Current medications were collected by questionnaire at each time point and reviewed by study staff. Medications were coded using the World Health Organization Collaborating Center for Drug Statistics Methodology Anatomical Therapeutic Chemical classification system (20). Antihypertensive (AH) medication use was considered positive at each visit if a participant reported using any drugs from the C02, C03, or C07–C09 Anatomical Therapeutic Chemical groups.

From baseline through 10 years, HbA₁ values were converted to DCCT-aligned HbA_1c values using a regression equation derived from duplicate assays (DCCT HbA_1c = 0.14 + 0.83[EDC HbA₁]) (21). At 18, 25, and 30 years, HbA_1c was measured using the DCA 2000 analyzer (Bayer HealthCare lLC, Elkhart, IN) and converted to DCCT-aligned HbA_1c using the equation DCCT HbA_1c = (EDC HbA_1c – 1.13)/0.81. Urinary albumin was measured using immunonephelometry in each of three timed urine samples (24-h, overnight, and 4-h collections) obtained over a 2-week period (22); the median of the three albumin excretion rates (AERs) was used in analyses. Overt nephropathy was defined as AER > 200 µg/min in at least two of the three timed urine samples. Total cholesterol and triglycerides were determined enzymatically (23,24), and HDL cholesterol was determined using a modified precipitation technique (25) based on the lipid research clinics method (26). Serum creatinine was measured using an Ektachem 400 Analyzer (Eastman Kodak Co., Rochester, NY). eGFR was estimated using the Chronic Kidney Disease Epidemiology Collaboration creatinine equation (27). Height and weight were measured using standard methods to calculate BMI. Insulin dose was calculated as total insulin units per day/body weight (kg). Past and current smoking status and highest level of education were obtained by self-administered questionnaire. Physical activity was assessed using the Paffenbarger Questionnaire (28), and average total weekly energy expenditure was calculated (kcal/week).

Participants were followed prospectively for 32 years to ascertain time to CVD incidence or censoring (31 December 2020 or last follow-up). Total CVD was defined as the first instance of CVD death, nonfatal myocardial infarction (MI), nonfatal stroke, coronary revascularization or coronary blockage ≥50%, ischemic electrocardiogram (Minnesota codes 1.3, 4.1–4.3, 5.1–5.3, or 7.1), or angina as determined by EDC study physician. CVD death was confirmed by medical records, death certificates, autopsy reports, and/or next of kin interview. Nonfatal MI was confirmed with medical records or subclinical MI on electrocardiogram (Minnesota codes 1.1 or 1.2). Nonfatal stroke and coronary revascularization or blockage ≥50% were confirmed with medical records. MACE was defined as the first instance of CVD death, nonfatal MI, or nonfatal stroke, and hCAD was the first instance of CAD death, nonfatal MI, coronary revascularization, or coronary blockage ≥ 50%; n = 25 women and n = 29 men with prevalent CVD at baseline were excluded; thus n = 300 women and n = 304 men were eligible for the analysis.

Mean longitudinal SBP, DBP, PP, and MAP trajectories were compared between women and men using linear mixed models with age in years as the timescale (PROC MIXED, SAS version 9.4; SAS Institute Inc., Cary, NC). Each BP metric was modeled as the dependent variable and sex as the independent variables of interest, adjusting for longitudinal time-varying AH medication use status. Models were fit with a random intercept and slope. Sex differences in BP metric slope were estimated using sex × age interaction terms.

Separately in women and men, longitudinal BP metrics and their associations with CVD end points (total CVD, MACE, and hCAD) were analyzed using joint models that simultaneously estimate the longitudinal repeated measures and time-to-event data processes. Longitudinal data were censored prior to the time of the first CVD event. For each outcome, separate multivariate joint models with random intercept and slope were fit for each BP metric with years of follow-up as the timescale and minimally adjusting for baseline age using the joineRML package version 0.4.6 in R version 4.1.2 (R Foundation for Statistical Computing, Vienna, Austria). Next, we sequentially adjusted for longitudinal time-varying AH medication use status, longitudinal HbA_1c, and longitudinal time-varying overt nephropathy status, retaining each in the model if P < 0.10. Because of the computationally intensive nature of joint models, the number of longitudinal variables that can be estimated in a single model without convergence problems is limited; thus we a priori selected HbA_1c and overt nephropathy for inclusion as the longitudinal covariates of interest. In the final set of “fully adjusted” models, we added the following traditional CVD risk factors at baseline as covariates: non-HDL cholesterol, HDL cholesterol, triglycerides, BMI, smoking status, more than a bachelor’s degree, estimated average weekly energy expenditure, white blood cell count, and eGFR. The best-fitting fully adjusted models were selected using backward selection, retaining baseline covariates with P < 0.05 and comparing the Akaike information criterion to assess model fit as covariates were removed. For all models, trace plots were examined to assess convergence of the Monte Carlo expectation maximization algorithm. Maximum number of Monte Carlo iterations was set at 20,000, using the recommended number of burn-in iterations = 100k, where k is the number of longitudinal parameters in the model; however, fewer burn-in iterations did not alter the results. Diabetes duration is highly correlated with age (r = 0.9, P < 0.0001) in this childhood-onset type 1 diabetes cohort, so both variables were not included together in the models. Results of alternative models adjusting for diabetes duration instead of age were nearly identical and thus not presented. Finally, we performed sensitivity analyses repeating the total CVD joint models in the subcohort of participants diagnosed 1965–1980 to examine evidence of potential cohort effects.

The data analyzed are not publicly available because of language in the consent forms of this legacy cohort that prohibited broad data sharing. Data may be requested from the corresponding author under a data sharing agreement in accordance with University of Pittsburgh Institutional Review Board policies.

Baseline characteristics by sex are provided in Table 1. SBP, DBP, PP, MAP, triglycerides, AER, and estimated energy expenditure per week were all significantly lower in women compared with men, while HDL cholesterol was higher in women. The median number of BP measures prior to any CVD incidence was six (interquartile range 3–8).

Sex-specific, model-estimated longitudinal BP metric trajectories across age are shown in Fig. 1. Corresponding parameter estimates are provided in Supplementary Table 1. Mean SBP over follow-up was 5.81 mmHg lower in women compared with men (P < 0.0001). SBP increased 0.49 mmHg per year of age overall (P < 0.0001), with no difference in longitudinal SBP slope by sex (P = 0.4809). Mean DBP was 6.19 mmHg lower in women compared with men (P < 0.0001). DBP decreased 0.09 mmHg per year of age overall (P < 0.0001); however, the decrease was significantly faster in women compared with men (β_{sex × age} [SE] = −0.26 [0.05], P < 0.0001). While there was no difference in average PP by sex (P = 0.4490), PP increased 0.47 mmHg per year of age overall (P < 0.0001), with a significantly faster increase in women (β_{sex × age} [SE] = 0.20 [0.07], P < 0.0001). Finally, mean MAP was 5.66 mmHg lower in women (P < 0.0001). MAP increased 0.12 mmHg per year of age overall (P = 0.0002), but the increase was significantly slower in women compared with men (β_{sex × age} [SE] = −0.18 (0.06), P = 0.0025). Rates of AH medication use were similar in women and men over time, with annual increases of 1.6% and 2.0% per year, respectively (P = 0.2477) (Supplementary Fig. 1).

Median CVD-free survival time was 30.0 years (95% CI 25.7, 31.7) for women and 28.4 years (95% CI 24.1, 20.8) for men. Cumulative total CVD incidence was 45.7% (n = 137) in women and 46.4% (n = 141) in men (P = 0.37) (Supplementary Fig. 2). Cumulative MACE incidence (31.3% and 31.6%, respectively) and hCAD (37.4% and 37.8%, respectively) were also nearly identical by sex. Of the 163 women who did not develop CVD and were censored, 23 (14%) died at (median) 14 years’ follow-up and 59 (36%) were lost to follow-up at (median) 21 years. Of the 163 men who did not develop CVD and were censored, 37 (23%) died at (median) 14 years’ follow-up and 51 (31%) were lost to follow-up at (median) 21 years. Associations between incidence of each CVD end point and SBP and DBP are shown in Table 2 (PP and MAP in Supplementary Table 2). For total CVD, in models minimally adjusted for age, higher longitudinal SBP, DBP, PP, and MAP were all similarly significantly associated with increased CVD incidence in both sexes. Associations were only slightly attenuated by adjustment for longitudinal AH medication use. After adjusting for longitudinal HbA_1c, the SBP, DBP, and MAP associations with total CVD remained similar, while the association with PP was attenuated in both sexes. Longitudinal overt nephropathy status was not associated with total CVD in women, and its inclusion did not improve model fit, so it was dropped from the multivariable models in that subgroup. In men, adjustment for longitudinal overt nephropathy status only slightly attenuated the BP metric HRs. In both sexes, adjustment for other risk factors attenuated the SBP association with total CVD, but DBP and MAP remained significantly associated (Fig. 2A and Supplementary Table 2).

For MACE, all BP metrics were more strongly associated in men compared with women when adjusted for age only. After adjustment for longitudinal AH medications, only DBP and MAP remained associated with MACE in women (Table 2 and Supplementary Table 2). Those associations were further attenuated after adjustment for other CVD risk factors (Fig. 2B); thus no BP metrics were independently associated with MACE in women. In contrast, in men, SBP, DBP, and MAP were all strongly associated with MACE even after adjustment for longitudinal AH medication use, HbA_1c, and overt nephropathy and other CVD risk factors (Fig. 2B).

For hCAD, associations with SBP, DBP, and MAP were significant and nearly identical by sex in the age-adjusted models and after adjustment for longitudinal AH medication use and HbA_1c (Table 2 and Supplementary Table 2). SBP was no longer associated with hCAD after adjustment for overt nephropathy in men, while, in women, SBP was no longer associated with hCAD after adjustment for HbA_1c. After adjustment for other CVD risk factors, DBP remained independently associated with hCAD in women, while no BP metrics remained associated in men (Fig. 2C).

The distributions of specific initial events comprising MACE and hCAD are provided in Supplementary Table 3. For both MACE and hCAD, women were more likely to have a fatal first event, while men were more likely to have a nonfatal MI. Women were also more likely than men to have a nonfatal stroke as their first MACE. The proportion with coronary revascularization and/or blockage of ≥50% as their initial event was nearly identical by sex. To determine whether the lack of association between BP and MACE in women was driven by a lack of BP-stroke association in women, we fit joint models estimating associations between each BP metric and time to stroke in a post hoc analysis. No BP metrics were associated with stroke incidence in women, while SBP, DBP, and MAP were significantly associated with stroke incidence in men (Supplementary Table 4). Finally, in sensitivity analyses restricted to participants diagnosed with diabetes in 1965–1980 (n = 206 women, 206 men, CVD incidence 34% in both), BP-CVD associations were similar to those observed in the overall cohort (Supplementary Table 5).

In this cohort of people living with type 1 diabetes, women had consistently lower SBP, DBP, and MAP over time compared with men. Most striking was the finding that women also had significantly faster average decline in DBP compared with men. While the observed mean differences of 5.8 mmHg lower SBP and 6.2 mmHg lower DBP in women compared with men exceed clinically meaningful thresholds for primary CVD prevention (29), the lower cumulative BP exposure in women in this type 1 diabetes cohort did not correspond to any apparent CVD benefit, as the 32-year incidence of all CVD end points was nearly identical in both sexes. Furthermore, despite the lower absolute levels in women, DBP and MAP had nearly identical associations with the relative risk of composite total CVD in women and men. However, upon examining MACE and hCAD, sex differences became apparent, with BP metrics more strongly associated with MACE in men but with hCAD in women.

While women in the general population have lower average BP compared with men across the life course (6), our study reveals important differences in age-related BP changes between this type 1 diabetes cohort and what has been reported in the general population. For example, in the Framingham Heart Study, women had a faster SBP increase but similar DBP decrease compared with men (30). Here, in people with type 1 diabetes, we observed the opposite: similar SBP increases in both sexes but a steady DBP decline in women only observed across the entire age range represented in our study (8–69 years old). In the general population, average DBP increases until the age of 50 to 60 years, after which it begins to decline in both sexes (6), with faster DBP decrease in women occurring around the time of the menopause transition (31). The mean age of women in EDC was 27 years at baseline, which is more than 20 years younger than their average age at menopause (32). Therefore, the observed DBP decrease in women in our study cannot be explained by menopause. Age-related decline in DBP is caused by increased arterial stiffening (33) and has been shown to occur regardless of prior diastolic hypertension (34). Our results suggest that, in type 1 diabetes, arterial stiffening may begin earlier in women compared with men, and additional studies to identify contributing factors early in the life course are critically needed.

While the relative risk of total CVD associated with BP was nearly identical by sex, we observed important differences for MACE and hCAD. In women, BP was weakly associated with MACE, while, in men, higher SBP, DBP, and MAP were strongly associated with increased hazard of MACE independent of other risk factors. In contrast, for hCAD, BP metrics were similarly associated with incidence in both sexes when adjusting only for age and time-varying AH use. However, after adjustment for other risk factors, higher DBP remained independently associated with hCAD incidence in women, while, in men, all BP-hCAD associations were attenuated after adjustment for overt nephropathy and further attenuated by adjustment for other CVD risk factors. Both MACE and hCAD include fatal CAD and nonfatal MI but are differentiated by the inclusion of fatal or nonfatal stroke in MACE and coronary revascularization and/or coronary blockage of >50% in hCAD. BP was a weaker risk factor for stroke in women compared with men, despite no difference in stroke incidence by sex (35). On average, both SBP and DBP were below the hypertensive range in EDC women, which may at least partially explain the lack of BP-stroke association, and concomitant lack of BP-MACE association, in women. Additionally, AH medication use is partially a proxy for the severity of hypertension; thus, adjusting for AH medication use over time may underestimate associations between continuous BP and stroke. On the other hand, associations between BP and hCAD in men were explained primarily by overt nephropathy, while DBP remained independently associated with hCAD in women. The observation that nephropathy is a stronger risk factor for long-term CAD risk in men confirms earlier EDC data on 10-year CAD incidence (36). Notably, PP was not associated with incidence of any of the CVD end points examined in either sex after adjustment for HbA_1c, consistent with prior findings that PP is a relatively weaker predictor of CVD than other BP metrics (37). Finally, our observation that women were more likely to have a fatal first event raises the possibility that disparities in health care delivery leading to delayed recognition of CVD in women could contribute to differences in BP-CVD associations, warranting further study.

Our results had similarities and differences compared with a recent DCCT/EDIC report. In both studies, women had the same 5.8 mmHg lower absolute SBP compared with men over time, however the absolute DBP difference by sex was nearly twice as large here (6.2 mmHg lower in women) than in DCCT/EDIC (3.2 mmHg lower in women). Additionally, DCCT/EDIC found most recent SBP and DBP were more strongly associated with total CVD and MACE incidence in women compared with men, while we observed similar associations by sex for total CVD but weaker BP associations with MACE in women. There are several key differences between EDC and DCCT/EDIC, including hypertension being an exclusion criterion for DCCT and the randomized treatment assignment, which have resulted in lower CVD incidence in DCCT/EDIC (<20%) compared with EDC (46%). Another difference is, while EDC and DCCT/EDIC have similar age distributions, EDC participants were younger at diabetes onset, thus any given age is associated with longer average exposure to diabetes in EDC. Therefore, the larger decrease in DBP observed in EDC women over time could be due to diabetes-related accelerated aging, a hypothesis warranting further research. Finally, our analysis incorporated the entire longitudinal BP trajectory prior to CVD incidence, while the DCCT/EDIC report examined time-varying most recent BP. Despite those differences, both studies are consistent in the conclusion that BP remains an important CVD risk factor in women with type 1 diabetes regardless of their lower absolute BP compared with men.

Our study has several strengths. EDC is a well-characterized, prospective study with long-term follow-up, designed to systematically ascertain incidence of and risk factors for complications of type 1 diabetes. The study used standardized clinical examinations and adjudication of CVD outcomes by EDC study physicians, thus reducing the gender bias in CVD assessment that may be present in studies relying on data from medical records alone (38). Our analysis minimized potential for reverse causality by only including BP and other risk factors measured prior to CVD incidence. The multivariate joint models incorporate all BP measures to model their longitudinal trajectory with respect to CVD incidence, accounting for within-subject variability and adjusting for key confounders. Another advantage of joint models is the ability to reduce bias associated with event-free survival far beyond the last measurement of a longitudinal covariate, by accounting for subject-specific random effects (39). More commonly used methods, such as modeling the two processes separately, may increase bias if participants survive far beyond the most recent follow-up (12). Limitations of our study include an inability to include all risk factors as longitudinal covariates, because of increasing computational intensity with each additional longitudinal factor; thus we only adjusted for baseline traditional CVD risk factors, except for longitudinal HbA_1c and overt nephropathy. Another limitation is the cohort is 98% White, because of the demographics of Allegheny County, PA (<15% African American/Black) and lower T1D incidence among African Americans during the diagnosis period of the cohort (40). Thus, more research is needed to determine whether these results apply to more diverse groups.

In summary, we observed important differences in longitudinal BP metrics and their associations with 32-year CVD incidence between women and men in a cohort of people living with type 1 diabetes diagnosed during childhood. Prior research has suggested CVD risk increases at lower BP in type 1 diabetes compared with the general population, so BP targets specific to this high-risk population may be warranted (7). Our current results suggest there are also important sex differences in BP and its relationship with CVD; thus, sex-specific guidelines should also be considered to reduce CVD risk in people living with type 1 diabetes. In addition, mechanisms underlying the earlier DBP decline in women, the weaker BP-stroke association in women, and the role of nephropathy in the BP-CAD association in men should be prioritized in future research.

Acknowledgments. The authors thank Jaia Gallegos and Delaney Asbury, University of Pittsburgh, for their assistance with data cleaning and literature review and Stella Miller, Pittsburgh, PA, for her assistance in creating the graphical abstract. The graphical abstract was created in BioRender, BioRender.com/m77i929 (R.G. Miller, 2024).

Funding. This study was supported by grants from the National Institutes of Health (R01-DK034818) and American Diabetes Association (grant #7-23-ICTSWH-19), and by the Rossi Memorial Fund (Pittsburgh, PA).

Duality of Interest. R.G.M. has received honoraria and travel and meeting expenses from the American Diabetes Association. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. R.G.M., T.J.O., and T.C. were involved in the conception, design, and conduct of the study and the interpretation of the results. R.G.M. performed the analysis and wrote the first draft of the manuscript. All authors edited, reviewed, and approved the final version of the manuscript. R.G.M. is the guarantor of this work and, as such, had full access to all data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Preliminary data on these findings were presented at the 84th Scientific Sessions of the American Diabetes Association, Orlando, FL, 22 June 2024.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Elizabeth Selvin and Neda Laiteerapong.