Sex Differences in Rates of Change and Burden of Metabolic Risk Factors Among Adults Who Did and Did Not Go On to Develop Diabetes: Two Decades of Follow-up From the Tehran Lipid and Glucose Study

Diabetes is a growing health challenge globally. The International Diabetes Federation estimated the global prevalence to be 451 million in 2017. These figures are expected to increase to 693 million by 2045 (1). The Middle East and North Africa (MENA) region had the second-highest prevalence (11%) of diabetes in 2017, globally (2). Accelerated urbanization in the MENA countries has led to a rapid transition from traditional to westernized diet, sedentary lifestyle, and lack of physical activity, which are the main factors in the rising prevalence of obesity (2,3). Generally, in 2016, the prevalence of obesity in MENA region was >30 and 20% in women and men, respectively. The most devastating outcome of obesity is type 2 diabetes. Recent studies have shown insulin resistance as the common link associated with obesity and diabetes (4). Moreover, overweight and obesity have been found to be modifiable risk factors for other cardiometabolic risk factors, including hypertension and dyslipidemia (5). During 1975–2015, MENA showed the highest prevalence of hypertension than the rest of the world (2). National studies among Iranian population in 2007 showed that the prevalence of obesity, physical inactivity, diabetes, hypertension, and hypercholesterolemia were 22.3, 40, 8.7, 26.6, and 42.9%, respectively, all of them were higher among female than in men (6,7). It has been well established that pathophysiological changes leading to the development of diabetes begin many years before its diagnosis (8). For example, there is evidence that progressive increase in fasting plasma glucose (FPG), 2-h postload plasma glucose (2-hPG), weight gain, and attenuation of insulin sensitivity may be detected 10 years before diagnosis of diabetes (9,10). A few studies have investigated gradual changes of several metabolic risk factors (MRFs) in adults before onset of diabetes, but with inconsistent results. The Whitehall II study showed unfavorable trajectories of systolic blood pressure (SBP) and HDL cholesterol (HDL-C) in adults who developed diabetes, compared with adults who did not develop disease over a period of 14 years (11). However, another small study over a 9-year period found no differences in blood pressure and HDL-C, but larger changes in BMI in men who developed diabetes compared with men in whom it did not develop (12). Importantly, the majority of existing studies in this area have been carried out in the U.S., U.K., and European countries (8,11,13), and thus, it is not possible to generalize their findings regarding the change in the MRFs before diabetes to other ethnic populations, especially to the MENA population with high burden of MRFs (2). Conducting separate longitudinal studies among different ethnic groups with longer follow-up period and inclusion of a broad range of MRFs may provide deeper insights into the physiological changes before diabetes onset. To our knowledge, no previous study has investigated the trajectory of multiple MRFs before the diagnosis of diabetes among the MENA population. By using repeated measurements of MRFs among participants of the Tehran Lipid and Glucose Study (TLGS), we aimed to identify change trajectories in MRFs prior to diabetes and to estimate linear rate of change and cumulative burden of MRFs among adult men and women who did and did not go on to develop diabetes during follow-up.

Longitudinal data were obtained from the TLGS cohort study. Full details regarding the sampling method and characteristics of participants have been previously published (14,15). Briefly, the TLGS is an ongoing longitudinal study of a random sample of residents living in Tehran, the capital of Iran, and designed to identify the risk factors and outcomes for noncommunicable disease. Six waves (phases) of examinations have been conducted until now. The first phase (1999–2001) comprised 15,010 individuals aged >3 years at baseline. In the second phase (2002–2005), 3,544 new participants were included. For the current study, we included 12,312 participants aged 20–70 years from the first (n = 9,967) and second (n = 2,345) phases as baseline population, and followed them in the next phases, including: phase 3 (2005–2008), phase 4 (2009–2011), phase 5 (2012–2015), and phase 6 (2015–2018). Of 12,312 participants, we excluded 1,244 individuals with prevalent diabetes at baseline. Of the remaining 11,068 eligible participants, we excluded those with missing data on diabetes status at baseline (n = 731) and those with no follow-up data (n = 1,857) after recruitment until the end of the study (18 April 2018). Finally, to identify nonlinear patterns of MRFs, individuals who participated in at least three phases (n = 7,163) were selected as the final study population (Supplementary Fig. 1). For individuals who developed diabetes during follow-up, all data before diabetes diagnosis were included. At each phase, written informed consent was obtained from the study participants. Study protocols were approved by the ethical committee of the Research Institute for Endocrine Sciences of Shahid Beheshti University of Medical Sciences.

All baseline and follow-up data were collected using standardized protocols. Information on age, medication use, smoking status, and family history of diabetes (FHD) was obtained via standardized questionnaires. Height was measured to the nearest 0.1 cm and weight to the nearest 0.1 kg. BMI was calculated as weight in kilograms divided by height in square meters. Blood pressure levels were measured using the average of two measurements taken on the right arm at 5-min resting intervals. Peripheral blood samples were collected in the morning after a 12-h fast for biochemical measurements, including FPG, 2-hPG, total cholesterol (TC), triglycerides (TG), and HDL-C. Physical activity level (PAL) was assessed using the Lipid Research Clinics questionnaire (16) in the first phase. From the second phase, PAL has been assessed by the Modifiable Activity Questionnaire (17).

Smoking status was categorized as current smoker versus nonsmoker. A current smoker was a person who smokes cigarettes or other smoking implements daily or occasionally. Nonsmokers included never-smokers and ex-smokers. FHD was defined as having diabetes in first-degree relatives. PAL was categorized as low and high. In the first phase, low PAL was defined as doing exercise or labor less than three times a week, and in the second phase, it was defined as achieving a score ≤600 MET-minutes per week (18). In all examinations, diagnosis of diabetes was based on FPG ≥7 mmol/L or 2-hPG ≥11.1 mmol/L or treatment with antidiabetic drugs.

Baseline characteristics of participants were compared between those who did and did not develop diabetes using the Student t test and χ² test, as appropriate. Also, we compared baseline characteristics of participants and nonparticipants. Nonparticipants were those with missing data on diabetes status at baseline, those without any follow-up data, and those individuals with less than three times of participation in the study. To remove positive skewness, the natural log of TG (Ln-TG) was used in all analyses. We used individual growth curve (IGC) analysis to estimate the sex-specific trajectory of MRFs measured at multiple times from young adulthood to old age before the diagnosis of diabetes and for those without this diagnosis. IGC is an advanced multilevel, linear mixed-effect regression technique that “focuses the study of change on interindividual differences in intraindividual change” (19). A mixed-effect model includes fixed and random effects and assumes that each individual has a unique pattern of change. It estimates the average change over time (fixed effect) and the individual differences around the fixed parameters (random effects). This approach allows for analysis of unequally spaced repeated measures. IGC is fitted in two levels: in level 1, the trajectories of individuals are estimated over time, and in level 2, intercept and slopes are measured as the overall population mean (20). To derive curve parameters (intercept and slopes) for the population, maximum likelihood estimates are used in the IGC analysis. The modeling strategy suggested by Mirman (21) and Singer and Willett (20) was followed in this study. We fitted three possible polynomial curves of the MRFs as a function of age: a linear, quadratic, and cubic to allow for nonlinear change of MRFs. The higher-order terms of age were included in models if their P values were significant (<0.05). We included random intercept and random slope in all models and evaluated sequentially if each additional random parameter improved the model fit. The goodness-of-fit of the models was assessed using likelihood ratio tests and Akaike information criterion (20). Because including the age and its higher-order terms in IGC produce collinearity, we centered the age at the grand mean age (47.2 years). Moreover, the term age² was divided by 10 and age³ by 20 to stabilize the variance terms (22). Cubic curves were fitted for BMI, SBP, diastolic blood pressure (DBP), TC, and Ln-TG. A quadratic curve was fitted for FPG and HDL-C, and a linear model was fitted for 2-hPG (Supplementary Tables 3–18). We computed the area under the curve (AUC) using the growth curve parameters as an overall measure of long-term burden of MRFs. Because individuals had different follow-up periods, the AUC was defined as the integral of the growth curves for each individual from 20 to 70 years of age divided by 50 to get the annual burden of each MRF (22). For each individual, the rate of change (linear change) of each MRF was defined as the combined (fixed plus random) coefficients of age term in IGC models. The differences in the mean of MRFs, AUCs, and curve parameters (rates of change and intercepts) were tested using two-way ANCOVA with sex and diabetes status as the main effects and an interaction term of sex and diabetes status. In fact, interactions assess simultaneous effects of diabetes status and sex on dependent variable and shows whether changes in means between those individuals in whom diabetes developed and those in whom it did not differed by sex. Due to the wide range of baseline age (20–70 years) among study participants, the means were adjusted for baseline age (age-adjusted means). Moreover, to examine the statistical interactions for sex by diabetes in the presence of potential confounders, further adjustment was applied for FHD, smoking status, and PAL. For multiple comparisons, the Bonferroni correction was applied to the P values. The ANCOVA was applied by general linear model, which has the additional benefit of allowing us to include an interaction term. All analyses were performed in R 3.6.2 (https://www.r-project.org), using the nlme package (23), and IBM SPSS Statistics for Windows, version 20 (IBM Corporation, Armonk, NY), and significance levels were set at a two-tailed P value of <0.05.

The study population included 7,163 individuals (3,069 men and 4,094 women), aged 20–70 years, with the mean (SD) ages of 40.3 (13.2) and 38.4 (12.4) years in men and women, respectively. Baseline characteristics of study population by diabetes status at follow-up are shown in Supplementary Table 1. In both men and women, individuals without incident diabetes were younger and had lower levels of BMI, FPG, SBP, DBP, Ln-TG, TC, and 2-hPG compared with individuals who developed diabetes. They also had lower probability of positive FHD compared with their counterparts with diabetes. Baseline characteristics of participants and nonparticipants are shown in Supplementary Table 2. There was no significant difference between participants and nonparticipants regarding sex, age, TC, FHD, and PAL. However, participants had lower levels of BMI, FPG, SBP, DBP, Ln-TG, and 2-hPG compared with nonparticipants. They had also lower probability of being a current smoker compared with nonparticipants. The median (interquartile range) follow-up of participants was 15.4 (12.5–16.5) years. The study sample included 743 (316 men) new case subjects with diabetes. The mean (SD) age of onset for diabetes was 58.9 (12.1) and 57.1 (11.5) years in men and women, respectively. Approximately 88% of study samples participated four to six times in the study.

Age-adjusted means of MRFs measured at baseline are presented in Table 1 by diabetes status, separately for men and women. Among both men and women, diabetes was associated with higher levels of BMI, FPG, SBP, DBP, TC, Ln-TG, and 2-hPG (P values <0.001), but a lower level of HDL-C (P < 0.001) in women but not in men. Testing for sex difference showed that among individuals in whom diabetes developed, women had higher levels of BMI, HDL-C, 2-hPG (all P < 0.001), and TC (P = 0.011), but lower Ln-TG (P = 0.024) compared with their male counterparts. Among those in whom diabetes did not develop, women had significantly higher BMI, HDL-C, TC, and 2-hPG, but lower levels of FPG, SBP, and Ln-TG (all P values <0.001) compared with their male counterparts.

The differences between those in whom diabetes developed and those in whom it did not develop were greater in women than in men for BMI, HDL-C, and Ln-TG (the P values for sex-diabetes interaction were 0.002 for BMI, 0.033 for HDL-C, and 0.038 for Ln-TG). As shown in Table 1, the sex-diabetes interaction for these MRFs remained significant even after further adjustment for smoking status, PAL, and FHD.

Supplementary Tables 3–18 represent detailed characteristics and estimates of the final models and their submodels for each MRF, separately for men and women.

The means for predicted value of each MRF are graphed in Supplementary Fig. 2, in which age is represented on the horizontal axis and change of MRFs is function of age, age squared, and age cubed. In each curve, the squiggled lines (solid and dotted) display the average of predicted values across individuals computed based on combined fixed and random effects of final models, and the blue solid lines have been smoothed using generalized additive model with polynomial function. In Fig. 1, we have shown only smoothed curves of each MRF in four groups (i.e., men and women with and without diabetes).

The AUC values and rates of change in studied MRFs extracted from final models are summarized in Table 2. Among both men and women, the age-adjusted mean of AUC values of all MRFs was significantly higher in those in whom diabetes developed compared with those in whom it did not develop, except for HDL-C. Testing for sex difference revealed that among individuals in whom diabetes did and did not develop, women had significantly higher AUC values of BMI, HDL-C, TC, and 2-hPG, but lower AUC values of FPG, SBP, DBP, and Ln-TG compared with their male counterparts. The magnitude of difference between those with and without diabetes were remarkably higher in women compared with men (P values for sex-diabetes interaction in age-adjusted model were 0.011 for AUC of BMI, 0.036 for AUC of FPG, <0.001 for AUC of HDL-C, and 0.031 for AUC of Ln-TG). The interactions remained significant after further adjustment for smoking status, PAL, and FHD (Table 2).

Age-adjusted mean of rates of change in MRFs is presented in Table 2. Among men, those individuals who developed diabetes had significantly faster rates of change in BMI, FPG, SBP, DBP, and 2-hPG, but showed slower rate of change in HDL-C. Among women, those individuals in whom diabetes developed had significantly faster rates of change in all MRFs except for Ln-TG and HDL-C. Among those individuals in whom diabetes did and did not develop, women showed a higher rate of change in all MRFs except for 2-hPG compared with their male counterparts. The differences in rates of change in BMI, FPG, HDL-C, and TC between those with and without diabetes were greater in women, compared with men, even after adjustment for smoking status, PAL, and FHD (the P values for sex-diabetes interaction were 0.006 for BMI, <0.001 for FPG and HDL-C, 0.002 for TC, and 0.055 for SBP) (Table 2).

As age was centered on the grand mean age of 47.2 years, the combined intercepts in each growth curve represent the mean of each MRF at the age of 47.2 years. Figure 2 shows the differences in age-adjusted mean values of MRFs at 47.2 years of age among those with and without diabetes. The differences in mean of BMI, FPG, HDL-C, and Ln-TG in whom diabetes developed versus those in whom it did not develop were greater in women compared with the comparable groups of men at the age of 47.2 years (all P values for sex-diabetes interaction were <0.05) (Fig. 2).

In this population-based longitudinal study with the follow-up period of 20 years, we tracked changes in major MRFs among 7,163 individuals who were between 20 and 70 years of age at the baseline. Using the multilevel IGC analysis, in both men and women, we found a greater burden of major MRFs and faster rates of change among those who did develop diabetes compared with those who did not. We also observed a sex difference in cumulative burden and rate of change of MRFs among individuals who did and did not develop diabetes; the differences between those in whom diabetes did and did not develop were more pronounced in burden of BMI, FPG, HDL-C, and TG and rates of change in BMI, FPG, HDL-C, TC, and SBP in women than in men.

To our knowledge, this is the first study in the MENA region to detail trajectory parameters of major MRFs before onset of diabetes among the adult population using IGC analysis. To date, only two studies among the U.S. population have examined the trajectory of cardiovascular disease (CVD) risk factors before the diagnosis of diabetes from childhood to midlife by this approach (24,25).

In this study, we found that differences in burden of BMI, FPG, Ln-TG, and HDL-C between those who did and did not develop diabetes were greater in women than in men. Our finding is broadly consistent with previous studies that confirmed that men attain less on average of several risk factors before onset of diabetes than do women. A population-based study in Scotland showed that the average of BMI before diagnosis of diabetes was lower in men than women across the age range (26). In a cross-sectional study of individuals without diabetes and with diabetes aged 60–79 years, it was found that levels of BMI, insulin resistance, HDL-C, and DBP differed more between women with and without diabetes women than between men with and without diabetes (27). Recently, Du et al. (24) have shown a higher burden of BMI, TC, LDL-C, and FPG from childhood to midlife among women who developed diabetes compared with their male counterparts.

Fundamentally, excess weight is the most important risk factor for diabetes in all individuals across the globe (28). When individuals gain weight, they become more insulin resistant. But adult men are more insulin resistant than women (29), and women have shown greater insulin sensitivity despite considerable weight gain compared with men (30). Consistent with previous literature, we observed significantly higher longitudinal exposure to BMI in women than men. However, the level of exposure to FPG and TG was significantly lower in women compared with their male counterparts. In fact, prior research has linked insulin sensitivity with body fat distribution (31). Women have greater subcutaneous fat stores and lower liver fat, and due to such body compositional differences, adult women are more insulin sensitive than their male counterparts (32); hence, they need to attain higher average of BMI in order to show an insulin-resistant state (30).

In the current study, we also found that differences in rates of change in BMI, FPG, HDL-C, and TC between those in whom diabetes did and did not develop were more marked in women compared with their male counterparts. Although this finding cannot support a causal association between rates of change of these MRFs and diabetes incidence, the association of rapid change in weight and risk of diabetes and CVD has been previously reported in a growing body of evidence (33,34).

A large number of studies have linked the greater adverse changes of CVD risk factors in women, compared with men, to increased risk for CVD outcomes when diabetes manifests (24,29,32,35). The small number of examinations (six waves) in this study did not allow us to investigate the causal association between AUC/rate of change of MRFs in development of diabetes and its complications; however, we have previously shown that the risk of coronary heart disease incidence was significantly higher in newly diagnosed women with no previous history of coronary heart disease than their male counterparts (36).

The graphs of IGC models (Fig. 1) in our study give an important insight into the pattern of change in MRFs. We found an inverse U-shape pattern for trajectory of BMI, TC, Ln-TG, and DBP in both men and women (i.e., increase after 20 years of age), in which they reach a peak and decline slightly thereafter. These findings are in line with current evidence that has shown young adults gain weight until middle age, stabilize, and begin to lose weight near 60 years of age (37). It has also been shown that DBP, plasma TC, and TG concentrations increase progressively, reaching peak values in middle age, and then start to decline (38). We found that among individuals in whom diabetes developed, TC and Ln-TG levels reached their peak at ∼45 and 60 years in men and women, respectively. But the peaks of the BMI trajectories occurred at earlier ages (40 years in men and 50 years in women) compared with those in whom diabetes did not develop. As BMI is one of the most important modifiable risk factors for diabetes and CVD, targeting adult men under the age of 40 and women under age of 50 years for the weight management could be the most effective intervention, at least among our study population.

In this study, we found a higher burden of BMI, TC, and 2-hPG and faster rates of change in BMI, FPG, SBP, DBP, TC, and Ln-TG among women in whom diabetes did not develop than their male counterparts. However, the burden of FPG, SBP, DBP, and Ln-TG was greater among men in whom diabetes did not develop than their female counterparts. These observations lead to two important questions. What do the aforementioned findings mean for public health? Should women in the general population receive more attention than men from the health care systems? The faster rates of change in MRFs observed in women without diabetes may lead to a greater burden of MRFs, as was found in this study for some of them. In contrast, a higher burden of MRFs in women in whom diabetes did not develop may increase both the relative risk for CVD events and complications of diabetes if it occurs (24,29). Given the high prevalence of obesity, physical inactivity, and other CVD risk factors among Iranian women (6,7), our study suggests that intervention programs for these MRFs might be effective in reducing the diabetes complications and CVD events after diagnosis of diabetes among women. Moreover, different patterns of change in multiple MRFs among women than men suggest that prevention interventions need to consider sex-specific strategies.

The current study has several strengths, including the large size of the cohort, uniform and systematic follow-up assessments, standardized direct (rather than reported) measurements of weight and height, FPG, and lipid profiles in each follow-up, and repeated measurements of other variables. However, some limitations also should be considered in the interpretation of its findings. First, participants were healthier than nonparticipants, and we might underestimate diabetes incidence. Second, data were from a Middle Eastern adult population, suggesting that our findings in this cohort may not be generalizable to other ethnic populations.

The current study showed that during the transition to diabetes, women experienced more adverse change in multiple MRFs than men. They had a higher burden of exposure to BMI, FPG, TG, and HDL-C and faster rates of change in BMI, FPG, HDL-C, TC, and SBP than men, suggesting that women attain a higher average of major MRFs before diabetes than do men. The greater exposure of women to and burden of MRFs before onset of diabetes may have implications for implementing sex-specific strategies to reduce diabetes complications.

Acknowledgments. The TLGS is a joint effort of many investigators, and the authors thank the large number of staff, consultants, and advisers for help and important contributions.

Funding. This study was supported by National Research Council of the Islamic Republic of Iran grant 121.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.R. researched data, contributed to the interpretation of results, and wrote the manuscript. F.A. contributed to the study design and discussion and reviewed the manuscript. F.H. contributed to the interpretation of results and discussion and reviewed the manuscript. F.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.