Mode of Onset of Type 2 Diabetes from Normal or Impaired Glucose Tolerance

  1. Ele Ferrannini1,
  2. Monica Nannipieri1,
  3. Ken Williams2,
  4. Clicerio Gonzales3,
  5. Steve M. Haffner2 and
  6. Michael P. Stern2
  1. 1Metabolism Unit, Department of Internal Medicine and C.N.R. Institute of Clinical Physiology, University of Pisa School of Medicine, Pisa, Italy
  2. 2Division of Clinical Epidemiology, Department of Medicine, University of Texas Health Science Center at San Antonio, Texas
  3. 3Centro de Estudios en Diabetes, American British Cowdray Hospital and Endocrinology and Metabolism Service, Division of Internal Medicine, Specialty Hospital of the National Medical Center, Mexican Social Security Institute, Mexico City, Mexico
  1. Address correspondence and reprint requests to Ele Ferrannini, MD, Department of Internal Medicine, Via Roma, 67, 56126 Pisa, Italy. E-mail: ferranni{at}


Fasting plasma glucose concentrations (FPG) predict development of type 2 diabetes. Whether hyperglycemia evolves from normoglycemia gradually over time or as a step increase is not known. We measured plasma glucose and insulin levels during oral glucose testing in 35- to 64-year-old men and nonpregnant women from a population-based survey (Mexico City Diabetes Study) at baseline (n = 2,279) and after 3.25 (n = 1,740) and 7 years (n = 1,711) of follow-up. In subjects with normal glucose tolerance (NGT) on all three occasions (nonconverters; n = 911), FPG increased only slightly (0.23 ± 0.79 mmol/l, mean ± SD; P < 0.0001) over 7 years. In contrast, conversion to diabetes among NGT subjects (n = 98) was marked by a large step-up in FPG regardless of time of conversion (3.06 ± 2.57 and 2.94 ± 3.11 mmol/l, respectively, at 3.25 and 7 years; P < 0.0001 vs. nonconverters). Likewise, in subjects who converted to diabetes from impaired glucose tolerance (n = 75), FPG rose by 3.14 ± 3.83 and 3.12 ± 3.61 mmol/l (P < 0.0001 vs. nonconverters). Three-quarters of converters had increments in FPG above the 90th percentile of the corresponding increments in nonconverters. Converters had higher baseline BMI (30.4 ± 4.9 vs. 27.3 ± 4.0 kg/m2; P < 0.001) and fasting plasma insulin values (120 ± 78 vs. 84 ± 84 pmol/l; P < 0.02) than nonconverters; however, no consistent change in either parameter had occurred before conversion. In contrast, changes in 2-h postglucose insulin levels between time of conversion and preceding measurement were significantly (P < 0.0001) related to the corresponding changes in FPG in an inverse manner. We conclude that, within a 3-year time frame, the onset of diabetes is very often rapid rather than gradual and is in part explained by a fall in glucose-stimulated insulin response.

Development of type 2 diabetes is, to some extent, a predictable event. Several studies in different populations have identified anthropometric and metabolic characteristics that increase the likelihood that a person with initially normal glucose tolerance (NGT) will progress to diabetes over a specified period of time (13). Thus, family history of diabetes and obesity are potent risk factors amplified by increasing age. In addition, both fasting hyperinsulinemia and the fasting plasma glucose concentration (FPG) itself mark, independent of one another, an enhanced risk of developing the disease (4,5). In the combined analysis of three prospective studies (6), the presence of one or more components of the metabolic syndrome, namely, hyperinsulinemia, dyslipidemia, hypertension, and glucose intolerance, predicted the emergence of diabetes over 8 years of follow-up. Among individuals with impaired glucose tolerance (IGT), the rate of progression to overt diabetes is higher than among people with NGT by a factor of 2–10 (7,8).

That a higher, if still normal, FPG shares with other risk markers the ability to predict future hyperglycemia has led to what probably is the concept prevailing among clinicians: that hyperglycemia evolves from euglycemia gradually over time. Thus, the spontaneous evolution of such a “tracking” trait in individuals who are destined to become diabetic would be that of a monotonic increment of FPG until it crosses the diagnostic threshold. In this view, the pre-diabetic phase is a slow progression toward hyperglycemia. Whether this is the case, however, is not firmly established. Possibly the only information relevant to this issue comes from a small cohort of Pima Indians (n = 34) whose plasma glucose concentrations were measured, at 2-year intervals, twice before and once after diagnosis. In these initially NGT subjects, glucose concentrations measured 2 h after an OGTT were noted to rise abruptly (by 6–8 mmol/l) during the 2 years before diagnosis from a perfectly normal level maintained for the preceding 2 years (9).

These findings prompted the following questions: 1) Is a rapid onset a common mode of diabetes development? 2) Does a rapid onset also characterize the emergence of diabetes from IGT? 3) If onset is relatively rapid, then what are the precipitating factors? In an attempt to answer these questions, we analyzed the information provided by the Mexico City Diabetes Study, a population-based survey in which glucose tolerance was tested three times over a period of 7 years.


In Mexico City, six low-income neighborhoods (colonias) were selected for the study (10). Complete enumerations of these colonias were carried out between February 1990 and October 1992, and 3,326 study-eligible individuals (35- to 64-year-old men and nonpregnant women) were identified. Of these, 2,813 (85%) completed a home interview, and 2,279 completed a medical examination (response rate, 69%). Subjects who attended the clinic examination were similar to those who provided a home interview only, in terms of age, sex, and self-reported history of diabetes. The protocol was approved by the Institutional Review Board of the University of Texas Health Science Center at San Antonio, and all subjects gave informed consent.

In April 1993, we began a first wave of follow-up (3.25 years) to determine the incidence of diabetes (11). The response rate to this first follow-up was 76% (1,740 of 2,279). Subjects who attended this follow-up examination were similar to those who did not attend it in terms of age, sex, and self-reported diabetes. In February 1997, a second wave of follow-up (7 years) was begun; the response rate to this was 75% of those who attended the baseline examination, again with no significant differences in age, sex, or history of diabetes between attenders and dropouts. The distribution of glucose tolerance status among subjects who were examined on the three occasions is shown in Table 1.

Height, weight, and waist and hip circumferences were measured using previously described methods (4). The waist circumference was used as a measure of body fat distribution, whereas BMI was used as a measure of overall adiposity. At baseline and follow-ups (3.25 and 7 years), blood samples were obtained after a 12- to 14-h fast for determination of plasma glucose and insulin concentrations. Glucose and insulin concentrations were also measured 2 h after a standardized 75-g oral glucose load. Blood samples were centrifuged, and the plasma was divided into aliquots and stored at −70°C. Plasma glucose was measured by the glucose oxidase reaction, and plasma insulin was measured by a solid-phase radioimmunoassay.

IGT and diabetes were classified at baseline and follow-up by American Diabetes Association criteria (12). Thus, IGT was classified as an FPG <7.0 mmol/l and a 2-h plasma glucose between 7.8 and 11.1 mmol/l. Diabetes was classified as an FPG ≥7.0 mmol/l or a 2-h glucose ≥11.1 mmol/l. Subjects who gave a history of diabetes and who at the time of their clinical examination were taking either insulin or oral antidiabetic agents were also considered to have diabetes regardless of their plasma glucose values. Diabetic subjects who were not taking insulin were considered to have type 2 diabetes; insulin-taking diabetic subjects whose age of onset was ≥40 years or whose BMI was >30 kg/m2 were also considered to have type 2 diabetes. The remaining insulin-taking diabetic subjects (n = 78) were considered to have type 1 diabetes or to be unclassifiable and were excluded from the analyses. Diabetes in at least one parent or sibling was classified as a positive family history of diabetes.

Subjects who developed diabetes from NGT or IGT at the first follow-up were denoted as NGT-D[1,2] and IGT-D[1,2], respectively. Subjects who progressed to diabetes from NGT or IGT between examinations 2 and 3 were denoted as NGT-D[2,3] and IGT-D[2,3], respectively. Subjects who tested normal on the OGTT on all three examinations were considered to be bona fide nonconverters over the time period of observation and therefore served as the control group.

Statistical analysis.

Group comparisons were performed with the use of ANOVA; post hoc comparisons were carried out by the Bonferroni-Dunn test. Proportions were compared by the χ2 test, and linear regression analysis was carried out by standard methods.


Between examinations 1 and 2, 49 NGT and 43 IGT subjects developed diabetes; similar numbers of subjects progressed to diabetes between examinations 2 and 3 (41 from NGT and 30 from IGT). By adding those few NGT (n = 8) and IGT (n = 2) subjects who converted between examinations 1 and 3 (i.e., those who missed examination 2), a total of 98 subjects converted from NGT and 75 from IGT over a period of 7 years, yielding incidence rates of 1.01 person-years (95% CI, 0.90–1.12) for the NGT population and 4.62 person-years (4.07–5.18) for the IGT population. However, 911 subjects remained NGT on all three occasions. Subjects who converted to diabetes from either NGT or IGT had similar sex and age distribution as nonconverters but were heavier and hyperinsulinemic (Table 2).

Before conversion, i.e., at the NGT stage, both fasting and 2-h plasma glucose concentrations were significantly higher in both groups of converters than in the nonconverters (Table 2). In the latter, FPG increased slightly and in an apparently linear manner over the 7 years of follow-up (by 0.23 ± 0.79 mmol/l; P < 0.0001). In contrast, conversion to diabetes among NGT subjects was marked by a large step-up in FPG (Fig. 1) regardless of the time of conversion (3.06 ± 2.57 and 2.94 ± 3.11 mmol/l, respectively, at examinations 2 and 3). FPG in IGT subjects was, at the IGT stage, higher than in nonconverters (Table 2). Having converted to diabetes, whether at examination 2 or 3, these individuals exhibited an increase in FPG (3.14 ± 3.83 and 3.12 ± 3.61 mmol/l, respectively) that was of similar magnitude as that of converters from NGT (Fig. 1). Two-hour plasma glucose concentrations showed a similar pattern (Fig. 1): a slight increase in nonconverters and a large increment in all converters groups, averaging 6 mmol/l in the 3.25 years that elapsed between the last nondiabetic examination and the examination at which conversion to diabetes was documented. Of note was that, in subjects who converted from NGT to diabetes between examinations 2 and 3, the changes in fasting and 2-h plasma glucose concentrations between examinations 1 and 2 (i.e., over the preceding 3.25 years) were not significantly different (NS for both) from the corresponding changes in nonconverters (Table 3). In contrast, in subjects who progressed from IGT to diabetes between examinations 2 and 3, the changes in fasting (0.89 ± 0.89 mmol/l) and 2-h glucose (2.17 ± 1.67 mmol/l) were larger than in nonconverters (P < 0.0001 for both). In the 21 individuals who progressed from NGT to IGT to diabetes at examinations 1, 2, and 3, respectively, FPG rose from 4.72 ± 0.67 to 5.72 ± 0.67 to 9.00 ± 3.89 mmol/l (the corresponding 2-h plasma glucose values were 6.39 ± 1.06, 9.39 ± 0.78, and 16.50 ± 3.67 mmol/l), again showing that the switch to diabetes was characterized by a steep rise in plasma glucose concentrations.

To provide a quantitative estimate of the rapidity of the glucose rise upon developing diabetes, we calculated the percentage of diabetes converters whose increase in FPG or postglucose plasma glucose levels was above the 90th percentile of the corresponding increments in comparison groups. As reported in Table 4, among subjects who converted at examination 2 (groups 2 and 4 in Table 3), more than three-fourths had a rise in fasting or postglucose plasma glucose concentration greater than the 90th percentile (1.33 mmol/l) of the corresponding rise in a control group consisting of groups 1, 3, and 5–14 in Table 3. Similar figures were obtained for subjects who converted at examination 3 (groups 3, 5, 6, and 7) compared with the appropriate comparison group (nonconverters plus groups 8–14). To generate a minimal estimate of the rapidity of glucose progression, we also compared diabetes converters (from NGT plus IGT; n = 92) with subjects who converted to IGT from NGT (groups 7, 8, and 14; n = 101) or were IGT at both examinations 1 and 2 (groups 5, 10, and 12; n = 42). In this comparison, 61% of diabetes converters had FPG increments above the 90th percentile (1.90 mmol/l) of the corresponding increments in the comparison group and 97% had 2-h glucose changes above the 90th percentile of the corresponding increments of the comparison group (4.67 mmol/l).

All subgroups of converters had higher baseline BMI values than nonconverters (Table 2); thus, 25% of subjects with a BMI above the population median (27.7 kg/m2) converted to diabetes versus 8% of subjects with a BMI below the median (P < 0.0001). However, there were no consistent changes in BMI at the time of conversion in any of the converter subgroups (Fig. 2). Increments in FPG were not significantly different in converters from NGT or IGT according to whether they had a BMI below or above the population median. In a similar manner, converters as a group showed fasting hyperinsulinemia in comparison with nonconverters (Table 2); 25% of hyperinsulinemic subjects (i.e., with fasting values above the population median of 75 pmol/l) progressed to diabetes versus 9% of normoinsulinemic subjects (P < 0.0001). Again, no consistent pattern of changes in fasting insulin concentrations was evident at the time of conversion to diabetes (Fig. 2). Finally, among converters, changes in fasting or 2-h plasma glucose concentration did not differ significantly between those with and those without a positive family history of diabetes.

In contrast to fasting insulin, changes in 2-h plasma insulin levels between time of conversion and preceding measurement were significantly (P < 0.0001) related to the corresponding changes in FPG in an inverse manner, with no difference in slope between subjects who progressed to diabetes from NGT or IGT (Fig. 3). The percentage of explained variance in this association was 40%. A similar, if weaker, relationship existed between the changes in 2-h plasma insulin and the corresponding changes in 2-h plasma glucose levels.


Mexicans who live in low-income suburbs of Mexico City have a relatively high prevalence of diabetes. The present data collected after 7 years of follow-up confirm the previously reported results for the 3.25-year examination (11): an incidence of diabetes that exceeds that of non-Hispanic whites who live in the United States (1) or Europeans (13) by a factor of 4–8. Moreover, the current results confirm that IGT, in this population, like in others (7,8), amplifies the risk of diabetes approximately sixfold. The main finding is that, in NGT and IGT subjects alike, conversion to diabetes is marked by a step increase in fasting and 2-h plasma glucose levels from previously relatively stable euglycemic levels. This pattern largely prevailed over that of a gradual, progressive rise in glycemia. In fact, more than three-fourths of subjects who converted to diabetes had an FPG increment exceeding the 90th percentile of the increment observed over the same time period in the nonconverters, a group with relatively stable NGT in whom the time-related increase in glycemia averaged 0.06 mmol/l per year. Thus, not only did converters generally have a higher baseline FPG than nonconverters, particularly if they started with IGT, but in most of them, the rise in FPG was 10-fold greater than in nonconverters over the 3.5 years preceding diagnosis. Indeed, such a rise may have been underestimated, as 25% of individuals with incident diabetes were already on antidiabetic treatment when they were examined. Conspicuously, this rapid onset of hyperglycemia was equally evident in 1) NGT and IGT subjects, 2) at the first and the second follow-up examinations, 3) in subjects with a BMI below or above the population median, and 4) in subjects with or without familial diabetes. In other words, rapid deterioration of glucose tolerance seemed to be an intrinsic feature of the natural history of diabetes in this population. Within the 3.5-year time frame, diabetes was not the gradual upward drift of glycemia through the normal range but rather a steep step into the diabetic range. Clearly, those individuals who progressed from NGT over 3 years may have gone through a brief phase of IGT. In this case, their FPG would have risen faster than in nonconverters, as was the case in the IGT converters group and in those who progressed through an IGT phase (Table 3). In any case, actual conversion to manifest diabetes meant rapidly emerging hyperglycemia. It should be stressed that more frequent scans of glucose tolerance would be necessary to characterize more precisely the exact time interval of rapid decompensation. In the context of the present evidence, rapid onset refers to the observed large increment in plasma glucose levels over 3.25 years or shorter time periods.

With regard to the potential mechanisms or correlates of the rapid onset of diabetes, the present data offer some clue. Thus, conversion was more frequent among obese and/or hyperinsulinemic subjects, as expected (4). However, neither weight changes nor changes in fasting plasma insulin concentration could be associated with conversion, not even when considering the entire 7-year time span (for those who converted at examination 3). In contrast, conversion was accompanied by a fall in 2-h plasma insulin concentrations. In quantitative terms (Fig. 3), an average increment in FPG of 3.3 mmol/l was associated with an average drop in 2-h insulin of 210 pmol/l. It should be noted, however, that the correlation between the two variables explained 42% of their variability. Thus, a rapid decline in β-cell competence was a likely correlate of the loss of glucose tolerance, but other, unknown factors may also have played a part. A single measurement of circulating insulin concentrations 2 h after a glucose load is a crude index of β-cell function. The latter is highly specialized and complex, with tonic components (baseline insulin release) and phasic components (glucose sensitivity, rate sensitivity, and potentiation [14]). Therefore, it is possible that a more complete description of β-cell function would be able to account better for the observed deterioration of glucose tolerance. It is relevant to consider that if converters had 2-h plasma insulin concentrations similar to those of nonconverters, then their predicted (from Fig. 3) fasting plasma glucose would be ∼5.6 mmol/l higher. This suggests that converters were severely insulin-resistant, in accordance with the prevailing view of the pathogenesis of type 2 diabetes. However, the present results suggest a more general conclusion. At the population level, type 2 diabetes is predicted by multiple traits, among which are obesity (4), visceral fat accumulation, insulin resistance (6), and hyperinsulinemia itself (5). This multifactorial genesis is fully compatible with the tight glucose homeostasis, which relies on multiple controls (to name only the main ones, insulin sensitivity of liver, fat, and skeletal muscle tissues as well as insulin and glucagon release). In the individual who is destined to become diabetic, the factors that control glucose tolerance all must be more or less altered, generating a critical state of instability. In such a condition, phase transition can be triggered by relatively small further changes and occur relatively rapidly. In the case of glucose tolerance, transition from obese, insulin-resistant, hyperinsulinemic NGT to overt diabetes may take the form of a large, rapid rise in glucose levels as a result of a relatively small further loss of β-cell competence. This instability paradigm has been advocated for type 1 diabetes (15) and immune responses to viral infections (16). Further serial prospective studies are needed to identify the triggering event(s) in the transition from NGT to type 2 diabetes with a view to prevent or reverse hyperglycemia.

FIG. 1.

Mean fasting (top) and 2-h postglucose (bottom) plasma glucose concentrations at each of three examinations in groups that converted to diabetes and in subjects with stable NGT. D, diabetes.

FIG. 2.

BMI (mean ± SD) and fasting plasma insulin concentrations (median and interquartile range) at each of the three examinations in subjects who converted to diabetes from NGT or IGT at examination 2 or 3 (NGT-D[1,2], IGT-D[1,2], NGT-D[2,3], and IGT-D[2,3], respectively). NNN, subjects who were NGT at all three examinations (nonconverters).

FIG. 3.

Reciprocal relationship between changes in FPG between the last nondiabetic examination and the examination at which conversion occurred and the corresponding changes in 2-h plasma insulin concentrations. The y-axis is a linear scale; the x-axis is the difference between log-transformed plasma insulin values. The open dots include the 95% CI of the best (logarithmic) fit of the plotted data.


The Mexico City Diabetes Study*


Baseline clinical phenotype of converters to diabetes


Changes in fasting and postglucose plasma glucose concentrations in subjects at examinations 2 and 3


Magnitude of plasma glucose changes in subjects who converted to diabetes


This work was supported by grants from the National Heart, Lung and Blood Institute (HL-24 799 and R37-HL-36820) and from the Italian Ministry of University and Scientific and Technological Research (MURST 40%).


    • Accepted September 29, 2003.
    • Received July 1, 2003.


| Table of Contents