Glucose Trajectory: More than Changing Glucose Tolerance with Age?

Abstract

While glucose tolerance is widely known to deteriorate with age, there are individuals whose borderline elevated glucose does not presage development of diabetes, but there are people who do develop overt diabetes. In addition, elevated glucose may also presage other morbidities, particularly for those who show progressive deterioration in glucose control over time. This concept of the glucose trajectory has taken on recent significance with sophisticated mathematical modeling that can identify several different arcs, primarily based on longitudinal changes in fasting plasma glucose. Other trajectories, calculated on changes in glycated hemoglobin, or integrated responses to oral glucose tolerance tests, are less well characterized. The author has reviewed the literature in an attempt to clarify these different themes of age-related deterioration in glucose control, highlight conflicting definitions of glucose trajectory, and potentially identify avenues of further investigation. Genetic contributions to the risk of development of type 2 diabetes, artificial intelligence and mathematical models of diabetes risk, and the discrepancy between fasting glucose and postprandial measures, including glycated hemoglobin, in risk prediction are also considered.

Introduction

Deterioration in glucose tolerance with age is well described.¹ Recognition of the gradual development of type 2 diabetes, coupled with changes in diagnostic criteria over time, have led to conditions such as prediabetes,^2,3 borderline diabetes, and chemical diabetes.⁴ Since prediabetes has been associated with micro- and macrovascular disorders more commonly linked to overt diabetes,⁵ attention has generally focused on interrupting the progression of prediabetes to overt diabetes. Attempts to block this progression have led to a variety of treatment strategies, including sulfonylurea therapy,⁶ lifestyle modifications, and other medications.^5,7,8

Surprisingly, there are scant data on the economic impact of diabetes prevention despite the intense interest.⁸ One study from Ireland, using a Markov simulation model, estimated a 0.5%–4.6% reduction in cases of diabetes in response to a theoretical prevention program.⁹ A similar study from Australia, using a different simulation technique, estimated an additional 753 person-years of productive labor for the male cohort and 890 person-years for the female cohort, with individual males estimated to earn an additional 44,600 Australian dollars per year, and females 31,000, had they been able to work longer by preventing diabetes.¹⁰

The progressive decline in glucose tolerance with age appears to be associated with deterioration in insulin sensitivity as well as reduced β-cell function; there may also be a possible connection to the genes coding for glucokinase (GCK).¹¹

Some investigators have tried to quantify the gradual decline in glucose control with the term “glucose trajectory.” A growing awareness of this concept is reflected in a superficial review by the author of PubMed: 486 search results for the keywords “blood glucose trajectory” between 1981 and January 3, 2022. The citation rate is increasing rapidly; for example, 15 citations in 2009, 28 in 2013, 64 in 2019, 59 in 2020, and 75 in 2021 (as of January 2022).

Merriam-Webster defines trajectory as “(T)he curve that a body (such as a planet or comet in its orbit or a rocket) describes in space. 2: a path, progression, or line of development resembling a physical trajectory.”¹² Thus, the trajectory of any metabolite, such as glucose, can be considered from that perspective.

It should be pointed out at the outset that there is a wide spectrum of what investigators consider a glucose trajectory. Some investigators refer to changes over years in long-term studies, while others refer to a pattern over a shorter term, such as a glucose response curve during an oral glucose tolerance test (OGTT). Even over the longer term, some authors have addressed changes in fasting glucose (FG), others in random glucose, and yet others in hemoglobin A1c (HbA1c). We will come back to this inconsistency shortly.

It is challenging to make sense of a field where the definition of the major outcome variable (diabetes) has evolved over time and where measures of the glucose trajectory are similarly diverse. For consistency, we will use the current diagnostic criteria defined by the American Diabetes Association³ for diabetes: HbA1c ≥6.5%, FG ≥126 mg/dL, or 2-hr postglucose (during an OGTT) ≥200 mg/dL; and prediabetes: HbA1c 5.7%–6.4%, FG 100–125 mg/dL, and 2-hr postprandial glucose 140–199 mg/dL.

The author accessed PubMed, using “blood glucose trajectory” as a search term and included only studies evaluating longitudinal glucose control in ambulatory patients, thus excluding inpatient glucose trajectory studies and other studies combining glucose and nonglucose trajectories (such as blood pressure or body mass index), as well as nonhuman studies. Furthermore, he limited his search to articles published since 2019. Other pertinent articles from the author's personal collection were also included, as well as relevant lay publications.

This article is not intended to be an exhaustive review, but rather a synopsis of the state of the art, clarifying the language currently in use, and a guide to future investigation.

Progression to Type 2 Diabetes Mellitus

The major interest in the glucose trajectory is in identifying those individuals whose glucose tolerance deteriorates to the point where overt diabetes develops. Not every otherwise healthy older adult with prediabetes manifests an inexorable decline into frank type 2 diabetes; in particular, older individuals (over age 66) with prediabetes appear to be protected from developing overt diabetes.^13

–16 This suggests that it may be possible to predict who will develop diabetes based on their prior glucose control.

It may also be possible, using glucose control as a marker for overall health, to predict who will develop other serious illness based on changes in blood sugar. For example, Zheng et al.¹⁷ and Anstey et al.¹⁸ have linked cognitive decline to changes in glucose tolerance.

Yashin and colleagues have reported that exceptional survivors (defined as “…individuals whose genetic construction adequately met the challenges of life resulting in life spans substantially exceeding those of less lucky members of the birth cohort. Exact definition of exceptional survivors differs from study to study and includes individuals survived to age 90 years, 100 years, or those who belong to the 10% percentile of longest lived individuals, etc.”) have flatter blood glucose trajectories based on casual fasting and random glucose values.¹⁹

It is debatable whether this linkage of stable glucose and longevity and good health reflects causation or concurrence. Because of uncertainty in the prediction of prediabetes to overt diabetes, current recommendations from the Endocrine Society call for a 2-hr glucose post-OGTT measurement in patients aged 65 years and older, without known diabetes, and who meet the criteria for prediabetes by fasting plasma glucose (FPG) (100–125 mg/dL) or HbA1c (5.7%–6.4%).²⁰

Pathophysiology of Type 2 Diabetes

One major unresolved issue is that while type 2 diabetes appears to be a single diagnosis based on elevated blood sugar, it results from at least two different pathophysiologic perturbations: progressive insulin resistance (itself complex and reflecting variable insulin action in the liver, fat, and muscle) and β-cell failure. In a recent review, Ha and Sherman²¹ recognized that type 2 diabetes is a chronic progressive disease.

Using data from the Baltimore Longitudinal Study of Aging and developing a longitudinal mathematical model based on Bergman's minimal model of insulin sensitivity,²² the authors distinguish between two extreme alternatives: impaired FG as a primary defect or impaired postprandial glucose tolerance as the primary lesion. They recognize that development of type 2 diabetes usually requires both components. They also address impaired insulin secretion and clearance, and noninsulin-mediated glucose uptake, as well as the role of impaired incretin signaling in development of diabetes. These authors point out that individuals with impaired FG have severe hepatic insulin resistance, but less impaired muscle insulin sensitivity, while those with impaired glucose tolerance have marked muscle insulin resistance and milder hepatic insulin resistance.²¹

Abdul-Ghani et al. concur with this distinction, but point out that both groups have reduced early-phase insulin secretion, and that the impaired glucose tolerance group has reduced late-phase insulin secretion as well.²³ In this context, it is important to point out that Reaven and coworkers have demonstrated insulin resistance in lean nondiabetic relatives of patients with type 2 diabetes, both at the level of muscle and liver, suggesting that insulin resistance may predate β-cell failure.^24
–26

Further complicating the picture is evidence that the degree of insulin secretion may also reflect inheritance.²⁷ It is clear that since diabetes may reflect independent pathways, trying to identify a single predictive factor will be difficult.

Role of Genetic Variants in the Development of Type 2 Diabetes

Recent scientific progress has allowed for expanded understanding of the role of genetic variants in the development of type 2 diabetes. While a detailed discussion is beyond the scope of this article, a brief review is warranted.

A recent meta-analysis based on 21 genome-wide association studies in cohorts of European descent confirmed an association of FG with genetic variants in or near the genes that encode GCK, the glucose-6-phosphatase catalytic subunit, and the melatonin receptor 1b. These genes influence signal transduction, cell proliferation, development, glucose sensing, and circadian regulation affecting type 2 diabetes risk and more modest elevations in glucose.²⁸ On the other hand, Florez and colleagues evaluated FG-associated single-nucleotide polymorphisms in the multiethnic cohort of the Diabetes Prevention Program and observed no detectable impact on diabetes incidence.²⁹

The validity of HbA1c as a diagnostic criterion for diabetes can be compromised by variation in glucose-6-phosphate dehydrogenase. Deficiency of this enzyme can lead to hemolysis in certain infections or in response to some medications (fluoroquinolones, sulfonylureas, and others) or other challenges (fava beans and henna exposure) and is widespread in malaria-endemic countries. There are variants associated with lower HbA1c in African Americans and Hispanic individuals in the United States^30
–32; similar variants are reported in Asians.³³ As a result, HbA1c may be disproportionally low and, if used as the sole diagnostic criterion for diabetes, may lead to an underdiagnosis of type 2 diabetes.

In a similar vein, a preliminary analysis found medication–gene interactions for glycemic outcome, for metformin, sulfonylureas, repaglinide, thiazolidinediones, and acarbose; data were insufficient for meta-analysis, but future understanding may lead to more personalized medication treatments.³⁴

Can Artificial Intelligence Help Predict the Development of Diabetes

Computer modeling has also been employed in an attempt to predict who will develop overt diabetes. Some approaches have used existing population datasets, while others have used new cohorts. For example, in studies based on the National Health and Nutrition Examination Survey (NHANES) dataset, support vector machine modeling³⁵ and other models were evaluated individually and then combined to identify at-risk patients.³⁶ Similar machine learning methods were applied using the well-established database of Pima Indians.^37
–39

A comparable attempt, using the bagged classification and regression tree (Bagged CART) approach, based on women in a Bangladeshi survey has been reported.⁴⁰ Choubey et al. reported that principal component analysis achieved maximum performance examining data of Pima Indians and a local dataset from India.⁴¹ Another study from India, based on hospital and community data, reported that a machine learning algorithm, termed “mixture of expert,” provided precise predictions.⁴² In a population-based cross-sectional study from China, Wang et al. reported that an artificial neural network could identify subjects at risk for diabetes, using demographic, lifestyle, and anthropometric, but not biochemical, data.⁴³

Silva et al., in a systematic review and meta-analysis, reported that while machine learning models predicted type 2 diabetes in the community with some success, improvements in methodology, reporting, and validation were needed before more general application was advisable.⁴⁴

In addition to the use of machine learning in predicting diabetes, there have been applications to diabetes classification that may have benefit for clinicians dealing with individual patients. Using continuous glucose monitoring (CGM) data from an endocrinology clinic at the People's Hospital of Henan Province, Wang and colleagues reported successful classification of type 1 and type 2 diabetes using an ensemble learning algorithm.⁴⁵ A similar approach to diabetes classification based on an individual patient's glucose profile from CGM and self-monitoring blood glucose (fingerstick) data was reported by Tolks et al.⁴⁶

Do Methods for Diagnosing Type 2 Diabetes Reflect Glucose Trajectories?

While overt hyperglycemia is clearly the major diagnostic criterion, recent revisions in threshold diagnostic criteria in nonpregnant adults have used HbA1c, FG, or postglucose challenge glucose levels as sufficient diagnostic criteria. These revisions have themselves been the result of a greater understanding of threshold glucose levels and subsequent development of diabetes complications such as retinopathy. Buysschaert and colleagues in 2016 have reviewed these diagnostic methods.⁴⁷

Hulman and colleagues have addressed the heterogeneity of responses to a standard OGTT and identified several distinct patterns, with the highest glucose trajectory during the test associated with the highest estimated cardiovascular risk.⁴⁸ Wagner and coworkers, using magnetic resonance imaging body composition data as well as glucose patterns, have identified pathophysiology-based subphenotype clusters with different predictive values for subsequent disease.¹⁵

How Is the Glucose Trajectory Calculated and Are There Characteristic Waveforms?

We have previously seen that the glucose trajectory covers a wide spectrum of glucose measures. In general, the glucose trajectory has been reported from three independent parameters of glucose tolerance: fasting blood glucose (FBG); glycosylated hemoglobin measures (usually HbA1c); and integrated glucose responses to oral glucose tolerance testing. In addition to reflecting a variety of glucose parameters, the glucose trajectory is usually expressed as a function of time (i.e., change in FG over years).

This variable has also been inconsistent from study to study and has ranged from 2- to 5-year intervals in terms of estimating long-term outcome. Thus, the language describing the trajectory is qualitative and usually refers to changes over time compared with some arbitrary baseline, for example, low, elevated, or moderate-stable trajectory, increasing or decreasing, or some similar language.

While a detailed discussion of statistical methods is beyond the scope of this review, it is valuable to consider how a trajectory could be estimated. Individuals can be evaluated by unobserved, or latent, trajectories by modeling techniques such as latent growth mixture modeling and latent class growth modeling; these differ by how variance within classes is considered.⁴⁹ Latent mixture modeling estimates heterogeneity by classifying individuals into unobserved groupings (latent classes) with similar (more homogeneous) patterns, as has been described by Jones et al., using the freely available Proc Traj approach.^50,51

Jin and colleagues, using this latent mixture modeling strategy, identified five discrete FBG trajectories according to the FBG range and changing patterns over time: elevated-stable, elevated-decreasing, moderate-increasing, moderate-stable, and low-stable trajectories.²⁶ These investigators reported a higher risk of myocardial infarction for patients who showed elevated-stable patterns and lower risk for those who showed an elevated-decreasing pattern, when compared with moderate-stable individuals, and when data were adjusted for potentially confounding variables, such as age, sex, smoking, alcohol and salt intake, and physical activity, among other factors.⁵²

Lee and colleagues in Taiwan explored the trajectories of both mean FPG and FPG variability (calculated as the coefficient of variation of FPG) in a longitudinal study also using a Proc Traj approach. They determined five trajectories of FPG variability (low, increasing, fluctuating, decreasing, and high) and means (well controlled, stable control, worsening control, improving control, and poor control). While the five trajectories of mean FPG shared the same mortality risk, when they were compared with the low FPG variability trajectory, the other four variability trajectories had significantly higher risk of mortality over the time frame of the study.⁵³

In contrast to studies based on FPG, Hulman and coworkers used the latent class trajectory analysis (a statistical method for identifying unmeasured class membership among subjects using categorical and/or continuous observed variables) derived from glucose response patterns during an OGTT; transitions between classes were analyzed with multinomial logistic regression models, and the concept of “trajectory” refers to the shape of the glucose response curve during the OGTT, not over time.⁵⁴

While short-term observations suggested that individuals tended to remain in their initial class trajectory, long-term data predicting morbidity or mortality are still pending. In a 2017 study, Hulman and colleagues observed several discrete glucose responses during an OGTT⁴⁸; a similar analysis of OGTT glucose curves by Abdul-Ghani et al. also described several discrete groups, and they opined that the more quickly the plasma glucose concentrations returned to fasting levels, the lower the future risk of diabetes.⁵⁵

Mason et al. reported two distinct and sequential patterns of 2-hr postchallenge glucose values in a longitudinal study of the Pima Indians: an initial gradual increase, followed by a sharper rise that typically occurred within 4.5 years before the onset of frank diabetes.⁵⁶ Qualitatively similar findings have been described in other populations.^57,58

Based on these and other studies, Dankner and Roth have recommended personalized normal ranges of glucose and HbA1c to better predict the development of diabetes in an individual.⁵⁹ This approach may help indicate transitions from normal glucose tolerance, through prediabetes, to overt diabetes,⁶⁰ even when an individual may still be within the normal ranges for glucose and HbA1c.^61,62

We have previously seen that overt diabetes requires deterioration in both FG and glucose tolerance. Let us take a closer look at the independence of the two. While, as stated before, overall glycemic control is linked to cognitive function,^17,18 the effect of either FG or glucose tolerance on predictions of neurocognitive decline is inconsistent.⁶³

As further illustration of the discordance between fasting and postprandial glucose levels, consider that in Cushing's syndrome, while FPG was lower in patients than in controls, postprandial glucose correlates with HbA1c.⁶⁴ On the other hand, in women with gestational diabetes, elevated FPG was a stronger predictor of adverse pregnancy outcomes such as gestational hypertension and infants large for gestational age than elevated postload glucose.⁶⁵

Another manifestation of the inconsistency between FBG and HbA1c, or overall glycemic control, is contained in the recent Progression of Early Subclinical Atherosclerosis Study, which reported a higher rate of subclinical atherosclerosis, documented by two-dimensional ultrasound and a coronary artery calcium score, in individuals with HbA1c in the subprediabetes band, between 5.0% and 5.5%.⁶⁶

In a study on the Relationship between Insulin Sensitivity and Cardiovascular Risk (RISC) cohort, involving over 1000 nondiabetic subjects, Mengozzi and colleagues reported that fasting and postload glucose homeostasis were independent and that different β-cell defects contributed to the disruption of each parameter.⁶⁷

In yet another variation on the term, trajectory, Ke and colleagues addressed the notion of glycemic trajectory referring to lifetime glucose exposure by evaluating the management of young-onset patients with type 2 diabetes (age of onset ∼34 years). These patients tended to have higher HbA1c throughout their life span than those with usual-onset diabetes (age of onset ∼57 years), leading both to worsened glycemic exposure and rapid glycemic deterioration; the suggestion is that these younger individuals warrant more personalized therapy.⁶⁸

Expansion of the Notion of Glucose Trajectory to Conditions Other than Diabetes Mellitus

Here too we see that ambiguity in the term trajectory can obscure the notion of trajectory being a time-dependent variable. For example, Ogata and colleagues, in a longitudinal analysis of changes in FPG, reported that there was an increase in cardiovascular disease (coronary heart disease and stroke) incidence in men and women where the FPG trajectory increased sharply.⁶⁹

Using data from the Coronary Artery Risk Development in Young Adults (CARDIA) Study, Bancks and colleagues characterized 25-year trajectories in FG and insulin resistance [derived by the homeostasis model assessment–estimated insulin resistance (HOMA-IR) derived from FG insulin values] and compared these trajectories with changes in cardiac structure and function determined using echocardiogram reports. Five FG trajectory groups were identified (low-stable, moderate-stable, moderate-increase, high-stable, and high-increase groups) along with three HOMA-IR trajectory groups (low-decreasing, moderate-stable, and high-increasing groups).

Both the moderate-increasing FG trajectory and the high-increasing IR trajectory were associated with adverse echocardiographic findings. It should be pointed out that this study was based on young adults (18–30 years of age at onset) and individuals with diabetes were excluded from the analysis, including those who developed diabetes over the course of the study.⁷⁰

Feng and coworkers applied the Proc Traj method to a longitudinal study (2006–2010) of an established Chinese cohort to explore the impact of changes in glucose tolerance associated with recent lifestyle changes in China. They identified five trajectories based on FBG: low-increasing, moderate-stable, moderate-increasing, elevated-decreasing, and elevated-stable trajectories.⁷¹ Despite a similar statistical approach, these trajectories are somewhat different from those described by Jin and colleagues.⁵²

These investigators observed that there was an increased risk for overall cancer in the low-increasing group and for gastrointestinal cancer in the elevated-stable group; BMI may modify these associations.⁷¹ While provocative, these studies need to be repeated in other populations and for longer periods; there also needs to be more consistency in the trajectory labels.

Table 1 summarizes the different criteria used to date to define glucose trajectory and the reported outcome data from each approach.

Table 1.

Different Criteria for the Glucose Trajectory and Reported Outcome

Criterion	Number of participants at baseline	Country of origin	Baseline age	Follow-up interval	Outcome	Reference
Casual glucose over time	5128 (original study); 3514 in offspring study	United States	30	Q 5 years, over 40 years.	Survival	¹⁹
FBG over time	3877	China	48.6–52.8 in different groups	Q 2 years, over 4 years.	Risk of myocardial infarction	⁵²
Mean FPG	3569	Taiwan	65.8–71.2 in different groups	Q 3 months, over 2 years.	Mortality	⁵³
Variability in mean FPG	3569	Taiwan	65.8–71.2 in different groups	Q 3 months, over 2 years.	Mortality	⁵³
Glucose during OGTT	1267	Denmark, the Netherlands, and United States	46–53	2-hr OGTT	Highest glucose correlates with increased calculated cardiovascular risk	⁴⁸
Glucose during OGTT	1443	Denmark, France, Germany, Ireland, and Italy	41–47	2-hr OGTT	There are discrete patterns during testing; long-term implications are unclear	⁵⁴
Return to FG level during OGTT	2445	Botnia	45–52 for different groups	OGTT at baseline, 7–8 years. f/u	Risk of development of diabetes	⁵⁵
Change in blood glucose 2-hr postglucose challenge	55	United States	16.1 at entry	Biennial examination, threshold at 4 years, prediagnosis.	Diabetes	⁵⁶
Lifetime glucose exposure (mean HbA1c)	328,199	Hong Kong	34.6 (young-onset diabetes), 59.3 (usual-onset diabetes)	Decades after diagnosis	Worsened diabetes control over time	⁶⁸
Fasting blood glucose	69,742	China	49.14–52.58 for different groups	20 years. f/u	Cancer risk	⁷¹
FG, HOMA-IR	5115	United States	24.7–25.3 for different groups	25 years. f/u	Echocardiographic deterioration	⁷⁰

FBG, fasting blood glucose; FG, fasting glucose; FPG, fasting plasma glucose; HbA1c, hemoglobin A1c; HOMA-IR, homeostasis model assessment–estimated insulin resistance; OGTT, oral glucose tolerance test.

In a different interpretation of the term “trajectory,” Hu and collaborators evaluated the association between preoperative FG and digestive tract cancer mortality in Chinese patients.⁷² A limited analysis identified a U-shaped risk trajectory; clearly, more work is needed.

Conclusions

Deterioration in glucose tolerance with age is not a new concept. With the development of overt diabetes, patients are subject to the well-known array of diabetic complications, impacting morbidity and mortality. However, attempts to prevent the progressive decline in glucose tolerance are hampered by uncertainty over which patients warrant aggressive attention.

Attempts to identify at-risk individuals by tracking the glucose trajectory are promising, but limited by inconsistent identification of the glucose assessment that needs attention: fasting or postprandial glucose or HbA1c. Discrepant impacts on the outcome of fasting versus postprandial measures of glycemic control, including glycated hemoglobin, are not understood. In addition, there is inconsistent labeling of glucose trajectories.

Early data show the potential value of trajectories of insulin resistance, but these studies, even those using simple measures such as HOMA-IR, require additional laboratory resources and costs. Mathematical modeling and the role of artificial intelligence are still nascent approaches.

Rapidly expanding genetic understanding should also provide clarification on the long-term impact of glycemic control and trajectory. Hopefully, future investigation will clarify where clinicians should focus their limited resources.

Footnotes

Acknowledgments

This material is the result of work supported with resources and the use of facilities at the VA Northern California Health Care System. The contents do not represent the views of the U.S. Department of Veterans Affairs or the U.S. Government. The author acknowledges the expert assistance of Ms. Alba Scott, MILS, without whose efforts this article would not have been possible.

Author Disclosure Statement

No conflicting financial interests exist.

Funding Information

No funding was received for this article.

References

DeFronzo

. Glucose intolerance and aging. Diabetes Care, 1981; 4:493–501.

Diabetes Canada Clinical Practice Guidelines Expert

Committee

, Punthakee

, Goldenberg

, et al. Definition, Classification and Diagnosis of Diabetes, Prediabetes and Metabolic Syndrome. Can J Diabetes, 2018; 42 Suppl 1:S10–S15.

American Diabetes Association. (2) Classification and diagnosis of diabetes. Diabetes Care, 2015; 38 Suppl:S8–S16.

Fajans

. I. Identification of chemical diabetes. The definition of chemical diabetes. Metabolism, 1973; 22:211–217.

Brannick

, Dagogo-Jack

. Prediabetes and cardiovascular disease: Pathophysiology and interventions for prevention and risk reduction. Endocrinol Metab Clin North Am, 2018; 47:33–50.

Fajans

. Treatment of chemical diabetes mellitus with sulfonylurea compounds. Metabolism, 1973; 22:373–376.

Tabák

, Herder

, Rathmann

, et al. Prediabetes: A high-risk state for diabetes development. Lancet, 2012; 379:2279–2290.

Roberts

, Barry

, Craig

, et al. Preventing type 2 diabetes: Systematic review of studies of cost-effectiveness of lifestyle programmes and metformin, with and without screening, for pre-diabetes. BMJ Open, 2017; 7:e017184.

Passey

, Shrestha

, Bertram

, et al. The impact of diabetes prevention on labour force participation and income of older Australians: An economic study. BMC Public Health, 2012; 12:16.

10.

Pierse

, O'Neill

, Dinneen

, et al. A simulation study of the economic and health impact of a diabetes prevention programme in Ireland. Diabet Med, 2021; 38:e14540.

11.

Ohn

, Kwak

, Cho

, et al. 10-year trajectory of β-cell function and insulin sensitivity in the development of type 2 diabetes: A community-based prospective cohort study. Lancet Diabetes Endocrinol, 2016; 4:27–34.

12.

Merriam-Webster.com. Definition of trajectory. https://www.merriam-webster.com/dictionary/trajectory (Last accessed April 22, 2021 ).

13.

Span

. How Meaningful Is Prediabetes for Older Adults?. New York, NY: New York Times, 2021.

14.

Rooney

, Rawlings

, Pankow

, et al. Risk of progression to diabetes among older adults with prediabetes. JAMA Intern Med, 2021; 181:511–519.

15.

Wagner

, Heni

, Tabák

, et al. Pathophysiology-based subphenotyping of individuals at elevated risk for type 2 diabetes. Nat Med, 2021; 27:49–57.

16.

Tang

, Matsushita

, Coresh

, et al. Mortality implications of prediabetes and diabetes in older adults. Diabetes Care, 2020; 43:382–388.

17.

Zheng

, Yan

, Yang

, et al. HbA_1c, diabetes and cognitive decline: The English Longitudinal Study of Ageing. Diabetologia, 2018; 61:839–848.

18.

Anstey

, Sargent-Cox

, Eramudugolla

, et al. Association of cognitive function with glucose tolerance and trajectories of glucose tolerance over 12 years in the AusDiab study. Alzheimers Res Ther, 2015; 7:48.

19.

Yashin

, Arbeev

, Akushevich

, et al. Exceptional survivors have lower age trajectories of blood glucose: Lessons from longitudinal data. Biogerontology, 2010; 11:257–265.

20.

LeRoith

, Biessels

, Braithwaite

, et al. Treatment of diabetes in older adults: An Endocrine Society Clinical Practice Guideline. J Clin Endocrinol Metab, 2019; 104:1520–1574.

21.

, Sherman

. Type 2 diabetes: One disease, many pathways. Am J Physiol Endocrinol Metab, 2020; 319:E410–E426.

22.

Bergman

, Ider

, Bowden

, et al. Quantitative estimation of insulin sensitivity. Am J Physiol, 1979; 236:E667–E677.

23.

Abdul-Ghani

, Tripathy

, DeFronzo

. Contributions of beta-cell dysfunction and insulin resistance to the pathogenesis of impaired glucose tolerance and impaired fasting glucose. Diabetes Care, 2006; 29:1130–1139.

24.

Zavaroni

, Reaven

. Insulin-resistance and associated risk factors for coronary heart disease as seen in families. Diabete Metab, 1991; 17(1 Pt 2):109–111. PMID: 1936465.

25.

Laws

, Stefanick

, Reaven

. Insulin resistance and hypertriglyceridemia in nondiabetic relatives of patients with noninsulin-dependent diabetes mellitus. J Clin Endocrinol Metab, 1989; 69:343–347.

26.

Facchini

, Chen

, Clinkingbeard

, et al. Insulin resistance, hyperinsulinemia, and dyslipidemia in nonobese individuals with a family history of hypertension. Am J Hypertens, 1992; 5:694–699.

27.

Elbein

, Hasstedt

, Wegner

, et al. Heritability of pancreatic beta-cell function among nondiabetic members of Caucasian familial type 2 diabetic kindreds. J Clin Endocrinol Metab, 1999; 84:1398–1403.

28.

Dupuis

, Langenberg

, Prokopenko

, et al. New genetic loci implicated in fasting glucose homeostasis and their impact on type 2 diabetes risk. Nat Genet, 2010; 42:105–116.

29.

Florez

, Jablonski

, McAteer

, et al.; Diabetes Prevention Program Research Group. Effects of genetic variants previously associated with fasting glucose and insulin in the Diabetes Prevention Program. PLoS One, 2012; 7:e44424.

30.

Wheeler

, Leong

, Liu

, et al. Impact of common genetic determinants of Hemoglobin A1c on type 2 diabetes risk and diagnosis in ancestrally diverse populations: A transethnic genome-wide meta-analysis. PLoS Med, 2017; 14:e1002383.

31.

Moon

, Louie

, Jain

, et al. A genome-wide association study identifies blood disorder-related variants influencing hemoglobin A_1c with implications for glycemic status in U.S. Hispanics/Latinos. Diabetes Care, 2019; 42:1784–1791.

32.

Sarnowski

, Leong

, Raffield

, et al.; TOPMed Diabetes Working Group; TOPMed Hematology Working Group; TOPMed Hemostasis Working Group; National Heart, Lung, and Blood Institute TOPMed Consortium. Impact of rare and common genetic variants on diabetes diagnosis by hemoglobin A1c in multi-ancestry cohorts: The Trans-Omics for Precision Medicine Program. Am J Hum Genet, 2019; 105:706–718.

33.

Leong

, Lim

VJY

, Wang

, et al. Association of G6PD variants with hemoglobin A1c and impact on diabetes diagnosis in East Asian individuals. BMJ Open Diabetes Res Care, 2020; 8:e001091.

34.

Maruthur

, Gribble

, Bennett

, et al. The pharmacogenetics of type 2 diabetes: A systematic review. Diabetes Care, 2014; 37:876–886.

35.

, Liu

, Valdez

, et al. Application of support vector machine modeling for prediction of common diseases: The case of diabetes and pre-diabetes. BMC Med Inform Decis Mak, 2010; 10:16.

36.

Dinh

, Miertschin

, Young

, et al. A data-driven approach to predicting diabetes and cardiovascular disease with machine learning. BMC Med Inform Decis Mak, 2019; 19:211.

37.

Nilashi

, Bin Ibrahim

, Mardani

, et al. A soft computing approach for diabetes disease classification. Health Informatics J, 2018; 24:379–393.

38.

Maniruzzaman

, Kumar

, Menhazul Abedin

, et al. Comparative approaches for classification of diabetes mellitus data: Machine learning paradigm. Comput Methods Programs Biomed, 2017; 152:23–34.

39.

Mishra

, Tripathy

, Mallick

, et al. EAGA-MLP-an enhanced and adaptive hybrid classification model for diabetes diagnosis. Sensors (Basel), 2020; 20:4036.

40.

Islam

, Rahman

, Chandra Roy

, et al. Automated detection and classification of diabetes disease based on Bangladesh demographic and health survey data, 2011 using machine learning approach. Diabetes Metab Syndr, 2020; 14:217–219.

41.

Choubey

, Kumar

, Shukla

, et al. Comparative analysis of classification methods with PCA and LDA for diabetes. Curr Diabetes Rev, 2020; 16:833–850.

42.

Shankaracharya, Odedra

, Samanta

, et al. Computational intelligence-based diagnosis tool for the detection of prediabetes and type 2 diabetes in India. Rev Diabet Stud, 2012 Spring; 9:55–62.

43.

Wang

, Li

, Wang

, et al. Evaluating the risk of type 2 diabetes mellitus using artificial neural network: An effective classification approach. Diabetes Res Clin Pract, 2013; 100:111–118.

44.

Silva

, Lee

, Forbes

, et al. Use and performance of machine learning models for type 2 diabetes prediction in community settings: A systematic review and meta-analysis. Int J Med Inform, 2020; 143:104268.

45.

Wang

, Liu

, Chen

, et al. A novel classification indicator of type 1 and type 2 diabetes in China. Sci Rep, 2017; 7:17420.

46.

Tolks

, Ament

, Eberle

. Model-driven classification of different diabetes types within a personalized diabetes management. Annu Int Conf IEEE Eng Med Biol Soc, 2018; 2018:5818–5821.

47.

Buysschaert

, Medina

, Buysschaert

, et al. Definitions (and current controversies) of diabetes and prediabetes. Curr Diabetes Rev, 2016; 12:8–13.

48.

Hulman

, Simmons

, Vistisen

, et al. Heterogeneity in glucose response curves during an oral glucose tolerance test and associated cardiometabolic risk. Endocrine, 2017; 55:427–434.

49.

van de Schoot

. Latent trajectory studies: The basics, how to interpret the results, and what to report. Eur J Psychotraumatol, 2015; 6:27514.

50.

Jones

, Nagin

, Roeder

. A SAS procedure based on mixture models for estimating developmental trajectories. Soc Methods Res, 2001; 29:374–393.

51.

Jones

, Nagin

. Advances in group-based trajectory modeling and an SAS procedure for estimating them. Soc Methods Res, 2007; 35:542–571.

52.

Jin

, Chen

, Vaidya

, et al. Longitudinal change in fasting blood glucose and myocardial infarction risk in a population without diabetes. Diabetes Care, 2017; 40:1565–1572.

53.

Lee

, Sheu

, Lee

, et al. Trajectories of fasting plasma glucose variability and mortality in type 2 diabetes. Diabetes Metab, 2018; 44:121–128.

54.

Hulman

, Witte

, Vistisen

, et al. Pathophysiological characteristics underlying different glucose response curves: A latent class trajectory analysis from the prospective EGIR-RISC study. Diabetes Care, 2018; 41:1740–1748.

55.

Abdul-Ghani

, Lyssenko

, Tuomi

, et al. The shape of plasma glucose concentration curve during OGTT predicts future risk of type 2 diabetes. Diabetes Metab Res Rev, 2010; 26:280–286.

56.

Mason

, Hanson

, Knowler

. Progression to type 2 diabetes characterized by moderate then rapid glucose increases. Diabetes, 2007; 56:2054–2061.

57.

Ferrannini

, Nannipieri

, Williams

, et al. Mode of onset of type 2 diabetes from normal or impaired glucose tolerance. Diabetes, 2004; 53:160–165.

58.

Tabák

, Jokela

, Akbaraly

, et al. Trajectories of glycaemia, insulin sensitivity, and insulin secretion before diagnosis of type 2 diabetes: An analysis from the Whitehall II study. Lancet, 2009; 373:2215–2221.

59.

Dankner

, Roth

. The personalized approach for detecting prediabetes and diabetes. Curr Diabetes Rev, 2016; 12:58–65.

60.

Dankner

, Danoff

, Roth

. Can “personalized diagnostics” promote earlier intervention for dysglycaemia? Hypothesis ready for testing. Diabetes Metab Res Rev, 2010; 26:7–9.

61.

Dankner

, Bergman

, Danoff

, et al. The metabolic deterioration that antedates diabetes: Personal trajectories of HbA(1c) and fasting glucose as early indicators and possible triggers for intervention. Diabetes Metab Res Rev, 2013; 29:1–7.

62.

Bergman

. The early diabetes intervention program—Is early actually late?. Diabetes Metab Res Rev, 2014; 30:654–658.

63.

Sims Wright

, Levy

, Katzel

, et al. Fasting glucose and glucose tolerance as potential predictors of neurocognitive function among nondiabetic older adults. J Clin Exp Neuropsychol, 2015; 37:49–60.

64.

Otsuki

, Kitamura

, Tamada

, et al. Incompatibility between fasting and postprandial plasma glucose in patients with Cushing's syndrome. Endocr J, 2016; 63:1017–1023.

65.

Ryan

, Savu

, Yeung

, et al. Elevated fasting vs post-load glucose levels and pregnancy outcomes in gestational diabetes: A population-based study. Diabet Med, 2020; 37:114–122.

66.

Rossello

, Raposeiras-Roubin

, Oliva

, et al. Glycated hemoglobin and subclinical atherosclerosis in people without diabetes. J Am Coll Cardiol, 2021; 77:2777–2791.

67.

Mengozzi

, Tricò D, Nesti L, et al.; RISC Investigators. Disruption of fasting and post-load glucose homeostasis are largely independent and sustained by distinct and early major beta-cell function defects: A cross-sectional and longitudinal analysis of the Relationship between Insulin Sensitivity and Cardiovascular risk (RISC) study cohort. Metabolism, 2020; 105:154185.

68.

, Stukel

, Shah

, et al. Age at diagnosis, glycemic trajectories, and responses to oral glucose-lowering drugs in type 2 diabetes in Hong Kong: A population-based observational study. PLoS Med, 2020; 17:e1003316.

69.

Ogata

, Watanabe

, Kokubo

, et al. Longitudinal trajectories of fasting plasma glucose and risks of cardiovascular diseases in middle age to elderly people within the general Japanese population: The Suita Study. J Am Heart Assoc, 2019; 8:e010628.

70.

Bancks

, Carnethon

, Chow

, et al. Fasting glucose and insulin resistance trajectories during young adulthood and mid-life cardiac structure and function. J Diabetes Complications, 2019; 33:356–362.

71.

Feng

, Wang

, Lyu

, et al. The association between fasting blood glucose trajectory and cancer risk in Chinese population without diabetes. Int J Cancer, 2020; 147:958–966.

72.

, Peng

, Lin

, et al. The risk trajectory between preoperative fasting glucose and common digestive tract cancer-specific mortality in the FIESTA cohort involving 6865 Chinese patients. J Cancer, 2019; 10:4596–4602.