Figures
Abstract
Aim
Subclasses of different glycaemic disturbances could explain the variation in characteristics of individuals with type 2 diabetes (T2D). We aimed to examine the association between subgroups based on their glucose curves during a five-point mixed-meal tolerance test (MMT) and metabolic traits at baseline and glycaemic deterioration in individuals with T2D.
Methods
The study included 787 individuals with newly diagnosed T2D from the Diabetes Research on Patient Stratification (IMI-DIRECT) Study. Latent class trajectory analysis (LCTA) was used to identify distinct glucose curve subgroups during a five-point MMT. Using general linear models, these subgroups were associated with metabolic traits at baseline and after 18 months of follow up, adjusted for potential confounders.
Results
At baseline, we identified three glucose curve subgroups, labelled in order of increasing glucose peak levels as subgroup 1–3. Individuals in subgroup 2 and 3 were more likely to have higher levels of HbA1c, triglycerides and BMI at baseline, compared to those in subgroup 1. At 18 months (n = 651), the beta coefficients (95% CI) for change in HbA1c (mmol/mol) increased across subgroups with 0.37 (-0.18–1.92) for subgroup 2 and 1.88 (-0.08–3.85) for subgroup 3, relative to subgroup 1. The same trend was observed for change in levels of triglycerides and fasting glucose.
Citation: Obura M, Beulens JWJ, Slieker R, Koopman ADM, Hoekstra T, Nijpels G, et al. (2020) Post-load glucose subgroups and associated metabolic traits in individuals with type 2 diabetes: An IMI-DIRECT study. PLoS ONE 15(11): e0242360. https://doi.org/10.1371/journal.pone.0242360
Editor: Domenico Tricò, University of Pisa, ITALY
Received: June 4, 2020; Accepted: October 31, 2020; Published: November 30, 2020
This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
Data Availability: This study analyzed already existing data obtained upon request and subject to license restrictions from a number of different sources and due to ethical and legal concerns i.e. it contains sensitive patient information, data cannot be made openly available. To access data one has to complete a Data access form and return it to the data access committee executive group - responsible for reviewing, approving and enabling access to data - (DIRECTdataaccess@Dundee.ac.uk).
Funding: The work leading to this publication was supported by the Innovative Medicines Initiative Joint Undertaking under grant agreement n°115317 (IMI-DIRECT ), resources of which are composed of financial contribution from the European Union's Seventh Framework Programme (FP7/2007- 2013) and EFPIA companies’ in kind contribution. The IMI-DIRECT grant was awarded to ML, IM, IF, BJ, HR, AM, MW, TJM, SB, GNG, ED, OP, JMS, JA, ERP, PWF, LMT and FR. The Amsterdam Public Health research institute provided support in the form of a PhD fellowship for MO. Eli Lilly Regional Operations GmbH, Vienna, Austria provided support in the form of a salary for IP, and Sanofi-Aventis Deutschland GmbH, R&D, Frankfurt am Main, Germany provided support in the form of salaries for BJ and HR. The specific roles of these authors are articulated in the ‘author contributions’ section. The funders played a role in data collection, but did not have any additional role in the study design, analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have read the journal’s policy and have the following competing interests: IP is employed by Eli Lilly Regional Operations GmbH, Vienna, Austria and BJ and HR are employees of Sanofi-Aventis Deutschland GmbH, R&D, Frankfurt am Main, Germany. This does not alter our adherence to PLOS ONE policies on sharing data and materials. There are no patents, products in development or marketed products associated with this research to declare.
Abbreviations: BIC, Bayesian Information Criterion; AIC, Akaike Information Criterion; IMI, Innovative Medicines Initiative; DIRECT, Diabetes Research on Patient Stratification; LCMM, Latent class mixed models; LCTA, Latent class trajectory analysis; MICE, Multivariate Imputation by Chained Equations; MMT, Mixed-meal tolerance
Introduction
Recent studies have shown significant heterogeneity in glucose response curves from oral glucose tolerance tests (OGTTs) in individuals with elevated risk of type 2 diabetes (T2D) [1–4]. Additionally, these glucose curves had distinct glycaemic risk profiles and were associated with future disease risk. Whether such distinct glycaemic profiles can also be identified in individuals with T2D has not been investigated to date. People with T2D differ in their rate of glycaemic deterioration and number and type of complications, indicating that T2D is a heterogeneous disease with different underlying pathophysiological mechanisms [5]. Therefore, studying the heterogeneity in glucose response curves and whether these glucose curves have distinct metabolic profiles may help improve our understanding of the different underlying pathophysiologies. We therefore aimed to examine the association between subgroups based on glucose curves during a five-point mixed-meal tolerance test (MMT) and metabolic traits at baseline and glycaemic deterioration in individuals with T2D.
Materials and methods
Study population
We used data from the Innovative Medicines Initiative-Diabetes Research on Patient Stratification (IMI-DIRECT) Study, which is described in detail elsewhere [6]. In brief, 789 individuals diagnosed with T2D within 6 to 24 months were recruited from six European clinical centers. Individuals were only recruited at baseline if they were using metformin therapy or on lifestyle management. If participants started using any diabetes medication or changed, this information was recorded at the next study visit. All participants signed informed consents and the IMI-DIRECT’s Data Access Committee—responsible for reviewing, approving and enabling access to data—approved this study. The study also conformed to the Declaration of Helsinki standards. We excluded those missing fasting glucose at baseline (n = 2) and those lost to follow-up (n = 119), thus 787 and 668 individuals with T2D were included in the analysis at baseline and follow-up respectively. At follow-up, MMT subgroups were identified in 651 individuals after excluding those missing fasting glucose (n = 17) (Fig 1). All individuals were white European adults, aged 35 to 75 years.
MMT assessment
MMTs were performed at baseline and at 18 months after a 10-hour overnight fast. Blood was sampled at 0, 30, 60, 90 and 120 min. Participants were asked to stop the use of metformin, when used, 24 hours before the study visit. Participants consumed a 250ml liquid drink (Fortisip; 18.4 g carbohydrate per 100 ml) over a period of 2–5 min. Plasma glucose (mmol/l) was determined using the enzymatic glucose hexokinase method (Konelab 20 XT Clinical Chemistry analyzer, Thermo Fisher Scientific, Vantaa, Finland). C-peptide (nmol/l) and plasma insulin (pmol/l) were determined using chemiluminometric immunoassay (Liaison Insulin and Liaison C-peptid, DiaSorin S.p.A, Saluggia, Italy). HbA1c (mmol/mol) was measured by ion-exchange high performance liquid chromatography on a TOSOH G8 analyser (Tosoh Bioscience, Inc, CA, USA). Triglycerides, total and high density lipoprotein (HDL) cholesterol were enzymatically assessed. More details on sample handling and biomarker determination are described elsewhere [6].
Assessment of covariates
Questionnaires were used to collect data on age, sex, parental diabetes status, disease history, study center, smoking status, alcohol consumption and medication for diabetes and other conditions. Additionally, waist circumference (cm), BMI (kg/m2), blood pressure (mmHg) and physical activity (mgs) were determined at baseline and follow-up.
Data analysis
Latent class trajectory analysis (LCTA) with cubic polynomials for the specification of time was used to identify glucose curve subgroups from MMT at baseline and at follow-up. The latent class mixed-effects models (lcmm) package in R (version 3.2.1) was used to conduct the LCTA. A random slope and random intercept were specified in the lcmm procedure. A random intercept was included to account for the correlation of measurements from the same individual. Optimal number of groups at baseline and follow-up was delineated by adding one more latent group at a time. To account for the chance of convergence to local maxima, the LCTA procedure was embedded in a gridsearch function. The best-fitting classification model was determined based on the Bayesian Information Criterion (BIC) and the Akaike Information Criterion (AIC). The lowest BIC and AIC suggesting the best fit and a difference of at least 10 points was regarded as sufficient improvement. The individuals were then assigned to a particular group based on their highest membership probability. Additionally, all mean membership probabilities for each class should be > 0.8 and this was also used to select the final number of groups. Furthermore, we produced plots of serum insulin levels corresponding to each subgroup’s glucose response curve at baseline.
We used general linear models to calculate coefficients (β) and 95% confidence intervals (CI) to estimate the association of identified subgroups with baseline metabolic traits and change in these metabolic traits at 18 months. We adjusted the p-values for multiple testing using the Bonferroni correction method.
Two multivariable models were formulated for the prospective analyses. Model 1 was adjusted for age, sex, follow-up, study center and respective baseline metabolic traits. Model 2 was additionally adjusted for smoking status, physical activity, family history of diabetes, diabetes duration and diabetes medication. Missing values were below 5% for all covariates with the exception of physical activity (8.4%). Therefore, we did a single imputation for physical activity values using the predictive mean matching method in the MICE (Multivariate Imputation by Chained Equations) package in R [7].
Furthermore, we compared baseline metabolic traits across individuals moving from subgroups at baseline to similar, lower or higher peak subgroups at follow-up.
For sensitivity analyses, we identified the baseline glucose curve subgroups stratified by study center and sex.
Results
The analysis included 787 individuals at baseline; 58% men, age 62.1 ±8.1, BMI 30.5 ±5.0 kg/m2 and HbA1c 46.5 ±5.8 mmol/mol (6.2 ±0.5%) (Table 1).
We identified three subgroups with different glucose patterns following an MMT, labelled in order of increasing glucose peak levels as subgroup 1–3 (Fig 2), consisting of 139 (18%), 466 (59%) and 182 (23%) individuals, respectively (Table 1). Subgroup 1 had the earliest and lowest glucose peak (8.4 mmol/l), while subgroup 3 took the longest to reach peak and had the highest glucose peak (13.1 mmol/l), and subgroup 2 had a glucose peak of 10.0 mmol/l. Membership probabilities were all above 0.80, ranging from 0.86 to 0.90. The four subgroup solution met all but one of the selection criteria, one of the membership probabilities was below 0.8 (S1 Fig and Table 2). Stratified by sex and center, three similar subgroups were identified. Heterogeneity was also evident in the insulin response curve subgroups. Subgroup 1 had the earliest and lowest insulin peak with the lowest 2-hour insulin value, while subgroup 3 had the latest insulin peak and highest 2-hour value (S2 Fig). Individuals in higher glucose peak subgroups i.e. subgroup 2 and 3, were more likely to have higher metabolic trait values such as higher BMI, insulin and glucose values compared to individuals in subgroup 1 (Table 1) and hence a less favourable metabolic risk profile compared to individuals in subgroup 1.
At 18 month follow-up, the beta-coefficients (95% CI) for change in HbA1c (mmol/mol) significantly increased across subgroups with 0.37 (-0.18–1.92) for subgroup 2 and 1.88 (-0.08–3.85) for subgroup 3, relative to subgroup 1 (Table 3). The same trend was observed for change in triglycerides and fasting glucose with the exception of HDL. Generally, individuals in subgroup 2 and 3 progressed faster in metabolic parameters than individuals in subgroup 1.
The optimum number of glucose curve subgroups identified at follow-up was four with the fourth subgroup having the highest peak, fasting and 2-hour glucose values (Fig 3). The percentage of individuals moving from subgroups at baseline to higher peak subgroups at follow-up was 38%, 15% and 7% for subgroup 1–3, respectively. Individuals from baseline subgroups moving to lower peak subgroups at follow-up were 27% and 46% for subgroup 2 and 3, respectively. In short, majority of participants in subgroups at baseline remained in the same subgroups at follow-up (Table 4). Individuals moving from baseline subgroups to higher peak subgroups at follow-up, were more likely to have higher BMI, waist circumference, diastolic pressure, hbA1c, fasting glucose and fasting insulin. They were also more likely to be males, use metformin and had the highest family history of T2D. Most of these characteristics including change in T2D medication were lowest in those moving from higher peak baseline subgroups to lower peak subgroups at follow-up (S1 Table).
Discussion
Using data from 787 individuals with T2D, who had undergone a frequently sampled MMT, combined with the LCTA approach, three glucose curve subgroups were identified. The individuals within these subgroups differed in their metabolic traits at baseline and their rates of subsequent glycaemic deterioration over 18 months. Individuals with T2D in the subgroups with the highest plasma glucose peaks and the highest 1-hour and 2-hour glucose levels (subgroups 2 and 3) had the worst metabolic traits profile, compared to individuals in the subgroup with the lowest and earliest glucose peak. Moreover, individuals moving from baseline subgroups to higher peak subgroups at follow-up had the worst metabolic traits profile compared to those moving from higher peak baseline subgroups to lower peak subgroups at follow-up.
To the best of our knowledge, there is no study that has exhaustively studied the heterogeneity in glucose curves following an MMT in individuals with T2D. However, several studies identified glucose response curves following an OGTT in those at elevated diabetes risk [1, 4, 8], and most of them identified four glucose response subgroups. The disparity in the number of subgroups between these studies and ours could be because we had a relatively more homogenous group of individuals with T2D (without insulin therapy) and hence less heterogeneity. Nevertheless, the subgroups with the highest peaks and 1-hour value in these studies had the worst metabolic risk profile which is in line with our findings.
The various plausible mechanisms that could explain the differences in the identified subgroups in our study include differences in insulin resistance and secretion. Subgroup 2 and 3 had higher BMI, insulin and glucose values, thereby indicating that they were more insulin resistant than those in subgroup 1. Subgroup 2 and 3 also had higher 1-hour glucose values compared to subgroup 1 and studies showed that higher 1-hour plasma glucose is associated with an increased risk of T2D complications [9–14]. High 1-hour glucose value could be a result of impaired early phase insulin secretion and reduced insulin sensitivity which could explain the heterogeneity we see in the subgroups.
Some limitations of our study include the limited generalizability of our results to other ethnicities due to the inclusion of only white European adults. Second, a short follow-up time of 18 months does not allow us to investigate associations with hard cardiovascular outcomes. Lastly, we did not have information on adherence to diabetes drugs which might influence progression within the 18 months period, however, we adjusted for the use of diabetes medication to try and mitigate this. Our study was strengthened by the use of a large sample with elaborate metabolic parameters at baseline and follow up.
In conclusion, different glycaemic profiles with different metabolic traits and different degrees of subsequent glycaemic deterioration can be identified using data from a frequently sampled MMT in individuals with T2D. This heterogeneity in glucose curves suggests different underlying pathophysiologies. However, more similar studies should be done to confirm the robustness of these results.
Supporting information
S1 Fig. Glucose curve subgroups following a MMT depicting estimated mean trajectories of the 4 group solution identified by the latent class trajectory analysis in 787 individuals with type 2 diabetes at baseline.
https://doi.org/10.1371/journal.pone.0242360.s001
(DOCX)
S2 Fig. Insulin curve subgroups corresponding to the 3 glucose curve groups following a MMT depicting estimated mean trajectories identified by the latent class trajectory analysis in 787 individuals with type 2 diabetes at baseline.
https://doi.org/10.1371/journal.pone.0242360.s002
(DOCX)
S1 Table. Characteristics of 651 individuals stratified by movement from baseline subgroups to similar, lower or higher peak subgroups at follow-up.
https://doi.org/10.1371/journal.pone.0242360.s003
(DOCX)
Acknowledgments
We are very grateful to all the participants in IMI-DIRECT: http://www.direct-diabetes.org/participants/
References
- 1. Hulman A, Gujral UP, Narayan KM, Pradeepa R, Mohan D, Anjana RM, et al. Glucose patterns during the OGTT and risk of future diabetes in an urban Indian population: The CARRS study. Diabetes Res Clin Pract. 2017;126:192–7. pmid:28259008
- 2. Hulman A, Simmons RK, Brunner EJ, Witte DR, Faerch K, Vistisen D, et al. Trajectories of glycaemia, insulin sensitivity and insulin secretion in South Asian and white individuals before diagnosis of type 2 diabetes: a longitudinal analysis from the Whitehall II cohort study. Diabetologia. 2017;60(7):1252–60. pmid:28409212
- 3. Hulman A, Simmons RK, Vistisen D, Tabak AG, Dekker JM, Alssema M, et al. Heterogeneity in glucose response curves during an oral glucose tolerance test and associated cardiometabolic risk. Endocrine. 2017;55(2):427–34. pmid:27699707
- 4. Hulman A, Vistisen D, Glumer C, Bergman M, Witte DR, Faerch K. Glucose patterns during an oral glucose tolerance test and associations with future diabetes, cardiovascular disease and all-cause mortality rate. Diabetologia. 2018;61(1):101–7. pmid:28983719
- 5. Abdul-Ghani MA, Tripathy D, DeFronzo RA. Contributions of beta-cell dysfunction and insulin resistance to the pathogenesis of impaired glucose tolerance and impaired fasting glucose. Diabetes Care. 2006;29(5):1130–9. pmid:16644654
- 6. Koivula RW, Heggie A, Barnett A, Cederberg H, Hansen TH, Koopman AD, et al. Discovery of biomarkers for glycaemic deterioration before and after the onset of type 2 diabetes: rationale and design of the epidemiological studies within the IMI DIRECT Consortium. Diabetologia. 2014;57(6):1132–42. pmid:24695864
- 7. Morris TP, White IR, Royston P. Tuning multiple imputation by predictive mean matching and local residual draws. BMC Med Res Methodol. 2014;14:75. pmid:24903709
- 8. Hulman A, Witte DR, Vistisen D, Balkau B, Dekker JM, Herder C, et al. Pathophysiological Characteristics Underlying Different Glucose Response Curves: A Latent Class Trajectory Analysis From the Prospective EGIR-RISC Study. Diabetes Care. 2018;41(8):1740–8. pmid:29853473
- 9. Abdul-Ghani MA, Lyssenko V, Tuomi T, DeFronzo RA, Groop L. Fasting versus postload plasma glucose concentration and the risk for future type 2 diabetes: results from the Botnia Study. Diabetes Care. 2009;32(2):281–6. pmid:19017778
- 10. Abdul-Ghani MA, Lyssenko V, Tuomi T, Defronzo RA, Groop L. The shape of plasma glucose concentration curve during OGTT predicts future risk of type 2 diabetes. Diabetes Metab Res Rev. 2010;26(4):280–6. pmid:20503260
- 11. Abdul-Ghani MA, Williams K, DeFronzo RA, Stern M. What is the best predictor of future type 2 diabetes? Diabetes Care. 2007;30(6):1544–8. pmid:17384342
- 12. Alyass A, Almgren P, Akerlund M, Dushoff J, Isomaa B, Nilsson P, et al. Modelling of OGTT curve identifies 1 h plasma glucose level as a strong predictor of incident type 2 diabetes: results from two prospective cohorts. Diabetologia. 2015;58(1):87–97. pmid:25292440
- 13. Bergman M, Chetrit A, Roth J, Dankner R. Dysglycemia and long-term mortality: observations from the Israel study of glucose intolerance, obesity and hypertension. Diabetes Metab Res Rev. 2015;31(4):368–75. pmid:25352076
- 14. Jagannathan R, Sevick MA, Li H, Fink D, Dankner R, Chetrit A, et al. Elevated 1-hour plasma glucose levels are associated with dysglycemia, impaired beta-cell function, and insulin sensitivity: a pilot study from a real world health care setting. Endocrine. 2016;52(1):172–5. pmid:26419850