Roberto Bizzotto (1), Azra Kurbasic (2), Chris Jennison (3), Angus Jones (4,5), Ewan R Pearson (6), Andrea Mari (1)
(1) Institute of Neuroscience, National Research Council, Padova, Italy; (2) Genetic and Molecular Epidemiology Unit, Lund University, Malmö, Sweden; (3) Department of Mathematical Sciences, University of Bath, Bath, UK; (4) National Institute for Health Research Exeter Clinical Research Facility, University of Exeter Medical School, Exeter, UK; (5) Royal Devon and Exeter NHS Foundation Trust, Exeter, UK; (6) Molecular and Clinical Medicine, University of Dundee, Dundee, UK.
Objectives: Research on personalization of treatment for type 2 diabetes (T2D) requires the study of the individual rate of disease progression, corrected for treatment and weight changes. Glycated hemoglobin (HbA1c) has been used as a marker of glucose tolerance deterioration, as it is considered to reflect the average glycaemia over the last two to three months. Fasting plasma glucose (FPG) is another important marker of glucose tolerance. However, the relationship between progression of HbA1c and FPG has not been studied. The aim of this work is the description of both HbA1c and FPG progression rates and the analysis of the differences in their individual values. A conditional linear mixed-effect model (Model C) is used for this purpose, as it is a tool for characterizing individual longitudinal patterns while minimizing the potential bias introduced by assumptions on and shrinkage of cross-sectional effects [1].
Methods: White European T2D patients enrolled within 24 months from diagnosis, treated only with lifestyle change or metformin until baseline visit, and with HbA1c <60 mmol/mol within previous 3 months, were recruited in the DIRECT multi-center study (N=736). HbA1c and FPG concentrations were collected at months 0, 9, 18, 27 (HbA1c only) and 36 after start of the study. Mixed-meal tolerance tests (MMTT) were performed at 0, 18 and 36 months. The Model C orthogonal longitudinal and cross-sectional components of the data were computed. The longitudinal components of HbA1c and FPG data were described as proportional functions of time, with normally distributed slopes describing underlying progression. As a term of comparison, the original untransformed HbA1c data were modelled as well, using linear time effects with normally (Model N) or uniformly distributed intercepts (Model U). Additive linear effects of the changes in BMI and in standardized dosage of the antidiabetic treatments were included in all models. Medications were considered effective when started at least 30 days before the HbA1c measurement, and 6 days before FPG measurement. Modelling was performed with MonolixSuite2016 R1.
Results: Model N produced 20% η-shrinkage of the individual estimates of HbA1c intercept. The HbA1c slope estimates of Model U or Model C were equal, and different from the slope estimates from Model N (mean absolute deviation = 0.32 mmol mol-1 y-1). Estimated slope was 0.70±1.33 mmol mol-1 y-1 (median±SD) (with median r.s.e. = 17%) for HbA1c and 0.22±0.27 mmol L-1 y-1 (12%) for FPG. BMI effect was 1.4 mmol mol-1 kg-1 m2 (8%) for HbA1c and 0.228 mmol L-1 kg-1 m2 (10%) for FPG. Linear correlation between individual slopes of HbA1c and FPG was 0.70. The number of subjects with positive FPG slopes was higher than that with positive HbA1c slopes (p<10-6, McNemar test). To understand this discrepancy, two groups of patients with discordant slopes were selected: group FPG+HbA1c- with FPG slope above the median and HbA1c slope below the median, and group FPG-HbA1c+ with FPG slope below the median and HbA1c slope above the median. The HbA1c and FPG trajectories were discordant in the two groups, as expected. The trajectories of mean incremental glucose during MMTT mirrored those of HbA1c and were discordant with FPG. The trajectories of mean absolute MMTT glucose were similar in the two groups and discordant with those of HbA1c and FPG. Group FPG+HbA1c- had robust baseline β-cell function but decreasing standardized fasting insulin secretion rate [2] and increasing HOMA insulin resistance. Group FPG-HbA1c+ was relatively β-cell deficient at baseline and had a fall over time in potentiation ratio [2] and rate sensitivity. Changes over time of cholesterol, LDL and triglycerides concentrations were correlated with those of insulin resistance and β-cell function.
Conclusions: Underlying progression of HbA1c and FPG were analyzed in T2D patients of recent diagnosis. Potential spurious correlations between the individual estimates of progression (slope) and baseline (intercept) were avoided using conditional modelling or, equivalently, uniformly distributed individual intercepts. On average, progression was faster for FPG than for HbA1c. The difference could not be explained in terms of mean glycaemia during the MMTT. Temporal trajectories of β-cell function and insulin sensitivity, together with lipid profiles, provided a possible explanation for the differences in the individual progression of HbA1c and FPG.
References:
[1] Verbake G. Conditional Linear Mixed Models. The American Statistician 55: 25–34, 2001.
[2] Mari A, Tura A, Gastaldelli A, Ferrannini E. Assessing Insulin Secretion by Modeling in Multiple-Meal Tests: Role of Potentiation. Diabetes 51: S221–S226, 2002.
Reference: PAGE 28 (2019) Abstr 9153 [www.page-meeting.org/?abstract=9153]
Poster: Drug/Disease Modelling - Endocrine