Hanna Kunina 1, Chengli Yu 1,2, Gustaf Wellhagen 1, Maria Kjellsson 1
1 Pharmacometrics Research Group, Department of Pharmacy, Uppsala University (Uppsala, Sweden), 2 Université Paris Cité and Université Sorbonne Paris Nord, Inserm, IAME (Paris, France)
Introduction/Objectives: Metabolic syndrome (MetS) is a cluster of metabolic disorders including obesity, dyslipidemia, hypertension, and insulin resistance [1]. Type 2 diabetes (T2D) commonly develops in the presence of MetS, and the accumulation of MetS components increases cardiovascular disease (CVD) risk among patients with T2D [2]. This combined metabolic burden leads to higher all-cause mortality rates that exceed the effect of hyperglycemia alone [3]. To quantify the underlying, unmeasurable disease burden, latent variable approaches, such as item response theory (IRT) [4], have been proposed. The primary aim of this study was to integrate multiple continuous cardiometabolic risk factors into an assessment of MetS severity and its longitudinal progression in patients with T2D.
Methods: Data from the Swedish National Diabetes Register (NDR) [5] were used to analyze a cohort of 33,396 adult patients newly diagnosed with T2D. Seven longitudinal biomarkers, i.e., systolic and diastolic blood pressure (sBP, dBP), low-density and high-density lipoproteins (LDL, HDL), triglycerides (TG), glycated hemoglobin (HbA1c), and body mass index (BMI), were categorized into four levels based on the established CVD risk thresholds [6-9]: 0 (low), 1 (elevated), 2 (moderately high), and 3 (high), serving as items in the IRT model. The IRT modelling was divided into three steps: (i) determination of the number of dimensions, (ii) estimation of item characteristic functions (ICFs), and (iii) modelling of the longitudinal change in the latent variables, with the baseline ICFs fixed. The optimal number of dimensions was determined by the Akaike information criterion (AIC) and Bayesian information criterion (BIC) for baseline IRT. Longitudinal progression was modelled by allowing the latent variables (LVs) to evolve over time, with ICFs fixed at baseline, and item parameters re-estimated longitudinally where needed. Analyses were performed in NONMEM v.7.5.1 [10] with LAPLACE via PsN v.5.3.1 [11]. Data management and graphical evaluation of the results were performed in R v.4.3.1 [12]. The computations were enabled by resources at the Uppsala Multidisciplinary Center for Advanced Computational Science (UPPMAX).
Results: A two-dimensional IRT model, selected over a unidimensional model based on AIC and BIC, was developed to capture baseline and longitudinal cardiometabolic changes in patients with T2D, with seven items informing two LVs. The first LV (LV1) was primarily informed by sBP and dBP, whereas the second LV (LV2) was primarily informed by HDL, BMI, and HbA1c, with dBP contributing to a lesser extent. Significant covariates included age at diabetes onset for both LVs and biological sex for LV2. The baseline model adequately described sBP, dBP, TG, and BMI, while time-varying effects were incorporated for HbA1c, LDL, and HDL. Both LV1 and LV2 declined at a similar annual rate of −0.03, with negative time effects (LV1: −0.0998; LV2: −0.2166) and negative age slopes (LV1: −0.02816; LV2: −0.02899) indicating a general decline in latent trait scores over time. The probability of HbA1c falling into the high-risk category initially decreased, then increased progressively, as indicated by a separate positive slope (0.4514), suggesting its distinct longitudinal progression. dBP showed a modest shift toward lower risk categories. Age at diabetes onset had a minor, but opposing effect on the two LVs (LV1: 0.02; LV2: −0.07), and women had lower LV2 scores compared to men (−0.43).
Conclusion: The two-dimensional IRT model effectively characterized baseline and longitudinal cardiometabolic dysfunction in patients with T2D, identifying a cardiovascular dimension primarily informed by sBP and dBP, and a metabolic dimension primarily informed by HDL, HbA1c, and BMI. HbA1c demonstrated a distinct longitudinal pattern, with initial improvement followed by progressive deterioration, indicating worsening glycemic control over time. The developed modeling framework offers a structured approach for quantifying multidimensional disease severity in T2D using routinely collected biomarkers.
References:
[1] Alberti KG et al. Diabet Med. 2006;23(5):469-480.
[2] Mottillo S et al. J Am Coll Cardiol. 2010;56(14):1113-1132.
[3] Einarson TR et al. Cardiovasc Diabetol. 2018;17(1):83.
[4] Ueckert S. CPT Pharmacometrics Syst Pharmacol. 2018;7(4):205-218.
[5] Gudbjörnsdottir S et al. Diabetes Care. 2003;26(4):1270-1276.
[6] Marx N et al. Eur Heart J. 2023;44(39):4043-4140.
[7] Whelton PK et al. Hypertension. 2018;71(6):1269-1324.
[8] Mach F et al. Eur Heart J. 2020;41(1):111-188.
[9] WHO Expert Committee. Physical Status: Use and Interpretation of Anthropometry. WHO; 1995.
[10] Beal SL et al. NONMEM User’s Guides. Icon Development Solutions; 2014.
[11] Lindbom L et al. Comput Methods Programs Biomed. 2005;79(3):241-257.
[12] R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing; 2023.
Reference: PAGE 34 (2026) Abstr 11979 [www.page-meeting.org/?abstract=11979]
Poster: Drug/Disease Modelling - Endocrine