Improving 10-year cardiovascular risk prediction in patients with type 2 diabetes with metabolomics

Abstract

Background: Existing cardiovascular risk prediction models still have room for improvement in patients with type 2 diabetes who represent a high-risk population. This study evaluated whether adding metabolomic biomarkers could enhance the 10-year prediction of major adverse cardiovascular events (MACE) in these patients.

Methods: Data from 10,257 to 1,039 patients with type 2 diabetes from the UK Biobank (UKB) and the German ESTHER cohort, respectively, were used for model derivation, internal and external validation. A total of 249 metabolites were measured with nuclear magnetic resonance (NMR) spectroscopy. Sex-specific LASSO regression with bootstrapping identified significant metabolites. The enhanced model's predictive performance was evaluated using Harrell's C-index.

Results: Seven metabolomic biomarkers were selected by LASSO regression for enhanced MACE risk prediction (three for both sexes, three male- and one female-specific metabolite(s)). Especially albumin and the omega-3-fatty-acids-to-total-fatty-acids-percentage among males and lactate among females improved the C-index. In internal validation with 30% of the UKB, adding the selected metabolites to the SCORE2-Diabetes model increased the C-index statistically significantly (P = 0.037) from 0.660 to 0.678 in the total sample. In external validation with ESTHER, the C-index increase was higher (+ 0.043) and remained statistically significant (P = 0.011).

Conclusions: Incorporating seven metabolomic biomarkers in the SCORE2-Diabetes model enhanced its ability to predict MACE in patients with type 2 diabetes. Given the latest cost reduction and standardization efforts, NMR metabolomics has the potential for translation into the clinical routine.

Keywords: Cardiovascular risk; Metabolomics; Prediction model; Type 2 diabetes.

Fig. 1

Associations between selected metabolites and major cardiovascular events across sexes in the derivation set (70% of UK biobank, N = 7,180). CI, confidence interval; FDR, false discovery rate; GlycA, glycoprotein acetyls; HR, hazard ratio; Omega-3-pct, Omega-3 fatty acids to total fatty acids percentage; SD, standard deviation; VLDL-size, average diameter for very-low-density lipoprotein particles. ^a Hazard ratios are expressed per 1 standard deviation of the respective metabolite concentration and are adjusted for age, systolic blood pressure, smoking status, diabetes age at diagnosis, glycated hemoglobin, and the estimated glomerular filtration rate. The standard deviations of creatinine, albumin, GlycA, acetate, omega-3-pct, VLDL-size, and lactate were 0.02 mmol/L, 3.60 mmol/L, 0.13 mmol/L, 0.04 mmol/L, 1.51%, 1.41 mmol/L, and 1.23 mmol/L, respectively

Fig. 2

Incremental discrimination of each metabolite for the model after the selected metabolites were added to SCORE2-Diabetes separately in different sexes in the internal validation set (30% of UK Biobank, N = 3,077). CI, confidence interval; GlycA, glycoprotein acetyls; HR, hazard ratio; Omega-3-pct, Omega-3 fatty acids to total fatty acids percentage; VLDL-size, average diameter for very-low-density lipoprotein particles

Fig. 3

Discriminative performance of four models for major adverse cardiovascular events (MACE) by sex and overall population in internal validation set (30% of UK Biobank, N = 3,077) and the external validation set (ESTHER study, N = 1,039). Note: Table 1 lists the variables included in each of the four models. NMR, nuclear magnetic resonance; CI, confidence interval; HDL, high-density lipoprotein