Choice of HbA1c threshold for identifying individuals at high risk of type 2 diabetes and implications for diabetes prevention programmes: a cohort study

Machine Learning Approach to Metabolomic Data Predicts Type 2 Diabetes Mellitus Incidence

Metabolomics, with its wealth of data, offers a valuable avenue for enhancing predictions and decision-making in diabetes. This observational study aimed to leverage machine learning (ML) algorithms to predict the 4-year risk of developing type 2 diabetes mellitus (T2DM) using targeted quantitative metabolomics data. A cohort of 279 cardiovascular risk patients who underwent coronary angiography and who were initially free of T2DM according to American Diabetes Association (ADA) criteria was analyzed at baseline, including anthropometric data and targeted metabolomics, using liquid chromatography (LC)-mass spectroscopy (MS) and flow injection analysis (FIA)-MS, respectively. All patients were followed for four years. During this time, 11.5% of the patients developed T2DM. After data preprocessing, 362 variables were used for ML, employing the Caret package in R. The dataset was divided into training and test sets (75:25 ratio) and we used an oversampling approach to address the classifier imbalance of T2DM incidence. After an additional recursive feature elimination step, identifying a set of 77 variables that were the most valuable for model generation, a Support Vector Machine (SVM) model with a linear kernel demonstrated the most promising predictive capabilities, exhibiting an F1 score of 50%, a specificity of 93%, and balanced and unbalanced accuracies of 72% and 88%, respectively. The top-ranked features were bile acids, ceramides, amino acids, and hexoses, whereas anthropometric features such as age, sex, waist circumference, or body mass index had no contribution. In conclusion, ML analysis of metabolomics data is a promising tool for identifying individuals at risk of developing T2DM and opens avenues for personalized and early intervention strategies.

Plasma metabolomic profiles of plant-based dietary indices reveal potential pathways for metabolic syndrome associations

BACKGROUND AND AIMS

Plant-based dietary patterns have been associated with improved health outcomes. This study aims to describe the metabolomic fingerprints of plant-based diet indices (PDI) and examine their association with metabolic syndrome (MetS) and its components in a Danish population.

METHODS

The MAX study comprised 676 participants (55% women, aged 18-67 y) from Copenhagen. Sociodemographic and dietary data were collected using questionnaires and three 24-h dietary recalls over one year (at baseline, and at 6 and 12 months). Mean dietary intakes were computed, as well as overall PDI, healthful (hPDI) and unhealthful (uPDI) scores, according to food groups for each plant-based index. Clinical variables were also collected at the same time points in a health examination that included complete blood tests. MetS was defined according to the International Diabetes Federation criteria. Plasma metabolites were measured using a targeted metabolomics approach. Metabolites associated with PDI were selected using random forest models and their relationships with PDIs and MetS were analyzed using generalized linear mixed models.

RESULTS

The mean prevalence of MetS was 10.8%. High, compared to low, hPDI and uPDI scores were associated with a lower and higher odd of MetS, respectively [odds ratio (95%CI); hPDI: 0.56 (0.43-0.74); uPDI: 1.61 (1.26-2.05)]. Out of 411 quantified plasma metabolites, machine-learning metabolomics fingerprinting revealed 13 metabolites, including food and food-related microbial metabolites, like hypaphorine, indolepropionic acid and lignan-derived enterolactones. These metabolites were associated with all PDIs and were inversely correlated with MetS components (p < 0.05). Furthermore, they had an explainable contribution of 12% and 14% for the association between hPDI or uPDI, respectively, and MetS only among participants with overweight/obesity.

CONCLUSIONS

Metabolites associated with PDIs were inversely associated with MetS and its components, and may partially explain the effects of plant-based diets on cardiometabolic risk factors.

Dietary polyphenols, metabolic syndrome and cardiometabolic risk factors: An observational study based on the DCH-NG subcohort

BACKGROUND AND AIMS

Polyphenol-rich foods have beneficial properties that may lower cardiometabolic risk. We aimed to prospectively investigate the relationship between intakes of dietary polyphenols, and metabolic syndrome (MetS) and its components, in 676 Danish residents from the MAX study, a subcohort of the Danish Diet, Cancer and Health-Next Generations (DCH-NG) cohort.

METHODS AND RESULTS

Dietary data were collected using web-based 24-h dietary recalls over one year (at baseline, and at 6 and 12 months). The Phenol-Explorer database was used to estimate dietary polyphenol intake. Clinical variables were also collected at the same time point. Generalized linear mixed models were used to investigate relationships between polyphenol intake and MetS. Participants had a mean age of 43.9y, a mean total polyphenol intake of 1368 mg/day, and 75 (11.6%) had MetS at baseline. Compared to individuals with MetS in Q1 and after adjusting for age, sex, lifestyle and dietary confounders, those in Q4 - for total polyphenols, flavonoids and phenolic acids-had a 50% [OR (95% CI): 0.50 (0.27, 0.91)], 51% [0.49 (0.26, 0.91)] and 45% [0.55 (0.30, 1.00)] lower odds of MetS, respectively. Higher total polyphenols, flavonoids and phenolic acids intakes as continuous variable were associated with lower risk for elevated systolic blood pressure (SBP) and low high-density lipoprotein cholesterol (HDL-c) (p < 0.05).

CONCLUSIONS

Total polyphenol, flavonoid and phenolic acid intakes were associated with lower odds of MetS. These intakes were also consistently and significantly associated with a lower risk for higher SBP and lower HDL-c concentrations.

Diagnostic Performance of Sex-Specific Modified Metabolite Patterns in Urine for Screening of Prediabetes

AIMS/HYPOTHESIS

Large-scale prediabetes screening is still a challenge since fasting blood glucose and HbA_1c as the long-standing, recommended analytes have only moderate diagnostic sensitivity, and the practicability of the oral glucose tolerance test for population-based strategies is limited. To tackle this issue and to identify reliable diagnostic patterns, we developed an innovative metabolomics-based strategy deviating from common concepts by employing urine instead of blood samples, searching for sex-specific biomarkers, and focusing on modified metabolites.

METHODS

Non-targeted, modification group-assisted metabolomics by liquid chromatography-mass spectrometry (LC-MS) was applied to second morning urine samples of 340 individuals from a prediabetes cohort. Normal (n = 208) and impaired glucose-tolerant (IGT; n = 132) individuals, matched for age and BMI, were randomly divided in discovery and validation cohorts. ReliefF, a feature selection algorithm, was used to extract sex-specific diagnostic patterns of modified metabolites for the detection of IGT. The diagnostic performance was compared with conventional screening parameters fasting plasma glucose (FPG), HbA_1c, and fasting insulin.

RESULTS

Female- and male-specific diagnostic patterns were identified in urine. Only three biomarkers were identical in both. The patterns showed better AUC and diagnostic sensitivity for prediabetes screening of IGT than FPG, HbA_1c, insulin, or a combination of FPG and HbA_1c. The AUC of the male-specific pattern in the validation cohort was 0.889 with a diagnostic sensitivity of 92.6% and increased to an AUC of 0.977 in combination with HbA_1c. In comparison, the AUCs of FPG, HbA_1c, and insulin alone reached 0.573, 0.668, and 0.571, respectively. Validation of the diagnostic pattern of female subjects showed an AUC of 0.722, which still exceeded the AUCs of FPG, HbA_1c, and insulin (0.595, 0.604, and 0.634, respectively). Modified metabolites in the urinary patterns include advanced glycation end products (pentosidine-glucuronide and glutamyl-lysine-sulfate) and microbiota-associated compounds (indoxyl sulfate and dihydroxyphenyl-gamma-valerolactone-glucuronide).

CONCLUSIONS/INTERPRETATION

Our results demonstrate that the sex-specific search for diagnostic metabolite biomarkers can be superior to common metabolomics strategies. The diagnostic performance for IGT detection was significantly better than routinely applied blood parameters. Together with recently developed fully automatic LC-MS systems, this opens up future perspectives for the application of sex-specific diagnostic patterns for prediabetes screening in urine.