Phenotype-based clusters, inflammation and cardiometabolic complications in older people before the diagnosis of type 2 diabetes: KORA F4/FF4 cohort study

Abstract

Background: Using a data-driven approach, six clusters with different risk profiles and burden of complications were recently identified in middle-aged people before the diagnosis of type 2 diabetes (T2D). We aimed to investigate whether these clusters could be generalised to older people and if subclinical inflammation was related to their cardiometabolic risk profiles.

Methods: We assigned 843 participants of the KORA F4 study aged 61-82 years without T2D to the six previously defined phenotype-based clusters. Based on 73 biomarkers of subclinical inflammation, we derived an inflammation-related score ("inflammatory load") using principal component analysis to assess subclinical inflammation. Risk factors, inflammatory load as well as prevalence and incidence of (pre)diabetes-related complications were compared between the clusters using pairwise comparisons and regression analyses.

Results: Clusters 1 and 2 had the lowest cardiometabolic risk, whereas clusters 5 and 6 the highest. T2D risk was highest in clusters 3, 4, 5, and 6 compared with the low-risk cluster 2 (age- and sex-adjusted ORs between 3.6 and 34.0). In cross-sectional analyses, there were significant between-cluster differences in chronic kidney disease (CKD), distal sensorimotor polyneuropathy (DSPN) and cardiovascular disease (all p < 0.045). In prospective analyses (mean follow-up time 6.5-8.3 years), clusters differed significantly in CKD and DSPN incidence, but not in incident CVD or all-cause mortality. The inflammatory load was highest in the high-risk cluster 5 and lowest in cluster 2. Adjustment for the inflammatory load had only a minor impact on the aforementioned differences in outcomes between clusters.

Conclusions: Our findings extend the knowledge about the previously identified six phenotype-based clusters in older people without T2D. Differences between clusters were more pronounced for T2D risk than for prevalent or incident (pre)diabetes-related complications and absent for mortality. The high cardiometabolic risk corresponded to the high inflammatory load in cluster 5 but not to the lower inflammatory load of high-risk clusters 3 and 6.

Keywords: Cardiovascular disease; Chronic kidney disease; Cohort; Diabetes; Heterogeneity; Inflammation; Mortality; Polyneuropathy; Prediabetes; Subtypes.

Fig. 1

Glucose tolerance categories and of the clustering variables in the six clusters. a Distribution of participants by glucose tolerance category and cluster. b Distribution of the clustering variables within each cluster. Medians are plotted for each cluster with the corresponding standardised level (mean = 0, standard deviation = 1) for each variable. AUC Gluc, area under the glucose curve during OGTT (2-point glucose area-under-curve was calculated as 120 * ((fasting blood glucose in mmol/L + blood glucose at 2 h in mmol/L)/2); BMI, body mass index (kg/m2); Fst Ins, fasting insulin (µIU/ml); hip, Hip circumference (cm); Insulin resistance: 1 / insulin sensitivity (Matsuda’s index for estimating insulin sensitivity during OGTT was calculated as 10,000/sqrt (fasting blood glucose in mmol/L * fasting insulin in pmol/L* (fasting insulin in pmol/L + insulin at 2 h in pmol/L)/2 * (fasting blood glucose in mmol/L + blood glucose at 2 h in mmol/L)/2)); Low HDL, HDL cholesterol in mmol/L * -1 (directionally flipped so that a higher area in the figure indicates a higher cardiometabolic risk); Secretion failure, Stumvoll’s first phase insulin secretion index calculated as 2503 + 6.476 * fasting insulin in pmolL—126.5 * blood glucose at 2 h in mmol/L + 0.954 * insulin at 2 h in pmol/L—293.3 * fasting blood glucose in mmol/L; Waist, waist circumference (cm)

Fig. 2

Incidence of T2D in the clusters. See also Supplementary Table 2, Additional file 1 for numbers of cases in each cluster. *p < 0.05, *** p < 0.001 for comparisons between each cluster to cluster 2 (reference)

Fig. 3

(Pre)diabetes-related complications in the clusters (a, prevalence; b, incidence). See also Supplementary Table 2, Additional file 1 for numbers of cases in each cluster. *p < 0.05, **p < 0.01, ***p < 0.001 for comparisons between each cluster to cluster 2 (reference)

Fig. 4

Comparison of the inflammatory load between the clusters. a Notched boxplots of the inflammatory load (PC score) for each cluster. The lower whisker indicates the smallest observation ≥ lower hinge-1.5*IQR. The upper whisker indicates the largest observation ≤ upper hinge + 1.5*IQR. b Pairwise comparison of the inflammatory load (PC score) between each cluster against all other clusters combined. Model 1, unadjusted; model 2, adjusted for age and sex; model 3, model 2 + BMI. The color of the heatmap indicates the difference in mean levels of the groups and the asterisks (*) indicate the significance (BH corrected p < 0.05). All p values were adjusted with Benjamini–Hochberg correction for all 6 comparisons. Clusters are named “C1”, “C2”, “C3”, “C4”, “C5”, and “C6”. PC score: principal component score

Fig. 5

Pairwise comparisons of biomarkers of inflammation between each cluster against all other clusters combined. a unadjusted; b adjusted for age and sex; c, adjusted for age, sex, and BMI. The colour of the heatmap indicates the difference in mean levels between the groups and the asterisks (*) indicate statistical significance (p_BH < 0.05). All p values were adjusted with Benjamini–Hochberg correction for all 438 tests (73 markers and 6 comparisons). Clusters are named “C1”, “C2”, “C3”, “C4”, “C5”, and “C6”