Gesundheitswesen 2017; 79(08/09): 656-804
DOI: 10.1055/s-0037-1605988
Poster
Georg Thieme Verlag KG Stuttgart · New York

Discrimination of metabolically healthy and unhealthy individuals using the triglyceride glucose index – distributional considerations

RS Peter
1   Institut für Epidemiologie und Medizinische Biometrie, Universität Ulm, Ulm
,
F Keller
2   Klinik für Kinder – und Jugendpsychiatrie, Universitätsklinikum Ulm, Ulm
,
B Föger
3   Arbeitskreis für Vorsorge- und Sozialmedizin, Bregenz
,
G Nagel
1   Institut für Epidemiologie und Medizinische Biometrie, Universität Ulm, Ulm
3   Arbeitskreis für Vorsorge- und Sozialmedizin, Bregenz
› Author Affiliations
Further Information

Publication History

Publication Date:
01 September 2017 (online)

Also available at
 

Introduction:

The triglyceride glucose index (TyG) is a surrogate measure of insulin resistance. It is calculated as Ln(triglycerides in mg/dl x glucose in mg/dl/2) and has been suggested for the identification of “normal weight but metabolic obese” individuals. Assuming that there are two groups, metabolic healthy (MH) and metabolic unhealthy (MUH), the TyG distribution will be a mixture of the distributions within these groups.

Methods:

For the present study we used data of the Vorarlberg health monitoring and promotion program including data of 174,112 (46.7% male, mean age: 46.1 years) persons with fasting glucose and triglyceride measurements. We applied Gaussian mixture modeling for identification of two TyG component distributions. Theoretical ROC curves were derived from the estimated probability distributions. Covariate profiles for MH and MUH groups were calculated using means, standard deviations or percentages, weighted by the probability of class membership. All calculations were performed separately for men and women, and body mass index (BMI) category (normal weight: 18.5 – 25.0 kg/m2, overweight: 25.0 – 30.0 kg/m2 and obese: > 30.0 kg/m2).

Results:

The TyG distribution could be represented by a mixture of two normal distributions by each BMI category in men and women. These two distributions differed in level and size. The estimated class size (prevalence) for the MUH group was higher in men than women (19.6, 15.6, and 16.2% vs. 12.3, 12.5, and 13.4%). ROC analysis based on the identified mixing distributions revealed an AUC of about 0.80 for any of the categories. The covariate profiles showed differences between MH and MUH groups of the same BMI category: average age, cholesterol and mean arterial pressure were all higher for MUH groups.

Conclusions:

We found MH and MUH groups for each BMI category. This is in line with the concepts of “metabolic healthy obese” and “metabolically obese but normal weight” subpopulations suggested in the literature.