Prediction of the presence of insulin resistance using general health checkup data in Japanese employees with metabolic risk factors

AIM

The aim of the current study was to develop a predictive model of insulin resistance using general health checkup data in Japanese employees with one or more metabolic risk factors.

METHODS

We used a database of 846 Japanese employees with one or more metabolic risk factors who underwent general health checkup and a 75-g oral glucose tolerance test (OGTT). Logistic regression models were developed to predict existing insulin resistance evaluated using the Matsuda index. The predictive performance of these models was assessed using the C statistic.

RESULTS

The C statistics of body mass index (BMI), waist circumference and their combined use were 0.743, 0.732 and 0.749, with no significant differences. The multivariate backward selection model, in which BMI, the levels of plasma glucose, high-density lipoprotein (HDL) cholesterol, log-transformed triglycerides and log-transformed alanine aminotransferase and hypertension under treatment remained, had a C statistic of 0.816, with a significant difference compared to the combined use of BMI and waist circumference (p＜0.01). The C statistic was not significantly reduced when the levels of log-transformed triglycerides and log-transformed alanine aminotransferase and hypertension under treatment were simultaneously excluded from the multivariate model (p=0.14). On the other hand, further exclusion of any of the remaining three variables significantly reduced the C statistic (all p＜0.01).

CONCLUSIONS

When predicting the presence of insulin resistance using general health checkup data in Japanese employees with metabolic risk factors, it is important to take into consideration the BMI and fasting plasma glucose and HDL cholesterol levels.

Predictive impact of adiponectin for detecting dysglycemia in non-fasting Japanese employees with metabolic risk factors

The screening of dysglycemia in the non-fasting state is a challenging issue in healthcare practice. We investigated whether the additional measurement of circulating adiponectin levels could improve screening performance. We used a database of 1190 Japanese employees with metabolic risk factors, who underwent a 75-g oral glucose tolerance test (OGTT), following non-fasting health check-ups. Dysglycemia was defined as fasting glucose levels ≥6.1 mmol/L or 2-hr glucose levels ≥7.8 mmol/L during the OGTT. Logistic regression analysis revealed that decreased adiponectin levels were associated with dysglycemia, independently of postprandial glucose (PG) and hemoglobin A1c (HbA1c) levels, as well as other health check-up data (p<0.01). The trivariate model with PG, HbA1c, and adiponectin levels (PG+A1c+ADN model) had a larger area under the receiver operating characteristic curves (AUC) than the bivariate model with only PG and HbA1c levels (0.746 vs. 0.729; p=0.01). However, the AUC of the multivariate model with PG, HbA1c, and other health check-up data, but not adiponectin levels (PG+A1c+Other model) was 0.815, much larger than that of the PG+A1c+ADN model (p<0.01). The addition of adiponectin levels to the PG+A1c+Other model had a significantly larger AUC than the A1c+PG+Other model only in the subgroup without abdominal obesity (p=0.01), but not in the overall population (p=0.06) or in the subgroup with abdominal obesity (p=0.62). In conclusion, circulating adiponectin levels were independently associated with dysglycemia in non-fasting Japanese employees with metabolic risk factors, but they improved the screening capacity only in those without abdominal obesity.