Long-Term Predictors of Insulin Resistance: Role of Lifestyle and Metabolic Factors in Middle-Aged Men

Ulf Risérus, MMED, PHD; Johan Ärnlöv, MD, PHD; Lars Berglund, BSC


Diabetes Care. 2007;30(11):2928-2933. 

In This Article


All relationships between the independent variables and insulin sensitivity were linear. Baseline data for the total original cohort and the study sample did not differ to any major degree, although subjects in the study sample were slightly healthier than those in the original population as a consequence of the exclusions ( Table 1 ).

Univariate correlation coefficients for all predictors are shown in Table 2 . All metabolic and lifestyle factors except smoking significantly predicted insulin sensitivity ( Table 2 and Table 3 ). BMI was the strongest predictor. For every 1 SD increase of BMI (~3 units), insulin sensitivity (M) decreased by nearly 1 unit, corresponding to a decrease in insulin sensitivity by 19% of the mean.

In the metabolic model, all variables were significant independent predictors of insulin sensitivity (overall R 2 = 0.21). BMI was the strongest predictor followed by triglycerides, HDL cholesterol, diastolic blood pressure, and glucose ( Table 3 ). The R 2 for BMI was 0.18, indicating that BMI was the sole most important factor explaining the variation in insulin sensitivity.

All metabolic and lifestyle variables significantly predicted insulin sensitivity independently of each other except for triglycerides and smoking (overall R 2 = 0.25). BMI was the strongest predictor followed by physical activity, HDL cholesterol, saturated fat index, socioeconomic status, diastolic blood pressure, and glucose ( Table 3 ). These results remained similar, if we also excluded men who developed diabetes (n = 106) or cardiovascular disease (n = 211) during follow-up. The ranking among predictors persisted with statistically significant regression coefficients, with only one exception: HDL cholesterol became a stronger predictor than physical activity (ß = 0.37, P < 0.0001, and ß = 0.27, P = 0.004, respectively). Furthermore, the results remained after subjects with malignant diseases at age 50 or 70 years were excluded. Because results were similar after these additional exclusions and to maintain a high statistical power (n = 453 vs. n = 770 for the lifestyle model), these data are not shown in the tables.

After baseline fasting insulin was added to the lifestyle model, insulin was, as expected, an independent predictor (ß = 0.57 [95% CI -0.72 to -0.41], P < 0.0001, n = 770; overall R 2 = 0.31) (data not shown in the tables). BMI, however, remained a strong predictor followed by socioeconomic status, HDL cholesterol, physical activity, and saturated fat. Smoking, triglycerides, diastolic blood pressure, and glucose levels were not significant predictors. In unadjusted analyses, BMI was a slightly stronger predictor ( Table 2 ) than insulin (-0.85 [-1.0 to -0.70], P < 0.0001). If Δ BMI (BMI at age 70 - BMI at age 50) was calculated and this variable was used instead of baseline BMI in the model, the results were similar, but the predictive capacity of Δ BMI was stronger (-0.84 [-0.96 to -0.71], P < 0.0001, R 2 = 0.42) than that for BMI alone. Further, the overall results remained similar, and the strong predictive capacity of BMI remained in all multivariate models after adjustment for Δ BMI (data not shown).

Another approach to adjust for subclinical insulin resistance at baseline is to exclude all insulin-resistant subjects, defined as the highest quartile of HOMA-IR. Results were strikingly similar to those presented in Table 2 .

To replicate the results from the clamp data (only available at follow-up), we also assessed HOMA-IR both at baseline and 20 years later in the study population. There was a 19% mean decrease in insulin sensitivity over 20 years. Using HOMA-IR as the outcome variable at follow-up, BMI was again the strongest predictor even after inclusion of baseline HOMA-IR in the multivariate model (ß = 0.13 [95%CI 0.08-0.19], P < 0.0001), followed by baseline HOMA-IR (0.11 [0.08-0.15], P < 0.0001) and HDL cholesterol (-0.08 [-0.14 to -0.03], P = 0.004).

Of the study population, 43% (330 of 770) were overweight or obese at baseline, yielding a subsample of 440 normal-weight men. Despite the lower sample size, the results from multivariate models were similar to those of the study sample ( Table 3 ), with the exception that smoking, but not physical activity, was an independent predictor (lifestyle model, Table 3 ). BMI, however, remained the strongest independent predictor in all models. In the insulin model, BMI (ß = -0.56) ( Table 3 ) was even a stronger predictor than insulin (ß = -0.53 [95% CI -0.74 to -0.33], P < 0.0001, n = 440) (data not shown in the tables).


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.