We have read with great interest the article by Ekblad et al. (1). The authors showed that in multivariable analysis insulin resistance (IR) and higher levels of insulin can be independent predictors of poorer verbal fluency performance and of a greater decline in verbal fluency during 11 years after adjusting for age, sex, and education (1). We would like to congratulate the authors on their published results in the high-quality journal Diabetes Care, which offer great insight into the association between diabetes and Alzheimer disease and cognitive impairment. However, it should be noted that some methodological and statistical issues were not addressed in their study. The effect of the studied predictors on cognitive decline may be different from those reported in the study because of regression dilution bias and collinearity. The authors used baseline measurements in the analysis; here, the important question is, “Is there an association between subsequent measurements of IR during follow-up and cognitive decline?” In other words, IR during follow-up may be a more important predictor of cognitive decline at the end of follow-up. Because of short-term and long-term within-person variation over time, the baseline measurement does not represent a valid and reliable association between the usual level of the risk factor and outcome. In other words, ignoring within-person variability in the analysis leads to regression dilution bias, as the magnitude of association between risk factor and outcome would be diluted (2). Several methods have been proposed to quantify and correct regression dilution bias (2).
The estimated parameters for the association of studied predictors and the outcomes in the study conducted by Ekblad et al. (1) have been degraded by collinearity among the predictors. As expected, HOMA-IR, glucose, and insulin are highly correlated, so including all of them in the same model resulted in inflation of standard error of the regression coefficients and imprecisely estimated the association of HOMA-IR, glucose, and insulin with cognitive decline.
Finally, the authors concluded that IR and higher levels of insulin are independent predictors of poorer verbal fluency performance and of a greater decline in verbal fluency. Such a conclusion is an optimistic impression of the predictive performance of IR because the model has not yet been validated. Validation models such as bootstrap resampling and cross-validation have been suggested to quantify and correct the optimism of the prediction model (3).
We acknowledge that the study provides very interesting results, but the above-mentioned issues have not been addressed or mentioned as limitations of the study. However, considering these methodological and statistical issues can help authors avoid misinterpretation when interpreting results.
Article Information
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
