We thank Drs. Ayubi and Safiri for their comments (1) on our article (2). First, they pointed out that because baseline measurements, instead of repeated measurements, of insulin resistance (IR), insulin, and glucose were used to evaluate associations between these risk factors and cognitive decline, the results might not be accurate because of regression dilution bias and collinearity. It is true that we did not have access to repeated measurements of IR, insulin, and glucose during the follow-up time, which is why we could not account for within-person variability in our analyses nor answer the question asked by Ayubi and Safiri (1): “Is there an association between subsequent measurements of IR during follow-up and cognitive decline?” To answer that question, further research, preferably with repeated measures of IR, would be needed. However, as stated by Ayubi and Safiri, “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.” Thus, the associations between IR and cognitive performance are likely to be underestimated, rather than overestimated, in our study.
Second, Ayubi and Safiri (1) suspect that the results of our study might have been degraded by collinearity, as they claim that IR, insulin, and glucose were included in the same model. However, we did not include IR, insulin, and glucose in the same regression model. Instead, the associations of IR, insulin, and glucose (and HbA1c and hs-CRP) were analyzed in separate models (first adjusted for age, sex, and education and then further adjusted for APOEε4 genotype and metabolic risk factors) to evaluate if these measures had independent value in predicting performance on the cognitive tests.
The third comment by Ayubi and Safiri considers the interpretation that IR and insulin are independent predictors of poorer verbal fluency performance at follow-up and a decline in verbal fluency. They state that because the model of adjustment used is not a validated predictive model such a conclusion is optimistic. Clinical prediction models are used to investigate the relationship between future or unknown outcomes and baseline health states, and the establishment of these models has recently been reviewed (3). The aim of our study was not to establish a clinical prediction model for cognitive decline but to examine the associations between baseline levels of IR, insulin, and glucose and cognitive performance at follow-up. Thus, the term “predictor” was used in our article as it is commonly used in regression models, where the association between a predictor variable (in this case IR) and a dependent variable (in this case cognitive test score) is determined.
To conclude, we appreciate the concerns expressed by Ayubi and Safiri (1) regarding our results. We hope that this letter of response provides clarification for future readers of the aims of our study and the methods used to assess the association between IR and cognitive decline.
Article Information
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
