*3.2. Variables Associated with Dementia and AD*

A double comparative analysis was carried out between the control group versus the group of cases with dementia and the group of cases with AD, by means of crude (unadjusted) binary logistic regression analysis, in order to know which study variables were associated with suffering from dementia in general, or AD in particular (Table 2).

Based on the results of Table 2, multivariate models were performed by binary logistic regression, for both dementia and Alzheimer's disease, including those independent variables with a statistical significance of *p* ≤ 0.2 (all variables except sex and hypertension), and considering the parsimony of modeling (no more than five independent variables). The multivariate models performed were as follows:


With regard to dementia, Table 3 shows the adjusted OR results for each explanatory variable in each of the proposed models. Table 3 shows that the age variable was involved in the three models with a high statistical significance (*p* < 0.001), as well as a stable adjusted OR value of approximately 1.16. That is, under equally adjusted variables, for each 1-year increase in the person's age, the risk of presenting dementia increases by 16%.


**Table 1.** Description of the sample according to sex.

<sup>a</sup> Type 2 diabetes mellitus; - No subject met that condition.


**Table 2.** Crude logistic regression (unadjusted) for dementia and AD.

cOR: crude Odds Ratio.

**Table 3.** Adjusted logistic regression models for dementia and AD outcome variables.


\* Confounding effect was tested and did not remain in the final model. \*\* Confounding effector was tested and modified 12.5% of the Beta coefficient of the variable Diabetes Mellitus. \*\*\* were left in the final model due to their low *p* value.

The variable depression was included in models 1 and 3, and it was significantly associated with the presence of dementia in both models, its adjusted OR ranging between 13.6 (4.8–38.7) in Model 1 and 15.6 (5.3–45) in Model 3. This means that under equal variables included in the model, those subjects who suffered an episode of depression presented between 13.6- and 15.6-fold higher prevalence of dementia than subjects without depression.

The behavior of the dyslipidemia variable was not included in Model 2 (*p* = 0.12 and non-confounding effect), but was included in Model 3, obtaining an adjusted OR of 0.54 (0.27–1.1) *p* = 0.089, considering it as a confounding variable by causing a variation of the crude-adjusted value of the beta coefficient of the T2DM variable of 12.5%. This means that, under equal variables included in the model, those subjects with dyslipidemia have had a 1.85-fold (1/0.54) lower prevalence of dementia than subjects without dyslipidemia.

Finally, T2DM was associated with dementia, adjusted OR= 2.6 (1.05–6.3) *p* < 0.05. That is, under the equal variables of Model 3, subjects with diabetes suffered a 2.6-fold higher prevalence of dementia than subjects without T2DM.

Regarding AD as a dependent variable (Table 3), the results obtained were similar to those found in the multivariate models for dementia (partly explained by the collinearity between dementia-AD, being the latter part of dementia). Variables significantly associated with the outcome variables, as well as the sign and adjusted OR values, were very similar in dementia and AD models. The dyslipidemia variable was maintained in Models 2 and 3 due to its small alpha error (*p* = 0.07 and 0.056, respectively).
