Assessing Alzheimer’s Disease Risk Among Latina/o/x Older Adults: A CART Analysis

Abstract

Introduction/Objective:

Alzheimer’s Disease (AD) presents a significant public health challenge in the U.S., with Latina/o/x elders being disproportionately affected. This study examines the key risk factors associated with AD in this population.

Methods:

We analyzed data from the National Alzheimer’s Coordinating Center (2017), focusing on 9,801 Latina/o/x older adults (32.7% males and 67.3% females). Statistical analyses conducted included Chi-square tests, t-tests, and Classification and Regression Tree (CART) analysis, which was used as the main statistical tool.

Results:

The CART model, trained on 70% of the sample and tested on the remaining 30% (N = 9,801), identified seven terminal nodes and selected seven key predictors from 16 candidate variables. The model demonstrated modest discriminative ability (AUC = 0.68 for both training and test sets; misclassification error ≈ 36%). Sensitivity was 75%, while specificity was 55% in the test set. The most important predictors included age, education, smoking history, BMI, hypertension, and use of antidepressant or antipsychotic medications. A critical threshold emerged at < 5.5 years of education, which, in interaction with age and smoking, was associated with notably increased AD risk.

Conclusion:

This study emphasizes the crucial role of sociodemographic factors-particularly gender, age, and education-in determining AD risk among Latina/o/x elders. CART analysis identified key thresholds for age and education levels impacting AD risk. The findings suggest the need for targeted interventions and policies, with a focus on education and lifestyle factors, to mitigate AD risk in this vulnerable population.

1. INTRODUCTION

In the United States, Alzheimer’s Disease (AD) presents a significant public health challenge, characterized by progressive memory loss and cognitive decline. The Centers for Disease Control and Prevention suggests that the prevalence of AD among Americans was 5.8 million in 2020, with projections suggesting a rise to nearly 14 million by 2060 [1]. This increase is particularly alarming for racial/ethnic minority elderly populations, with Latina/o/x older adults disproportionately affected. The U.S. Census forecasts a 328% growth in the number of older racial/ethnic minority populations, including Latinas/os/x, from 1990 to 2030 [2]. However, accurate diagnosis of AD, especially in its early stages when treatments are most effective, remains challenging. This issue is especially acute within the Latina/o/x community, where AD is significantly underdiagnosed and often goes undertreated, if treated at all [2].

The elevated risk of AD among Latinas/os/x is attributed to a combination of factors, including genetic predisposition, higher prevalence of age-related chronic diseases, suboptimal social behaviors, and the intersection of low socioeconomic status and education levels [3–5]. Consequently, addressing these disparities is crucial for ensuring accurate diagnosis, effective treatment, and the development of potential preventive strategies for this demographic.

Despite the growing prevalence of AD and its disparate impact across racial groups, there remains a significant research gap, particularly concerning the Latina/o/x community. Existing research often generalizes AD prevalence and risk factors across diverse populations, failing to account for the unique socio-cultural, genetic, and environmental factors affecting Latinas/os/x [3, 6]. Moreover, in clinical trials on AD and related dementias, Latinas/os/x comprise only 2% of study participants [7]. This oversight underscores the need for more detailed studies that consider the complex interplay of socio-demographic variables, cardiovascular risk factors, and AD within the Latina/o/x population [3, 8–10].

The present study aims to bridge the existing research gap by methodically investigating the impact of various factors-such as educational attainment, antidepressant usage, Body Mass Index (BMI), and the age at which smoking cessation occurs-on the risk of AD among older Latina/o/x adults. Leveraging the Classification and Regression Tree (CART) methodology, this research transcends the limitations of previous studies, which often fell short in delineating the complex interrelationships and risk factors contributing to AD susceptibility within this specific ethnic group [3, 10]. By conducting an in-depth examination of how these variables interact and correlate within the Latina/o/x community, this study sheds light on the multifaceted nature of AD risk. This analysis is important for the creation of customized interventions, healthcare policies, and strategies that specifically address the distinct needs and challenges faced by the Latina/o/x community regarding AD.

Sex disparities in AD prevalence are notable, with women constituting most diagnosed cases. The research suggests that while AD rates among men and women are similar up to age 79, women exhibit higher rates after that, a difference that may be influenced by biological factors such as genetics and hormones [11, 12]. Additionally, sociocultural factors like education, occupation, and access to medical treatment play a role in these gender disparities [13].

The role of race in AD underscores the complex interplay of genetics and social constructs. Despite the lack of a strict biological basis for race, genetic variations affecting AD prevalence differ among racial and ethnic groups. However, research has primarily focused on non-Hispanic white populations, highlighting a significant gap in our understanding of AD within diverse racial and ethnic contexts [14, 15].

Education emerges as a critical demographic variable, with research suggesting it may be a protective factor against AD. Higher levels of education have been associated with a reduced risk of clinical AD, potentially due to an increased cognitive reserve that helps individuals cope with brain atrophy [16, 17].

Exploring cardiovascular risk factors reveals their intricate connection to AD. Hypertension, hypercholesterolemia, smoking history, BMI, heart disease, diabetes, and stroke history have all been identified as contributors to AD risk [18, 19]. For example, hypertension is a well-established risk factor affecting memory and vascular health, influencing cerebral blood flow and AD risk [12, 18]. Similarly, obesity during midlife and specific genotypes, like the apolipoprotein E (APOE) genotype, have been linked to increased AD risk. Specifically, the relationship between obesity and neural vulnerability is modulated by the APOE genotype. Furthermore, people with APOE4 are more susceptible to metabolic syndrome and cardiovascular disease [20].

The potential influence of medication history, particularly the use of antidepressants, anxiolytics, and antipsychotics on AD, also warrants attention. While these medications are commonly used to manage AD symptoms, their direct effects on AD risk and progression remain underexplored. Higher usage of antidepressants among AD patients compared to non-AD individuals has been documented, suggesting a relationship that merits further investigation [21].

The present study aims to bridge this gap by methodically investigating the impact of multiple risk and protective factors-such as educational attainment, antidepressant usage, and Body Mass Index (BMI)-on Alzheimer’s disease among Hispanic older adults. To achieve this, we applied Classification and Regression Tree (CART) modeling, complemented by logistic regression benchmarks, to explore both individual predictors and their complex interactions.

2. METHODS

2.1. Data Source and Sample

This study utilized 2017 data from the National Alzheimer’s Coordinating Center (NACC). The NACC data, a comprehensive, nationwide clinical dataset, aggregates standardized clinical and neuropathological data from 39 current and former Alzheimer’s disease research centers across the United States (naccdata.org). Each participating center obtained approval from its institutional review board and secured informed consent from participants. The dataset included cognitively normal individuals at baseline who were followed until either study withdrawal or death. For this study, we focused on 9,801 self-identified Latina/o/x older adults, comprising 3,206 males (32.7%) and 6,595 females (67.3%).

2.2. Classification of AD

Participants’ AD diagnosis and its type were confirmed either by AD diagnoses and consensus teams or by qualified physicians based on structured clinical history, neuropsychological testing, and validated cognitive and functional status assessments (naccdata.org). Diagnostic criteria followed the National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) guidelines before 2015 and the National Institute on Aging-Alzheimer’s Association (NIA-AA) guidelines thereafter. Both probable and possible AD cases were included. Participants were diagnosed with AD if it was identified as the primary or contributing cause of cognitive impairment.

2.3. Measures

Table 1 provides a comprehensive overview of variables relevant to AD. The dependent variable for this study was AD status, coded as “0 = no,” and “1 = yes.” Among the independent variables, sex was treated as a categorical variable as “1 = Male,” and “2 = Female.” Age was recorded at the time of the participants’ first visit to the AD center; however, the dataset did not include the age at which the diagnosis of AD was made. Years of education were treated as a continuous variable.

Table 1.

Study Participants Demographics (N = 9,801).

Variables	Value
Demographic
Sex, female	6,595 (67.3%)
Age	76.69 (7.25)
Years of education	11.56 (5.16)
Cardiovascular Risk Factors
Hypertension	2,243 (22.9%)
Hypercholesterolemia	2,282 (23.3%)
Age of quitting smoking	46.4 (15.12)
Total years smoked cigarettes	9.11 (15.70)
BMI	27.97 (5.08)
Congestive heart failure	116 (1.2%)
Diabetes	1,899 (19.4%)
History of stroke	557 (5.7%)
PTSD	28 (0.3%)
Medication History
Antidepressant use	2,875 (29.3%)
Anxiolytic use	1,193 (12.2%)
Antipsychotic use	707 (7.2%)
Incident AD	4,197 (42.8%)

Cardiovascular diseases-hypertension, hypercholesterolemia, and Congestive Heart Failure (CHF)-were originally coded as “0 = Absent,” “1 = Recent/Active,” “2 = Remote/Inactive,” and “9 = Unknown.” For the current analysis, categories 1 and 2 were recoded as “1 = Yes,” while category 9 was treated as missing.

Age at smoking quit and number of smoking years were both treated as continuous variables, as was Body Mass Index (BMI). Diabetes was originally coded in the same manner as cardiovascular diseases; categories 1 and 2 were recorded as “1 = Yes,” and category 9 was treated as missing.

Stroke history was initially coded as “0 = Absent” and “2 = Present.” For consistency, the “2 = Present” category was recoded as “1 = Yes.” Use of antidepressant and/or anxiolytic drug(s), and antipsychotic use were coded as “0 = not reported at visit” and “1 = reported at visit.” Apnea was recoded in the same way as diabetes. Presumptive PTSD diagnosis was treated as a binary variable (“0 = no” and “1 = yes”).

2.4. Analytic Strategy

We employed a multi-step analytic approach to examine the risk factors associated with AD among Latina/o/x older adults. Initially, descriptive statistics were used to summarize the demographic and clinical characteristics of the sample. We conducted bivariate analyses, including chi-square tests for categorical variables and t-tests for continuous variables, to explore associations between AD risk and various sociodemographic, lifestyle, and health-related variables.

Classification and Regression Tree (CART) analysis was used for the primary analysis to identify interaction effects and critical thresholds within the predictor variables associated with AD. CART is a non-parametric decision tree approach ideal for identifying complex interactions and stratifications among variables.

All predictors, including age and sex, were entered into the CART model. Unlike regression models that adjust for covariates through coefficients, CART evaluates all predictors simultaneously and identifies those that optimally partition the outcome. Thus, the model considers potential confounding effects of age and sex by allowing them to emerge as splitting variables, either independently or in interaction with other factors.

The CART model was trained on a randomly partitioned dataset (70% training, 30% testing), with model complexity determined using cost-complexity pruning under the one-standard-error (1-SE) rule. Predictive performance was evaluated using stratified 10-fold cross-validation to preserve the proportion of AD and non-AD cases across folds. Feature importance scores were derived from the final pruned tree to indicate the relative contribution of each predictor to classification. While cross-validation was used to estimate model accuracy and stability, it was not applied to generate fold-specific variability in feature importance. This method provided a hierarchical classification structure to reveal AD risk patterns based on sociodemographic factors, medication use, lifestyle choices, and health conditions, offering actionable insights into population-specific risk factors.

Missing values were handled using CART’s surrogate split procedure. When a primary splitting variable was missing for a given case, the algorithm identified the most correlated predictor(s) and applied the corresponding split rule. This approach allows all available cases to contribute to the model without listwise deletion.

3. RESULTS

3.1. Prevalence of Alzheimer’s Disease and Demographic Breakdown

This study included a total of 9,801 self-identified Latina/o/x older adults: 3,206 (32.7%) males and 6,595 (67.3%) females. The average age of the research participants was 76.69 ± 7.25. The average years of education were 11.56 ± 5.16, which is below the U.S. national average, indicating a lower level of education within this group [22]. Other medical and medication factors are included in Table 1.

3.2. Bivariate Analysis Results

Results of the bivariate analysis examining the association between sociodemographic variables and AD are presented in Table 2. Effect sizes were calculated to supplement statistical significance testing. For continuous variables, Cohen’s d was used; for categorical predictors, Cramer’s V was reported. Effect sizes were interpreted according to conventional thresholds (small, medium, large).

Table 2.

Differential Characteristics of Risk Factors According to AD Status (N = 9,801).

Variables	AD Status: No AD	AD Status: Yes AD	Chi-square / t-value (p-value)	Effect size (Cramer’s V / Cohen’s d)
Sex	-	-	12.153 (<.001)	.035
Male	1,753 (31.3%)	1,453 (34.6%)	-	-
Female	3,851 (68.7%)	2,744 (65.4%)	-	-
Total	5,604 (100%)	4,197 (100%)	-	-
Hypert	-	-	20.80 (<.001)	.074
No	1,074 (43.6%)	481 (36.0%)	-	-
Yes	1,388 (56.4%)	855 (64.0%)	-	-
Total	2,462 (100%)	1,336 (100%)	-	-
Hypchol	-	-	2.78 (.096)	.027
No	927 (39.1%)	476 (36.3%)	-	-
Yes	1,446 (60.9%)	836 (63.7%)	-	-
Total	2,373 (100%)	1,312 (100%)	-	-
Cong. Hrt	-	-	5.77 (.016)	.039
No	2,398 (97.4%)	1,284 (96.0%)	-	-
Yes	63 (2.6%)	53 (4.0%)	-	-
Total	2,461 (100%)	1,337 (100%)	-	-
Diabetes	-	-	.687 (.407)	.010
No	2,726 (73.0%)	2,309 (72.1%)	-	-
Yes	1,007 (27.0%)	892 (27.9%)	-	-
Total	3,733 (100%)	3,201 (100%)	-	-
Hx Stroke	-	-	13.87 (<.001)	.050
No	2,725 (91.4%)	2,281 (88.4%)	-	-
Yes	257 (8.6%)	300 (11.6%)	-	-
Total	2,982 (100%)	2,581(100%)	-	-
Adep	-	-	218.45 (<.001)	.150
No	4,265 (76.5%)	2,624 (62.7%)	-	-
Yes	1,313 (23.5%)	1,562 (37.3%)	-	-
Total	5,578 (100%)	4,186 (100%)	-	-
Aanx	-	-	5.21 (.023)	.023
No	4,933 (88.4%)	3,638 (86.9%)	-	-
Yes	645 (11.6%)	548 (13.1%)	-	-
Total	5,578 (100%)	4,186 (100%)	-	-
Apsy	-	-	414.08 (<.001)	.206
No	5,432 (97.4%)	3,625 (86.6%)	-	-
Yes	146 (2.6%)	561 (13.4%)	-	-
Total	5,578 (100%)	4,186 (100%)	-	-
Apnea	-	-	.488 (.485)	.023
No	450 (80.4%)	269 (82.3%)	-	-
Yes	110 (19.6%)	58 (17.7%)	-	-
Total	560 (100%)	327 (100%)	-	-
PTSD Dx	-	-	5.36 (.021)	0.37
No	2,452 (99.0%)	1,334 (99.7%)	-	-
Yes	24 (1.0%)	4 (0.3%)	-	-
Total	2,476 (100%)	1,338 (100%)	-	-
Age (Mean, SD)	74.97 (6.54)	78.98 (7.51)	−28.16 (<.001)	−.575
Years of Education (Mean, SD)	12.38 (4.80)	10.46 (5.41)	18.46 (<.001)	.378
Age of Quitting Smoking (Mean, SD)	44.63 (14.46)	48.68 (15.62)	−6.39 (<.001)	−.270
Total Years of Smoking (Mean, SD)	8.94 (15.12)	9.32 (16.35)	−1.00 (.318)	−.024
BMI (Mean, SD)	28.34 (5.23)	27.44 (4.79)	7.76 (<.001)	.178

Note: Cong. Hrt, Congestive heart failure; Hypchol, Hypercholesterolemia; Hypert, Hypertension; Hx Stroke, History of stroke; Adep, Use of antidepressant; Aanx, Use of anxiolytic; Apsy, Use of antipsychotic agent; PTSD Dx, Presumptive etiologic diagnosis of the cognitive disorder - PTSD.

Sex significantly correlated with AD (χ² = 12.153, p < .001, Cramer’s V = .035), indicating a sex-based predisposition. Age, a critical determinant, showed a significant association (t = −28.16, p < .001, Cohen’s d = −.575), underscoring its influence on AD risk. Education level, indicating cognitive reserve, was significantly associated with AD (t = 18.46, p < .001, Cohen’s d = .378). Smoking cessation age (t = −6.39, p < .001, Cohen’s d = −.270) was also associated with AD; however, smoking duration (t = −1.00, p = .318, Cohen’s d = −.024) was not. BMI was also inversely correlated with AD risk (t = 7.76, p < .001, Cohen’s d = .178). Hypertension (χ² = 20.80, p < .001, Cramer’s V = .074) and stroke history significantly affected AD, diverging from hypercholesterolemia, heart failure, or diabetes. Antidepressant (χ² = 218.45, p < .001, Cramer’s V = .150) and antipsychotic use (χ² = 414.08, p < .001, Cramer’s V = .206) were notably associated with AD, contrasting with the non-significant associations of anxiolytics, PTSD, or apnea with AD.

3.3. Decision Tree Analysis Results

The CART analysis uses a detailed evaluation of predictive accuracy for AD among Latina/o/x older adults, leveraging specific statistical measures. The CART model demonstrated a modest discriminative ability, with an AUC of 0.682 on the test set. While performance was above chance (AUC = 0.50), the overall discriminative capacity is limited and should be interpreted cautiously. These values are within a 95% confidence interval ranging from 0.209 to 1 for the training set and from 0.211 to 1 for the test set, suggesting a degree of uncertainty but overall effectiveness in classification. The confusion matrix summarizes the model’s performance regarding actual versus predicted classifications. In the training set, out of 2,917 actual events (cases with AD), 2,218 were correctly predicted (true positives), and 699 were misclassified (false negatives), resulting in a true positive rate (sensitivity) of 76.0%. For non-events (cases without AD), out of 3,941, 1,814 were misclassified as events (false positives), and 2,127 were correctly identified (true negatives), leading to a true negative rate (specificity) of 54.0%.

The overall accuracy, calculated as the percentage of correct predictions, stood at 63.4% for the training set. In the test set, the model maintained similar performance levels. Out of 1,280 actual events, 960 were correctly identified (true positives), and 320 were misclassified (false negatives), resulting in a sensitivity of 75.0%. Out of 1,663 cases for non-events, 744 were incorrectly classified as events (false positives), while 919 were accurately identified (true negatives), with a specificity of 55.3%. The overall accuracy was slightly better at 63.8%.

We attempted to benchmark CART performance against logistic regression models adjusted for age and sex; however, logistic regression required complete data across all predictors, and substantial missingness led to excessive case loss. Given these limitations, the logistic regression results were not reliable, and thus, the CART findings are presented as the primary analyses (Fig. 1).

Fig. (1).

Classification and Regression Tree of Alzheimer’s Disease among Latinas/os/x. Abbreviations: APSY, Use of antipsychotic agent; ADEP, Use of antidepressant; EDUC, Years of education; QUITSMOK, Age of quit smoking; SMOKYRS, Years of smoking. (A higher resolution / colour version of this figure is available in the electronic copy of the article).

The CART model methodically identified age as a critical factor, drawing a clear line at 79.5 years. Beyond this threshold, the prevalence of AD rises significantly to 59.4%, underscoring age’s profound impact on AD risk. In stark contrast, for those below this age limit, AD prevalence markedly decreases to 33.7%, paving the way for detailed analysis within the younger demographic group. For participants younger than this cutoff, the subsequent branch intersects with the use of antipsychotic medications. In this segment, the model reveals a striking disparity: 68.8% of individuals on antipsychotic medication are found to have AD, casting a spotlight on the significant link between antipsychotic use and heightened risk of AD. This finding accentuates the critical role of antipsychotic medications in the analysis, suggesting their substantial influence on elevating the likelihood of AD.

Subsequently, for individuals under the age threshold of 79.5 years, the decision tree advances to examine antidepressant usage, segmenting this younger demographic further. It suggests that within this subset, 42.6% of antidepressant users are diagnosed with AD, illustrating a significant correlation between antidepressant use and AD incidence among the younger cohort. Conversely, those in the same age group not using antidepressants show a substantially lower AD prevalence, at 26.9%, offering nuanced insights into how antidepressant consumption among those under 79.5 years influences AD risk differently.

Within the cohort of individuals younger than 79.5 years and not using antipsychotic and antidepressant medications, the CART model introduces another layer of differentiation based on education, with a pivotal cutoff at 5.5 years. This stratification reveals a significant contrast in AD vulnerability: individuals with less than 5.5 years of education exhibit a significantly higher AD prevalence of 46.1% compared to their counterparts with more than 5.5 years of education, who have a lower prevalence rate of 24.6%. This division underscores levels of education as a critical factor in AD risk among this subgroup.

Further refining the analysis for individuals under 79.5 years who utilized antidepressants without antipsychotics, the model delineates a subgroup based on the age of smoking cessation, using 49.5 years as a critical threshold. Among this group, individuals who quit smoking after the age of 49.5 exhibit a marginally higher susceptibility to AD, with a 44% prevalence, compared to those who quit earlier (42.4%). This segmentation sheds light on the subtle yet significant impact that the timing of smoking cessation has on AD risk, illustrating the model’s comprehensive approach to dissecting complex risk factor interplays.

In the final segment of the decision tree, our attention shifts to individuals under the age of 79.5 years who fall into a specific category: those who have not used antipsychotics but have taken antidepressants and ceased smoking before reaching 79.5 years. A critical smoking duration cutoff further categorizes this group at 12 years. Intriguingly, those with a smoking history shorter than 12 years show an elevated AD risk of 44.3%, in contrast to those who smoked longer, whose risk diminishes to 27.5%. This distinction emphasizes the significant impact that smoking duration has on AD risk, highlighting the precision of data-driven analysis in identifying nuanced risk profiles.

Seven predictors emerged as important in the final CART model, with variables such as age, sex, and hypertension consistently contributing to classification. While fold-specific variability in feature importance was not estimated, the repeated identification of these predictors across resampling procedures provides some reassurance of their robustness.

4. DISCUSSION

The present study on AD among 9,801 Latina/o/x older adults reveals significant demographic trends and risk factors. The bivariate analysis reveals significant associations between sociodemographic variables and AD. Sex, age, and education level are strongly correlated with AD risk, highlighting sex predisposition, the impact of age, and the protective role of education. Smoking habits and BMI also show significant links to AD, with hypertension and stroke history further influencing risk. Notably, antidepressant and antipsychotic use are associated with higher AD incidence, while anxiolytics, PTSD, and apnea do not show significant correlations.

An important consideration is that logistic regression requires complete data across all predictors. In our dataset, several variables exhibited substantial missingness, which would have resulted in excessive case loss if logistic regression had been applied. By contrast, CART accommodates missing values through surrogate splits, thereby retaining the full sample for analysis. Although this feature highlights one of the strengths of CART, it also prevented a direct performance comparison with logistic regression. Future research with more complete data or robust imputation strategies should benchmark CART findings against regression models to provide complementary validation.

Furthermore, CART analysis unveiled nuanced interactions among some factors within specific subgroups, highlighting unique findings. Notably, the study identified a distinct threshold for age and education level, significantly influencing AD risk. For instance, individuals younger than 79.5 years, not using antipsychotic or antidepressant medications, showed varying AD risks based on their education levels-a cutoff at 5.5 years of education markedly differentiated AD prevalence. Additionally, within the same demographic, smoking cessation age and total years of smoking emerged as significant factors, further delineating AD risk. This layered approach to the analysis offers new insights into how combinations of lifestyle, demographic, and health-related factors interplay to affect AD susceptibility in Latina/o/x older adults.

Our findings affirm the critical role of sex and educational attainment in AD prevalence, echoing prior research that highlights a higher diagnosis rate in women and the protective effect of education against AD [11, 23, 24]. However, females outnumbered males in the sample by a ratio of 2:1, which may bias specific statistical tests and increase the chance of finding statistically significant differences that are not clinically meaningful. This limitation should be taken into account when interpreting sex-related results, and future research with more balanced samples is warranted.

The study also corroborates the significant impact of cardiovascular risk factors, such as hypertension and BMI, on AD development, aligning with recent literature [25, 26]. The association between antipsychotic medication use and increased AD risk raises concerns consistent with current discussions in the field [27]. However, the lack of significant links between diabetes, PTSD, and AD in our cohort diverges from some existing studies, suggesting that population-specific factors may influence these relationships [28]. This discrepancy underscores the complexity of AD risk factors and the importance of context-specific research in uncovering the multifaceted nature of AD prevalence and its potential variances across different demographics.

Our study underscores the necessity for targeted practice and policy changes. It advocates for personalized risk assessments, emphasizing factors such as sex, education level, and cardiovascular health to tailor interventions. The protective role of education against AD suggests the development of lifelong learning programs, mainly aimed at reducing disparities in racial/ethnic minority communities. Addressing cardiovascular health comprehensively, alongside promoting lifestyle adjustments, can significantly lower AD and cardiovascular risks, particularly for Latinas/os/x with lower levels of educational attainment.

For healthcare practitioners, the association between antipsychotic medication use and increased AD risk necessitates cautious prescribing practices. It should be noted that one study revealed that antipsychotic medication is used by Latinas/os/x at nearly twice the rate compared to non-Hispanic whites [29]. This finding may call for stricter monitoring of antipsychotic medication drugs that are prescribed to Latina/o/x older adults, as these drugs may adversely interact with cognitive decline risk factors. On the policy front, devising healthcare initiatives and strategies to promote holistic health, with a focus on cardiovascular health, regulation of medication safety, and targeted research funding, is crucial. These actions aim to address disparities, inform safer medication practices, and foster environments conducive to reducing AD risk, particularly within vulnerable populations, such as Latinas/os/x and individuals with limited educational attainment.

5. STUDY LIMITATIONS

The present study on AD among Latina/o/x older adults yields significant insights but faces methodological and scope limitations. Specifically, focusing solely on the Latina/o/x population may not allow our findings to be generalized across different ethnic and racial groups. Excluding genetic factors from this analysis represents a significant limitation, given their known impact alongside sociodemographic and cardiovascular variables.

Another limitation of this study is the lack of inter-country control. Due to the lack of consistent country-level information, we were unable to adjust for or stratify our analyses by country. As such, results should be interpreted cautiously, and future research should validate these findings across countries to account for potential cultural and contextual heterogeneity.

The study’s cross-sectional design and reliance on self-reported data also constrain our ability to infer causality and may introduce biases. Due to the lack of consistent available data on the exact age of AD diagnosis, this study used participants’ age at first assessment rather than age at AD diagnosis or the age at the last visit. While this approach allows for standardized comparisons among participants, it may not fully capture important contextual information related to the timing of diagnosis or disease progression. Last, we did not control for ancestry or country of origin. This is an important limitation given that the Latina/o/x population is not a monolithic group; there are significant cultural and socioeconomic differences (income, education, and homeownership) among its various subgroups, which could alter the findings [29].

Future research directions include employing longitudinal study designs to trace the evolution of AD and its risk factors over time, broadening the demographic scope to enhance generalizability, integrating genetic analyses, acculturation levels, and nativity status for a more comprehensive risk assessment, adopting objective measures for increased accuracy, and exploring the role of social and cultural determinants in AD prevalence. This expanded approach will enhance our understanding of AD complexities, particularly within underrepresented communities.

CONCLUSION

This study, based on 9,801 Latina/o/x older adults, highlights the complex interplay of demographic, lifestyle, and medical factors in Alzheimer’s disease risk. Consistent with prior research, sex, age, and education emerged as strong predictors, with higher education offering a protective effect. Lifestyle and health-related factors such as smoking, BMI, hypertension, and stroke history were also linked to increased risk, while antidepressant and antipsychotic use were associated with higher AD prevalence.

Importantly, CART analysis revealed critical thresholds and interactions, notably the pronounced effect of having fewer than 5.5 years of education, in combination with age, smoking behaviors, and medication use. Although CART demonstrated only modest predictive accuracy (AUC ≈ 0.68; misclassification ≈ 36%), it uncovered subgroup-specific risk patterns that traditional methods may overlook. These findings highlight the need for targeted prevention and intervention strategies within the Latina/o/x population and underscore the value of combining traditional statistical approaches with machine learning methods to better capture heterogeneity in AD risk.

LIST OF ABBREVIATIONS

Cong. Hrt

Congestive Heart Failure

Hx Stroke

History of Stroke

PTSD Dx

Post-Traumatic Stress Disorder Diagnosis

AD

Alzheimer’s Disease

AVAILABILITY OF DATA AND MATERIALS

The dataset utilized in this study is available upon request through the NACC’s website (naccdata.org).