Exploring sex differences in Alzheimer’s disease: a comprehensive analysis of a large patient cohort from a memory unit

Abstract

Background

Alzheimer’s disease (AD) stands as the leading cause of dementia worldwide, and projections estimate over 150 million patients by 2050. AD prevalence is notably higher in women, nearly twice that of men, with discernible sex differences in certain risk factors. To enhance our understanding of how sex influences the characteristics of AD patients and its potential impact on the disease trajectory, we conducted a comprehensive analysis of demographic, clinical, cognitive, and genetic data from a sizable and well-characterized cohort of AD dementia patients at a memory clinic in Barcelona, Spain.

Methods

The study cohort comprised individuals with probable and possible AD dementia with a Clinical Dementia Rating (CDR) score between 1 and 3 diagnosed at the Memory Unit from Ace Alzheimer Center Barcelona, Spain, between 2008 and 2018. We obtained cognitive baseline data and follow up scores for the Mini-Mental State Examination (MMSE), the CDR scale, and the neuropsychological battery used in our center (NBACE). We employed various statistical techniques to assess the impact of sex on cognitive evolution in these dementia patients, accounting for other sex-related risk factors identified through Machine Learning methods.

Results

The study cohort comprised a total of 6108 individuals diagnosed with AD dementia during the study period (28.4% males and 71.6% females). MMSE scores exhibited an average decline of approximately two units per year, unaffected by sex. Similarly, the decline in most neuropsychological functions assessed by NBACE did not exhibit significant differences between males and females. However, we observed that women diagnosed with mild AD dementia progressed more rapidly based on their CDR score (HR = 2.57, 95%CI:2.33–2.84) than men (HR = 2.03, 95%CI: 1.71–2.41) (p-interaction = 0.01).

Conclusions

Our findings do not strongly support the notion that sex significantly modifies the clinical progression of AD dementia based on cognitive data. Further research is essential to validate whether women with mild AD dementia indeed progress more rapidly than men at a similar stage and to delve into the potential underlying reasons for this finding.

Background

Alzheimer’s disease (AD) stands as the leading cause of dementia worldwide, with a projected 150 million patients expected to suffer from AD in 2050 [1]. Globally, the prevalence of AD is nearly twice as high in women compared to men [2], accompanied by sex-related differences in key risk factors of the disease. Age, the most significant determinant of AD, is influenced by differences in life expectancy between men and women due to higher incidence of heart disease, cancer, and trauma, among others [3]. Cerebrovascular and cardiovascular diseases, more prevalent in men, particularly those below 60 years of age, increase the risk of AD and share with it common risk factors such as hypertension, diabetes, smoking, and alcohol abuse [4].

Despite shifts in the frequency of modifiable risk factors such as midlife obesity, physical inactivity, and low educational attainment in recent decades, substantial differences persist between men and women [5]. Psychological disorders such as depression, anxiety, sleep problems, and chronic pain, which might be associated to cognitive decline, are more prevalent among women. Furthermore, medications used for treating these conditions (including antidepressants, antipsychotics, benzodiazepines and opioids, among others) may have deleterious effects on cognitive functions, potentially increasing the risk of AD [6] and worsening its prognosis [7].

Female hormonal and reproductive changes have been associated to depression, cognitive impairment, and AD [8]. Additionally, heightened immunoreactivity in women may underlie increased neuroinflammatory responses that could ultimately be linked to AD [9]. Finally, socioeconomic and cultural factors such as education level or domestic responsibilities also differ significantly between sexes, further contributing to an increased risk of AD among women [4, 10, 11].

While the impact of sex on AD has traditionally been overlooked, it is now recognized as a pivotal factor in both research and clinical practice. Neglecting sex differences has been suggested to lead to significant delays in progressing toward better detection, treatment, and care of AD patients. Some authors hypothesize that focusing more on these differences will results in improved patient outcomes in both cognition and behavior [12].

Precision medicine, advocating for the consideration of individual differences in genes, environments, and lifestyles when diagnosing or treating patients, holds significant potential for AD [13]. Therefore, the integration of sex into these efforts is deemed crucial.

To deepen our understanding of how sex influences the characteristics of AD patients and its potential impact on the disease course, we conducted a comprehensive analysis of demographic, social, clinical, cognitive, and genetic data from a large and well-characterized cohort of AD dementia patients at a memory clinic in Barcelona, Spain.

Methods

Study cohort

The study cohort comprised individuals diagnosed with probable and possible AD dementia at the Memory Unit from Ace Alzheimer Center Barcelona, Spain, between 2008 and 2018. This center follows an integrated care model for AD patients and their families that includes diagnosis, therapy, and follow-up care for both patients and their family members [14].

AD dementia diagnosis adhered to NINCDS/ADRDA criteria [15] up to 2014 and subsequently transitioned to NIA-AA criteria [16] thereafter. Individuals with suspected causes of dementia other than AD (i.e. Lewy body disease [17], Parkinson’s disease [18], vascular dementia [19], and frontotemporal dementia [20, 21]) were excluded from the analysis. Only patients with a Clinical Dementia Rating (CDR) score between 1 and 3 were included [22]. The study period extended from baseline, defined as the date of the initial AD dementia diagnosis, to the conclusion of the study in April 2022.

Variables analyzed

Data extracted from each patient’s medical records underwent thorough review. Baseline social and demographic information encompassed sex, age at first AD dementia diagnosis, years of formal education, marital status, cohabitation status, and type of caregiver.

Medical comorbidities were meticulously ascertained, including smoking status, alcohol abuse, depression, epilepsy, psychosis, chronic obstructive pulmonary disease (COPD), arthrosis, traumatic brain injury, thyroid disease, cerebrovascular disease (CeVD), heart disease, hyperlipidemia, diabetes mellitus, hypertension, kidney disease, and peptic ulcer. Additionally, we investigated family medical history, including dementia, Huntington’s disease, Parkinson’s disease, psychiatric disorders, and Down syndrome.

Cognitive data were comprehensively documented, capturing baseline and follow up scores for the Mini-Mental State Examination (MMSE), the memory part of the 7-Minute screen test (7MS) [23], the Neuropsychiatric Inventory Questionnaire (NPI-Q) [24], the CDR scale [22], and the neuropsychological battery used in our center (NBACE) [25]. The NBACE is a comprehensive battery which assesses various cognitive functions including Attention (digit backward, and digit forward), Automatic inhibition (SKT errors and SKT time), Executive functions (category fluency and letter fluency), Language (verbal comprehension and visual naming), Orientation (global and temporal), Praxis (Block Design WAIS-III, imitation, and ideomotor), Verbal memory (delayed recall, learning, and recognition memory), and Visual perception (Luria’s clock and Poppelreuter-type tests).

Study subjects were further characterized based on APOE genotype.

Statistical analyses

The distribution of social/demographic factors, comorbidities, and cognitive tests scores within the cohort were described and compared between male and female patients.

Subsequently, Machine Learning (ML) methods were employed to identify those factors more closely related to patient’s sex. For this purpose, a random forest model was trained with sex differentiation as the target variable. The SHAP (Shapley Additive Explanations) [26] method was then applied to assess the importance of the input variables in the model’s predictions. Variables identified as most influential by SHAP were chosen as covariates for subsequent analyses. Model performance and SHAP attributions were evaluated on the test using ten repetitions of five-fold cross-validation with class stratification. Further details regarding ML modeling can be found in Appendix 1.

Multivariable linear or logistic regression models were utilized to estimate adjusted associations between sex and selected baseline factors identified through ML analysis, accounting for potential confounding variables.

Linear mixed models were employed to examine the temporal variation in MMSE and NBACE results, constructing separate models for each test score encompassed in NBACE. This approach provided insights into how these cognitive measures evolved over time. Spaghetti plots were used to show changes with time during the first five years of follow up, along with linear mixed models’ predictions with 95%CI.

Lastly, survival analyses were conducted to scrutinize the progression of Clinical Dementia Rating (CDR) over time, commencing from baseline values and differentiating between sexes. Specifically, the time until experiencing an increase in CDR and the time until reaching a CDR of 3 were investigated. Cox proportional hazards models were employed to obtain adjusted hazard ratios, 95% confidence intervals, and p-values, with statistical significance set at p < 0.05. Given the large number of comparisons, we used Bonferroni correction for an alternative multiple-comparison adjusted threshold of 5 × 10^− 4. Note that p-values of statistically significant results based on this alternative threshold (p < 0.0005) appear in bold in the tables.

All ML models were implemented in Python (v3.11.5) using the Scikit-Learn (v1.4.0) and SHAP (v0.44.0) libraries. Stata 18.0 was used for all other statistical analyses.

Results

Cohort description

The study cohort consisted of 6108 individuals diagnosed with AD dementia during the study period (Table 1), with 1735 (28.4%) males and 4373 (71.6%) females.

Table 1 Patient’s characteristics at baseline by sex

The ML algorithm selected the covariates that were more closely related to patients’ sex. These ML models demonstrated an average accuracy of 74.9% in differentiating between males and females over the test set, assessed through ten repetitions of five-fold cross-validation. The variables with the greatest influence in differentiating between sexes were the following: age, education, marital status, cohabitation status, type of caregiver, arthrosis, and depression (Supplementary Fig. 1). Thus, all these factors were introduced along with age in all subsequent adjusted analyses (both baseline and follow-up).

At the time of the initial AD dementia diagnosis, women were older (mean age 81.2 years, SD = 7.2) compared to men (mean age 80.2 years, SD = 7.7). Women presented with a more severe Clinical Dementia Rating (CDR) stage (4.8% with stage 3 CDR, compared to 4.0% in men) and lower MMSE scores (mean MMSE 18.4, SD = 5.0, vs. mean MMSE 19.7, SD = 5.2).

Notable differences in educational attainment were observed, with 23.3% of women having no formal education compared to 11.0% of men (p < 0.01). Marital status also varied between sexes, as the majority of women were widowed (51.4%), while most men were married or partnered (76.5%). Although a similar proportion of individuals had no caregiver at the time of the initial AD dementia diagnosis in both sexes (36.5% in women, 37.9% in men), the type of caregiver differed among those who had one. Specifically, the spouse or partner most commonly served as the caregiver for men (31.9%) compared to women (11.0%).

The most prevalent comorbidities in the cohort included hypertension (men: 62.0%, women: 65.4%, p < 0.01), arthrosis (men: 35.1%, women: 62.8%, p < 0.01), and hyperlipidemia (men: 43.2%, women: 46.2%, p = 0.67). History of depression, arthrosis, thyroid disease, and hypertension was more common among women, while history of COPD, alcohol abuse, CeVD, heart disease, diabetes, kidney disease, and peptic ulcer was more frequent among men (Table 1). Family history of dementia was similar in both sexes (men: 42.8%, women: 40.9%).

Neuropsychiatric symptoms assessed by the NPI-Q showed variations between sexes (Table 2). Symptoms were either more frequent among women (hallucinations, anxiety, delusions, depression/dysphoria, appetite/eating disorders, and sleep/night-time behavior) or among men (apathy/indifference, agitation/aggression, disinhibition, and irritability/lability).

Table 2 Frequency of neuropsychiatric symptoms at baseline by sex

Regarding genetic information, 42.0% of men and 38.8% of women carried at least one copy of the ε4 allele of the APOE gene, but this difference was not statistically significant (Table 3). When stratified by age, we observed that APOE ε4 carriers were most frequent among patients up to 70 years of age (men:51.9%; female:52.8%) compared to those over 70 years (men:40.7%; female:37.3%), but no sex differences were detected.

Table 3 APOE Genotype at baseline by sex

The results of the 7MS test revealed notable sex differences as outlined in Table 4. Specifically, the total score exhibited a consistent elevation in men compare to women (men: 8.2, SD = 3.8; women: 7.5, SD = 3.7). This trend persisted across all individual categories of the test, encompassing naming, immediate recall, free recall, and cued recall. Additional data of age-stratified results of the 7MS by sex at baseline can be found in Supplementary Table 2.

Table 4 Other baseline cognitive tests results by sex

Baseline values for neuropsychological tests included in NBACE (Table 4) were consistently higher in men compared to women, with statistically significant differences observed in twelve out of nineteen tests (Digit Backward: p < 0.01; Category fluency: p < 0.01; Letter fluency: p = 0.01; Visual naming: p < 0.01; Global orientation: p < 0.01; Temporal: p < 0.01; Block Design WAIS-III: p = 0.01; Imitation: p = 0.04; Ideomotor: p < 0.01; Delayed recall: p = 0.01; Learning: p = 0.01; Luria’s clock test: p < 0.01).

Follow-up analyses

Follow-up data for the MMSE was available for 6054 individuals, averaging of 3.9 MMSE measures per patient throughout the study period (Table 5). Employing a linear mixed model and obtaining adjusted estimates from these data, we observed that MMSE scores declined at a consistent rate for both men (-2.00 units/year, 95%CI: -2.11,-1.90) and women (-2.06 units/year, 95%CI: -2.12,-2.00) (Supplementary Fig. 5). In contrast, age at baseline emerged as a substantial influencer of MMSE changes over time. Individuals diagnosed before the age of 60 experienced the most pronounced decline (-3.68 units/year, 95%CI: -4.11,-3.25), while those diagnosed at 85 years or older exhibited a decrease of less than 2 units in MMSE score per year (-1.92 units/year, 95%CI: -2.02,-1.82).

Table 5 Changes in MMSE per year during follow-up

Similarly, education demonstrated a substantial impact on the evolution of MMSE scores. The rate of decline exhibited notable variations based on education levels, with individuals having no formal education experiencing a substantially lower decline (-1.77 units/year, 95%CI: -1.88,-1.65) compared to those with a university degree (-2.66 units/year, 95%CI: -2.86,-2.46). While marital status, cohabitation, and the type of caregiver did not appear to exert a significant influence MMSE evolution (Table 5), intriguingly, patients with a history of depression seemed to demonstrate a comparatively better performance (-1.87 units/year, 95%CI: -1.97,-1.78) than those without history of depression (-2.11 units/year, 95%CI: -2.17,-2.05).

It is noteworthy that although these factors are strongly associated with sex (as detailed above, female patients were older, less educated, and more frequently presented with depression) none of these effects were found to be modified by or related to sex when explored in the analyses of interaction (Table 5).

The availability of follow-up data for the NBACE varied across the different neuropsychological tests included in this battery. The following tests had widespread follow-up data: Digit Backward (n = 5070), Digit Forward (n = 5075), Category fluency (n = 5074), Letter fluency (n = 4888), Similarities (n = 5022), Verbal comprehension (n = 5081), Visual naming (n = 5051), Global orientation (n = 5055), Temporal (n = 5057), Ideomotor (n = 5047), Delayed recall (n = 5080), Learning total (n = 5081), Recognition memory (n = 5054), Luria’s clock test (n = 4982), and Poppelreuter-type test (n = 4949). Data on other tests, such as SKT – Errors (n = 3563), SKT-Time (n = 3563), and Block Design WAIS-III (n = 3174) were available only for part of the study population. On average, individuals had around two measures for each test, limiting the evaluation of time trends. The decline in most neuropsychological functions assessed by these tests did not exhibit significant differences between males and females (Table 6, Supplementary Figs. 6–24). However, we observed that verbal memory tended to decline more rapidly among women. Specifically, the Learning total score decreased by 0.38 units per year (95%CI:-0.47,-0.28) among women, compared to 0.14 units per year (95%CI:-0.28,-0.00) among men (p = 0.007). Similarly, Recognition memory scores decreased by 0.28 units per year (95%CI:-0.35,-0.21) among women compared to 0.11 units (95%CI:-0.22,-0.01) among men (p = 0.014).

Table 6 Changes in NBACE test results (units per year) during follow-up

On average, study participants had 4.3 recorded CDR scores over an average follow-up period of 3 years. As previously noted, a baseline CDR score of 3 (severe dementia) was more prevalent among women than men. Among individuals with a baseline score below 3 (n = 4701), we employed survival methods to investigate the determinants of experiencing any increase in CDR score (Table 7). Notably, the small group (n = 66) of individuals below the age of 60 years at baseline (i.e. initial diagnosis) exhibited a higher likelihood of experiencing an increase in CDR score (HR = 1.44, 95%CI: 1.10–1.89).

Table 7 Time to any CDR increase during follow-up

Additionally, we observed that patients with a baseline CDR score of 1 were more likely to progress during the study period than those with a CDR score of 2 (HR = 2.42, 95%CI:2.22–2.64). Conversely, individuals without a designated caregiver were less likely to experience a CDR increase (HR = 0.78, 95%CI: 0.72–0.84). Despite these differences, overall we found that CDR scores increased as frequently in women as in men (HR = 0.96, 95%CI: 0.89–1.04) (Fig. 1).

Fig. 1

Time to any CDR increase during follow-up: Kaplan-Meier curve by sex. Abbreviations: CDR = Clinical Dementia Rating; CI = confidence interval

While sex did not modify the observed effects of age (p-interaction = 0.8) and caregiver (p-interaction = 0.11) on the evolution of CDR score, sex was found to modify the effect of baseline CDR score (p-interaction = 0.01). Thus, the risk of CDR score progression among those with a baseline CDR score of 1 was higher among women (HR = 2.57, 95%CI:2.33–2.84) than among men (HR = 2.03, 95%CI: 1.71–2.41). Sub-analyses, such as those stratified by baseline CDR or those exploring time to a CDR score of 3, yielded similar results (Supplementary Tables 3–5, Supplementary Figs. 2–4).

Discusion

In this study, our comprehensive investigation delved into the intricate landscape of sex differences in AD dementia, encompassing demographic, social, clinical, cognitive and genetic data from a large and well-characterized single-site cohort of patients from a Memory Clinic in Barcelona, Spain.

Overall, our study substantiated previously documented sex differences in AD dementia, aligning with existing literature. Consistent with prior findings, we observed a higher prevalence of cardiovascular and cerebrovascular disease among men, while a history of depression, arthrosis, or illiteracy was more prevalent among females [27]. While some reports suggest a potentially stronger impact of known AD risk factors, such as hypertension, in females [28] (i.e., effect modification), this evidence remains relatively weak and inconsistent [29, 30]. Although our study focused on analyzing disease progression among dementia patients rather than exploring AD risk in healthy individuals, our findings indicate little or no evidence of any effect modification of known AD risk factors by sex.

Regarding educational achievement, our findings indicated a prevalence of higher academic qualifications among males. It is noteworthy the absence of improvement in educational attainment among females during the 10-year study period (data now shown). We hypothesize that this is related to the limited percentage of females in our cohort born after 1945, when societal changes in women education became apparent in Spain [31]. Furthermore, we believe these differences in educational attainment might explain some observed crude differences in baseline test scores by sex.

Despite a slightly higher frequency of the APOE ε4 allele among males (though not statistically significant), we found no sex differences in the overall frequency of APOE ε4 carriers, consistent with previous reports [32]. However, notable sex differences emerged in the baseline neuropsychiatric evaluation, aligning with outcomes from a previous publication from our group that demonstrated distinct patterns of neuropsychiatric symptoms between males and females [33].

In our cohort, MMSE scores exhibited an average decline of approximately two units per year, consistent with recent estimates from a study encompassing 616 AD patients across Spain, Germany, and the UK [34], as well as from an earlier study in France with a comparable sample size [35]. Noteworthy predictors of MMSE decline in our study included factors such as age at diagnosis, education, or history of depression. However, our analysis revealed that sex did not influence the MMSE changes over time and did not modify the effects of identified predictors, highlighting the robustness and consistency of MMSE decline patterns across various demographic and clinical factors, and emphasizing the need for nuanced understanding of individual predictors in the trajectory of cognitive decline.

Exploring AD dementia progression through changes in CDR score over time unveiled several noteworthy associations. Younger age at diagnosis was linked to a faster functional progression, a finding consistent with previous reports suggesting that more aggressive AD cases tend to be diagnosed at younger ages [36]. Notably, having a designated caregiver was also associated with faster AD progression, as measured by CDR. We hypothesize that having a caregiver might be a marker of disease severity (initially poorer cognitive status in those patients, prompting the need for assistance). Furthermore, caregivers typically exhibit greater awareness of symptoms, possess enhanced insight into cognitive decline, and provide more precise information to healthcare providers. Conversely, the absence of a caregiver may be associated with an increased likelihood of attending a day-care center for cognitive stimulation (to ensure the patient’s supervision for several hours a day) potentially resulting in a decelerated rate of decline among these patients.

Notably, we observed differences in progression based on baseline CDR, with individuals having mild dementia (i.e. CDR 1) progressing more rapidly than those with moderate dementia (i.e. CDR 2). Although sex, overall, was not associated with progression using this endpoint, we discovered a modification of the relationship between baseline CDR and progression by sex. Specifically, women diagnosed with mild dementia progressed more rapidly than men. One possible explanation for this finding could be increased resilience among women in the initial stages of cognitive decline, potentially delaying the first diagnosis [37] and leading to faster progression once the dementia staged is reached.

While modest effect sizes were observed for various components of the NBACE battery and 7MS between sexes, caution is warranted in their interpretation. Even when statistical significance was achieved, the clinical relevance of these small effects was relatively limited, primarily attributable to the small sample sizes and restricted follow-up durations for most of the tests. Recent evidence highlighting significant sex differences in neuropsychological test performance in AD patients without dementia [38] raises question about the significance of the observed small differences in our longitudinal analysis. Careful interpretation is essential, underscoring the complexities involved in understanding sex-related variations in cognitive functions over time.

Strengths and limitations

Our study has several limitations, mostly due to data availability, which is not always independent of patient evolution. For instance, patients with very low MMSE scores were no longer assessed with the NBACE in follow-up visits, constraining the assessment of time trends. We also lacked data related to female reproductive health (including menopausal age). Furthermore, the absence of information on neuroimaging and cerebrospinal AD-related biomarkers, partly due to the cohort’s initiation in 2008, constitutes a significant limitation. Regarding the latter, we anticipate that due to the absence of biomarker confirmation, around 20% of our clinical AD dementia cohort might not have underlying AD pathology, consistent with previous studies [39]. In Memory Clinics, most of the diagnosis of dementia are solely based on clinical evaluations, as CSF biomarker confirmation is not feasible for all patients due to various constraints (clinical contraindications, patient or family refusal, budgetary and facility limitations). Also, we lacked data about medications (including anticholinesterase inhibitors, memantine, antidepressants and antipsychotics) taken by our patients. Lastly, our study involves many hypothesis tests, possibly inflating the study-wide type I error. To overcome this problem, we used an alternative multiple-comparison adjusted-threshold. Note that using this alternative threshold does not impact greatly our conclusions. Furthermore, this study presents mainly exploratory analyses, and we should be cautious interpreting statistical significance. Further replication of any relevant finding is warranted. However, our paramount strength lies in the extensive size of the study cohort, derived from a single memory unit where all patients are managed using an identical care model. The comprehensive information encompassing sociodemographic characteristics, medical comorbidities, cognition, neuropsychiatric symptoms and APOE, along with yearly follow-ups, ensures a detailed characterization of our cohort.

Conclusion

In summary, our study successfully replicated sex differences in established demographics, comorbidities, neuropsychiatric symptoms, and severity features at initial AD dementia diagnosis. Nevertheless, our findings did not provide support for the idea that sex could modify the progression of the disease based on cognitive data. However, we observed that women with mild dementia might progress more rapidly than men at similar stage using CDR data. Further research is needed to confirm and validate this observation and delve into the potential underlying reasons for this finding.