Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The present study investigated the associations between endogenous estradiol, cognition, behaviour and functional outcomes in a sample of postmenopausal women. The authors hypothesized that longer endogenous estradiol would be associated with reduced severity of cognitive, behavioural and functional symptoms in later life, suggesting a potential protective effect of prolonged estradiol exposure on the aging female brain.

Cross-sectional evidence from a sample of dementia-free postmenopausal women was shown indicating that longer endogenous estradiol may be associated with reduced severity of subjective behavioural and functional complaints. These findings underscore the potential role of LEE2 in supporting brain health in later life. Moreover, they highlight the importance of broadening dementia risk assessment in women beyond cognitive measures to also encompass behavioural and functional changes.

Please improve the quality of figures

Considering the substantial amount of data available, a more thorough analysis would be warranted.

Author Response

Comment 1: Please improve the quality of figures

Response 1: We thank the reviewer for noting this issue. All figures have been updated to high-resolution images (600 DPI) to improve clarity and readability.

Comment 2: Considering the substantial amount of data available, a more thorough analysis would be warranted.

Response 2: We appreciate this important suggestion. To strengthen the rigour of our analyses, we have incorporated additional covariates that may influence LEE2 and dementia risk markers, including age at first childbirth and body mass index.

We agree that further analyses may provide additional insight, including domain-specific associations with neuropsychological, ECog-II, MBI-C, and SAGEA scores. However, the primary objective of the present study was to establish whether LEE2 is associated with global cognitive, behavioural, and functional ability scores. We have clarified this aim in the Introduction and Discussion. We intend to conduct more granular, domain-level analyses in the future.

Lines 96-99: The present study examined the associations between LEE2, global measures of cognition, behaviour, and function in a sample of postmenopausal females.

Lines 394-397: Future studies should aim to examine these findings using biomarker and hormonal data in diverse cohorts, and extend findings to domain-level cognitive, behavioural, and functional outcomes to reveal more nuanced associations not captured in global scores.

Reviewer 2 Report

Comments and Suggestions for Authors

This study examines associations between LEE2 and dementia risk markers from the CAN-PROTECT cohort. While the topic is relevant, several limitations undermine their contribution.

Equating each pregnancy to a uniform 0.75-year elevation, regardless of trimester-specific estrogen levels, complications, or parity, is a substantial oversimplification with weak biological justification.
The study neglects to include APOE genotype, a major confounder in dementia and estrogen-neurocognition research.
SES across the lifespan is a paramount omitted variable. SES influences menarche age, and later-life brain health, creating a likely confounding pathway that is not addressed.
The image quality is low, please give high resolution images.

Author Response

Comment 1: Equating each pregnancy to a uniform 0.75-year elevation, regardless of trimester-specific estrogen levels, complications, or parity, is a substantial oversimplification with weak biological justification.

Response 1: We appreciate and completely agree with the reviewer’s comment. We address the oversimplification of our parity adjustment in our LEE2 measure within the limitation section.

At present, however, there is no standardized approach for incorporating pregnancy into LEE2 estimates. Existing studies have adopted several approaches, including: (1) adjusting reproductive span (menopause age minus menarche age) by subtracting duration of breastfeeding, or, if breastfeeding data are unavailable, applying an estimated reduction per pregnancy (e.g., 1.5 months per pregnancy) to account for ovulatory suppression (Fix et al., 2013; Matyi et al., 2019); (2) adjusting reproductive span by adding cumulative gestational duration (i.e., assigning 0.75 years per pregnancy; Cote et al. 2023); and (3) reproductive span adjusted for the covariate, parity, modeled as a categorical variable (e.g., parity ≥3 versus parity <3). Our approach is consistent with prior studies that approximate pregnancy-related E2 exposure as a cumulative gestational duration.

Given expanded interest in measuring LEE2 with later-life health outcomes, expert groups, which include one of our co-authors, are actively working on standardizing definitions and clarifying approaches for inferring LEE2 based on pregnancy. While we await more standardized procedures for estimating LEE2, we hope this manuscript will contribute to the field based on current methodology.

Within CAN-PROTECT, data on breastfeeding history and pregnancy complications are not available, precluding more detailed modeling of pregnancy-related hormonal variability. To address this limitation, we have clarified our operationalization of LEE2 in the Methods, expanded the Discussion to explicitly acknowledge this limitation, and highlighted the need for more detailed reproductive data in future studies to estimate LEE2.

Additionally, given evidence that age at first childbirth and BMI may associate with dementia risk, we have now included these two variables as covariates in all models, with the Methods and Results updated accordingly.

Lines 134-138: To account for elevated E2 during pregnancy¹³, an additional 0.75 years (equivalent to nine months) was added for each reported biological child, an approach previously used to examine LEE2 and white matter hyperintensities¹⁴. This approach represents a simplified proxy of cumulative endogenous E2 and does not capture variability in E2 levels across pregnancies or postpartum factors, which were not available in CAN-PROTECT.

Lines 293-299: While our measure of L_EE2incorporated reproductive span and parity, other studies account for factors such as breastfeeding^15,36, a factor not captured in CAN-PROTECT. Notably, studies incorporating breastfeeding^15,36 into L_EE2calculations have reported positive associations with cognition, while studies limited to reproductive span alone have not consistently observed such relationships⁴⁴. These findings suggest that L_EE2should be interpreted as an approximation of E2 exposure, and differences in its construction may contribute to variability in observed associations.

Lines 366-376: Second, our L_EE2measure represents a simplified proxy of cumulative endogenous E2 exposure and does not capture important contributors to hormonal variability, including breastfeeding history, menstrual cycle characteristics, or pregnancy-related factors such as trimester-specific changes and complications. The incorporation of pregnancy as a fixed duration also assumes uniform exposure, which does not reflect biological variability. These unmeasured factors may influence observed associations. Third, MHT use was categorical and lacked critical details about timing of initiation, duration, dosage, or formulation -factors that may modify the study outcomes^30,64. Relatedly, information on hormonal contraceptive use (i.e., birth control pills) during reproductive years was unavailable, despite widespread use⁶⁵ of estrogen-containing birth control pills that may alter L_EE2.

Lines 154-157: Age at first childbirth was measured continuously and included as a covariate, given prior evidence linking later age at first childbirth to lower AD risk³⁶. BMI, a factor linked to endogenous E2 levels³⁷, was also included and calculated from self-reported height and weight.

References:

Fox M, Berzuini C, Knapp LA. Cumulative estrogen exposure, number of menstrual cycles, and Alzheimer's risk in a cohort of British women. Psychoneuroendocrinology. 2013;38(12):2973-2982.

Matyi JM, Rattinger GB, Schwartz S, Buhusi M, Tschanz JT. Lifetime estrogen exposure and cognition in late life: the Cache County Study. Menopause. 2019;26(12):1366-1374.

Cote S, Perron T-L, Baillargeon J-P, Bocti C, Lepage J-F, Whittingstall K. Association of cumulative lifetime exposure to female hormones with cerebral small vessel disease in postmenopausal women in the UK biobank. Neurology. 2023;101(20):e1970-e1978.

Park HK, Marston L, Mukadam N. The effects of estrogen on the risk of developing dementia: a cohort study using the UK biobank data. The American Journal of Geriatric Psychiatry. 2024;32(7):792-805.

Comment 2: The study neglects to include APOE genotype, a major confounder in dementia and estrogen-neurocognition research.

Response 2: We agree that APOE genotype is an important factor in dementia risk and may modify associations between E2 exposure and brain health outcomes. APOE genotype is not currently available in CAN-PROTECT and therefore could not be included in our analyses. We have now explicitly acknowledged this in the limitations section and highlighted the importance of incorporating genetic data in future work. We also note that genotyping is planned in future waves of the CAN-PROTECT study.

Lines 376-383: Fourth, several important variables were not adjusted for in our analyses that may influence results. Apolipoprotein E genotype, a well-established genetic risk factor for AD that may also modify associations between E2 and brain health⁶⁶, was not assessed. Socioeconomic status, which may influence both reproductive history and dementia risk⁶⁷, was also unavailable. In addition, other factors –including vascular risk profiles, psychiatric history, and other medication use– were not comprehensively captured. The absence of these variables introduces the possibility of residual confounding, which may bias observed associations.

Comment 3: SES across the lifespan is a paramount omitted variable. SES influences menarche age, and later-life brain health, creating a likely confounding pathway that is not addressed.

Response 3: We agree completely that socioeconomic status (SES) is an important potential confounder influencing both reproductive factors and later-life brain health.

SES data are not currently available in CAN-PROTECT, limiting our ability to adjust for this factor directly. While we adjusted for years of education and ethnocultural background –variables related to SES– we acknowledge that these do not fully or directly capture life-course SES. We have now explicitly included this as a limitation in the Discussion and emphasize the need for future studies incorporating comprehensive SES measures. We also note that a social, systemic, and structural determinants of health module will be implemented in CAN-PROTECT later this year.

Lines 302-304: Higher educational attainment may contribute to attenuating detectable associations between L_EE2and cognitive performance and may also reflect broader socioeconomic differences not captured in the present study.

Lines 379-380: Socioeconomic status, which may influence both reproductive history and dementia risk⁶⁷, was also unavailable. In addition, other factors –including vascular risk profiles, psychiatric history, and other medication use– were not comprehensively captured. The absence of these variables introduces the possibility of residual confounding, which may bias observed associations.

Comment 4: The image quality is low, please give high resolution images.

Response 4: We thank the reviewer for noting this issue and apologize for the low quality. All figures have been updated to high-resolution (600 DPI) images.

Reviewer 3 Report

Comments and Suggestions for Authors

This is a timely study with a large sample and multidimensional outcomes, but due to its cross-sectional design, crude LEE2 estimation, reliance on self-reported data, and potential residual confounding, I recommend major revision.

This manuscript investigates the association between lifetime exposure to endogenous estradiol (LEE2) and later-life cognitive, behavioural, and functional outcomes in a large cohort of postmenopausal women from the CAN-PROTECT study. The topic is timely and relevant, particularly given the growing interest in sex-specific risk factors and early markers of neurodegeneration.

The study is strengthened by a relatively large sample size, the inclusion of multidimensional outcome measures (cognitive, behavioural, and functional), and generally appropriate statistical modelling.

However, several important methodological and interpretational limitations need to be addressed before the manuscript can be considered for publication. In particular, the cross-sectional design, the crude estimation of LEE2, reliance on self-reported data, and potential residual confounding limit the strength of the conclusions.

Overall, I recommend a major revision.

Major comments

Given the cross-sectional design of the study, causal inferences and temporal directionality cannot be established, however, the manuscript repeatedly uses language suggesting protective effects of longer LEE2 and implications for dementia risk, therefore I recommend tempering such interpretations throughout the text by consistently framing the findings as associations, clearly emphasizing their correlational and exploratory nature, and revising the conclusions to avoid any implication of risk
The operationalization of LEE2 as reproductive span plus pregnancy duration represents a simplified proxy of cumulative estradiol exposure and does not account for several important determinants, such as breastfeeding history, hormonal contraceptive use, menstrual cycle characteristics, or endogenous hormonal variability; therefore, I recommend that the authors expand the limitations section to more explicitly acknowledge the absence of these data and consider how this may influence or attenuate the observed associations.
The key exposure variables, including age at menarche, age at menopause, and reproductive history, are based on self-reported data and may therefore be subject to recall bias, particularly in an older population, accordingly, I recommend that the authors explicitly address this limitation in the manuscript and consider discussing the potential direction and impact of such bias (e.g., whether it is likely to be non-differential or differential) on the observed associations.
The CAN-PROTECT cohort represents an online, volunteer-based sample requiring internet access and is characterized by a highly educated and predominantly European-background population, which raises concerns regarding selection bias and limits the generalizability of the findings, therefore, I recommend that the authors expand the discussion of this issue and more clearly acknowledge that the results may not be generalizable to populations with lower educational attainment or more diverse sociodemographic backgrounds.

The behavioural (MBI-C) and functional (SAGEA) outcomes are based on self-reported measures rather than objective or clinician-rated assessments, which may introduce reporting bias, therefore, I recommend that the authors explicitly discuss this limitation and acknowledge that subjective complaints may not necessarily reflect objective

Several potentially important confounders, including genetic risk factors (e.g., APOE genotype), vascular risk factors, BMI, psychiatric history, and medication use, are not accounted for in the analyses, raising the possibility of residual confounding, therefore, I recommend that the authors more explicitly acknowledge this limitation and discuss how unmeasured confounding may influence the observed associations.

Minor comments

I recommend that higher-quality, higher-resolution versions of Figures 1 and 2 be provided to improve clarity and readability in the manuscript.

Author Response

Comment 1: Given the cross-sectional design of the study, causal inferences and temporal directionality cannot be established, however, the manuscript repeatedly uses language suggesting protective effects of longer LEE2 and implications for dementia risk, therefore I recommend tempering such interpretations throughout the text by consistently framing the findings as associations, clearly emphasizing their correlational and exploratory nature, and revising the conclusions to avoid any implication of risk.

Response 1: We appreciate this important point made by the reviewer and apologize for not attending to this issue a bit more carefully prior to submission. We have carefully revised the manuscript to ensure that all interpretations are framed in correlational terms, avoiding any implication of causality or risk prediction.

Comment 2: The operationalization of LEE2 as reproductive span plus pregnancy duration represents a simplified proxy of cumulative estradiol exposure and does not account for several important determinants, such as breastfeeding history, hormonal contraceptive use, menstrual cycle characteristics, or endogenous hormonal variability; therefore, I recommend that the authors expand the limitations section to more explicitly acknowledge the absence of these data and consider how this may influence or attenuate the observed associations.

Response 2: We agree that our LEE2 measure is a simplified proxy of cumulative endogenous E2 exposure, an issue also brought up by Reviewer 2. We have added further clarification in the Methods regarding how LEE2 was constructed and expanded the Discussion to explicitly acknowledge missing factors (e.g., breastfeeding, hormonal contraceptive use, menstrual characteristics and disorders) that may influence E2. We also more clearly emphasize that LEE2 is an approximation, and further adjustments may strengthen or attenuate observed associations.

Comment 3: The key exposure variables, including age at menarche, age at menopause, and reproductive history, are based on self-reported data and may therefore be subject to recall bias, particularly in an older population, accordingly, I recommend that the authors explicitly address this limitation in the manuscript and consider discussing the potential direction and impact of such bias (e.g., whether it is likely to be non-differential or differential) on the observed associations.

Response 3: We agree with the reviewer that this is an important limitation. Our reported average ages at menarche and menopause align with current epidemiological evidence, which is reassuring, although misremembering specific ages related to reproductive history is still very possible. However, we have expanded the Discussion to explicitly address recall bias in self-reported reproductive history and note that such misclassification is likely non-differential, which may bias results toward the null.

Lines 383-388: Finally, recall bias in self-reported reproductive history may affect LEE2 accuracy. This limitation extends to much of the CAN-PROTECT dataset, which relies on retrospective self-reporting. Although mean reported ages at menarche and menopause were consistent with global averages⁶⁸, recall error is possible, particularly in older participants. Such error is likely non-differential with respect to outcomes and may bias associations toward the null.

Comment 4: The CAN-PROTECT cohort represents an online, volunteer-based sample requiring internet access and is characterized by a highly educated and predominantly European-background population, which raises concerns regarding selection bias and limits the generalizability of the findings, therefore, I recommend that the authors expand the discussion of this issue and more clearly acknowledge that the results may not be generalizable to populations with lower educational attainment or more diverse sociodemographic backgrounds.

Response 4: We agree that the CAN-PROTECT cohort may not be fully representative of the general population, given its online, volunteer-based design, relatively high educational attainment, and European ethnocultural backgrounds. We have since expanded the Discussion to acknowledge this limitation and note that findings may not generalize to populations with lower educational attainment or more diverse socioeconomic and ethnocultural backgrounds. We also note ongoing efforts to improve cohort diversity with CAN-PROTECT.

Lines 391-397: Additionally, the CAN-PROTECT cohort is characterized by relatively high educational attainment and predominately European ethnocultural backgrounds. As such, findings may not generalize to populations with lower educational attainment or more diverse socioeconomic and ethnocultural profiles. Future studies should aim to examine these findings using biomarker and hormonal data in diverse cohorts, and extend findings to domain-level cognitive, behavioural, and functional outcomes to reveal more nuanced associations not captured in global scores.

Comment 5: The behavioural (MBI-C) and functional (SAGEA) outcomes are based on self-reported measures rather than objective or clinician-rated assessments, which may introduce reporting bias, therefore, I recommend that the authors explicitly discuss this limitation and acknowledge that subjective complaints may not necessarily reflect objective.

Response 5: We appreciate the reviewer’s concern regarding reliance on self-reported measures. While self-report potentially introduces bias, there is a growing body of evidence supporting the reliability, validity, and clinical relevance of subjective measures of behaviour and function. These subjective measures have shown associations with Alzheimer disease biomarkers (e.g., amyloid, phosphorylated tau; Johansson et al., 2021; Elefante et al., 2025; Leon et al., 2025; Ghahremani et al., 2025a; Ghahremani et al., 2025b), neurodegeneration (Guan et al., 2024; Ghahremani et al., 2026a), and cognitive decline (Guan et al., 2023; Ghahremani et al., 2026b). Additionally, prior work has demonstrated meaningful concordance between self-reported and informant-reported measures of the MBI-C, especially in relatively cognitively healthy samples like those in CAN-PROTECT (Chen et al., 2022).

To clarify this line of work but also identify the important risk of self-reported measures, we have strengthened the rationale for using these measures in the Introduction and Methods and explicitly acknowledged potential reporting bias in the Discussion.

Lines 182-183: Behavioural symptoms were assessed using the Mild Behavioral Impairment Checklist (MBI-C)⁴⁰, a 34-item measure design to capture late-life onset, persistent (≥6 months), and impactful behavioural changes not explained by established psychiatric conditions or other diagnoses. The MBI-C demonstrates validity as a marker of clinical and biological AD risk^19-21,41.

Lines 191: Functional ability was measured using the Standard Assessment of Global Everyday Activities (SAGEA) scale⁴², a 15-item, multidomain measure of functional capacity over the past month. The SAGEA has demonstrated validity as a dementia screening tool^24,25,43.

Lines 388-390: Similarly, while instruments such as the ECog-II, MBI-C, and SAGEA are validated and clinically informative of AD risk, subjective reports may be influenced by individual perception or reporting tendencies.

References:

Johansson M, Stomrud E, Insel PS, Leuzy A, Johansson PM, Smith R, Ismail Z, Janelidze S, Palmqvist S, van Westen D, Mattsson-Carlgren N, Hansson O. Mild behavioral impairment and its relation to tau pathology in preclinical Alzheimer's disease. Transl Psychiatry. 2021; 11(1):76.
Elefante C, Beatino MF, Fustini C, Lepri V, Acierno D, Scalzo A, Petrozzi L, Giampietri L, Tognoni G, Baldacci F, Lattanzi L, Siciliano G, Ismail Z, Ceravolo R, Perugi G, Mild Behavioral Impairment and Plasma Biomarkers of Neurodegeneration as Predictors of Cognitive Decline in Geriatric Patients With Psychiatric Disorders, Journal of Geriatric Psychiatry and Neurology, 10.1177/08919887251394747, (2025).
Leon R, Ghahremani M, Guan D.X, Smith E.E, Zetterberg H, Ismail Z. Enhancing Alzheimer Disease Detection Using Neuropsychiatric Symptoms: The Role of Mild Behavioural Impairment in the Revised NIA-AA Research Framework. Journal of Geriatric Psychiatry and Neurology. 2025;10.1177/08919887251366634.
Ghahremani M, Smith EE, Ismail Z. Persistent Functional Impairment as an Early Indicator of Alzheimer Disease Pathology and Progression. Journal of the American Geriatrics Society. 2025; 74(2): 447-457.
Ghahremani M, Leon R, Smith E.E, Ismail Z. Exploring the association between mild behavioral impairment and plasma p-tau217: Implications for early detection of Alzheimer disease. Journal of Alzheimer’s & Dementia. 2025;17(3):e70119.
Guan D.X, Rehamn T, Nathan S, Durrani R, Potvin O, Duchesne S, Pike G.B, Smith E.E, Ismail Z. Neuropsychiatric symptoms: Risk factor or disease marker? A study of structural imaging biomarkers of Alzheimer’s disease and incident cognitive decline. Journal of Alzheimer’s & Dementia. 2024; 45(13): e70016.
Guan D.X, Smith E.E, Pike G.B, Ismail Z. Persistence of neuropsychiatric symptoms and dementia prognostication: A comparison of three operational case definitions of mild behavioral impairment. Journal of Alzheimer’s & Dementia. 2023; 15(4): e12483.
Ghahremani M, Smith E.E, Ismail Z. Persistent functional impairment as an early indicator of cognitive decline and dementia in cognitively normal older adults. Journal of Alzheimer’s disease. 2026;109(3): 1415-1427.
Ghahremani M, Smith E.E, Ismail Z. Persistent functional impairment as an early indicator of Alzheimer disease pathology and progression. Journal of American Geriatrics Society. 2026; 74(2): 447-457.
Chen T-H, Yeh Y-C, Huang M-F, et al. Validation and Comparison of the Informant-Rated and Self-Rated Versions of the Mild Behavioral Impairment Checklist. Journal of Alzheimer’s Disease. 2022;90(3):1203-1213. doi:10.3233/JAD-220006

Comment 6: Several potentially important confounders, including genetic risk factors (e.g., APOE genotype), vascular risk factors, BMI, psychiatric history, and medication use, are not accounted for in the analyses, raising the possibility of residual confounding, therefore, I recommend that the authors more explicitly acknowledge this limitation and discuss how unmeasured confounding may influence the observed associations.

Response 6: We thank the reviewer for highlighting the importance of potential residual confounding. Where possible, we have strengthened our analyses incorporating relevant covariates. Specifically, we have now included BMI and age at first childbirth in all models, given their associations with endogenous E2 levels and broader implications for later brain health:

Several other variables were noted by the reviewer, and we agree that future analyses must explicitly examine the confounding role of each of these factors. Each variable may plausibly influence LEE2 and later-life cognitive, behavioural, and functional outcomes. For example, APOE genotype may modify estrogen-related neurobiological pathways, vascular risk factors are closely linked to menopause timing and dementia risk, and psychiatric conditions may influence both subjective symptoms reporting and neurocognitive trajectories. We have now expanded the limitations section of the Discussion to explicitly acknowledge these sources of potential residual confounding and their possible impact on the observed associations. We note that unmeasured confounding may bias results in either direction or, in some cases, may attenuate true associations. Future studies incorporating genetic data, detailed vascular and psychiatric profiling, and comprehensive medication histories will be essential to disentangle these relationships and refine estimates of the associations between LEE2 and dementia risk markers.

We also want to note that APOE genotype is not currently available in CAN-PROTECT; however, collection of genetic data is planned in future waves of the study, which will enable more detailed investigation of gene-hormone interactions.

Comment 7: I recommend that higher-quality, higher-resolution versions of Figures 1 and 2 be provided to improve clarity and readability in the manuscript.

Response 7: We thank the reviewer for this comment and apologize for the low quality. All figures have been updated to high-resolution (600 DPI).

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have effectively addressed my concerns. I recommend publication.

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you for the corrections. I am satisfied with the changes and recommend that the manuscript be accepted.

Review Reports

Major comments

Minor comments